Toroidal structure and DNA cleavage by the CRISPR-associated [4Fe-4S]-cluster containing Cas4 nuclease SSO0001 from Sulfolobus solfataricus

Nuclease activity of Cas4 proteins

The purified Cas4 proteins SSO0001 and SSO1391 from S. solfataricus and Pcal_0546 from Pyrobaculum calidifontis exhibited a brown color indicating the presence of functional Fe-S clusters (Fig. 1a). Their absorption spectra showed the presence of a broad shoulder at 380–420 nm typical of 4Fe-4S cluster containing proteins (Fig. 1a). Nuclease assays using the 5'- and 3'-[³²P]-labeled linear ssDNA revealed the presence of high metal-dependent nuclease activity in all three proteins (Fig. 1b–d) including SSO1391 which was reported to be catalytically inactive in a previous work²⁰. The nuclease activity of SSO0001 was maximal in the presence of Mg²⁺ or Mn²⁺ (5 mM), KCl (20 – 100 mM), at alkaline pH (9 – 10), and at high temperatures (65 – 70 °C) (Supplementary Fig. S2). In contrast to the HNH-like and GIY-YIG-like nucleases, Ca²⁺ did not support the nuclease activity of SSO0001 (Supplementary Fig. S2a). The rates of cleavage of ssRNA and dsDNA with blunt ends by SSO0001 were approximately 20 times and 200 times lower, respectively, compared to ssDNA (Fig. 2b, Supplementary Fig. S2d).

With the 5'-labeled ssDNA as substrate, SSO0001 produced mononucleotides as products, whereas the 3'-labeled substrate was cleaved with the formation of a series of oligonucleotides with the shortest product containing four nucleotides (Fig. 1b). These results confirm that SSO0001 is a 5' → 3' exonuclease that cleaves one nucleotide at a time from the 5'-end of the substrate. The short oligonucleotides (17 nt) were cleaved by SSO0001 faster than longer substrates (62 nt) (Supplementary Fig. S2j), whereas phosphorylation or biotinylation of substrates at the 5'-terminus had no effect on activity (Supplementary Fig. S2h, i). The 5' to 3' exonuclease activity was also observed in the Cas4 protein Pcal_0546 (also DUF83), whereas SSO1391 (DUF911) produced a range of cleavage products with both substrates indicating that it cleaves ssDNA in both directions (5' → 3' and 3' → 5') (Fig. 1c and 1d, respectively). The three proteins also showed Mn²⁺-dependent endonuclease activity with circular ssDNA of the M13mp18 phage as a substrate (Fig. 1e–g), but no activity against circular dsDNA (Supplementary Fig. S3a). Interestingly, the E. coli RecB nuclease domain is able to cleave ssDNA with dual polarity and endonucleolytically, whereas the RecB nuclease domain of the B. subtilis AddB is a 5' to 3' exonuclease^17,21. Thus, our work together with the previous report on SSO0001²⁰ has demonstrated that the CRISPR-associated RecB-like proteins represent a diverse group of exonucleases, which can cleave ssDNA in one or both directions and also exhibit endonuclease activity against ssDNA.

Most known nucleases cleave the phosphodiester bond with the formation of 5'-phosphates and 3'-OH groups²². To determine the phosphodiester bond cleavage site of SSO0001, we carried out two reactions using the unlabeled ssDNA as substrate and T4 polynucleotide kinase (PNK): direct phosphorylation of cleavage products at 5'-end (requires a 5'-hydroxyl) and phosphate exchange reaction at 5'-end (requires a 5'-phosphate end). The results show high product labelling in the PNK phosphorylation reaction and no labelling in the PNK exchange reaction suggesting the absence of phosphate at the product 5'-end (Supplementary Fig. S3d). These results were confirmed using an unlabeled poly-dC₁₀ ssDNA substrate and a liquid chromatography-mass spectrometry analysis (LC-MS) of the cleavage products of the wild type SSO0001 and its inactive mutant D99A (Supplementary Fig. S3e). Thus, SSO0001 cleaves ssDNA substrates with the formation of products containing 5'-hydroxyls and 3'-phosphates. In contrast, the characterized RecB-like nucleases and λ exonuclease have been shown to generate the products with 5'-phosphates and 3'-hydroxyls^23,24 suggesting that these enzymes and SSO0001 might use different catalytic mechanisms.

SSO0001 exhibits ATP-independent DNA unwinding activity

Further biochemical characterization of Cas4 proteins was focused on SSO0001, since we solved its crystal structure. In electrophoretic mobility shift assays, the purified SSO0001 showed strong binding to ssDNA (42 nt) (Fig. 2a). With dsDNA substrates, detectable binding was observed only for substrates containing 5'-overhangs (10 nt), whereas no significant binding was observed to dsDNA with 3'-overhangs or blunt ends. This is consistent with the 5' to 3' polarity of ssDNA cleavage by SSO0001 (Fig. 1b). In nuclease assays, SSO0001 was able to cleave these dsDNA substrates, although with lower rates than ssDNA (Fig. 2b). With dsDNA substrates, the highest activity was observed against the dsDNA with 5'-overhangs and the lowest activity toward dsDNA with blunt ends. These results also suggest the presence of DNA unwinding activity in SSO0001. This was confirmed using the splayed arm duplex substrates, which are the intermediates of DNA recombination and integration and also might represent a potential intermediate of CRISPR spacer acquisition. Splayed arm structures are known substrates for RecB-like nucleases and are cleaved by them through a combination of exonuclease, endonuclease, and ATP-dependent helicase activities¹⁵. Two types of splayed arm substrates were used: containing both dsDNA (20 nt) and ssDNA (21 nt) areas with a [³²P]-label located at the 3'-end of the bottom (in the dsDNA area) or top (in the ssDNA area) strands (Fig. 2c, S3b). The time point experiments with the first substrate revealed the formation of a series of products with different lengths with the transitional accumulation of the 20 nt long products suggesting that the cleavage of labeled strand by SSO0001 was slowed upon reaching the substrate duplex (Fig. 2c). Longer incubation times (25 min) resulted in the formation of short products (1–4 nt), which can be produced only after complete unwinding of DNA duplex. The formation of short products was also observed with the second substrate containing the label at the 3'-end of the top strand, but this required longer incubation times consistent with the requirement to unwind and degrade the unlabeled strand first (Fig. 2c). With a series of splayed arm substrates with different arm lengths (5 to 30 nt), SSO0001 showed higher activity toward substrates with short arms (5–15 nt) (Supplementary Fig. S2k). ATP addition had no effect on the unwinding and cleavage of splayed arm substrates by SSO0001 suggesting that SSO0001 is an ATP-independent DNA unwinding enzyme (Supplementary Fig. S2l). Thus, our results indicate that in addition to endonuclease and 5' → 3' exonuclease activities toward ssDNA SSO0001 also exhibits the unwinding activity against dsDNA.

SSO0001 activity against CRISPR-related substrates

The CRISPR adaptation step has been proposed to involve the recognition of the leader, repeat, and PAM sequences and the formation of dsDNA breaks and 3′ ssDNA overhangs^3,5,12,20. To determine if SSO0001 exhibits a preference for substrates containing the S. solfataricus CRISPR sequences, we analyzed its activity against 14 ssDNA substrates (60–63 nt) containing various combinations of the repeat, spacer-1, and leader sequences from the S. solfataricus CRISPR locus B, which is the biggest (and probably most active) CRISPR locus in this organism (Supplementary Table S1). SSO0001 cleaved all these substrates with some preference for the repeat-spacer-repeat substrate, which also supported higher activity of SSO0001 in DNA binding assay suggesting that this protein might exhibit some level of sequence preference (Fig. 3a, b). It has been shown that the Sulfolobus CRISPR systems preferentially target protospacers located at the 3'-end of the PAM sequences CCN, TCN, and GTN or at the 5'-end of the PAM NGG^19,25. We designed 32 ssDNA substrates (CRISPR-A spacer-1, 41 nt) containing these PAMs located at the substrate 5'-end, in the middle, or at the 3'-end, as well as 27 substrates with non-PAM sequences (ACA, CAA, GAA) (Supplementary Table S1). SSO0001 showed similar cleavage and binding activities with both PAM and non-PAM substrates suggesting that the purified SSO0001 is not strictly specific to the PAM-containing ssDNA substrates in vitro (Fig. 3c, 3d; Supplementary Fig. S4, S5). The effect of PAMs on the DNA unwinding activity of SSO0001 was tested using several splayed arm substrates (42 nt) containing the sequences of the S. solfataricus CRISPR-B leader or spacer-1 and the TCC PAM placed at the 5'-end, in the middle, or 3'-end of the bottom strand, which was 3'-[³²P]-labeled (Fig. 3e; Supplementary Fig. S3c). SSO0001 exhibited different levels of unwinding and cleavage activities against these substrates with the absence of the noticeable correlation with the presence or location of PAM (Fig. 3e; Supplementary Fig. S3c). Thus, our results are consistent with a lack of significant sequence preference in SSO0001 toward the substrates containing the S. solfataricus PAMs in DNA binding and cleavage reactions in vitro.

The crystal structure of SSO0001 reveals bound Mn²⁺ and a 4Fe-4S cluster

Purified SSO001 was crystallized and produced brown-coloured crystals suggesting the presence of intact Fe-S clusters (Fig. 4a). The structure of the seleno-methionine substituted protein was solved at 2.35 Å resolution (Supplementary Table S3) and revealed the core αβα protein domain capped by a small, mostly α-helical sub-domain (Fig. 4b). The core domain contains the V-shaped β-sheet, which is surrounded by α-helices α1, α2, α3, α4, and α5 creating a large elongated opening (12–24 Å) on one side and a smaller round hole (9 Å) on the other side. The small sub-domain is formed by three helices (α3, α8, and α9) and the long loop 177–186. This domain coordinates a cubane-like [4Fe-4S] cluster with the typical Fe-Fe (2.9 Å) and Fe-S (2.1 Å) distances (Fig. 4b, 5d). The [4Fe-4S] cluster is coordinated by the side chains of four Cys residues (Cys32, Cys188, Cys191, Cys197), which are conserved in all Cas4 proteins and AddB-like nucleases (Supplementary Fig. S1). Thus, the SSO0001 structure confirms that Cas4 proteins are members of a class of “iron-staple” nucleases, which also include AddB and Dna2^17,26.

A Dali search²⁷ for the SSO0001 structural homologues identified several exonucleases including the alkaline exonuclease LHK-Exo from Laribacter honkongensis (PDB code 3SZ4; Z-score 11.9, root mean square deviation 3.2 Å), λ exonuclease (PDB 3SLP; Z-score 11.0, r.m.s.d. 3.2 Å), and the C-terminal RecB-like nuclease domain of the B. subtilis AddAB helicase-nuclease complex (PDB 3U44, Z-score 10.6, r.m.s.d. 3.3 Å). The nuclease domain of the E. coli RecE was also highly similar to SSO0001 (3H4R; Z-score 10.5, r.m.s.d. 3.7 Å), whereas the E. coli RecB nuclease domain showed lower similarity (1W36; Z-score 6.9, r.m.s.d. 3.9 Å). All these enzymes are metal-dependent 5' to 3' exonucleases that bind to dsDNA ends and processively digest the 5'-strand to produce long 3'-overhangs^{12,16,17,28,29}.

SSO0001 protomers form decameric toroids

The final model of SSO0001 revealed ten protomers assembled as a pentamer of “dimers” creating a toroidal structure (105 Å × 67 Å) with a large central channel (55 Å diameter) (Fig. 5). The tightly sealed “dimers” are formed through the interactions between the two long helices α5 with the contribution from the residues located on the α1 helix and the β-strands β6 and β7 (Fig. 4c). Analysis of the crystal contacts in the SSO0001 structure using quaternary prediction server PISA also suggested a decameric state arranged as a pentamer of “dimers”. The decameric state of SSO0001 in solution was experimentally confirmed using analytical ultra-centrifugation (243,138 Da or 9.5 protomers; predicted molecular mass for the His-tagged protomer 25.6 kDa) and size-exclusion chromatography (216–268 kDa or 9–10.7 protomers). The SSO0001 toroid is formed by five vertically positioned and slightly twisted “dimers” creating a complex with two pentameric rings with D5 symmetry (Fig. 5). The C-terminal part of each protomer coordinates a [4Fe-4S] cluster and also interacts with two protomers from a neighbour “dimer” suggesting that the intact state of the Fe-S cluster might be important for the maintenance of the SSO0001 toroid. This is supported by a gel-filtration analysis of the oligomeric state of the SSO0001 Cys mutants (Cys32A, Cys188A, Cys191A, and Cys197A), which showed no brown color after purification and existed as a mixture of high molecular weight aggregates (385–400 kDa; 58–73%), pentamers (117–123 kDa; 20–30%), and monomers (27–29kDa; 7–14%) in solution (Fig. S3f).

SSO0001 active site is located in a tunnel

The position of the SSO0001 active site is indicated by the location of a metal ion (identified as Mn²⁺ using an X-ray fluorescence analysis), which is bound at the bottom of the elongated opening (between α5 and β5) close to the residues of the RecB-like sequence motif (Asp99, Glu113, Lys115, and Tyr148) (Fig. 6, Supplementary Fig. S6). In the SSO0001 toroid, all ten active sites are exposed to the interior side of the central channel, where the Fe-S clusters are located and whose surface is dominated by positively charged residues (Fig. 6, Supplementary Fig. S7). The opposite side of the SSO0001 protomers (exposed to the external side of the ring) has a predominantly negatively charged surface with small portals (Supplementary Fig. S7). Inside the SSO0001 protomer, the two openings are connected by the internal tunnel with ~ 10–12 Å diameter accommodating the Mn²⁺ ion (Fig. 6a). Compared to the λ exonuclease trimeric and RecE tetrameric rings, the SSO0001 toroid has a much larger central channel and is composed of two pentameric rings.

The Mn²⁺ ion is located 20 Å away from the [4Fe-4S]-cluster and is coordinated by the side chains of conserved or semi-conserved residues His62 (2.3 Å), Asp99 (2.1 Å), Glu113 (2.1 Å), as well as the main chain carbonyl oxygen of Ile114 (2.1 Å) (Fig. 6b, Supplementary Fig. S6). Similar coordination of the metal ion was found in the λ exonuclease (Asp119, Glu129, and the backbone carbonyl of Leu130), E. coli RecB (His956, Asp1067, Asp1080, and the backbone carbonyl of Tyr1081), and E. coli RecE (His652, Asp748, Asp759, and the backbone carbonyl of Val760)^16,30,31 (Supplementary Fig. S8). In SSO0001, Asp99 and Glu113 are part of the RecB motifs II and III, which also include the closely located Lys115 (motif III), Gln131, and Tyr135 (QxxxY motif) (Fig. 6b, Supplementary Fig. S1). Based on the SSO0001 structure, its RecB motif I residue is Glu78 (7.1 Å from the motif II Asp99), which is located near the exit from the internal channel and might contribute to product release (Fig. 6). The bottom part of the active site (between Mn²⁺ and Fe-S cluster) is occupied by the side chains of semi-conserved Arg31, Arg35, His127, His128, and Trp186 representing the potential binding site for a nucleic acid substrate, whereas the closely located Lys189 and Tyr190 might form a wedge structure involved in DNA strand separation (Fig. 6a).

Site-directed mutagenesis and potential catalytic mechanism of SSO0001

The four SSO0001 mutant proteins with conserved cysteines mutated to Ala (C32A, C188A, C191A, and C197A) produced colorless preparations, whose gel-filtration profiles showed the formation of various protein aggregates (Supplementary Fig. S3f). These mutant proteins were inactive in both binding and cleavage of ssDNA and splayed arm substrate indicating that the Fe-S cluster domain contributes to the stabilization of the SSO0001 active site (Fig. 7). In addition, the Fe-S cluster domains are oriented toward the central channel of the SSO0001 toroid providing a potential “wedge” structure for DNA strand separation (Fig. 5b). Similarly, alanine replacement of the Fe-S cluster coordinating cysteines of AddB eliminated the ability of this protein to bind duplex DNA²⁶. In SSO0001, a strong negative effect of mutation on both substrate binding and cleavage was also observed in the K115A and Y148A proteins, whereas D99A and E113A showed significant binding, but very low or no nuclease activity confirming the importance of Mn²⁺ coordination and RecB motif residues for nuclease activity of SSO0001 (Fig. 7). The absence of nuclease activity in the SSO0001 D99A protein was also reported by Zhang et al.²⁰. The H62A and H128A proteins retained high nuclease activity against ssDNA, but showed reduced ssDNA binding suggesting that these residues contribute to substrate coordination. The mutant proteins H127A, Q131A, and Y135A exhibited both significant DNA binding and nuclease activity against ssDNA (Fig. 7). Interestingly, the mutant proteins E113A and Y148A appear to be affected in both activity and ability to cleave short ssDNA, as they produced only a limited number of long products and no short products (Fig. 7). In the SSO0001 structure (Fig. 6b), the highly conserved Tyr148 is positioned close to the catalytic Lys115 (motif-III) and likely contributes to the coordination of the 5'-DNA strand in the active site. In addition, the R31A, R35A, H127A, H128A, W186A, K189A, and Y190A proteins retained high nuclease activity against ssDNA with the formation of short products, but they were unable to produce short products from the splayed arm substrate confirming our hypothesis that these residues are involved in the unwinding activity of SSO0001 and supporting our assumption that the closely located large opening represents the entry gate into the SSO0001 active site (Fig. 7).

Most characterized nucleases including the RecB-like enzymes have been proposed to use a catalytic mechanism based on the two metal ion centers²². Although the apo (nonliganded) structure of the λ exonuclease showed the presence of one metal ion³⁰, the structure of this enzyme in complex with dsDNA revealed two Mg²⁺ ions (Mg^A and Mg^B) bridged by the side chain of Asp119 on one side and the scissile phosphate of the DNA strand on the other²⁹. Mg^A (which is present in the apo-structure) is bound to Glu129, the main chain carbonyl of Leu130, and two water molecules, one of which is also coordinated by Lys131 and is positioned for in-line attack on the scissile phosphate²⁹ (Supplementary Fig. S8). Mg^B is bound only in the presence of DNA and is proposed to stabilize the transition state through the bidentate coordination to the scissile phosphate. A similar catalytic mechanism can also be proposed for SSO0001. In this model, the catalytic water molecule is coordinated by the Mn²⁺ ion (Supplementary Fig. 6a), Lys115, and phosphate, whereas the second metal ion (bound to Asp99), Gln131, Tyr135, and Tyr148 might interact with the DNA phosphates.

Protein expression, purification, and mutagenesis

Cloning and purification of the 6His-tagged SSO0001, SSO1391, and Pcal_0546, as well as site-directed mutagenesis of SSO0001 were performed as described previously³⁹. The oligomeric state of purified proteins was analyzed using size-exclusion chromatography on a Superdex 200 HiLoad 16/60 column (Amersham Biosciences) and analytical ultracentrifugation as described previously⁴⁰.

Preparation of nucleic acid substrates

The ssDNA and RNA oligonucleotides used in this work (Table S1) were purchased from IDT (USA). The oligonucleotides were [³²P]-labelled at the 3'-end using the calf thymus Terminal Transferase TdT or at the 5'-end using T4 polynucleotide kinase (PNK, BioLabs) and purified as previously described⁴¹. The synthetic dsDNA or splayed arm substrates were prepared by annealing the oligonucleotides shown in Tables S1 and S2. The 5′-phosphorylated and 5′-biotynilated substrates were purchased from IDT (USA).

Enzymatic and DNA binding assays

The reaction mixture for multiple-turnover ssDNase assays with SSO0001 contained 50 mM CAPS (pH 9.5), 100 mM KCl, 10 mM MgCl₂ (or 0.5 mM for other divalent cations), 1 mM DTT (for exonuclease assay), 0.1 µM 3'- or 5'-[³²P]-labelled substrate, or 5 nM ssDNA of the M13mp18 phage, or 20 nM pUC19 dsDNA. The reaction mixture for ssRNase assays contained 50 mM Tris-HCl (pH 7.5), 100 mM KCl, 5 mM MgCl₂, 1 mM DTT, 0.1 µM 3'-[³²P]-labelled RNA, and SSO0001 (as indicated). The reaction mixtures for ssDNase assays with Pcal_0546 and SSO1391 contained 50 mM Tris-HCl (pH 8.0), 5 mM MgCl₂ (or 2 mM of other divalent metal ion), 50 mM KCl (for SSO1391), 1 mM DTT (for exonuclease assays) and 0.1 µM 3′- or 5′-[³²P]-labeled substrate (or 5 nM of M13mp18 ssDNA or 20 nM pUC19 dsDNA). The solutions were incubated at 42–50 °C and processed as previously described⁴¹. The analysis of the ssDNA product ends was carried out using T4 PNK as previously described⁴¹. The reaction for the LC-MS-based analysis of SSO0001 products was performed using optimal reaction conditions with 6 µM SSO0001 and poly-dC₁₀ substrate (10 µg). The LC-MS analysis of the products of poly-dC₁₀ cleavage by SSO0001 was performed using an Exactive orbitrap mass spectrometer with an electrospray ionization source and an Accela 1250 U-HPLC system (all from Thermo Scientific) essentially as previously described⁴². The reaction mixture for DNA-binding assays contained 50 mM Tris-HCl (pH 8.0), 5 mM KCl, 0.5 mM CaCl₂, 1 mM DTT, and SSO0001 (as indicated). The solutions were incubated at 25 °C for 10 min and analyzed by native gel electrophoresis (10% acrylamide). These reaction conditions are optimal for DNA binding, whereas the DNA cleavage was prevented by low incubation temperature, short incubation time, and Ca²⁺ ions.

Crystallization and structure determination of SSO0001

SSO0001 was crystallized at room temperature using the sitting drop vapour diffusion protocol by mixing 1 µl of the selenomethionine substituted protein (10 mg/ml) with 1 µl of the crystallization solution containing 25 % (w/v) PEG 2000 MME, 0.2 M NaCl, 100 mM Na-acetate (pH 5.6), and 0.3 M NDSB 201. The crystals were stabilized with the crystallization buffer containing 3 M NaCl followed by cryoprotection in Paratone-N prior to flash-freezing in liquid nitrogen. Data collections were carried out at the beamlines 19-ID and 19-BM of the Structural Biology Center, Advanced Photon Source, Argonne National Laboratory⁴³. Prior to data collection, an X-ray fluorescence spectrum was recorded, which identified the presence of Se, Fe, and Mn in protein crystals. Data set was collected from a single crystal to 2.35 Å at the wavelength of 0.9794-Å and processed using the program HKL3000⁴⁴ (Table S3). The structure of SSO0001 was determined by the Se-methionine SAD phasing, density modification, and initial model building as implemented in the Auto-Rickshaw package⁴⁵. The initial models (~60% complete) were further built manually using the program COOT⁴⁶ and refined with the PHENIX⁴⁷ and REFMAC⁴⁸ programs. The presence and identity of metals were verified by collecting data sets (from the same crystal) at the iron X-ray K absorption edge and just above the Mn absorption edge and calculating anomalous Fourier difference maps (Supplementary Fig. S6). Analysis and validation of structures were performed using MOLPROBITY⁴⁹ and COOT⁴⁶ validation tools. The final model was refined to R_work/R_free = 0.205/0.249, and it shows good geometry with no outliers in the Ramachandran plot. Data collection and refinement statistics are summarized in Supplementary Table S3. The atomic coordinates have been deposited in the Protein Data Bank, with accession code 4IC1.