Plasmids and cell lines used in this study
See Supplementary Table 2 for plasmids used in this study. Mutant variants of TNKS2 were either obtained by PCR-based site-directed mutagenesis on pLP-dMYC SD-Hs TNKS2 (NM_025235)34 or, for a subset of combination mutants, by gene synthesis (GenScript). See Supplementary Table 3 for nomenclature of combination mutant variants. The pcDNA5-FRT/TO-mCitrine vector was generated by first PCR-amplifying an mCitrine-coding fragment from pLP-mCitrine C1 SD (V3534)35. The template-derived fragment included a 5′ Kozak sequence and a 3′ linker (coding for GSGRA). PCR introduced a 5′ HindIII restriction site and 3′ AscI, PacI and BamHI restriction sites. The PCR product was inserted into pcDNA5-FRT/TO using the HindIII and BamHI sites. The TNKS2 cDNA, excised from pLP-dMYC SD-Hs TNKS2 constructs4,34, was cloned into the vector using the AscI and PacI restriction sites.
Sf9 (Spodoptera frugiperda) insect cells were obtained from Thermo Fisher Scientific.
HEK293T cells were obtained from C. Lord (ICR, London). HeLa Flp-In T-Rex cells were a gift from S. Taylor (University of Manchester)36. Cell lines were authenticated at source and confirmed to be free of Mycoplasma contamination. HEK293T and HeLa Flp-In T-Rex cells (and derivatives) were maintained in a humidified incubator at 37 °C with 5% CO2 in DMEM supplemented with antibiotics (streptomycin sulfate and benzylpenicillin), 2 mM glutamine and 10% FBS (F7524, Sigma). To generate HeLa Flp-In T-Rex cell lines inducibly expressing mCitrine–TNKS2, parental cells were cultured under pre-selection by zeocin (50 μg ml−1) and blasticidin (4 μg ml−1). Parental HeLa Flp-In T-Rex were seeded at 8 × 104 cells per well in six-well plates. Twenty-four hours later, cells were co-transfected with pOG44 (encoding Flp recombinase) and pcDNA5-FRT/TO-mCitrine–TNKS2, at a ratio of 9:1 (total of 2 μg), using Lipofectamine 2000 (Thermo Fisher Scientific), with a DNA:lipofectamine ratio of 1:2. Twenty-four hours post-transfection, cells were re-plated on a 10-cm dish and selected with hygromycin (200 μg ml−1) and blasticidin (4 μg ml−1), to allow resistant colonies to form. Colonies were then pooled and expanded to create a polyclonal cell line with stably integrated mCitrine–TNKS2 under a tetracycline-inducible promoter.
Expression and purification of TNKS2 SAM–PARPG1032W (867–1162) for EM
TNKS2 SAM–PARPG1032W (867–1162) was produced recombinantly as a His6-MBP-Asn10-TEV fusion protein in Escherichia coli BL21-Codon Plus (DE3)-RIL (Stratagene) grown in Terrific Broth media. Expression was induced at an optical density at 600 nm (OD600) of 2.0 with 0.5 mM IPTG overnight at 18 °C. Cells were collected by centrifugation and resuspended in a buffer containing 50 mM Tris-HCl (pH 7.5), 0.5 M NaCl, 5 mM imidazole (pH 7.5; to enable nickel affinity purification with reduced background from the same lysate should MBP affinity purification give insufficient yields), 10 mM β-mercaptoethanol and protease inhibitors (1 mM PMSF, 1 μg ml−1 leupeptin, 1 μg ml−1 aprotinin and 1 μg ml−1 pepstatin A; 50 ml lysis buffer was added to a pellet from each 1 l of culture). The cells were lysed by sonication using a Vibra-Cell sonicator (Sonics & Materials) and centrifuged to remove insoluble cellular debris. The lysate was filtered through a 0.45-μm filter and loaded onto a 5-ml MBPTrap HP affinity column (GE Healthcare/Cytiva). The column was washed with at least 5 column volumes (CV) of wash buffer (as lysis buffer, but without protease inhibitors). The His6-MBP-Asn10 fusion protein was eluted with a linear α-methyl glucoside (AMG) gradient (0–2 M) in a buffer also containing 50 mM Tris-HCl (pH 7.5), 0.5 M NaCl and 10 mM β-mercaptoethanol. The protein was incubated with recombinant MBP-tagged TEV protease overnight to cleave the His6-MBP-Asn10 tag, and dialysed against 50 mM Tris-HCl (pH 7.5), 0.5 M NaCl and 10 mM β-mercaptoethanol. As removal of the cleaved tag by a subtractive affinity purification reduced filament yield, the cleaved tag was not removed from the sample prior to imaging. Aliquots of TNKS2 SAM–PARPG1032W were flash-frozen in liquid nitrogen and stored at −80 °C.
Cryo-EM grid preparation and data collection
To prepare grids for cryo-EM, 3 μl of purified protein at approximately 25 μM, in a buffer containing 50 mM Tris-HCl (pH 7.5), 0.5 M NaCl and 10 mM β-mercaptoethanol, were applied onto previously glow-discharged Quantifoil Cu R 1.2/1.3 400 mesh carbon-coated grids for 30 s in a humidity-controlled Vitrobot Mark IV automatic plunge freezer (Thermo Fisher Scientific). Humidity was set to 100%, but regulation was turned off 10 min before grid preparation, and temperature was set to 18 °C. After a 30-s incubation, 2 μl of water were pipetted onto the protein solution and removed, and the process was repeated ten times to gradually lower the salt concentration. The grids were blotted using the Vitrobot Mark IV (Thermo Fisher Scientific) and plunged into liquid ethane cooled by liquid nitrogen. Vitrified grids were imaged in separate data collection sessions using Titan KRIOS transmission electron microscopes (Thermo Fisher Scientific) operated at 300 keV at the Electron Bio-Imaging Centre, Diamond Light Source, UK. A total of five datasets were collected with direct-detector cameras, four with a Gatan K2 Summit and one with a Gatan K3 camera in super-resolution counting mode using EPU software (Thermo Fisher Scientific, v184.108.40.206). Cryo-EM micrographs were collected as movies with a nominal pixel size of 1.06 Å (super-resolution dataset with a pixel size of 0.53 Å), a total accumulated dose of 40 electrons per Å2 and applied defocus values from −1.2 to −3.5 μm.
Movies were processed using RELION (v2.10 and 3.08)37,38. Images were motion-corrected and dose-weighted using MotionCor2 (ref. 39). Movies from the K3 super-resolution counting dataset were 2× binned at this stage, resulting in a pixel size of 1.06 Å. The contrast transfer function (CTF) was estimated using CTFFIND4 (ref. 40). Straight portions of filaments were manually selected as start–end coordinates and extracted with a box size of 400 pixels and an inter-box distance of 27.2 Å (asymmetric unit 2). Reference-free 2D classification was performed, and segments from suboptimal 2D class averages were discarded. Selected particles from good 2D classes from each dataset were combined (139,880 particles in total). To estimate the helical symmetry parameters, the best 2D class averages were padded in a larger box of 1,200 × 1,200 pixels for finer Fourier sampling, and their power spectra generated in Bshow (from Bsoft v1.9.5)41,42. The individual power spectra were then iteratively rotationally aligned using e2align2d.py (from EMAN2 v2.31)43, using the sum of three already vertically well-aligned power spectra as initial reference, and the sum of aligned spectra as reference for further alignment iterations. The final sum of power spectra was inspected using HELIXPLORER (http://rico.ibs.fr/helixplorer/; version March 2018) for determination of initial symmetry parameters (Extended Data Fig. 2b). The power spectra showed regularly spaced layer lines, indicating a pseudo-repeat of 94 Å. The first layer line from the equator showed a maximum near the meridian and indicated a pitch of approximately 94 Å, which was confirmed by inspecting the real-space 2D class average. A layer line with a peak on the meridian indicated an axial rise of approximately 13.5 Å, therefore enabling the number of units per turn to be calculated as approximately 7 (94/13.5). This symmetry was tested on a well-defined and verticalized 2D class average using Segclassreconstruct from SPRING (v0.86.1661)44.
Helical reconstructions were refined using 3D auto-refinement in RELION (v2.10 and 3.08)45. The first 3D auto-refinement step used a cylinder with an outer diameter of 152 Å as the initial reference. The helical parameters converged to a helical twist of −52.3° and a helical rise of 13.6 Å. The resulting 3D reconstruction suggested the presence of twofold symmetry, and applying D1 point group symmetry during 3D auto-refinement increased the resolution of the map. The subsequent refinement steps used the previously generated 3D reconstruction, low-pass filtered to 10 Å, as the initial reference, with D1 symmetry. Particles were subjected to per-particle CTF refinement38 and Bayesian polishing46 before the final refinement. Using standard post-processing in RELION, the reconstructed map was automatically sharpened with a B-factor of −82.96 Å2 and masked with a soft-edge solvent mask enclosing 40% of the box in the filament axis direction. The global resolution was estimated based on the gold-standard Fourier shell correlation (FSC) 0.143 criterion between two independently refined half-maps. The local resolution was calculated using RELION. Cryo-EM data collection and 3D reconstruction statistics are shown in Extended Data Table 1.
Model building, refinement and validation
To aid model building, the post-processed map was locally sharpened using Phenix Autosharpen (from Phenix v1.18.2-3874)47 using a high-resolution cut-off of 3.0 Å. An initial model of a single TNKS2 SAM–PARPG1032W protomer was built in Coot48 after fitting the TNKS2 PARP domain (PDB code 5NWG, chain IB)33 and SAM domain (PDB code 5JRT)4 into the central region of the sharpened cryo-EM map where resolution was highest, using UCSF Chimera (v1.14)49. A TNKS2 SAM–PARPG1032W filament of ten protomers per protofilament was obtained by applying helical and D1 symmetry to the protomer (equivalent to chain E of the presented model) in UCSF Chimera. The resulting filament model, containing ten protomers in each protofilament (A to J in sense direction, and K to T in antiparallel direction; see Extended Data Fig. 3j), was real-space refined using Phenix Refine (from Phenix v1.18.2-3874)47 (maximum of 100 iterations, 5 macrocycles, target bond root-mean-square deviation of 0.01, target angles root-mean-square deviation of 1.0, secondary structure restraints, non-crystallographic symmetry (NCS) detection and refinement, and Ramachandran restraints). Chain E was extracted and subjected to multiple iterations of model building in Coot, symmetrization in UCSF Chimera and real-space refinement in Phenix Refine. The final model was validated using MolProbity50, implemented in the Phenix software package. The FSC map versus model plot was calculated using mtriage in Phenix51. Model refinement and validation statistics are shown in Extended Data Table 1.
TNKS2 SAM negative-stain EM and image processing
Wild-type TNKS2 SAM (867–940) was produced recombinantly and purified as previously described4. Of purified TNKS2 SAM, 3 µl at 104 μM was applied to the clean side of carbon on a carbon–mica interface, after which the carbon was floated on 2% uranyl acetate. A 400-mesh Cu grid (EMS) was placed on top of the carbon, then picked up with newsprint and laid flat on filter paper to drain the stain. Sixty-seven micrographs were collected on a Tecnai F20 microscope (Thermo Fisher Scientific) equipped with a F416 CMOS detector (TVIPS GmbH), operated at 200 kV with a pixel size of 1.732 Å per pixel.
CTF estimation was performed with GCTF (v1.06)52. Particles were manually selected using e2helixboxer.py in EMAN2 (v2.9)43, and 9,921 overlapping segments with a box size of 176 pixels were extracted. All further processing steps were carried out in RELION (v3.1.0)38. Several rounds of reference-free 2D classification were conducted, and the best class averages were used as templates for reference-based auto-picking. Particles (n = 20,131) were subsequently extracted and subjected to 2D classification, resulting in a cleaned dataset of 18,720 particles. A first round of asymmetric 3D refinement was carried out against a featureless cylinder, resulting in a map with a clear D1 symmetry axis. A second round of refinement was then carried out against this map with applied D1 symmetry, yielding a final map that showed clear features of a two-start helix. Projections of the final map were compared with projections of both the SAM domain filament from the SAM–PARP model and the single-stranded TNKS2 SAM domain filament crystal structure (PDB ID 5JRT4), low-pass filtered to 20 Å.
Interface analysis using PDBePISA
Interfaces were characterized using PDBePISA (Proteins, Interfaces, Structures and Assemblies; https://www.ebi.ac.uk/pdbe/pisa/)21. Similarity searches for PARP–PARP head-like and tail-like interfaces were performed with the E–R and E–Q chain pairs, respectively, and the following parameters: 70% sequence similarity, multimeric assembly found or not found, head or tail interface found, and any other interface found or not found. Complementary searches for chains not in head-like or tail-like contacts were performed with otherwise identical parameters. All non-tankyrase hits were removed for the subsequent analysis. Domain pairs were superimposed onto the E–R or E–Q PARP–PARP domain pairs using MatchMaker in UCSF Chimera (v1.14)49. To compare domain conformations (Extended Data Fig. 5c and Supplementary Table 1), all redundant chains in the set (arising from crystallographic symmetry) were removed; that is, molecules related by crystallographic symmetry were counted only once. In cases in which single domains were modelled as two separate chains, the chains missing from the PDBePISA interface file were added manually. Domains were superimposed onto the PARP domain from chain E of the TNKS2 SAM–PARP filament. The similarity search for the tail interface using the E–Q chain pair missed a number of clearly identifiable tail-like pairs. To include this population of domain pairs, an additional search was performed with the A–A chain pair from PDB 5DCZ (which had the highest Q score in the similarity search with the E–Q chain pair) and results from both searches combined.
To explore whether small-molecule tankyrase inhibitors could determine the D-loop base conformation, we compared the conformations of 239 PARP domains not engaged in a head-like interaction. We reasoned that within this subset, the D-loop would remain unrestrained (whereas a PARP–PARP head interface would stabilize the open D-loop base). An A-site (or dual-site) binder was invariably associated with an open D-loop base. Of the 18 domains with an open D-loop base, 13 featured small-molecule inhibitors occupying the A-site. The remaining five domains with an open D-loop base featured N-site binders, but also electron densities within the A-site, which were often partially modelled as glycerol (Extended Data Fig. 5c and Supplementary Table 1). In the two domains without a bound small-molecule inhibitor, the D-loop base was closed (Supplementary Table 1). Chain B of 3MHJ31 (with an N-site binder) displayed alternate D-loop base conformations (Extended Data Fig. 5d), suggesting that the D-loop can sample both conformations in the absence of an A-site binder. Together, these observations support the notion that A-site binders can induce and/or stabilize an open D-loop base. We speculate that an open D-loop base may encourage the PARP–PARP head interaction and thereby support tankyrase polymerization.
Structure representation and analysis
For TNKS2 SAM–PARP, secondary structure was assigned using the STRIDE algorithm53, provided on a web interface (http://webclu.bio.wzw.tum.de/cgi-bin/stride/stridecgi.py)54. The output from the web interface was converted to PDB format using http://www.canoz.com/sdh/STRIDEtoPDBsecondarystruct.pl. Figures were generated using UCSF Chimera (v1.14) and ChimeraX (v1.3), developed by the Resource for Biocomputing, Visualization, and Informatics at the University of California, San Francisco, with support from NIH P41-GM103311, R01-GM129325, and the Office of Cyber Infrastructure and Computational Biology, National Institute of Allergy and Infectious Diseases49,55. Contacts were defined by atoms with a van der Waals (VDW) overlap of −0.5 Å or greater.
Luciferase reporter assays
Luciferase reporter assays were performed as previously described4, using the TOPFlash reporter containing six TCF/LEF-binding sites, which respond to active β-catenin56. HEK293T cells were seeded on white 96-well plates at 26,000 cells per well. Twenty-four hours later, cells were transfected in technical quadruplicate with pLP-dMYC-SD-TNKS2 (16 ng per well), TOPFlash (10 ng per well) and ptkRL (2 ng per well). DNA was filled up to a total amount of 50 ng per well using pDNR-MCS-SA35. Cell media were changed for 100 μl Opti-MEM I (Thermo Fisher Scientific/Gibco), and cells were transfected using Lipofectamine 2000 (Thermo Fisher Scientific/Invitrogen) using a DNA:transfectant ratio of 1:3 in Opti-MEM I. Four hours after complex addition, media were changed for DMEM with 0.3% FBS. Twenty hours following media change, cells from two technical replicates were lysed using passive lysis buffer (Promega) and processed for luminometry using the Dual-Luciferase Reporter Assay system (Promega). Plates were read using a POLARstar Omega plate reader (BMG Labtech), using the Omega software (v5.70). Data were analysed using Microsoft Excel for Mac (v16.57) and GraphPad Prism (v9.3.1). Upon background subtraction, ratios of firefly luciferase to Renilla luciferase signals were calculated for each of the two technical replicates. The means of the technical replicates were further analysed as indicated in the figure legends. Cells from the third and fourth technical replicates were lysed in 2× SDS sample buffer and used to monitor the expression levels of TNKS2 by western blotting. Data shown are from at least three independent experiments, as detailed in the figure legends.
Full-length TNKS2 PARP activity assays
HEK293T cells were seeded on 10-cm dishes at 3.5 × 106 cells per dish. Twenty-four hours later, cells were transfected with a pLP-dMYC-SD empty vector or the indicated TNKS2 constructs (two dishes for WT, K1042A and K1014A and one dish for other constructs) using calcium phosphate. Twenty-four hours post-transfection, cells were scraped in ice-cold PBS and cell pellets snap-frozen in liquid nitrogen. Cell pellets were lysed in 1 ml high-salt RIPA buffer (50 mM HEPES-NaOH (pH 7.5), 750 mM NaCl, 1% Triton X-100, 0.5% sodium deoxycholate, 0.1% SDS, 1 mM DTT, 1 μM PDD000172173 PARG inhibitor (Merck) and protease inhibitors (Pierce protease inhibitor tablets, EDTA-free, Thermo Fisher Scientific)). Cell lysates were briefly sonicated on ice to shear DNA and cleared by centrifugation (20,817g for 15 min) at 4 °C. The cleared cell lysates were incubated with 30 μl (50% slurry) of pre-equilibrated anti-MYC magnetic resin (9E10, Thermo Fisher Scientific) for 2 h, rotating at 4 °C. The magnetic resin was washed nine times with 1 ml lysis buffer and three times with 1 ml PARP assay buffer (50 mM HEPES-NaOH (pH 7.5), 150 mM NaCl, 0.01% Triton X-100, 10% glycerol and 1 mM DTT). After the final wash step, each sample was split into two (for incubation with or without NAD+), and resin was resuspended in 41 μl of PARP assay buffer to achieve a final volume of 45 μl. The resuspension volume was based on the estimation that half the volume of slurry is made up of resin, and the resin traps roughly half its volume in buffer. Each reaction was incubated with 200 μM NAD+, or with H2O as control, for 30 min at 30 °C with shaking (800 rpm). The reaction was stopped by adding 25 μl of 4× SDS sample buffer and boiled for 5 min at 100 °C. Samples were analysed by western blotting using pan-ADP-ribose binding reagent (MABE1016, Millipore). Levels of pan-ADP-ribose were quantified using ImageJ software (NIH, v2.3.0/1.53f), which involved background correction (using an empty lane) and normalizing to the levels of TNKS2 variants quantified from the without-NAD+ conditions. For quantitation of in vitro PARylation, the portion of the lane corresponding to TNKS2 and above was measured (primarily auto-PARylation, see examples in Supplementary Fig. 1 (1) and (5)). Where background correction gave rise to negative values, these were constrained to zero. Data were expressed relative to WT TNKS2.
Expression and purification of His6-MBP–TNKS2 SAM–PARP variants for biochemical and biophysical assays
His6-MBP–TNKS2 SAM–PARP variants were expressed either in E. coli or Sf9 insect cells.
Proteins expressed in E. coli were prepared as described above for TNKS2 SAM–PARPG1032W (867–1162), except for WT His6-MBP–TNKS2 SAM–PARP, for which 1% (w/v) glucose was added to the preculture media to limit leaky expression that resulted in toxicity. Cells were resuspended in a buffer containing 50 mM HEPES-NaOH (pH 7.5), 0.5 M NaCl, 10 mM β-mercaptoethanol and 10% glycerol with cOmplete EDTA-free protease inhibitor cocktail tablets (Roche) (50 ml of lysis buffer and one protease inhibitor tablet added for each litre of culture). Cells were lysed by sonication in a Vibra-Cell sonicator (Sonics & Materials) and centrifuged to remove insoluble cellular debris. The lysate was filtered with a 0.45-μm filter and loaded onto a 5-ml MBPTrap HP affinity column (GE Healthcare/Cytiva). The column was washed with at least 5 CV of wash buffer (as lysis buffer without protease inhibitors). Proteins were eluted with a linear maltose gradient (0–50 mM) in a buffer also containing 50 mM HEPES-NaOH (pH 7.5), 0.5 M NaCl, 10 mM β-mercaptoethanol and 10% glycerol. The protein was dialysed overnight against 50 mM HEPES-NaOH (pH 7.5), 0.5 M NaCl, 2 mM TCEP and 10% glycerol without the addition of TEV protease. Protein samples were flash-frozen in liquid nitrogen and stored at −80 °C.
For NAD+-binding assays, the WT, VY903/920WA and G1032W mutant variants of His6-MBP–TNKS2 SAM–PARP variants were expressed in Sf9 insect cells as we observed a higher ADP-ribosylation activity and lower basal auto-PARylation levels for proteins produced in insect cells as compared with E. coli. We expressed proteins from codon-optimized cDNAs (GenScript) subcloned into the 4C pFastBac vector (Supplementary Table 2). Viral bacmids for the protein expression were generated using Tn7 transposition in chemically competent DH10 MultiBacTurbo E. coli cells. Recombinant baculoviruses were generated in Sf9 cells by transfection with purified bacmids using Cellfectin II reagent (cat. no. 10362100, Thermo Scientific). Proteins were expressed in 400 ml of Sf9 cell cultures at a density of 1 × 106 cells per millilitre. Cells were infected with the amplified recombinant baculoviruses and incubated at 27 °C under mild shaking (270 rpm) until the viability dropped below 80% (typically after 96 h). Cells were collected by centrifugation at 250g for 10 min at 4 °C and cell pellets were frozen at −80 °C. Cell lysates were prepared as described for proteins expressed in E. coli. Filtered lysates were loaded onto a 5-ml HisTrap column (cat. no. 17524802, Cytiva), pre-equilibrated in wash buffer (50 mM HEPES-NaOH (pH 7.5), 0.5 M NaCl, 10 mM β-mercaptoethanol, 20 mM imidazole and 10% glycerol), and the column was washed with 10 CV of wash buffer or until baseline UV absorption at 280 nm was observed. Proteins were eluted using a linear imidazole gradient (20–350 mM) in a buffer containing 50 mM HEPES-NaOH (pH 7.5), 0.5 M NaCl, 10 mM β-mercaptoethanol and 10% glycerol. Fractions containing the fusion protein were pooled and dialysed against a final buffer containing 50 mM HEPES-NaOH (pH 7.5), 0.5 M NaCl, 2 mM TCEP and 10% glycerol. Proteins were concentrated using Vivaspin Turbo concentrators (cat. no. VS15T21, Sartorius) or small-volume (0.5 ml) Pierce Protein Concentrators (cat. no. 88513, Pierce).
His6-MBP–TNKS2 SAM–PARP activity assays
Equal amounts (1 µM) of purified recombinant proteins in a final purification buffer (50 mM HEPES-NaOH (pH 7.5), 500 mM NaCl and 2 mM TCEP and 10% glycerol) were incubated with or without 0.2 mM NAD+ for 30 min at 30 °C in a thermoblock (VWR) with mild shaking (180 rpm). A higher concentration of NAD+ (0.5 mM) was used for the experiments shown in Extended Data Fig. 1. The reaction was stopped by adding 4× SDS loading buffer. Samples were resolved by SDS–PAGE, and ADP-ribosylation was detected using an anti-pan-ADP-ribose binding reagent (1:10,000 dilution; MABE1016, Millipore). Protein loading was assessed by Ponceau S (cat. no. P7170, Sigma-Aldrich) or Revert 700 total protein stain (cat. no. 926-11021, LI-COR).
For mammalian-cell based assays, proteins were extracted by boiling cell pellets or resin with immunoprecipitated proteins in SDS sample buffer. Proteins were then resolved by SDS–PAGE and electroblotted onto nitrocellulose membranes. Following blocking in 5% dried skimmed milk powder dissolved in Tris-buffered saline with 0.1% Tween 20 (TBST), membranes were incubated with primary antibodies overnight at 4 °C. Primary antibodies were anti-MYC (1:1,000; 9E10, Abcam), anti-pan-ADP-ribose (immunoprecipitates: 1:4,000, purified proteins: 1:10,000; MABE1016, Millipore), anti-α-tubulin (1:1,000; MA119162, Thermo Fisher) and anti-β-actin (1:1,000; A00702-40, GenScript). Membranes were then washed in TBST and incubated for at least 1 h with appropriate fluorescently conjugated secondary antibodies (immunoprecipitates: 1:15,000, purified proteins 1:30,000; IRDye 680RD/IRDye 800CW, LI-COR). After washing in TBST, bound secondary antibodies were detected using the LI-COR Odyssey imaging system with LI-COR Image Studio software (v5.2.5).
Microscope coverslips (High Precision, no. 1.5, 24 × 50 mm, #630-2187, VWR) were cleaned twice, sequentially with Milli-Q water, 100% isopropanol, Milli-Q water. Washed coverslips were dried with compressed air. Reusable silicone CultureWell gaskets (#GBL103250, Sigma-Aldrich) were cleaned sequentially with Milli-Q water, 100% isopropanol, Milli-Q water and dried with compressed air. Cleaned gaskets were placed on the cleaned coverslips and mounted on the sample stage of a Refeyn OneMP mass photometer (Refeyn Ltd). All measurements were performed in acquisition buffer containing 50 mM HEPES-NaOH (pH 7.5), 500 mM NaCl and 2 mM TCEP, and filtered through a 0.2-μm filter.
Purified His6-MBP–TNKS2 SAM–PARP variants were diluted in acquisition buffer to a concentration of 10 μM and incubated on ice for 30 min. Just before mass photometry measurements, 12 μl of buffer were pipetted into a well and used to identify the focal position for data collection using the autofocus function. Samples were diluted to an intermediate concentration of 250 nM and then immediately to a final concentration of 50 nM in the well (3 μl of 250 nM protein added to the 12-μl drop of buffer), except for the Y920A/G1032W variant, which was diluted to a final concentration of 10 nM because of the high number of binding events in the field of view due its monomeric state. Data acquisition was performed using the Refeyn AcquireMP software (v2.3.1). Mass photometry movies of 6,000 frames were recorded from a 10 × 10 μm instrument field of view at a frame rate of 362.4 Hz, with a frame binning value of 4, resulting in an effective frame rate of 90.6 Hz.
Data were processed and analysed using Refeyn DiscoverMP software (v2.3.0) by performing three main steps: (1) background removal, (2) identification of particle landing events on the coverslip acquisition field, and (3) particle fitting to extract maximum contrast. Default parameters were used for all steps. Individual particle contrasts from each individual movie were converted to mass using a contrast-to-mass calibration. Five datasets for each His6-MBP–TNKS2 SAM–PARP variant were merged to give a single kernel density estimate curve, using a Gaussian kernel with a fixed bandwidth of 5 kDa.
The minimum between monomer and dimer peaks on the kernel density estimate curve for each variant was used as a cut-off to estimate the proportion of monomeric/oligomeric species in the sample. Three variants (PARP–PARP tail combination, PARP–PARP head–tail combination and SAM/linker–PARP combination [PARP]) gave lower-quality data, characterized by multiple ‘sticky’ binding/unbinding species, low numbers of fitted events and poor definition between peaks. For these variants, the monomer/oligomer cut-off was instead taken from the G1032W variant. These cut-offs were then applied to the five datasets for each variant to calculate the mean percentage of monomeric and >monomeric events, with s.e.m., using Microsoft Excel for Mac (v16.57) and GraphPad Prism (v9.3.1). Note that the quantifications in Fig. 5e and Extended Data Fig. 9b show the percentages of particles that are monomeric or multimeric; this is distinct from the percentages of protein molecules in different assembly states (monomeric or >monomeric). For example, a mass photometry peak corresponding to dimers and of equal height to the monomer peak contains twice as many proteins.
Contrast-to-mass calibration was performed in acquisition buffer. NativeMark unstained protein standard (#LC0725, Thermo Fisher Scientific), which contains proteins in the range of 20–1,200 kDa, was diluted 1:100 in acquisition buffer, and 5 μl were added to a 12-μl drop in a well for measurement. The following masses were used to generate a standard calibration curve in the DiscoverMP software: 146, 480 and 1,048 kDa.
As polymers will dissociate upon dilution, measurements probably reflect dissociation products. This explains how mutations in the PARP–PARP head interface affect polymerization, although the detected polymeric species are smaller than the shortest polymers that would enable PARP–PARP domain contacts.
Fluorescence polarization (FP) experiments were performed with the non-hydrolysable NAD+ analogue BAD22, labelled at the primary amine of the adenine group with 6-carboxyfluorescein (6-FAM) via a 1,6-diaminohexane linker (synthesized by Jena Bioscience). WT and mutant variants of His6-MBP–TNKS2 SAM–PARP protein at increasing concentrations were incubated with 50 nM of BAD in FP buffer (50 mM HEPES-NaOH (pH 7.5), 500 mM NaCl, 2 mM TCEP and 0.01% CHAPS) for 30 min at room temperature in a 96-well half-area plate (cat. no. 675076, Greiner). Fluorescence intensities were read on a BMG Pherastar plate reader (with a 485-nm excitation and two 520-nm emission filters; PHERAstar FSX software (BMG, v5.70 R4)) and background-corrected. FP values were calculated in the MARS data analysis software (BMG, v220.127.116.11). Three separate experiments were performed in technical duplicate, and averages of the technical duplicates were subsequently analysed. To obtain ΔFP values, baseline correction was performed by subtracting the zero-protein well FP value from each of the other FP values. To calculate Kd values, a non-linear regression was performed using the one-site-specific binding model in GraphPad Prism (v9.3.1). The ΔFP curves for polymerizing proteins probably reflect a combination of NAD+ binding and concentration-dependent polymerization; hence, the Kd values reflect apparent affinities. Note that the assay assumes that the filaments formed by WT His6-MBP–TNKS2 SAM–PARP under the experimental conditions are of sufficient length to enable PARP–PARP head interactions. As mass photometry requires substantial dilution, this is difficult to confirm. WT His6-MBP–TNKS2 SAM–PARP showed evidence of protein aggregation at the highest concentration points (13 µM and 26 μM). Therefore, the titration of the WT protein was limited to a maximum of 6.5 μM.
HeLa Flp-In T-Rex mCitrine–TNKS2 cell lines were seeded directly onto coverslips (18 × 18 mm) at 6 × 104 cells per coverslip. Thirty-two hours later, cells were treated with 100 ng ml−1 doxycycline with or without 1 μM TNKS1/2 inhibitor, Compound 21 (ref. 57). Forty-eight hours post-seeding, coverslips were incubated with CellMask (Thermo Fisher Scientific) for 10 min at 37 °C. Coverslips were then washed twice with PBS, and cells were fixed with 4% formaldehyde for 10 min at 37 °C. Coverslips were washed three times with PBS, and cells were stained with Hoechst 33258 (5 μg ml−1; Merck/Sigma) for 30 min at room temperature. Coverslips were next washed three times with PBS and mounted onto slides using VECTASHIELD (VECTOR Laboratories). Cells were imaged 24–48 h post-fixation using a Zeiss Axio Observer Z1 Marianas microscope. Z-stacks were imported into ImageJ/FIJI software (NIH, v2.3.0/1.53f) and used to generate maximum intensity projections for analysis. At least two coverslips were analysed per condition. CellProfiler (Broad Institute, v3.1.9) was used to quantitate the number of puncta per cell and the size (area) of each individual punctum from image projections. The analysis pipeline was created to (1) identify cytoplasm using CellMask staining, (2) identify objects within this area using the IdentifyPrimaryObjects function, and (3) measure the number of puncta per cell and their size (area) in pixels. The average number of puncta per cell and the average size of puncta were calculated for three independent experiments, with a minimum of 150 cells quantified per condition in each experiment (Supplementary Table 4). A one-way ANOVA with a Tukey test for multiple comparisons was performed in GraphPad Prism (v9.3.1) (Supplementary Table 5). Images used in figures are maximum intensity projections generated from Z-stacks. A uniform exposure adjustment across all panels in Photoshop (Adobe, v2021) was used to enhance features in the figures.
Differential scanning fluorimetry
Thermal stability of the His6-MBP–TNKS2 SAM–PARP WT and mutant proteins was assessed using differential scanning fluorimetry58,59. Five µM of WT or mutant protein in purification buffer (20 mM HEPES-NaOH (pH 7.5), 500 mM NaCl and 2 mM TCEP) were mixed with 5x SYPRO Orange (Invitrogen; relative to 5,000 × stock of undisclosed concentration), and 20 μl of each setup were transferred onto a 96-well PCR plate. Melting curves were obtained on a QuantStudio 6 Flex PCR machine (Thermo Fisher Scientific) with QuantStudio real-time PCR software (v1.7.1). Temperature increases were linear from 22 °C to 95 °C with 0.5 °C per 15-s increments. Fluorescence intensity was measured in real time using the ROX channel. Average fluorescence intensities from three parallel experiments for each protein were normalized (0–100%), smoothed and plotted against the temperature using GraphPad Prism (v9.3.1). The melting temperatures (Tm; defined as 50% of the maximum fluorescence intensity) were obtained directly from the resulting melting curves and showed good agreement with those obtained from the peaks of the first derivative curves.
Multiple sequence alignments were generated with Clustal Omega60 using the web services from the EMBL-EBI61 (version January 2021). UniProt entry names for human ARTD family members are PARP1_HUMAN, PARP2_HUMAN, PARP3_HUMAN, PARP4_HUMAN, PAR15_HUMAN, PAR14_HUMAN, PARP9_HUMAN, PAR10_HUMAN, PAR11_HUMAN, PAR12_HUMAN, ZCCHV_HUMAN, PARPT_HUMAN, PAR16_HUMAN, PARP8_HUMAN and PARP6_HUMAN. NCBI accession numbers for TNKS1 orthologues are NP_003738.2 (Homo sapiens), NP_780300.2 (Mus musculus), NP_989671.1 (Gallus gallus), XP_035410112.1 (Cygnus atratus), XP_019389515.1 (Crocodylus porosus), XP_037752450.1 (Chelonia mydas), XP_026528659.1 (Notechis scutatus), XP_018099067.1 (Xenopus laevis), XP_005451454 (Oreochromis niloticus), ENSDART00000111694.5 (Danio rerio), XP_005171802.1 (D. rerio), XP_014351007.1 (Latimeria chalumnae), XP_036384182.1 (Megalops cyprinoides) and XP_032873798.1 (Amblyraja radiata). NCBI accession numbers for TNKS2 orthologues are NP_079511.1 (H. sapiens), NP_001157107.1 (M. musculus), NP_989672.1 (G. gallus), XP_035404219.1 (C. atratus), XP_019411065.1 (C. porosus), XP_007059469.2 (C. mydas), XP_026526558.1 (N. scutatus), XP_018082988.1 (X. laevis), XP_005471626.1 (O. niloticus), XP_006006371.1 (L. chalumnae), XP_036385759.1 (M. cyprinoides), XP_032889463.1 (A. radiata) and XP_020371197.1 (Rhincodon typus). Two tankyrase paralogues appear to emerge in cartilaginous fish (Chondrichthyes). NCBI accession numbers for all other tankyrase orthologues are XP_032806710.1 (Petromyzon marinus), XP_002121662.3 (Ciona intestinalis), XP_019641281.1 (Branchiostoma belcheri), XP_789260.4 (Strongylocentrotus purpuratus), XP_022094330.1 (Acanthaster planci), NP_651410.1 (Drosophila melanogaster), XP_321116.5 (Anopheles gambiae), XP_032783964.1 (Daphnia magna), XP_023333893.1 (Eurytemora affinis), XP_029842287.1 (Ixodes scapularis), GBM18725.1 (Araneus ventricosus), XP_022240762.1 (Limulus polyphemus), CAD5118347.1 (Dimorphilus gyrociliatus), XP_005099438.1 (Aplysia californica), XP_022340347.1 (Crassostrea virginica), KRZ50196.1 (Trichinella nativa), KHN72016.1 (Toxocara canis), CDS23197.1 (Echinococcus granulosus), TNN12026.1 (Schistosoma japonicum), XP_012563232.1 (Hydra vulgaris) and XP_019848937.1 (Amphimedon queenslandica). Alignments were visualized and coloured in Jalview (v2.10.5)62. For Extended Data Fig. 6, the stretches of sequences that were not aligning with the reference sequence (human TNKS2) were deleted for clarity. Percentage identity across tankyrase paralogues and orthologues, as per the alignment omitting the non-tankyrase ARTD family members, was mapped onto TNKS2 SAM–PARP using UCSF Chimera (v1.14)49, using a linear red-to-white colour gradient from 100% to 90% and white for identity lower than 90%.
Generation of figures
Publication figures were generated using Adobe Photoshop and Illustrator (v2021).
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.