To understand the molecular evolution of inteins, we are interested in elucidating three-dimensional structures of various inteins with a presumable HEN domain. The first intein was identified as an intervening sequence within the yeast vacuolar membrane ATPase (VMA), subunit A (Hirata et al., 1990). The majority of inteins among eukaryotes reside at the highly conserved insertion site within the Vacuolar ATPase (VMA-a insertion site) (Swithers et al., 2009). The extensively investigated VMA intein from Saccharomyces cerevisiae (SceVMA) defines a proto-typical intein possessing homing endonuclease activity, also called PI-SceI, as a rare cutting DNA endonuclease (Grindl et al., 1998). Whereas yeast inteins are inserted at the highly conserved insertion site (VMA-a site), archaeal inteins commonly target a region approximately 20-residue downstream of the VMA-a insertion site (VMA-b insertion site), located at the P-loop motif of ATPases (Swithers et al., 2009). The VMA intein from P. horikoshii (PhoVMA) consists of 376 amino acids, which is considerably smaller than canonical HEN-associated inteins, e.g., SceVMA consisting of 454 residues but more similar to the size of the TFIIB intein from Methanococcus jannaschii (MjaTFIIB, 335 residues). The structure of the MjaTFIIB intein could previously not be determined together with the HEN domain by protein crystallography (Iwaï et al., 2017). Inteins share conserved amino acid sequence stretches designated as Blocks A-G (Pietrokovski, 1994; Perler et al., 1997) (Figure 2A). Blocks C, D, E, and H denote the HEN domain, out of which Blocks C and E represent the eponymous conserved LAGLIDADG helices bearing the acidic catalytic residues (Perler et al., 1997). The sequence alignment of the archaeal inteins also suggests that PhoVMA intein lacks homing endonuclease activity due to the absence of the active site residues in Blocks C and E (Figure 2B). We were successful in producing the PhoVMA intein and obtaining diffracting crystals. We solved the crystal structure of the PhoVMA intein at the 2.5 Å-resolution (Figure 2C; Supplementary Table S1). The crystal structure of PhoVMA intein revealed the typical HINT domain of thermophilic inteins, which contains a β-strand insertion and the HEN domain structure (Figure 2C) (Aranko et al., 2014). As expected from the primary structure, the PhoVMA intein lacks the presumed HEN active site residues in both usually conserved LAGLIDADG helices (Blocks C and E). It shows a partial truncation in Block E, hypothetically indicating progressive HEN degeneration (Figures 2A–C).

The length of inteins considerably varies from 123 to >1,000 residues due to various insertions such as HENs and sequence deletions (Green et al., 2018). Large intein sequences generally indicate the presence of an active or inactive nested HEN. Therefore, we were interested in elucidating the structures of other VMA inteins inserted at the same VMA-b site to reveal a possible structural basis directing inteins of diverse sizes to the same target insertion site within host genomes. Among VMA inteins inserted at the VMA-b insertion site, we could obtain crystals of the VMA intein from Thermococcus litoralis (Tli). The TliVMA intein comprises 429 residues and is larger than the PhoVMA intein (376 residues) but similar to the size of PI-SceI (454 residues). To prevent self-cleavages during protein production, we expressed both inteins in E. coli with alanine substitutions of the catalytic cysteines 1 (Cys1). We used the N-terminal small ubiquitin-like modifier (SUMO) fusion to facilitate protein purification of PhoVMA intein (Guerrero et al., 2015). However, TliVMA intein required an N-terminal MBP fusion in addition to SUMO (H₆-MBP-SUMO-TliVMA intein) for successful soluble expression (Guerrero et al., 2015). Unlike PhoVMA intein with the presumably degenerated HEN domain, TliVMA intein also required a high salt buffer composition, compensating for the lack of nucleic acids to maintain solubility after proteolytic removal of the fusion tag.

We solved the structure of TliVMA intein at the 1.6 Å-resolution (Figure 2D, Supplementary Table S1). The crystal structures of the TliVMA intein revealed a very similar overall structure as found in the PhoVMA intein, including the three-domain architecture known from the three other reported HEN-containing inteins (Figures 2C–E) (Duan et al., 1997; Ichiyanagi et al., 2000; Moure et al., 2002; Matsumura et al., 2006). The Hedgehog/Intein domains (HINT, gray) are composed of the N- and C-terminal fragments (Int_N and Int_C) with the HEN domains (yellow) inserted into the common intein split-site located between the two pseudo-two-fold symmetrical units forming a horseshoe-like fold common to all HINT domains (Eryilmaz et al., 2014; Iwaï et al., 2017). The HINT domain of the TliVMA intein also contains the β-strand extension found among thermophilic inteins (Figure 2C) (Aranko et al., 2014; Beyer et al., 2019; Hiltunen et al., 2021).

Not surprisingly, the HINT domains of PhoVMA andTliVMA inteins show a virtually identical three-dimensional structure with a 86% sequence identity (Figure 2; Supplementary Table S4). The HINT domains connect to the first of the two LAGLIDADG helices of the HEN domains via unstructured loops of 28–32 residues, located distant from the DNA-binding interfaces. We have identified “accessory domains” (ACD, shown in blue) residing between the HEN domains and the C-terminal part of the HINT domains, where we observed the most notable differences. The striking contrast between both inteins is the divergence of their ACDs, showing the least structural homology between the two molecules (33% pairwise sequence identity, Figures 2C,D; Supplementary Figure S3). The 53-residue difference in the lengths between the PhoVMA andTliVMA inteins can be mainly attributed to the difference in the ACDs. Even though ACDs at the intersections between HINT and HEN domain were identified in other reported HEN-associated inteins, their biological functions remain unclear (Figure 2E) (Duan et al., 1997; Ichiyanagi et al., 2000; Moure et al., 2002; Matsumura et al., 2006).

As for the nested HEN domains, the deletion in Block E in the PhoVMA intein (Figure 2B) causes a truncation of the second LAGLIDADG helix by one turn, thereby removing one of the catalytic aspartate residues (Figures 2B–D). Another obvious consequence of the degeneration in the PhoVMA intein appeared in the structure of the DNA-binding interfaces of the HEN mediated by two stretches of β-sheets, each originating from one copy of the two-fold pseudo symmetric LAGLIDADG elements (Supplementary Figures S1A,B) (Moure et al., 2002). The electrostatic surface potential of the HEN domains is very different between the two VMA inteins, which is in line with their binding to DNA fragments (see below).

Based on the three-dimensional structures, we deleted the HEN domain (residues 123-335 for PhoVMA intein and 123-388 for TliVMA intein) and connected with an “NG” sequence linker, resulting in 165-residue cis-splicing PhoVMA^ΔHEN and TliVMA^ΔHEN inteins. We modeled the structure of the two deletion variants with the RoseTTAFold software using the deep-learning algorithm (Supplementary Figures S1C,D) (Baek et al., 2021). Both structures appear identical with high confidence scores and an r.m.s.d. of 0.7 Å for 165 Cα atoms.

The PhoVMA^ΔHEN intein still retained the protein splicing activity, indicating that the HINT domain of PhoVMA intein is functionally independent of the nested HEN domain without having developed a mutualism (Supplementary Figure S1C) (Iwaï et al., 2017). However, the protein splicing activity of the TliVMA^ΔHEN intein largely diminished, presumably because of the mutualism developed between the HINT and HEN domains (Supplementary Figure S1D) (Iwaï et al., 2017). Even though the two three-dimensional structures are predicted to be almost identical to the original HINT domain (Supplementary Figures S1C,D), the HEN domain of TliVMA intein likely contributes to the protein splicing activity, as it has also been suggested for MvuTFIIB intein (Iwaï et al., 2017).

The primary structures and the three-dimensional crystal structures of the VMA inteins suggest that the HEN activity of PhoVMA intein has most likely degenerated during evolution and is inactive due to the lack of active site aspartate residues. However, TliVMA intein probably carries a catalytically active HEN domain capable of binding to DNA and introducing DNA double-strand breaks (Figures 2A,B).

To experimentally validate these assumptions, we performed in vitro DNA-binding and cleavage studies. First, we generated DNA substrates containing the theoretical homing sites of the inteins, that is, the coding DNA sequence of the Tli and Pho vma genes without the intein coding region (Figure 3A). These reconstructed intein-less DNA fragments should resemble the allelic situation before invasion by the inteins (Figure 3A). We generated 750-bp linear double-strand DNA fragments asymmetrically harboring the reconstituted homing site by amplifying their respective sequences from the genomic DNA by PCR and tested the cleavage of the DNA fragments by the inteins. Indeed, we observed that TliVMA intein cleaved its reconstituted homing site accompanied by a strong DNA binding affinity as indicated by an electrophoretic mobility shift (EMSA) of the substrate- and product-DNA fragments (Figure 3B). In contrast, PhoVMA intein was neither able to process its theoretical homing site, nor did it show any detectable DNA binding affinity (Figure 3C). Thus, the DNA substrates with the reconstituted homing sites validated our assumptions derived from the structures of Tli and Pho VMA inteins.

We attributed the catalytic inactivity of the HEN in PhoVMA intein to the loss of presumed active-site residues. The differences in the electrostatic surface potential distributions of the HEN domains between Tli and Pho VMA inteins further might support the weaker DNA-binding affinity of PhoVMA compared with TliVMA intein (Figures 3B,C). Because the architecture of the HEN domain in PhoVMA intein was retained intact despite degenerative mutations and deletion, we wondered whether the nuclease activity of PhoVMA intein could be restored by protein engineering to reverse the evolutional process. To validate our hypothesis, we engineered the inactive PhoVMA intein by grafting the active sites in the LAGLIDADG helices from the sequences of the TliVMA intein (Figures 2B, 3D). Indeed, the engineered PhoVMA intein (PhoVMA_Act intein) with the restored catalytic residues cleaved the DNA substrate containing the reconstituted homing site, albeit less efficiently not attaining the complete substrate digestion as observed with TliVMA intein (TliVMA intein, Figure 3B; PhoVMA_Act intein, Figure 3D).

We designed and generated the DNA substrates for the DNA cleavage assay from Tli and Pho genomic DNA using PCR. Removing the intein coding sequences from the vma genes restored the theoretical homing site within a linear DNA of 750 bp containing the 250- and 500-bp fragments of the genomic sequences upstream and downstream of the reconstituted homing site, respectively (Figure 3A). The DNA cleavage at the homing site by the VMA inteins should produce 250- and 500-bp products. While the engineered PhoVMA_Act intein produced the expected two fragments (Figure 3D), TliVMA intein exhibited an unexpected pattern of the products (Figure 3B). The disappearance of the DNA fragments at higher concentrations of TliVMA intein without SDS-treated denaturation is presumably due to the strong affinity to the DNA molecule (“end-holding”). Interestingly, besides the expected 500-bp fragment, a product of ∼200 bp and a third one shorter than 75 bp appeared with TliVMA intein. The analysis of the cleavage products by DNA sequencing revealed that PhoVMA_Act intein cleaved precisely at the expected homing site (Supplementary Figure S2A).

In contrast, TliVMA intein cleaved at two different sites. One site was indeed at the theoretical homing site (HS) with the central four base pairs of the sequence 5′-AAAA-3′, while the other alternative site (AS) is located 52 bp upstream of the reconstituted homing site and contains the central sequence 5′-TCTT-3′ (Supplementary Figure S2B). We assume that recognition and cleavage of the AS by TliVMA intein occur on the opposite strand of the HS. The sequence on the opposite strand corresponds to the sequence of 5′-AAGA-3′, reminiscent of the reconstituted homing site of PhoVMA_Act intein (Figures 3E,F; Supplementary Figure S2C) and bearing a single substitution to the central four base pairs of the homing site of TliVMA intein. Indeed, the alignment of the DNA sequence against the reverse strand of the alternative site revealed a striking 63% identity encompassing a 30 bp region surrounding the two cleavage sites (Figure 3F). Overall, the DNA substrates reconstituted from Tli and Pho genomic DNA have sufficient similarity to assume that both contain the AS next to the HS (Supplementary Figure S2C). However, PhoVMA_Act intein could exclusively process the homing site (HS), leaving the AS unaffected (Figure 3D). We could conclude that the activated PhoVMA_Act intein is more specific toward recognizing the reconstituted homing site (HS) despite its lower affinity.

The lengthy DNA sequences recognized by homing endonucleases (HENs) attracted protein engineering of HENs for genomic application because the high specificity of HENs could facilitate various in vitro and in vivo applications (Stoddard, 2011). However, the number of HENs that recognize different DNA sequences which could be used for broad applications is small. Although dozens of intein structures have been deposited to the protein data bank (PDB), only three of those contain a nested HEN domain. Moreover, exclusively the intein structure of PI-SceI from Saccharomyces cerevisiae was elucidated as the DNA/intein complex (Duan et al., 1997; Moure et al., 2002). The limited structural information of HEN-associated inteins hinders our understanding of inteins as site-specific DNA endonucleases, impeding further development of HEN-associated inteins by protein engineering as genetic engineering tools. Other reported HEN-associated intein structures are archaeal inteins from Thermococcus kodakaraensis (PI-TkoII) (Matsumura et al., 2006) and Pyrococcus furiosus (PI-PfuI) (Ichiyanagi et al., 2000). Just like the VMA inteins from T. litoralis and P. horikoshii, these inteins have an accessory domain (ACD) in addition to HINT and HEN domains (Figure 2D). Furthermore, in the case of PI-TkoII, an additional domain, termed domain IV, was reported (Matsumura et al., 2006).

It is believed that ACDs in HEN-associated inteins might generally contribute to interactions with DNA. For PI-SceI, where the ACD is referred to as DNA recognition region (DRR), this role has been demonstrated, although the location of the ACD in PI-SceI is different from other reported HEN-associated intein structures (Figures 2C–E; Supplementary Figure S3). The ACD (DDR) can be seen as an insertion into the HINT domain rather than a connection of HEN and HINT domains (Moure et al., 2002) (Figure 2E; Supplementary Figure S3). However, HENs also exist free-standing without being embedded in inteins or introns. They are known to be among the most sequence-specific endonucleases due to their relatively long sequence recognition motif (Chevalier and Stoddard, 2001). Some of such HENs do not contain ACDs. Thus, it remains elusive why some HEN-associate inteins require ACDs and cannot define sufficient DNA sequence specificity with their intrinsic DNA recognition capability.

To investigate the structural and functional roles of ACDs in HEN-associated inteins, we decided to delete the ACD region from the TliVMA intein based on the three-dimensional structure. We could also validate our deletion design by determining the crystal structure of the deletion variant, termed TliVMA^ΔACD intein (Figure 4A; Supplementary Table S1). The crystal structure of TliVMA^ΔACD intein confirmed that the deletion of the ACD did not influence the HINT and HEN domain folds, nor their relative orientation toward each other (Figure 4A). In the DNA cleavage and binding assays, TliVMA^ΔACD intein similarly cleaved the same substrate as the wild-type TliVMA intein did, albeit with reduced DNA-binding (Figures 3B, 4B). To our surprise, the deletion of the ACD from the TliVMA intein changed the cleavage profile. The substrate cleavage profile by TliVMA^ΔACD intein resembled that of the PhoVMA_Act intein, producing two main products as opposed to three products generated by TliVMA intein (Figure 4C). The DNA sequencing chromatogram of the smaller cleavage product generated by TliVMA^ΔACD supported that TliVMA^ΔACD intein did not cleave the alternative site as observed for the wild-type TliVMA intein, similar to the digestion pattern of the PhoVMA_Act intein (Figure 4F). Moreover, we found that TliVMA intein cleaved the reconstituted DNA substrate from the Pyrococcus horikoshii (Pho) genome at the alternative cleavage site (Figure 4E).

Similarly, PhoVMA_Act intein was able to digest the homing site within the Tli genome, albeit less efficiently as its cognate homing site (Figure 4C; Supplementary Figure S4). This cross-activity between TliVMA and PhoVMA_Act inteins is presumably due to the close homology between the two substrate sequences created from Pho and Tli genome (Supplementary Figures S2C, S6A). The deletion of ACDs suggests that ACDs could play critical roles in increasing the cleavage specificity of HEN-associated inteins as well as making them more promiscuous by adding the capability to recognize an alternative cleavage site.

Next, we were interested in how the ACD in TliVMA intein influences the DNA recognition specificity. We speculated two possible scenarios: direct recognition of the alternative site sequence mediated by the ACD or indirect recognition via a cooperative binding effect. The binding of the intein to the homing site could guide the recognition of the alternative site separated by only 52 bp from the homing site by cooperative domain interaction with a second intein molecule involving the ACD. The DNA substrate containing only the alternative cleavage (AS) site (intein coding sequence remained inserted into the homing site) indicated that only TliVMA intein bearing the ACD was capable of digesting the DNA substrate, whereas TliVMA^ΔACD intein was not (Figure 4D). These results revealed that the TliVMA intein cleaved the alternative site (AS) independent of the homing site (HS) but depended on the presence of the ACD domain. We also performed DNA-binding tests using the isolated ACD domain of the TliVMA intein and revealed that the ACD seemingly does not contribute to the overall DNA affinity (Supplementary Figure S5).

To further validate the role of the ACD in TliVMA intein as a modulator of the DNA recognition responsible for the alternative site, we tested whether grafting of the ACD in PhoVMA_Act intein from TliVMA intein would confer alternative site recognition. We thus created PhoVMA_Act-ACD(Tli) intein having the grafted ACD from TliVMA intein (ACD (Tli)). Indeed, PhoVMA_Act-ACD(Tli) could process both homing and alternative sites of the DNA substrate generated from the Pho genome, reminiscent of the cleavage profile produced by the TliVMA intein (Figure 4E). Furthermore, the swapping of the ACD rendered PhoVMA_Act-ACD(Tli) intein more efficient in processing the DNA substrate without altering the apparent overall DNA affinity (Figure 3D; Supplementary Figure S6B). The weaker activity of PhoVMA_Act-ACD(Tli) intein also allowed resolving a preferentiality of the homing site over the alternative site as the latter required a higher enzyme concentration (Supplementary Figure S6B).

The crystal structures of Tli and PhoVMA inteins, inserted at the same VMA-b insertion site of their host proteins, revealed a notable structural difference in their ACDs, largely deviating from each other (Figures 2C, 5A,B). The structural difference prompted us to investigate the functional role of ACDs. Our results demonstrated that the ACD in TliVMA intein induced a second cleavage site in addition to the theoretical homing site (Figure 4C). Interestingly, engineering the reactivated PhoVMA intein by grafting the ACD from TliVMA intein triggered cleavage at the alternative site (AS) adjacent to the homing site (HS), suggesting that the ACD is responsible for the cleavage at the AS (Figure 4E). The ACD of TliVMA intein strongly resembles the helix-turn-helix motif common for many DNA binding proteins, such as transcriptional regulator proteins (Figures 5A,B; Supplementary Table S5) (Anderson et al., 1981; Brennan and Matthews, 1989). The homology to DNA binding proteins suggests that ACDs mediate contacts with the DNA substrate.

Next, we wanted to test whether TliVMA intein is a promiscuous endonuclease and could cut unrelated substrates. We, therefore, tested digestion of λ-phage DNA by incubating overnight with TliVMA intein or TliVMA^ΔACD intein lacking the ACD (Figures 5C,D). To our surprise, we identified multiple cleavages in line with our observations using the model DNA substrates generated from T. litoralis genomic DNA. Furthermore, similar to our model DNA substrate, deletion of the ACD in TliVMA intein indeed reduced cleavage of λ-phage DNA, supporting our hypothesis that the ACD renders the intein endonuclease more promiscuous. In contrast, the activated PhoVMA_Act intein with the endogenous ACD and the activated PhoVMA_Act-ACD(Tli) with the grafted ACD from TliVMA intein (ACD (Tli)) did not produce any detectable λ-phage DNA cleavage, presumably due to the much lower affinity to the DNA substrate (Figures 3D, 4E, 5D).

We wondered how ACDs from other homologous inteins and an unrelated DNA-binding domain of the bacteriophage 434 repressor (434R) would affect cleavages of λ-phage DNA by TliVMA intein (Aggarwal et al., 1988). We engineered TliVMA intein by ACD-swapping the 434R domain and found that the engineered TliVMA intein decreased λ-phage DNA processing. However, we could still detect some extent of cleavages (Figure 5D). Replacing the ACD inTliVMA intein with an ACD from the more related inteins like VMA inteins from P. furiosus (PfuVMA) and P. abyssi (PabVMA) had a milder effect on the cleavage of λ-phage DNA (Figure 5D). Whereas the TliVMA intein variant carrying the ACD from PfuVMA intein (TliVMA_ACD(Pfu)) produced a restriction pattern very similar to the wild-type TliVMA intein, the variant with the ACD from Pab (TliVMA_ACD(Pab)) exhibited a less similar pattern (Figure 5D). The difference in the digestion profiles might arise from the fact that the ACD from PfuVMA intein has eight mutations, while the ACD from PabVMA intein contains 12-residue changes relative to the 55-residue region of the ACD in the TliVMA intein. Interestingly, replacing the ACD in the TliVMA intein with an unrelated DNA binding domain of phage 434R nearly abolished the cleavage of λ-phage DNA by the TliVMA intein (TliVMA₄₃₄), indicating that grafting of 434R might disrupt the functional structure completely or create steric hindrances due to the poor protein engineering.

All plasmids, oligonucleotides, and synthetic DNA substrate molecules used in this study are described in Supplementary Table S2. All recombinant proteins were produced in the E. coli strain T7 Express (New England Biolabs, USA). Expression details are given in Supplementary Table S3. All inteins carry a substitution of the catalytic cysteine 1 to alanine (C1A) to enable purification as fusion proteins except for those used in protein splicing tests. Residue numbering starts with 1 for this catalytic intein amino acid position and proceeds toward the C-terminus. Intein preceding residues are given negative indices.

Expression cultures were harvested by centrifugation at 4,700g for 10 min, 4°C. Pelleted cells from 1 or 2 L cultures were lysed in buffer A (50 mM sodium phosphate, pH 8.0, 300 mM NaCl) using continuous passaging through an EmulsiFlex-C3 homogenizer (Avestin, Canada) at 15,000 psi, 4°C for 10 min. Lysates were cleared by centrifugation at 38,000g for 60 min, 4°C. Proteins were purified in two steps using 5 ml HisTrap HP columns (GE Healthcare Life Sciences, USA) as previously described, including the removal of the hexahistidine tag and MBP and SUMO fusion domains (Guerrero et al., 2015).

After proteolytic removal of the fusion domains using Ubiquitin-like-specific protease 1 (UlpI), thermostable PhoVMA_Act intein (pHBRSF067) and TliVMA^ΔACD intein (pHBRSF075) were heat fractionated at 80°C for 20 min before application to the Ni²⁺-NTA HisTrap column. After purification, all proteins were dialyzed overnight at 8°C against the following buffers: TliVMA intein (pHBRSF063) was first dialyzed against 10 mM Tris-HCl pH 8.0, resulting in precipitation of the target protein. Precipitated protein was resolubilized by the addition of 500 mM KCl followed by dialysis against 10 mM Tris-HCl pH 8.0, 400 mM KCl. TliVMA^ΔACD intein was treated like TliVMA intein, but 800 mM KCl was used for resolubilization, and the last step of dialysis was omitted. PhoVMA intein (pCARSF54) was dialyzed against deionized water. PhoVMA_Act and PhoVMA_Act-ACD(Tli) intein (pHBRSF079) were dialyzed against 10 mM Tris-HCl pH 8.0, 300 mM KCl. The isolated ACD of Tli (pHBRSF082) was dialyzed against 20 mM Tris-HCl pH 8.0, 300 mM KCl. The TliVMA intein variants containing exchanged ACDs (pHBRSF083, PfuVMA ACD; pHBRSF084, PabVMA ACD; pHBRSF161, 434-repressor domain) were purified like the TliVMA intein (pHBRSF063). After proteolytic removal of the H₆-MBP-SUMO purification tag, sample and purification buffers were supplemented with 350 and 200 mM KCl, respectively. After dialysis against salt-free buffers, the three proteins were resolubilized by addition of 475, 600, and 700 mM KCl, respectively.

Proteins were subsequently concentrated using Macrosep^® Advance Centrifugal Devices 10K MWCO (PALL). For enzymatic assays, proteins were diluted to 50 µM with 50 mM Tris-HCl pH 8.0, 300 mM KCl, 10% (v/v) glycerol, 1 mM dithiothreitol (DTT) and stored at −80°C for further use.

All diffracting crystals were obtained at room temperature using the sitting drop vapour diffusion method by mixing 100 nL concentrated PhoVMA, TliVMA, and TliVMA^ΔACD inteins with 100 nL of the following mother liquors: PhoVMA, (100 mM HEPES pH 8.0, 5 mM cadmium chloride, 5 mM magnesium chloride, 5 mM nickel (II) chloride, 10% (w/v) polyethylene glycol (PEG) 3,350); TliVMA, (100 mM magnesium formate, 15% (w/v) PEG 3350); TliVMA^ΔACD, (100 mM HEPES pH 7.5, 70% (v/v) 2-methyl-2,4-pentanediol (MPD)). Data were collected at beamline i04 (PhoVMA and TliVMA^ΔACD inteins) at Diamond Light Source (Didcot, United Kingdom) and beamline ID30A-1 (MASSIF-1, TliVMA intein) at ESRF (Grenoble, France) (Bowler et al., 2015). Data were processed using XDS (Kabsch, 2010). Structures were solved by molecular replacement (MR) starting from a PhoVMA intein search model generated using SWISS-MODEL (Arnold et al., 2006) with the intein homing endonuclease II of Thermococcus kodakarensis DNA polymerase (PDB: 2cw7) as a template. Initial models were obtained using the MR pipeline of Auto-Rickshaw (Panjikar et al., 2005) and ARP/wARP (Langer et al., 2008). Model-building and refinement were performed with COOT (Emsley et al., 2010) and PHENIX (Adams et al., 2002). The obtained structures were validated with Molprobity (Chen et al., 2010). Please refer to the Supporting Materials and Methods for a detailed description. Figures presenting three-dimensional coordinates were generated using PyMOL (The PyMOL Molecular Graphics System, Version 2.2.0, Schrödinger, LLC.). Electrostatic surface distributions were calculated and visualized using UCSF Chimera 1.13.1 (Pettersen et al., 2004) with ABS 1.3 (Jurrus et al., 2018) after model preparation with PDB2PQR 2.2.1 (Dolinsky et al., 2004) using a PARSE force field. Sequence alignments were performed with Clustal Omega 1.2.4. (Sievers et al., 2011).

If not indicated otherwise, electrophoretic mobility shift and enzymatic cleavage assays were performed in a total volume of 10 µL in 10 mM Tris-HCl pH 8.0, 100 mM KCl, 10 mM MgCl₂, 1 mM DTT using 0.05 µM (750 bp), or 0.01 µM (2037 bp) linear dsDNA substrates containing the desired intein homing site (Supplementary Table S2). Typically, reactions contained 0.5 µM of TliVMA- or 8 µM of PhoVMA-derived inteins and were incubated at 80°C for 2 h. Where indicated, reactions were stopped by the addition of 1 µL endonuclease stop solution (5% (w/v) SDS, 250 mM EDTA, 100 mM Tris-HCl pH 7.5). Mobility shifts and cleavage products were visualized on 1.2% agarose gels.

For the determination of HEN cleavage sites, 1.5 µg of substrate DNA were digested overnight using the respective endonuclease with the above-described buffers, temperature, and concentrations. For the TliVMA intein, a stop solution was used to dissociate the HEN from the restriction products. Products were gel-purified and sequenced via Eurofins Genomics GmbH using the exterior oligonucleotides as used for the generation of the DNA substrates (Supplementary Table S2). For the digestion of λ-phage DNA, 1 µg substrate was incubated overnight with the indicated intein variants as described above.

Protein cis-splicing of intein variants was tested by expressing the indicated intein variants flanked by two B1 domains of the IgG-binding protein G in 5 ml cultures of E. coli strain T7 Express (New England Biolabs, USA) and purified using immobilized metal affinity chromatography as described elsewhere (Beyer et al., 2020). The used plasmids are listed in Supplementary Table S2. The experiments were performed at 30–37°C and the expression period lasted 3–4 h. Protein splicing was analyzed by SDS-PAGE using 16.5% gels and Coomassie Blue staining.