key: cord-0021111-8i7vq439 authors: Yosaatmadja, Yuliana; Baddock, Hannah T; Newman, Joseph A; Bielinski, Marcin; Gavard, Angeline E; Mukhopadhyay, Shubhashish M M; Dannerfjord, Adam A; Schofield, Christopher J; McHugh, Peter J; Gileadi, Opher title: Structural and mechanistic insights into the Artemis endonuclease and strategies for its inhibition date: 2021-08-13 journal: Nucleic Acids Res DOI: 10.1093/nar/gkab693 sha: d3122bb494505ac08ab32247025bd6c807f7fe9b doc_id: 21111 cord_uid: 8i7vq439 Artemis (SNM1C/DCLRE1C) is an endonuclease that plays a key role in development of B- and T-lymphocytes and in dsDNA break repair by non-homologous end-joining (NHEJ). Artemis is phosphorylated by DNA-PKcs and acts to open DNA hairpin intermediates generated during V(D)J and class-switch recombination. Artemis deficiency leads to congenital radiosensitive severe acquired immune deficiency (RS-SCID). Artemis belongs to a superfamily of nucleases containing metallo-β-lactamase (MBL) and β-CASP (CPSF-Artemis-SNM1-Pso2) domains. We present crystal structures of the catalytic domain of wildtype and variant forms of Artemis, including one causing RS-SCID Omenn syndrome. The catalytic domain of the Artemis has similar endonuclease activity to the phosphorylated full-length protein. Our structures help explain the predominantly endonucleolytic activity of Artemis, which contrasts with the predominantly exonuclease activity of the closely related SNM1A and SNM1B MBL fold nucleases. The structures reveal a second metal binding site in its β-CASP domain unique to Artemis, which is amenable to inhibition by compounds including ebselen. By combining our structural data with that from a recently reported Artemis structure, we were able model the interaction of Artemis with DNA substrates. The structures, including one of Artemis with the cephalosporin ceftriaxone, will help enable the rational development of selective SNM1 nuclease inhibitors. Nucleases catalyse the hydrolysis of the phosphodiester bonds in nucleic acids and are broadly classified as exonucleases or endonucleases. Exonucleases are often non sequence-specific, while endonucleases can be further grouped into sequence-specific endonucleases, such as restriction enzymes, and structure-selective endonucleases (1) . Artemis (SNM1C or DCLRE1C), along with SNM1A (DCLRE1A) and SNM1B (Apollo or DCLRE1B), are human nucleases that are members of the extended structural family of metallo-␤-lactamase (MBL) fold enzymes (2, 3) . The N-terminal region of Artemis has a core MBL fold (aa 1-170, 319-361) with an inserted ␤-CASP (CPSF73, Artemis, SNM1 and PSO2) domain (aa 170-318) (4) . ␤-CASP domains are present within the larger family of eukaryotic nucleic acid processing MBLs and confer both DNA/RNA binding and nuclease activity (2, 5) . The Cterminal region of Artemis (aa 362-692) mediates proteinprotein interactions, contains post translational modification (PTM) sites, directs subcellular localization, and may modulate catalysis (6) (7) (8) (9) . Although SNM1A, SNM1B and Artemis have similar structures of their core catalytic domains, each has distinct functions and selectivities. While SNM1A and SNM1B are exclusively 5 to 3 exonucleases, Artemis is an endonuclease (7, 9) , although a minor 5 to 3 exonuclease activity has been reported (10) . Human SNM1A localizes to sites of DNA damage, can digest past DNA damage lesions in vitro, and is involved in inter-strand crosslink (ICL) repair (11) (12) (13) . SNM1B is a shelterin-associated protein required for resection at newly-replicated leading-strand telomeres to generate the 3 -overhang necessary for telomere loop (tloop) formation and telomere protection (14) (15) (16) . SNM1A and SNM1B prefer ssDNA substrates in vitro, with an absolute requirement for a free 5 -phosphate (3, 17) . By contrast, cloning (LIC) (38) ) into the baculovirus expression vector pBF-6HZB (GenBank™ accession number KP233213.1), which combines an N-terminal His 6 sequence,the Z-basic tag and a TEV protease cleavage site for efficient purification. Site-directed mutagenesis was carried out using an inverse PCR method whereby an entire plasmid is amplified using complementary mutagenic primers (oligonucleotides) with minimal cloning steps (39) , using the Herculase II Fusion DNA Polymerase (Agilent) for amplification and the KLD enzyme mix (NEB). Baculovirus generation was performed as described (3) . Recombinant proteins were produced by infecting Sf9 cells at 2 × 10 6 cells/ ml with 1.5 ml of P2 virus for WT and 3 ml of P2 virus for mutants respectively. Infected Sf9 cells were harvested 70 h after infection by centrifugation (900 × g, 20 min). The cell pellets were resuspended in 30 ml/l lysis buffer (50 mM HEPES pH 7.5, 500 mM NaCl, 10 mM imidazole, 5% (v/v) glycerol and 1 mM TCEP), snap frozen in liquid nitrogen, then stored at −80 • C for later use. Thawed cell aliquots were lysed by sonication. The lysates were centrifuged (40 000 × g, 30 min); the supernatant was passed through a 0.80 m filter (Millipore), then loaded onto an immobilized metal affinity chromatography column (IMAC) (Ni-NTA Superflow Cartridge, Qiagen) equilibrated in lysis buffer. The column was washed with lysis buffer, then eluted using a linear gradient to elution buffer (50 mM HEPES pH 7.5, 500 mM NaCl, 300 mM imidazole, 5% (v/v) glycerol and 1 mM TCEP). Protein-containing fractions were pooled and passed through an ion exchange column (HiTrap ® SP FF GE Healthcare Life Sciences) preequilibrated in the SP buffer A (25 mM HEPES pH 7.5, 300 mM NaCl, 5% (v/v) glycerol and 1 mM TCEP). Protein was eluted using a linear gradient to SP buffer B (SP buffer A with 1 M NaCl), and fractions containing the ZB-tagged Artemis were identified by electrophoresis. Artemis containing fractions were pooled and dialysed overnight at 4 • C in SP buffer A, supplemented with recombinant tobacco etch virus (TEV) protease to cleavage the His 6 -ZB tag. The protein was loaded into an ion exchange column (HiTrap ® SP FF GE Healthcare Life Sciences), pre-equilibrated in the SP buffer A. The protein was eluted using a linear gradient to SP buffer B; fractions containing tag-free Artemis were identified by electrophoresis. Artemis-containing fractions from the SP column elution were combined and concentrated to 1 ml using a 30 kDa MWCO centrifugal concentrator. The protein was then loaded on to a Superdex 75 increase 10/300 GL equilibrated with SEC buffer (25 mM HEPES pH 7.5, 300 mM NaCl, 5% (v/v) glycerol, 2 mM TCEP). Mass spectrometric analysis of the purified proteins revealed masses of 41716. 5 The truncated Artemis protein was produced in a similar manner except that the first purification step (IMAC) was omitted. The tight binding of the ZB-tagged protein to the ion exchange column and the early elution of the tag-free protein allowed to achieve effective purification. After the second ion exchange step, the protein was further purified by size exclusion chromatography (Highload® 16/200 Superdex ® 200). The full-length Artemis encoding construct was cloned into pFB-CT10HF-LIC, a baculovirus transfer vector containing a C-terminal His 10 and FLAG tag, using ligation independent cloning (LIC) (38) . pFB-CT10HF-LIC was a gift from Nicola Burgess-Brown (Addgene plasmid # 39191; http://n2t.net/addgene:39191; RRID: Addgene 39191). The baculovirus mediated expression of the full length DCLRE1C/ Artemis gene was performed in a manner similar to that for the truncated protein using 3.0 ml of P2 virus to infect Sf9 cells at 2 × 10 6 cells/ml. Cell harvesting and the initial IMAC purification steps were performed as described for the catalytic domain. Following IMAC purification and TEV cleavage overnight in dialysis buffer (50 mM HEPES pH 7.5, 0.5 M NaCl, 5% glycerol and 1 mM TCEP) the protein was passed through a 5 ml Ni-sepharose column to trap the tag and other metalbinding contaminants; the flowthrough fractions were collected. The Artemis protein was then concentrated using a centrifugal concentrator (Centricon, MWCO 30 kDa) before loading on a Superdex S200 HR 16/60 gel filtration column in dialysis buffer. Fractions containing purified Artemis protein were pooled and concentrated to 10 mg/ml. Reversed-phase chromatography was performed prior to mass spectrometry using an Agilent 1290 uHPLC system (Agilent Technologies Inc., Palo Alto, CA, USA). Concentrated protein samples were diluted to 0.02 mg/ml in 0.1% aqueous formic acid and 50 l was injected on to a 2.1 mm × 12.5 mm Zorbax 5um 300SB-C3 guard column housed in a column oven set at 40 • C. The solvent system used consisted of 0.1% aqueous formic acid in ultrahigh purity water (Millipore) (solvent A) and 0.1% aqueous formic acid in methanol (LC-MS grade, Chromasolve) (solvent B). Initial chromatography conditions were 90% A and 10% B and a flow rate of 1.0 ml/min. A linear gradient from 10% B to 80% B was applied over 35 s. Elution then proceeded isocratically at 95% B for 40 s followed by equilibration at initial conditions for a further 15 s. Protein intact mass was determined using a 6530-electrospray ionization quadrupole time-of-flight mass spectrometer (Agilent Technologies Inc., Palo Alto, CA, USA). The instrument was configured with the standard ESI source and operated in positive ion mode. The ion source was operated with the capillary voltage at 4000 V, nebulizer pressure at 60 psig, drying gas at 350 • C and drying gas flow rate at 12 l/min. The instrument ion optic voltages were as follows: fragmentor 250 V, skimmer 60 V and octopole RF 250 V. Artemis (PDB: 6TT5) was crystallized using the sitting drop vapour diffusion method by mixing 50 nl protein with 50 nl crystallization solution comprising 0.2 M ammonium chloride, 20% (v/v) PEG 3350. Crystals appeared after 2 weeks and reached maximum size within 3 weeks. The crystals were soaked in cryoprotectant solution (mother liquor supplemented with 20% (v/v) ethylene glycol), then flash frozen in liquid nitrogen. The non-IMAC purified Artemis (PDB: 7AF1) was crystallized in a similar manner, with addition of 20 nl of crystal seed solution obtained from previous crystallization experiment. Crystals were grown in a solution comprising 0.25 M ammonium chloride and 30% (v/v) PEG 3350 at 4 • C. Crystals appeared after one day and reached a maximum size within one week. Artemis variants (mutants H33A and H35D) were crystallized using the sitting drop vapour diffusion method by mixing 50 nl protein with 50 nl crystallization solution comprising 0.1 M sodium citrate pH 5.5, 20% PEG 3350; the D37A variant was crystalized in 0.2 M ammonium acetate, 0.1 M bis-Tris pH 5.5, 25% PEG 3350. All Artemis variants were crystalized in the presence of 20 nl of crystal seed solution obtained from previous crystallization experiment. Crystals grew after one day at 4 • C. and reached maximum size within 1 week. Data were collected at Diamond Light Source beamlines I04, I03 and I24. Diffraction data were processed using DIALS (40) and structures were solved by molecular replacement using PHASER (41) and the PDB coordinates 5Q2A. Model building and the addition of water molecules were performed in COOT (42) and structures refined using REFMAC (43) . Data collection and refinement statistics are given in Table I . The X-ray fluorescence data was collected at Diamond Light Source I03 (6TT5) using 100% transmission and 12.7 eV, and I24 (7AF1) using 1% transmission and 12.8 eV (Supplementary Figure S1 ). 10 pmol of single-stranded DNA (Eurofins MWG Operon, Germany) was labelled with 3.3 pmol of ␣-32 P-dATP (Perkin Elmer) by incubation with terminal deoxynucleotidyl transferase (TdT, 20 U; ThermoFisher Scientific), at 37 • C for 1 h. This solution was then passed through a P6 Micro Bio-Spin chromatography column (BioRad), and the radiolabeled DNA was annealed with the appropriate unlabeled oligonucleotides (1:1.5 molar ratio of labelled to unlabeled oligonucleotide) (Supplementary Table S1 for sequences) by heating to 95 • C for 5 min and gradual cooling to below 30 • C in annealing buffer (10 mM Tris-HCl; pH 7.5, 100 mM NaCl, 0.1 mM EDTA). Nucleic Acids Research, 2021, Vol. 49, No. 16 9313 Gel-based nuclease assays Standard nuclease assays were carried out in reactions containing 20 mM HEPES-KOH, pH 7.5, 50 mM KCl, 10 mM MgCl 2 , 0.05% (v/v) Triton X-100, 5% (v/v) glycerol (final volume: 10 l), and the indicated concentrations of Artemis. Reactions were started by addition of substrate (10 nM), incubated at 37 • C for the indicated time, then quenched by addition of 10 l stop solution (95% formamide, 10 mM EDTA, 0.25% (v/v) xylene cyanole, 0.25% (v/v) bromophenol blue) with incubation at 95 • C for 3 min. Reaction products were analysed by 20% denaturing polyacrylamide gel electrophoresis (made from 40% solution of 19:1 acrylamide:bis-acrylamide, BioRad) and 7 M urea) in 1× TBE (Tris-borate-EDTA) buffer. Electrophoresis was carried out at 700 V for 75 min; gels were subsequently fixed for 40 min in a 50% methanol, 10% acetic acid solution, and dried at 80 • C for 2 h under a vacuum. Dried gels were exposed to a Kodak phosphor imager screen and scanned using a Typhoon 9500 instrument (GE). The protocol of Lee et al. (44) was adapted for structurespecific endonuclease activity. A ssDNA substrate was utilized containing a 5 FITC-conjugated T and a 3 BHQ-1 (black hole quencher)-conjugated T (Supplementary Table S1 ). As the FITC and BHQ-1 are located proximal to one another, prior to endonucleolytic incision, the intact substrate does not fluoresce. Following endonucleolytic incision by SNM1C/Artemis, uncoupling of FITC from BHQ-1 causes an increase in fluorescence. Inhibitors (at increasing concentrations) were incubated with Artemis (50 nM) for 10 min at room temperature, before starting the reaction by adding the DNA substrate (25 nM). Assays were carried out in a 384-well format, in 25 l reaction volumes. The buffer was the same as for the gel-based nuclease assays., Fluorescence spectra were measured using a PHERAstar FSX (excitation: 495 nm; emission: 525 nm) with readings taken every 150 s, for 35 min, at 37 • C. The catalytic domain of Artemis (aa 3-361) fused to the basic His 6 -Zb tag, which confers tight binding to cation exchange columns, was produced in baculovirus-infected Sf9 cells. The protein was purified using immobilized metal affinity chromatography (IMAC) on a nickel-Sepharose column as the initial step. Subsequent preparations were performed without the use of IMAC, to avoid introduction of Ni 2+ ions. Artemis crystals were grown and diffracted to 1.6Å resolution (Table 1) ; the structure was solved using a structure of SNM1A (PDB: 5Q2A) as a molecular replacement model. The resultant structure (PDB: 7AF1; space group P1) contains a single molecule in the asymmetric unit, with two active site zinc ions (i.e. at both M1 and M2 sites) ( Figure 1 ). Copurifying metal ions were identified using Xray fluorescence (XRF) analysis. With the protein purified using IMAC, the structure was refined with a full occupancy nickel ion at the M1 site and a partial occupancy zinc ion at the M2 site (PDB: 6TT5) ( Figure 1I ); the identification of a nickel ion (at M1 and/or M2 sites) was supported by XRF (Supplementary Figure S1) . A similar overall metal ion coordination pattern has been observed with other members of the family, such as SNM1A and SNM1B/Apollo (3, 11) . The overall fold of the catalytic core of Artemis is similar to that of SNM1A and SNM1B (2Å backbone RMSD); it has all the key structural characteristics of human MBL fold nucleases, with the di-metal containing active site at the interface between the MBL and ␤-CASP domains ( Figure 1A ). Its MBL domain ( Figure 1A , B) has the typical ␣/␤-␤/␣ sandwich MBL fold (45) and contains the conserved motifs 1-4 ( Figure 1C and D, Supplementary Figure S2 ) typical of the MBL superfamily and motifs A-C typical of the ␤-CASP fold family (2, 4, 29, 46) . Motifs 1-4 ( Figure 1D -J) are responsible for metal ion coordination in both DNA and RNA processing MBL fold enzymes (4) . As observed in structures of SNM1A and SNM1B, Artemis can coordinate one or two metal ions in its active site. In the di-zinc complexed structure, one zinc ion (M1) is coordinated by four residues (His33, His35, His115 and Asp116) and two water molecules (H 2 O 506 and 611) in an octahedral manner ( Figure 1H ). The second zinc ion (M2) was refined with 30% occupancy and is coordinated by Asp37, His38 and Asp136 and two waters. The low occupancy of the second zinc ion, together with the two conformations (0.5 occupancy for each conformation) observed for Asp37 ( Figure 1I and J) suggest that the M2 site binds a metal ion less tightly than the M1 site, consistent with studies on other human MBL fold nucleases (3, Baddock et al. The structure of human SNM1A has been solved with a single zinc ion coordinated in its active site ( Figure 1E Figure 1F ). In summary, the coordination for the M1 site zinc ion involves three histidines, one aspartate, and either water molecules or a phosphate oxygen of the substrate. The M2 site metal ion is more weakly coordinated in the SNM1 family, with one histidine and two aspartates, with the remaining three positions occupied by water or a phosphate oxygen of the substrate. Thus, the presence of the substrate at the active site may help promote full metal ion occupancy at the M2 site. We propose that Artemis coordinates a phosphate group of its substrate in a similar manner to that proposed for SNM1B. A structure of human CPSF-73 (PDB: 2I7V), an MBL RNA processing nuclease, has been solved with two active site bound zinc ions, an adjacent sulphate, and a bridging water molecule proposed to play an important role in hydrolysis ( Figure 1G ) (47) . The CPSF-73 structure shows that the two zinc ions are coordinated in a very similar geometry with the human MBL DNA processing enzymes (48, 49) . A striking difference between the MBL RNA and DNA nucleases is that the second metal ion (M2) in the RNA processing nucleases is coordinated by an additional histidine (His418 for CPSF-73) (47) , which has no counterpart in the DNA processing enzymes. Data in parentheses is for the high-resolution shell. Proteins with a ␤-CASP fold form a distinct sub-group in the MBL-superfamily that specifically act on nucleic acids (2) . Artemis' ␤-CASP domain (aa 156-384) is inserted within the MBL fold sequence between ␣-helices 6 and 7 ( Figure 1 ). Notably, a second metal ion binding site, unique to Artemis, with similarity to the canonical Cys 2 His 2 zincfinger motif, is present in its ␤-CASP domain (50, 51) . Many DNA binding proteins, including transcription factors and DNA repair factors (including those involved in NHEJ), possess a Cys 2 His 2 zinc finger motif that stabilizes the DNA binding domain (50) (51) (52) (53) . A typical Cys 2 His 2 zinc coordinating finger ( Figure 2A ) has a ␤␤␣ motif, wherein the zinc ion is coordinated between an ␣-helix and two antiparallel ␤-sheets. Hydrophobic residues located at the sides of the zinc coordination site enable specific binding of the zinc finger in the major groove of the DNA (50, 51, 54, 55) . Similar to the canonical Cys 2 His 2 zinc finger motif, the zinc ion coordination in Artemis' ␤-CASP domain adopts a tetrahedral geometry, with coordination by two cysteine (Cys256, Cys272) and two histidine (His228, His254) residues (Figure 2B ). However, in the case of Artemis the metal ion coordination site is sandwiched between two ␤-sheets instead of an ␣-helix and two antiparallel ␤-sheets. All but one of the residues in the zinc-finger-like motif (His228, Cys256 and Cys227) are distinct to Artemis relative to SNM1A/B (Supplementary Figure S2) , with only His254 seen in all three enzymes. However, the four residues that form the zinc-finger like motif are conserved in Artemis across the animal kingdom (from humans to sponges), implying functional importance (Supplementary Figure S3) . Consistently, substitution of His228 and His254 (H228N and H254L), two of the zinc coordinating residues in the ␤-CASP domain of Artemis, cause RS-SCID in humans (4, 36, 56) . Patients with these inherited mutations suffer from impaired V(D)J recombination, leading to underdeveloped B and T lymphocytes. The full-length H254A Artemis variant is unable to carry out V(D)J recombination in vivo and has no discernible endonucleolytic activity in vitro (4). During our work two Artemis structures (PDB: 6WO0 and 6WNL) similar to our structure (PDB: 7AF1) (backbone RMSDs of 0.48 and 0.54Å, respectively) were reported (37), with identical relative positioning of the MBL and ␤-CASP domains ( Figure 3A and Supplementary Figure S4A ). The only significant difference was that whilst we refined our structure PDB: 7AF1 with two zinc ions in the active site, both of the crystal forms reported by Karim et al. (37) were modelled with a single active site zinc ion (Zn1), reinforcing the proposal of weaker metal ion binding at the Zn2/M2 site. 1SP2 ). This has a ␤␤␣ fold, where two Cys-and two His-residues are involved in zinc ion coordination and the sidechains of three conserved hydrophobic residues are shown. (B) The ␤-CASP region of Artemis has a novel zinc-finger like motif. The inset shows the four residues (two His and two Cys) coordinating the zinc ion (grey). The F o -F c electrondensity map (scaled to 2.5 in PyMOL) surrounding the zinc ion before it was included in refinement is shown. A striking aspect of the reported (37) structures is that both crystal forms were obtained in the presence of DNA and were reported to require DNA for their growth; the crystals showed a fluorescence signal supporting the presence of DNA (the oligonucleotides used contained a cyanine dye fluorophore), yet neither of the models presented contain DNA. The authors referred to some broken stacking electron density in 6WNL in a solvent channel and a patch of unsolved density approaching the active site in 6WO0, but state that the DNA 'did not bind to the protein in a physiological way, and likely bound promiscuously to promote crystallization' (37) . We re-examined these structures looking closely at the residual electron density. For the 6WNL structure we found evidence for a distorted duplex DNA of around 13 base pairs which we propose may be the product of duplex annealing of the oligonucleotide used for crystallization (a semi-palindromic 13-mer that was designed to form a hairpin with phospho-thioate linkages in the single-stranded region) (Supplementary Figure S4B) . For this structure, we are in general agreement with Karim et al. that the DNA does not appear to make meaningful interactions with the protein that inform on the mechanism of nuclease activity, although this mode of association with DNA may be relevant to alternative binding modes relating to higher order complexes containing Artemis. For the PDB: 6WO0 structure we were able to confidently build a DNA molecule that contacts the Artemis active site in a manner that we believe to be relevant to the Artemis nuclease activity. The model contains an 8-nucleotide 5single-stranded extension with a short 2-base pair region of duplex DNA that reaches into the Artemis active site, mak-ing close contacts with the metal ion centre in a manner consistent with the proposed catalytic mechanism ( Figure 3 ). The sequence of the longest strand corresponds to the 10-nucleotide cy-5 labelled strand (cy5-GCGATCAGCT) with residual density at the 5 -end that may be attributed to the cyanine fluorophore which we did not include in our model. The complementary strand used in crystallization was 13-nucleotides long and was intended to produce a 5overhang, but only two bases and three phosphates could be located in the density. The abrupt manner in which the electron density apparently disappears from either end of this strand suggests that this is the product of a cleavage reaction, although it is possible that remaining nucleotides are not located due to disorder. The analysis of electron density at this site is complicated by the proximity to a crystallographic 2-fold symmetry axis, which brings a symmetry copy of the DNA molecule into a position where atoms partially overlap and the extended 5 strands form a pseudo duplex (Supplementary Figure S5A) . The occupancy of the entire DNA molecule is thus limited to 0.5, and the lower occupancy is reflected in the electron density map which requires a lower contour level than would usually be applied (Supplementary Figure S5B) . After carefully building and refining the afore-described DNA bound model, significant positive electron density was revealed for the second metal ion (M2 site) which we included in the model with the same occupancy (0.5) as the DNA. Our model was refined to similar crystallographic R-factors as 6WO0 and has been deposited with PDB accession number 7ABS (refinement statistics are given in Table I) . Nucleic Acids Research, 2021, Vol. 49, No. 16 9317 Using the crystallographically observed DNA as a template we were able make a model for Artemis binding to a longer section of double-stranded DNA by complementing unpaired bases on the single-stranded DNA overhang with canonical base pairs, whilst maintaining acceptable geometry of the sugar phosphate backbone ( Figure 4A ). The duplex section of this model deviates slightly from the ideal Bform geometry (57) , in a manner that is reminiscent of some transcription factor DNA complexes (58, 59) . We have also modelled an extension of the metal ion contacting strand by three nucleotides to form a 5 -overhang; the positioning of the overhang nucleotides is more speculative, nevertheless it was possible to avoid clashes with residues whist maintaining a relaxed geometry ( Figure 4E ). In the extended DNA complex model, Artemis contacts both strands of the DNA in several areas; notably a single phosphate lies above the di-metal ion bearing active site and ligates to both metal ions in the same manner as in structures of related SNM1 enzymes (Baddock et al. 2021, accompanying paper) (60) . The two downstream nucleotides on this strand pass close to the protein surface, forming possible interactions with the backbone NH of Asp37 and the sidechain of Lys36; subsequent nucleotides are not close to the protein ( Figure 4C ). The overhang portion of this strand continues with a slightly altered trajectory, potentially contacting Artemis in the vicinity of the cleft separating the MBL and ␤-CASP domains, with the possibility of forming favourable interactions with both positively charged (Arg 172) and aromatic residues (Phe173, Trp293) ( Figure 4E ). The surface between the MBL and ␤-CASP domains of Artemis contains a belt of positively charged residues (Figure 4A) , which likely facilitates productive DNA binding. The complementary strand interacts with the protein via backbone contacts that span a 4-nucleotide sequence between 5-and 8-bases from the 3 -terminus with positively charged sidechains in the MBL domain (Lys36, Lys40, Arg43 and Lys74) ( Figure 4C ). The 3 -end of this strand apparently terminates directly above a cluster of polar or positively charged residues in the ␤-CASP domain (Lys207, Lys288, Asn205) ( Figure 4D ). Whilst the experimental (as used in co-crystallization) DNA substrate and our model both contain a 3 -hydroxyl group, the model implies that an additional 3 -phosphate could be accommodated and may be expected to make favourable interactions with the basic cluster of residues. Thus, our model suggests Artemis preferentially binds DNA with a 5 -overhang binding at the junction between double-and single-stranded regions; the expected product of this reaction would be a blunt ended DNA with a 5 -phosphate. In the case of hairpin DNA substrates our model indicates Artemis may accommodate a 4-nucleotide loop connecting the two paired strands, with the cleavage product being DNA with a 3 -overhang cleaved from the last paired base of the duplex. In an accompanying paper [Baddock et al. 2021 ] we report a structure of SNM1B/Apollo in complex with two deoxyadenosine monophosphate nucleotides and suggest a model for SNM1B binding to DNA containing a 3overhang (one of its preferred substrates). The overall mode of DNA binding is similar in the two models ( Figure 4A and B), with the two DNA duplexes being roughly paral- lel and forming contacts to similar regions on the MBL domain. The most important differences lie in the nature of the contacts formed to the active site and the paths of the various overhangs. In the SNM1B model, extensive contacts are made to the 5 -phosphate in a well-defined phosphate binding pocket. Both SNM1A and SNM1B are exclusively 5phosphate exonucleases, with most of these phosphate binding residues being conserved (Supplementary Figure S2) . Artemis lacks these key phosphate binding residues and the 5 -phosphate binding pocket of SNM1A and SNM1B. Instead, in Artemis this pocket is partially filled by the Phe318 sidechain, which may inhibit the binding of the substrate 5phosphate. These differences rationalize major differences in nuclease activities within the family, i.e. SNM1A and SNM1B are exonucleases and Artemis is an endonuclease. Further differences between Artemis and SNM1B/SNM1A are found in the loop connecting ␤-strands G and H (using Artemis numbering), which in Artemis is significantly longer and occupies a different position contacting residues in the MBL domain ( Figure 4D ), compared to the loops in SNM1B and SNM1A that form part of the phosphate binding pocket and make potential contacts with the 3 -overhang. This loop displacement in Artemis may contribute to its ability to accommodate substrates with either 5 -overhangs or hairpins. Another striking difference, at least in the available structures, is that the active site of Artemis is more open compared to those of SNM1A or SNM1B. This openness may reflect an ability to accommodate different substrate conformations including hairpins, 3 -and 5 -overhangs, as well as DNA flaps and gaps. Both human SNM1A and SNM1B appear to have a more sequestered active site that would only fit a single strand of DNA, which is consistent with previous findings on their preferred substrate selectivity (10) . To investigate the activity of different versions of Artemis, we performed assays using radiolabelled DNA. We compared the catalytic domain purified using IMAC (which contained Ni 2+ in the active site) with catalytic domain protein purified using ion exchange chromatography (avoiding IMAC), which contained predominantly Zn 2+ . We also tested the activity of full-length phosphorylated Artemis. The results show that both truncated enzymes have similar activity to the full-length enzyme ( Supplementary Figure S6 ). One notable difference between our full-length protein and that of Ma et al. (20) , is that our protein is active in the absence of DNA-PKcs. Intact mass spectrometric analysis of full-length protein shows that it has undergone up to five phosphorylations (Supplementary Figure S7) . Poinsignon et al. have shown that Artemis is constitutively phosphorylated in cultured mammalian cells and is further phosphorylated in response to induced DNA damage (61) ; it is interesting that the capacity to phosphorylate Artemis to produce an active form is also conserved in insect cells. We observed exonuclease activity with full-length Artemis at 10 nM enzyme concentration (Supplementary Figure S8) , though this was weak compared to its endonuclease activity at the same concentration. We observed no exonuclease activity for the truncated Artemis construct, it is possible that phosphorylation alters the balance between endonuclease and exonuclease activity, though the biological relevance of this, if any, remains to be validated. Both human SNM1A and SNM1B require a 5phosphate for activity (3, 11, 17) . To investigate whether there is a similar requirement for Artemis, we tested the activity of truncated Artemis against single-stranded and overhang DNA substrates with different 5 -end groups, including a phosphate, hydroxyl and biotin groups ( Figure 5A ). The results obtained imply that, at least under the tested conditions, Artemis is agnostic to the different end modifications, exhibiting comparable digestion of all substrates. Extensive evidence demonstrates that full-length Artemis in complex with DNA-PKcs is a structure-selective endonuclease (19, 20, 29) . These studies report Artemis can digest substrates including overhangs, hairpins, stem-loops, and splayed arms (pseudo-Y). To investigate the activity of truncated Artemis catalytic domain (aa 1-361) we performed assays using a variety of radio-labelled DNA substrates ( Figure 5B ). The results show that truncated Artemis has substrate-selective endonuclease activity, with a preference for single-stranded DNA substrates, and those that contain single stranded character (e.g. 5 -and 3 -overhangs, splayed arms, and a lagging flap structure), compared with double-stranded DNA structures (e.g. ds DNA and a replication fork). This is in accord with previous reports, where Artemis is reported to cleave around ss-to dsDNA junctions in DNA substrates (62) . The truncated Artemis catalytic domain also exhibits hairpin opening activity, in accordance with what has previously been reported (Supplementary Figure S9 ). On a duplex substrate (YM117 from Ma et al.) (20) with a 20 nt hairpin region, Artemis cleaves adjacent to the hairpin, consistent with previous data. It is clear that truncated form of Artemis exhibits nuclease activity closely comparable to the phosphorylated full-length Artemis protein (20) , indicating that the structural studies presented here reveal mechanistic insights of direct relevance to the DNA-PKcs-associated form of Artemis that engages in end-processing reactions in vivo. Previous mutagenesis studies targeting metal ion coordinating residues (D17N, H33A, H35A, D37N ) of full-length Artemis (aa 1-692) established the importance of motifs 1-4 for activity (29) . Each of these substitutions markedly reduced or abolished Artemis' ability to carry out its role in V(D)J recombination in vitro. We crystallized three forms of truncated Artemis (aa 1-361) with substitutions in several of these metal ion co-ordinating residues, i.e. D37A, H33A, and the Omenn Syndrome patient mutation H35D (33, 56) . The overall architecture of the three variants is almost identical to WT Artemis ( Figure 6A ). The D37A structure crystallizes with only one nickel ion (M1) ( Figure 6B ); Both the H33A and H35D variants exhibited loss of the M1 ion (Figure 6C and D) . By contrast, all three variants retained the Zn ion in the zinc finger-like motif of the ␤-CASP domain. Our D37A, H33A, and H35D variants lost the ability to digest DNA substrate in vitro ( Figure 6E ), in accord with results with full-length variants (29, 33) . Differential scanning fluorimetry (DSF) analyses showed that both H33A and H35D variants are substantially destabilized with T m around -13 • C compared to WT Artemis (Supplementary Figure S10) . By contrast the D37A variant has similar thermal stability as the WT Artemis, suggesting that it is folded in the presence of a single metal ion in the active site. Asp37 can adopt two conformations and the coordinated zinc ion is present at about 30% occupancy in both of the WT Artemis structures (PDB 6TT5 and 7AF1). It is thus unsurprising that mutation of Asp37 to alanine results in the loss of Zn2. Histidines 33 and 35 are the first two histidine residues in the HxHxDH motif (M1 binding) in the active site. In the absence of active site metal ions, the loop comprising residues 113-119 moves away from the active site ( Figure 6C and D) and a small rearrangement occurs in ␣8 (residues 348-358). In both the H33A and H35D variants, ␣8 moves slightly closer towards ␤14, compared to the WT and D37A variant. Larger differences occur in ␤-strand E (residues 268-270) and ␣-helix E (residues 261-267), both located near the zinc finger motif in the ␤-CASP domain ( Figure 6A ). In H33A and H35D variants, both ␤-strand E and ␣-helix E shifted upward and away from the zinc finger like motif. These conformational changes suggest induced fit during substrate binding. The combined results with the variants thus reveal the importance of the HxHxDH motif and highlight the importance of the di-metal catalytic core in the SNM1 family, not only in directly catalysing hydrolysis, but also likely in conformational changes involved in catalysis (46) . Artemis, along with SNM1A and SNM1B, possess a conserved MBL-fold domain similar to that of the true bacterial MBLs, suggesting that they may be inhibited by ␤lactams. Studies on SNM1A/SNM1B, showed that ceftriaxone (Rocephin), a widely used ␤-lactam (third generation cephalosporin) inhibits SNM1A/SNM1B (63) . To investigate if it inhibits Artemis we performed assays with ceftriaxone, cefotaxime, and 7-aminocephalosporanic acid (Figure 7A) . The results show that neither cefotaxime nor 7aminocephalosporanic acid inhibit Artemis, whilst ceftriaxone inhibits with a modest IC 50 of 65 M ( Figure 7B ). We solved a structure of ceftriaxone bound to Artemis (purified by IMAC) at 1.9Å resolution ( Figure 7C ) in the space group P1 with one protein molecule in the asymmetric unit. As before, in this structure Artemis possesses the canonical bilobar MBL and ␤-CASP fold with an active site containing one nickel ion, likely reflecting the purification method. Ceftriaxone binds to the protein surface in an extended manner making interactions with the active site, towards the ␤-CASP domain ( Figure 7C ). There is no evidence for cleavage of the ␤-lactam, nor of loss of the C-3 cephalosporin side chain, reactions that can occur during 'true' MBL catalysed cephalosporin hydrolysis. Electron density at the active site clearly reveals the ceftriaxone side chain coordinates the M1 site nickel ion replacing two waters (72 and 106) in the uncomplexed structure ( Figure 7D ). Despite the conservation of key elements of the active site of the MBL fold nucleases and the 'true' ␤−lactam hydrolysing MBLs (64), ceftriaxone, does not interact with the nickel ion via its ␤-lactam carbonyl (as occurs for the true MBLs), but via both carbonyl oxygens of the cyclic 1, 2 diamide in its sidechain (Ni-O distances: 2 and 2.2Å), i.e. it is not positioned for productive ␤-lactam hydrolysis. The amino-thiazole group (N7) of ceftriaxone forms hydrogen bonds with the side chain of Asn205, while the S1 of the 7-aminocephalosporanic acid core of the compound interacts with the hydroxyl of Tyr212 through an ethylene glycol molecule. The rest of the molecule appears to be flexible. The binding mode of ceftriaxone to Artemis shown in Figure 7C is near identical to that observed for ceftriaxone with SNM1A (PDB: 5NZW) structure (Supplementary Figure S11) . A notable difference between the ceftriaxone-bound Artemis structure and the non-complexed structure is the loss of a metal ion at M2 site, normally coordinated by Asp37, His38 and Asp130 ( Figure 7D ). Without a metal ion at the M2 site, as in the ceftriaxone bound structure, the Asp37 side chain is positioned away from the active site ( Figure 7D ), as seen when nickel is bound in the M1 site ( Figure 1I ). To investigate the possibility of inhibiting Artemis through binding to the zinc finger motif in the ␤-CASP domain, we used the fluorescence-based nuclease assay to test compounds known to react with thiol groups present in zinc fingers and which result in zinc ion displacement, i.e. ebselen (65) , auranofin (66) and disulfiram (67) . Both ebselen and disulfiram inhibit Artemis with IC 50 values around 8.5 M and 10.8 M respectively, whilst auranofin inhibits less potently (IC 50 Nucleic Acids Research, 2021, Vol. 49, No. 16 9323 The DCLRE1C/Artemis gene was first discovered in studies of children with a radiosensitive form of severe combined immunodeficiency disease (RS-SCID) (68) . Subsequent work has shown Artemis is a structure-specific endonuclease involved in V(D)J recombination (18, 19, 69) and the c-NHEJ DNA repair pathway (4, 23, 70) (2). Our structures of wild-type and catalytic site mutants of Artemis protein show that, like SNM1A and SNM1B/Apollo and the RNA processing enzyme CPSF73, Artemis has a typical ␣/␤-␤/␣ sandwich fold in its MBL domain and has a ␤-CASP domain, the latter a characteristic of MBL fold nucleases. Both our structures and those of Karim et al (37) reveal a unique and conserved structural feature of Artemis in its ␤-CASP domain that is not reported in other human MBL enzymes, i.e. a conserved (in Artemis orthologues) classical zinc-finger like motif. Zinc-finger motifs are common in DNA binding proteins such as transcription factors (51, 55) and are observed in required and accessory of NHEJ proteins (53) . They are reported to both provide stability and enhance substrate selectivity rather than being directly involved in catalysis; we propose that this is likely the case for Artemis. Furthermore, point mutations in His 228 and His 254 (H228N and H254L) have been reported in patients with a SCID phenotype (56) . The role of the zinc-finger motif in Artemis is unknown and the subject of ongoing investigations. We have mutated the zinc finger residues (His 228, His 254 Cys 256 and Cys 272), but none of the mutated constructs produced purified protein. Although inconclusive, this may indicate that the Zinc finger is important for the structural integrity or stability of the protein. The presence of one or two metal ions coordinated by the HxHxDH motif at the Artemis active site is a hallmark of the SNM1 family (11, 71) ; the available evidence implies that metal ion binding at one site (M1 site in standard MBL nomenclature) is stronger than at the other (M2 site). By analogy with studies on the true MBLs, these metal ions are proposed to activate a water / hydroxide ion that act as the nucleophile in the phosphodiester cleavage (72) . This hydroxide ion is also present in the crystal structure of CPSF-73 ( Figure 1G ), along with an adjacent sulphate ion (47) , which is proposed to mimic the phosphate group of the pre-mRNA at the cleavage site. Our structure (PDB: 7AF1) suggests that the native metal ion(s) residing in the M1 site of Artemis is zinc, although a nickel ion can also occupy the M1 site depending on the how the protein was purified (PDB: 6TT5). Neither the presence of copurifying Ni ion in the active site, nor the truncation of the C-terminal tail, appear to substantially inhibit the activity of Artemis. Thus, using radio-labelled gel-based nuclease assays, we showed that the truncated Artemis catalytic domain (aa 1-361) with either copurifying Zn or Ni ions (as observed crystallographically in the same preparations) have similar activity with the full-length Artemis construct (aa 1-693). Therefore, it seems likely that nickel ions are able to replace zinc ions in solution; catalysis of MBL fold enzymes, including hydrolytic reactions, with metal ions other than zinc is wellprecedented (64, 73) We solved structures of three Artemis variants D37A, H33A and an Omenn syndrome patient mutation, H35D. Using gel-based nuclease assays, we showed that these variants are inactive. Overall, the three variant structures are similar to the WT structure, even though H33A and H35D entirely lack any metal ions in their active site, although zinc was present in the zinc finger. Mutation of Asp37 to alanine results in the loss of the second metal in the catalytic site, likely explaining the loss of activity, although the first metal ion is still present. Note that some MBL fold hydrolases uses two metal ions (e.g. B1 and B3 subfamilies of the true MBLs and RNase J1 from Bacillus subtilis) ( Supplementary Figure S13A and C) (64, 74) whereas others, sometimes with apparently similar active sites, only use one metal ion at the M1 site (e.g. the B2 subfamily of the true MBLs and RNase J2 from Staphylococcus epidermis) (64,75) (Supplementary Figure S13B and D) . Thus, whilst our results support the importance of having both metals for the nuclease activity by Artemis, subtle features can influence MBL fold enzyme activity (64, 71) . Following re-analysis of the Karim et al. structure (PDB code 6WO0), we generated a model of a DNA overhang in complex with Artemis that informs on the substrate binding mode. Our model shows that Artemis interacts with its DNA substrate at the interface between the MBL and the ␤-CASP domains. This interaction is mediated through a combination of polar or positive residues and aromatic residues of Artemis and the DNA substrate ( Figure 6) . A related interaction was also observed in a cryo-EM structure of the pre-mRNA processing enzyme CPSF-73 with substrate bound, where residues R34, R174, K24 and F176 (CPSF-73 numbering) form direct contacts with the substrate (76) . The Artemis DNA bound structure (PDB code 7ABS) contains a 2-base pair region of duplex DNA that makes close contact with the metal centre in the Artemis active site, which could potentially be catalytically relevant ( Figure 3 ). This phosphate ion coordinates the di-metal centre in a similar manner to the AMP molecule bound to the active site of SNM1B/Apollo [Baddock et al. 2021, accompanying paper] , and the phosphate backbone of the pre-mRNA substrate bound in CPSF-73 (PDB code 6V4X) (76) . By analogy with proposals for true MBLs (71, 77) , the mechanism of SNM1A/B and Artemis catalysed phosphate hydrolysis thus likely involves metal ion mediated reaction of a hydroxide ion with the phosphate and activation of the phosphate by metal ion chelation. However, the precise identity of the in vivo relevant SNM1 active site metal ions and whether there are one or two metal ions is uncertain. The true MBLs can be active with two metal ions (B1 and B2 subfamily MBLs) or one metal ion (B2 MBLs); these metal ions are likely zinc ions, but even this has not been unequivocally demonstrated within a human disease context and it is possible that metal ions other than zinc are relevant (77, 78) . The available evidence is that the SNM1 MBL nucleases employ two (zinc) metal ions, with one (at the M2 site) being less tightly bound than the other (M1 site) -though considerable uncertainty remains as to the in vivo relevant metal ions. Binding of a metal ion at the M2 site appears to be promoted by the DNA substrate and it cannot be ruled out that it is (sometimes) delivered by the substrate. Artemis is the only identified MBL/␤-CASP DNA processing enzyme that possesses substantive endonuclease activity. By contrast both SNM1A and SNM1B/Apollo are strictly 5 -phosphate exonucleases (3, 11, 17) . The recent structure of SNM1B/Apollo in complex with two deoxyadenosine monophosphate nucleotides reveals a cluster of residues that form a 5 -phosphate binding pocket, adjacent to the metal centre [Baddock et al., 2021, accompanying paper] . Structural sequence alignments of the three SNM1 proteins shows that these residues are conserved in SNM1A and SNM1B (Supplementary Figure S1) . Apart from Ser317, none of these conserved phosphate binding pocket residues are present in Artemis. Instead, the pocket is partially occupied by the Phe318 side chain, which is absent in both SNM1A and SNM1B. Artemis also possesses a longer and more flexible loop connecting ␤-strands 27 and 28 ( Figure 6B ), compared to the same loop in SNM1A and SNM1B that make up part of the 5 -phosphate binding pocket. The flexibility of this loop could enable accommodation with different types of DNA structures, such as hairpins, and 5 -overhangs. These differences plausibly explain Artemis' substrate preferences and its primary activity as a structure-selective endonuclease. Radiotherapy is a mainstay of cancer therapy; its effectiveness relies on inducing DNA double-strand breaks (DSBs) that contain complex, chemically modified ends that must be processed prior to repair (79, 80) . The canonical non-homologous end-joining (c-NHEJ) pathway repairs around 80% of DNA double-strand breaks in mammalian cells (22, 25) . Combining radiation therapy in conjunction with SNM1 inhibitors could selectively radiosensitize tumours. Of the three of the ␤-lactam antibacterial compounds previously shown to inhibit SNM1A (63), we only observed inhibition of Artemis with ceftriaxone. Although the potency of inhibition is moderate (IC 50 65 M), we were able to solve the structure of ceftriaxone in complex with Artemis. Notably, ceftriaxone does not bind with its ␤lactam carbonyl located at the active site where it ligates to one zinc (or other metal) ion, but instead binds the single nickel ion in a bidentate manner via the carbonyls of the cyclic 1,2 diamide on its C-3 sidechain (71, 81) . Studies with the true MBLs have shown that appropriate derivatization of weakly binding molecules can lead to highly potent and selective inhibitors. In an attempt to inhibit Artemis though its unique zincfinger like motif, we tested inhibitors with thiol-reactive groups. Ebselen, disulfiram and auranofin have the potential to interact with zinc fingers, including via zinc ejection with consequent protein destabilization (66, 67, 82, 83) . Both ebselen and auranofin are reported have antimicrobial properties (84) and ebselen is in clinical trials for conditions, ranging from stroke to bipolar disorder (85) , and auranofin is used for treatment of rheumatoid arthritis (86) . Recent studies have also shown that ebselen inhibits enzymes from SARS-CoV-2, i.e. the main protease (Mpro) and the exonuclease ExoN (nsp14 ExoN -nsp10) complex (87, 88) . Disulfiram is a known acetaldehyde dehydrogenase inhibitor used in treatment for alcohol abuse disorder (89) . Our re-sults show that both ebselen and disulfiram inhibit Artemis (IC 50 s 8.5 M and 10.8 M, respectively), whilst auranofin is less potent (IC 50 46 M) . Studies focussed on inhibiting the MBL fold nucleases are at an early stage compared with work on the true MBLs (78) The structures and assays results presented here provide starting points with established drugs, from which it might be possible to generate selective Artemis inhibitors, either binding at the active site or elsewhere (including the apparently unique zinc finger of Artemis), in order to radiosensitize cells. Coordinates and structure factors are deposited in the Protein Data Bank under codes 6TT5, 7AF1, 7AFS, 7AFU, 7AGI, 7APV and 7ABS. Nucleases: diversity of structure, function and mechanism Metallo-beta-lactamase fold within nucleic acids processing enzymes: the beta-CASP family The structures of the SNM1A and SNM1B/Apollo nuclease domains reveal a potential basis for their distinct DNA processing activities A histidine in the ␤-CASP domain of Artemis is critical for its full in vitro and in vivo functions Crystal Structure of TTHA0252 from Thermus thermophilus HB8, a RNA degradation protein of the metallo-b -lactamase superfamily DNA-PK autophosphorylation facilitates Artemis endonuclease activity Artemis C-terminal region facilitates V(D)J recombination through its interactions with DNA ligase IV and DNA-pkcs DNA-PKcs dependence of artemis endonucleolytic activity, differences between hairpins and 5 or 3 overhangs Autoinhibition of the nuclease ARTEMIS is mediated by a physical interaction between its catalytic and C-terminal domains Evidence that the DNA endonuclease ARTEMIS also has intrinsic 5 -exonuclease activity The SNM1A DNA repair nuclease The multifunctional SNM1 gene family: not just nucleases Human SNM1A and XPF-ERCC1 collaborate to initiate DNA interstrand cross-link repair The apollo 5 exonuclease functions together with TRF2 to protect telomeres from DNA repair Apollo, an Artemis-related nuclease, interacts with TRF2 and protects human telomeres in S phase Human SNM1B is required for normal cellular response to both DNA interstrand crosslink-inducing agents and ionizing radiation Characterization of the human SNM1A and SNM1B/Apollo DNA repair exonucleases VDJ recombination: Artemis and its in vivo role in hairpin opening Hairpin opening and overhang processing by an Artemis/DNA-dependent protein kinase complex in nonhomologous end joining and V(D)J recombination The Artemis:DNA-PKcs endonuclease cleaves DNA loops, flaps, and gaps Artemis, a novel DNA double-strand break repair/V(D)J recombination protein, is mutated in human severe combined immune deficiency The mechanism of DSB repair by the NHEJ DNA-PKcs regulates a single-stranded DNA endonuclease activity of Artemis DNA double-strand break repair inhibitors as cancer therapeutics Nonhomologous DNA end-joining for repair of DNA double-strand breaks DNA hairpin opening mediated by the RAG1 and RAG2 proteins DNA cleavage activity of the V(D)J recombination protein RAG1 is autoregulated Crystal structure of the V(D)J recombinase RAG1-RAG2 Functional and biochemical dissection of the structure-specific nuclease ARTEMIS Targeted disruption of the gene encoding DNA ligase IV leads to lethality in embryonic mice V (D) J recombination: broken DNA molecules with covalently sealed (hairpin) coding ends in scid mouse thymocytes The mechanism and regulation of chromosomal V(D)J recombination Plenary paper Omenn syndrome due to ARTEMIS mutations DCLRE1C (ARTEMIS) mutations causing phenotypes ranging from atypical severe combined immunodeficiency to mere antibody deficiency A founder mutation in Artemis, an SNM1-like protein, causes SCID in athabascan-speaking native Americans Functional analysis of naturally occurring DCLRE1C mutations and correlation with the clinical phenotype of ARTEMIS deficiency Structural analysis of the catalytic domain of Artemis endonuclease/SNM1C reveals distinct structural features High-throughput production of human proteins for crystallization: The SGC experience Site-directed mutagenesis by inverse PCR DIALS: implementation and evaluation of a new integration package Phaser crystallographic software Coot: model-building tools for molecular graphics REFMAC5 for the refinement of macromolecular crystal structures Cephalosporins inhibit human metallo ␤-lactamase fold DNA repair nucleases SNM1A and SNM1B/apollo The 3-D structure of a zinc metallo-␤-lactamase from Bacillus cereus reveals a new type of protein fold The ␤-lactamase motif in Snm1 is required for repair of DNA double-strand breaks caused by interstrand crosslinks in S. cerevisiae Polyadenylation factor CPSF-73 is the pre-mRNA 3 -end-processing endonuclease Analysis of zinc binding sites in protein crystal structures Zinc coordination geometry and ligand Binding affinity: the structural and kinetic analysis of the second-shell serine 228 residue and the methionine 180 residue of the aminopeptidase from Vibrio proteolyticus Zinc fingers -folds for many occasions DNA Recognition by Cys 2 His 2 Zinc Finger Proteins Structural classification of zinc fingers DNA double-strand break repair: Putting zinc fingers on the sore spot Physical and functional sensitivity of zinc finger transcription factors to redox change Zinc finger proteins: new insights into structural and functional diversity The most frequent DCLRE1C (ARTEMIS) mutations are based on homologous recombination events DNA structure: A-, B-and Z-DNA helix families Zif268 protein-DNA complex refined at 1.6Å: A model system for understanding zinc finger-DNA interactions Signatures of protein-DNA recognition in free DNA binding sites Polyadenylation factor CPSF-73 is the pre-mRNA 3 -end-processing endonuclease The metallo-␤-lactamase/␤-CASP domain of Artemis constitutes the catalytic core for V(D)J recombination Structure-specific nuclease activities of Artemis and the Artemis: DNA-PKcs complex Cephalosporins inhibit human metallo ␤-lactamase fold DNA repair nucleases SNM1A and SNM1B/apollo Metallo-␤-lactamase structure and function Supramolecular organization of the repetitive backbone unit of the Streptococcus pneumoniae pilus Zinc finger domains as therapeutic targets for metal-based compounds -an update Zinc finger-containing cellular transcription corepressor ZBTB25 promotes influenza virus RNA transcription and is a target for zinc ejector drugs Artemis: guarding small children and, now, the genome Redox active thiol sensors of oxidative and nitrosative stress Comparison of nonhomologous end joining and homologous recombination in human cells The chemical biology of human metallo-b-lactamase fold proteins Metallo-␤-lactamase structure and function Use of ferrous iron by metallo-␤-lactamases Unusual, dual endo-and exonuclease activity in the degradosome explained by crystal structure analysis of RNase J1 Structural and biochemical characteristics of two Staphylococcus epidermidis RNase J paralogues RNase J1 and RNase J2 Structure of an active human histone pre-mRNA 3 -end processing machinery The catalytic mechanisms of binuclear metallohydrolases ␤ -Lactamases and ␤ -Lactamase Inhibitors in the 21st century Chemotherapeutic compounds targeting the DNA double-strand break repair pathways: the good, the bad, and the promising A historical reflection on our understanding of radiation-induced DNA double strand break repair in somatic mammalian cells; interfacing the past with the present The enzymes of B-lactam biosynthesis Density functional theory study of the attack of ebselen on a zinc-finger model Identification of labile Zn sites in drug-target proteins Repurposing auranofin, ebselen, and PX-12 as antimicrobial agents targeting the thioredoxin system Ebselen, a useful tool for understanding cellular redox biology and a promising drug candidate for use in human diseases Auranofin: repurposing an old drug for a golden new age Structure of Mpro from SARS-CoV-2 and discovery of its inhibitors Characterisation of the SARS-CoV-2 ExoN (nsp14ExoN-nsp10) complex: implications for its role in viral genome stability and inhibitor identification Disulfiram efficacy in the treatment of alcohol dependence: a meta-analysis We are very grateful to Dr Rod Chalk and Tiago Moreira for mass spectrometry, and Dr Neil Patterson for the helpful discussions and data collections at Diamond Light Source. We acknowledge Diamond Light Source for time and technical support on Beamlines I03, I04 and I24 under Proposal MX19301. Supplementary Data are available at NAR Online.