GeoCas9 is a 1087 amino acid polypeptide, thus we employed a ‘divide and concur’ approach for NMR studies, which we previously showed to be useful for quantifying allosteric structure and motion in SpyCas9 (Skeens et al., 2024; East et al., 2020; Nierzwicki et al., 2021; Nierzwicki et al., 2022). The GeoRec lobe is comprised of subdomains GeoRec1 and GeoRec2, which likely work together to recognize nucleic acids. We engineered constructs of the GeoRec1 (136 residues, 16 kDa) and GeoRec2 (212 residues, 25 kDa) subdomains and solved the X-ray crystal structure of GeoRec2 at 1.49 Å, which aligns remarkably well with the structure of the GeoRec2 domain within the AlphaFold model (RMSD 1.03 Å) and new cryo-EM structure of GeoCas9 (RMSD 1.10 Å, Figure 1A). We were neither able to crystallize GeoRec1 nor full-length GeoCas9 in the apo state, but our GeoRec2 crystal structure represents the structure of the subdomain within the full-length GeoCas9 protein quite well. Our previous studies of GeoHNH also show identical superpositions of X-ray crystal structures with full-length Cas complexes (Belato et al., 2022a). In addition to the individual subdomains, we also generated an NMR construct of the intact GeoRec (370 residues, 43 kDa).

Despite only 22% sequence identity, the structure of SpyRec3 and GeoRec2 are highly similar (RMSD 2.00 Å, Figure 1—figure supplement 1). The structure of GeoRec1, in contrast, does not align perfectly with SpyRec1, instead, it partially aligns with both SpyRec1 and SpyRec2 (Figure 1—figure supplement 1). Thus, the nearly identical SpyRec3 and GeoRec2 architectures and their intrinsic dynamics may be a common thread among Type II Cas9s of different size and PAM preference. To capture atomic-level signatures of GeoRec, we obtained well-resolved ¹H-¹⁵N NMR fingerprint spectra for all three protein constructs and assigned the amide backbones (Figure 1—figure supplement 2). ¹H-¹⁵N amide and ¹H-¹³CH₃ Ile, Leu, and Val (ILV)-methyl NMR spectra (Figure 1B, Figure 1—figure supplement 3) of GeoRec overlay very well with those of its individual subdomains, suggesting that the linkage of subdomains within the full-length GeoRec polypeptide does not alter their individual folds. Consistent with this observation, circular dichroism (CD) thermal unfolding profiles of GeoRec1 (T_m ~34 °C) and GeoRec2 (T_m = 61.50 °C) are distinct and occur as separate events in the unfolding profile of GeoRec (Figure 1—figure supplement 4). The dumbbell shape of GeoRec, with its two globular subdomains connected by a short flexible linker, is a likely contributor to these biophysical properties.

To understand how the structure and gRNA interactions of GeoCas9 can be modulated at the level of GeoRec, we engineered two charge-altering point mutants in the GeoRec2 subdomain, K267E and R332A. Based on the AlphaFold2 model of GeoCas9, both of these residues are <5 Å from the bound RNA:DNA hybrid and were predicted to interface with the nucleic acids directly (Figure 2A/B). A new experimental cryo-EM structure of GeoCas9 confirmed the interaction between K267 and the gRNA, but does not report a<5 Å interaction of R332 with the gRNA (Shen et al., 2024). The rationale for our designed mutations was also that removal of positive charge would weaken the interactions between GeoCas9 and the gRNA, affecting K_d via the electrostatics or dynamics of the GeoRec lobe. Studies of SpyCas9 revealed that interaction of SpyRec3 (analogous to GeoRec2) with its RNA:DNA hybrid triggers conformational rearrangements that allow the catalytic HNH domain to sample its active conformation (Dagdas et al., 2017). Thus, SpyRec3 acts as an allosteric effector that recognizes the RNA:DNA hybrid to activate HNH. Mismatches (i.e. off-target DNA sequences) in the target DNA generally prevent SpyRec3 from undergoing the full extent of its required conformational rearrangements, leaving HNH in a ‘proofreading’ state with its catalytic residues too far from the DNA cleavage site. Off-target DNA cleavage by Cas9 remains an area of intense study and substantial effort from various groups has gone into mitigating such effects (Chen et al., 2017; Slaymaker et al., 2016; Ricci et al., 2019; Eggers et al., 2023; Doudna, 2020). Indeed, many high-specificity SpyCas9 variants contain mutations within SpyRec3 that increase the threshold for its conformational activation, reducing the propensity for HNH to sample its active state in the presence of off-target DNA sequences (Dagdas et al., 2017; Chen et al., 2017; Slaymaker et al., 2016). Studies of flexibility within Rec itself, as well as its gRNA interactions in the presence of mutations, are therefore essential to connecting biophysical properties to function and specificity in related Cas9s.

The K267E GeoRec2 variant is sequentially and structurally similar to a specificity enhancing site in SpyCas9 (K526E), within the evoCas9 system (Casini et al., 2018). The SpyCas9 K526E mutation substantially reduced off-target activity alone, but was even more effective in conjunction with three other single-point mutations in SpyRec3 (Casini et al., 2018). The R332A GeoRec2 variant also resembles one mutation within a high-specificity SpyCas9 variant, an early iteration of HiFi SpyCas9 called HiFi Cas9-R691A (Vakulskas et al., 2018). We assessed mutation-induced changes to local structure in GeoRec via NMR chemical shift perturbations in ¹H-¹⁵N HSQC backbone amide spectra. Consistent with experiments using GeoRec2 alone, chemical shift perturbations and line broadening are highly localized to the mutation sites. (Figure 2C–F). Perturbation profiles of the GeoRec2 subdomain and intact GeoRec also implicate the same residues as sensitive to the mutations (Figure 2—figure supplement 1).

CD spectroscopy revealed that wild-type (WT), K267E, and R332A GeoRec2 maintained similar alpha-helical secondary structure, although the thermostability of both variants was slightly reduced from that of WT GeoRec2 (Figure 2—figure supplement 2). The T_m of WT GeoRec2 is ~62 °C, consistent with the T_m of the full-length GeoCas9, while that of K267E GeoRec2 was decreased to ~55 °C. Though the R332A GeoRec2 T_m remains ~62 °C, this variant underwent a smaller unfolding event near 40 °C before completely unfolding. These data suggest that despite small structural perturbations, both mutations are destabilizing to GeoRec2, which led us to expect a change in NMR-detectable protein dynamics.

Due to the high molecular weight of the intact GeoRec lobe, decays in NMR signal associated with spin relaxation experiments were significant and hampered data quality. Thus, we focused on quantifying the molecular motions of the GeoRec2 subdomain, where the K267E and R332A mutations reside, and the chemical shift perturbations are most apparent. To obtain high-quality per-residue information representative of GeoRec, we measured longitudinal (R₁) and transverse (R₂) relaxation rates and heteronuclear ¹H-[¹⁵N] NOEs (Figure 3—figure supplement 1), then used these data in a Model-free analysis of per-residue order parameters (S²). Previous measurements of S² across the adjacent GeoHNH nuclease revealed substantial ps-ns timescale flexibility, (Belato et al., 2022a) leading us to wonder whether a similar observation would be made for GeoRec2, which abuts GeoHNH. Such a finding could suggest that HNH-Rec2 crosstalk in GeoCas9 is driven primarily by rapid bond vector fluctuations. However, unlike GeoHNH, S² values for GeoRec2 are globally elevated, suggesting that the ps-ns motions of this subdomain arise primarily from global tumbling of the protein in solution. We therefore carried out Carr-Purcell-Meiboom-Gill (CPMG) relaxation dispersion NMR experiments to assess the flexibility of GeoRec2 on slower timescales, which has been linked to chemical information transfer in the well-studied SpyCas9 (Palermo et al., 2018; Skeens et al., 2024; East et al., 2020; Nierzwicki et al., 2021). Evidence of μs-ms motions (i.e. curved relaxation dispersion profiles) is observed in 17 residues within the GeoRec2 core, spanning its interfaces to Rec1 and HNH (Figure 3A and B, Supplementary file 1). Such motions are completely absent from GeoHNH, thus two neighboring domains, GeoRec2 and GeoHNH, diverge in their intrinsic flexibility (at least in isolation), raising questions about the functional implications of these motions in GeoRec2. We previously showed that heightened flexibility of SpyRec3 via specificity-enhancing mutations concomitantly narrowed the conformational space sampled by SpyHNH, highlighting a ‘motional trade-off’ between the domains. Manipulation of the flexibility of SpyCas9 and GeoCas9 domains by mutagenesis also impacts aspects of nucleic acid binding and cleavage, (Belato et al., 2022b; Palermo et al., 2018; Sternberg et al., 2015; Chen et al., 2017; Nierzwicki et al., 2021) which led us to investigate similar perturbations in GeoRec2.

Since SpyRec3 and GeoRec2 have similar structures and μs-ms flexibility, we speculated that charge-altering mutations would modulate the biophysical properties of GeoRec and the function of GeoCas9, as observed for SpyCas9. We investigated K267E and R332A GeoRec2 with NMR spin relaxation, as described for WT GeoRec2 (vide supra). An analysis of chemical exchange rates, k_ex, derived from dual-field CPMG relaxation dispersion show a global k_ex for WT GeoRec2 of 147±41 s^–1. The K267E mutation, which directly contacts the nucleic acids, shifts the globally fitted k_ex to 376±89 s^–1, while the R332A variant maintains a global k_ex similar to that of WT GeoRec2 (142±28 s^–1) and consistent with its similar thermal stability. The global fit of the K267E variant is based on CPMG profiles of 33 residues, while that of R332A is derived from 18 residues (Figure 3—figure supplements 2–4, Supplementary file 1). Interestingly, the residues participating in the global motions of both variants are distinct from those of WT GeoRec2, demonstrating that residue-specific flexibility is redistributed throughout GeoRec2, which suggests an altered intradomain molecular crosstalk within the larger GeoRec. Indeed, perturbation to NMR-detectable motions in SpyCas9 rewired its allosteric signaling and enzymatic function (Skeens et al., 2024; Nierzwicki et al., 2021). A similar dynamic modulation of GeoCas9 may fine-tune its DNA cleavage, which has been demonstrated within the GeoHNH nuclease (Belato et al., 2022b) and wedge (WED) domains (Eggers et al., 2023). We also assessed the ps-ns fluctuations of GeoRec2 variants (a negligible contribution to the WT GeoRec2 dynamic profile) and calculated order parameters from R₁, R₂ and ¹H-[¹⁵N] NOE relaxation measurements (Figure 3C). Bond vector fluctuations on the ps-ns timescale are only locally altered, thus the mutation-induced reshuffling of these motions is negligible (<∆S²>≤0.1) and suggests that, like WT GeoRec2, ps-ns motion arises primarily from global tumbling in solution.

The role of the Rec lobe in orienting the RNA:DNA hybrid within Cas9 is crucial to its function (Palermo et al., 2018; Dagdas et al., 2017; Mir et al., 2018; Jiang et al., 2015). Thus, the structure, motions, and nucleic acid interactions of Rec represent a critical piece of the Cas9 signaling machinery. Previous studies of SpyCas9 revealed that gRNA binding to the Rec lobe induces a global structural rearrangement of the protein that positions the adjacent HNH into its ‘proofreading’ state (Chen et al., 2017) after which target DNA binding positions the nucleases into active conformations for cleavage (Chen et al., 2017; Palermo et al., 2017). We wondered if the atomistic details of the apo GeoCas9-to-RNP transition could be captured by NMR using the GeoRec construct. In our previous studies, we used an in vitro DNA cleavage assay with GeoCas9 and a 141nt gRNA containing a 21nt spacer targeting the mouse Tnnt2 gene locus (Belato et al., 2022b). Since this assay was already established, we utilized the same gRNA sequence. However, truncating this gRNA was necessary to optimize binding studies for NMR analysis. We focused on the 5’ end of the gRNA, which includes the spacer sequence, based on the GeoCas9 AlphaFold2 model and structural data from NmeCas9 and SpyCas9 showing interactions between the Rec lobe and this region of the gRNA. The subsequent cryo-EM structure of GeoCas9 corroborated this interaction (Shen et al., 2024). Initial attempts using a truncated 101nt portion of the full gRNA resulted in poor NMR spectra. An overlay of the ¹H-¹⁵N HSQC NMR spectra of apo GeoRec and GeoRec-RNP at a 1:1 molar ratio showed extensive line broadening (Figure 4—figure supplement 1), likely due to the large size of the complex (75.5 kDa). To mitigate this issue, a 39nt RNA containing the 21 bp spacer sequence was selected for its ability to maintain the NMR signal while being long enough to interact fully with the Rec lobe, as suggested by prior structures (Figure 4—figure supplement 2). When bound to GeoRec, this complex is 55.6 kDa and a ¹H-¹⁵N NMR spectral overlay of apo GeoRec and the domain bound to 39nt RNA shows clear, resolved resonances with significant chemical shift perturbations and line broadening (Figure 4A/B, Figure 4—figure supplement 1). The strongest chemical shift perturbations are localized to the GeoRec2 subdomain that interfaces with the RNA:DNA hybrid at the PAM distal end, where previous studies of specificity-enhancing variants of SpyCas9 have identified alterations in nucleic acid binding to SpyRec3 (Skeens et al., 2024). It is not known whether specific residues at the PAM distal binding interface of GeoRec2 play a similar role. Line broadening is evident in both GeoRec1 and GeoRec2, primarily localized to the RNA:DNA hybrid interface revealed in recent GeoCas9 structures. Microscale thermophoresis (MST) experiments quantified the affinity of GeoRec for this RNA, producing a K_d = 3.3 ± 1.5 µM that is consistent with the concentration-dependent NMR chemical shift perturbations (Figure 5A).

To understand how the K267E and R332A mutants impact RNA binding to GeoRec, we conducted RNA titration experiments via NMR and observed that chemical shift perturbations were attenuated in both variants, relative to WT GeoRec. Despite this muted structural effect, the impact from RNA-induced line broadening remains substantial in the GeoRec1 subdomain. Our NMR data revealed that a three-fold greater concentration of RNA was required to induce the maximal structural and dynamic effects in the variants than is required for WT GeoRec (Figure 4A/C), suggesting that the variants have a reduced RNA affinity. MST experiments showed statistically significant reductions in RNA affinity for the K267E and R332A constructs, relative to WT GeoRec, where K267E GeoRec produced a K_d = 7.2 ± 1.0 µM and R332A GeoRec produced a K_d = 7.2 ± 1.5 µM (Figure 5B, Supplementary file 2). The ~twofold increase in K_d may also be due, in part, to a change in the binding mode of the RNA, such as a faster k_off. Collectively, these data reveal that mutations within GeoRec primarily alter its structure around the mutation site with weaker distal effects, but more significantly impact protein dynamics and in turn, the RNA interaction. NMR experiments also demonstrate that the presence of RNA impacts both subdomains of GeoRec, providing a significant structural interface for additional molecular tuning of nucleic acid binding.

To investigate the impact of RNA binding on GeoRec2 in greater detail, we conducted NMR titration experiments using the isolated domain, which yielded even clearer NMR spectra. Figure 4—figure supplement 3 shows NMR spectra of WT, K267E, and R332A GeoRec2 overlaid with their corresponding RNA-bound spectra (39nt Tnnt2 RNA). At protein:RNA molar ratios used for full-length GeoRec studies, the WT GeoRec2 spectrum exhibited significant line broadening across the GeoRec2 sequence. Plots of NMR peak intensities (I_bound/I_free) show substantial resonance intensity losses (Figure 4—figure supplement 3), with many residues likely in the intermediate exchange regime, in addition to the assumed changes in rotational correlation of the domain. In comparison, spectra of the RNA-bound K267E and R332A GeoRec2 variants showed less pronounced signal decay at the same levels of titrant, retaining nearly double the I_bound/I_free ratio across these spectra (Figure 4—figure supplement 3). These data are consistent with the results of NMR experiments with the 43 kDa GeoRec, supporting the premise that GeoRec2 mutations weaken its interaction with RNA. The less crowded NMR spectrum of isolated GeoRec2 facilitated the resolution of distinct structural features that explain the impact of the mutations on RNA binding (Figure 4—figure supplement 3). For example, residue I53 adopts a similar conformation in RNA-bound WT and K267E GeoRec2 but assumes a different structural state in R332A. Conversely, residue R25 populates a WT-like structure in RNA-bound R332A GeoRec2, unlike K267E. Additionally, two resonances are observed for residue K71 in the RNA-bound R332A NMR spectrum, indicating real-time equilibration between two structural states. This effect is unique to the R332A variant and underscores subtle structural and dynamic changes to GeoRec during RNA binding.

We further examined the NMR data to attempt to identify residues most critical for RNA binding to GeoRec. In an overlay of the WT and mutant RNA-induced chemical shift perturbations (Δδ, Figure 4—figure supplement 4), it became clear that the effect of RNA binding to GeoRec variants was muted, where even at saturating concentrations, the chemical shift perturbations across the K267E and R332A GeoRec2 sequences were weaker than those of same residues in WT GeoRec. The residual Δδ (WT - mutant) was plotted (Figure 4—figure supplement 4), where positive values indicate that residues in a GeoRec variant are weakly affected by RNA, relative to WT. Negative residual Δδ denote sites where GeoRec variants experience a greater structural impact from RNA than corresponding sites in WT. Of particular interest are the positive residuals that hint at the sites in GeoRec most critical for tight RNA binding. These residues were mapped onto the GeoRec structure (Figure 4—figure supplement 4) and termed allosteric hotspots, as many are not at the RNA interface. Mutations of these hotspots in future studies offers a potential means of precisely tuning the affinity of GeoRec to its gRNA. Notably, residues with positive residual Δδ (suggested as critical for tight RNA binding) largely overlap in the analysis of both variants. Specifically, residues F170, R192, H264, R269, L270, L279, H300, D301, E368, D376, D403, E405, E408, and I429 appear as allosteric hotspots (with CPMG relaxation dispersion) critical to WT-like RNA interaction.

Having observed a reduced affinity of GeoRec variants for RNA by NMR and MST, we next quantified the impact of the K267E and R332A mutations on RNP formation and stability in full-length GeoCas9. The thermal unfolding midpoint of full-length WT GeoCas9 determined by CD is ~60 °C and the K267E and R332A mutations do not change the T_m of the apo protein (Figure 5—figure supplement 1). Upon formation of an RNP (using a full-length gRNA), the T_m of WT GeoCas9 increases to 73 °C. K267E GeoCas9 retains a similar T_m increase to 70 °C, while R332A GeoCas9 forms a less stable RNP with T_m of 61 °C. The trend of these data is consistent with NMR and MST, which highlight that although K267E and R332A mutations within GeoRec have somewhat muted structural effects, these changes alter protein dynamics and the interaction with gRNA.

To further investigate the effects of mutations on protein dynamics, we performed molecular dynamics (MD) simulations based on the cryo-EM structure of full-length GeoCas9 (PDB: 8UZA) in complex with gRNA and target DNA. We simulated the full-length WT GeoCas9 and its K267E and R332A mutants as well as a double mutant combining K267E and R332A (Figure 6A), in three replicates of approximately 2 μs each. Multi-microsecond simulations revealed substantial changes in the dynamics of the GeoCas9 mutants compared to the WT (Figure 6—figure supplement 1). Specifically, we observed that mutations in the REC domain significantly altered the dynamics of both the Rec and the adjacent HNH (Figure 6—figure supplement 1). Differential root-mean-square fluctuations (∆RMSF) analysis of protein residues between the WT and variants further highlighted these alterations, showing increased dynamics in the HNH and Rec domains induced by the mutations (Figure 6B). To quantify the impact of these mutations, we analyzed protein-RNA interactions by calculating the number of contacts between GeoCas9 and gRNA in the WT and mutant systems. A contact was defined as a distance between two atoms of ≤4.5 Å. The number of contacts was significantly reduced in all variants compared to the WT, with the most pronounced reduction observed in the K267E variant (Figure 6C). We next quantified gRNA-Rec domain binding by calculating the binding free energy using the MM-GBSA method over ~200 ns of stable simulation trajectories (details in Materials and methods). Consistent with a reduction in gRNA contacts, variants with the K267E mutation exhibited a substantial reduction in binding free energy (>60 kcal mol^–1) relative to the WT, whereas the R332A variant displayed a smaller reduction (<20 kcal mol^–1, Figure 6D). Notably, protein-DNA interactions remained largely unaffected, suggesting that these mutations do not impair GeoCas9 DNA cleavage ability.

Additionally, we simulated a novel variant, iGeoCas9 (PDB: 8UZB), containing mutations in the Rec1 and WED domains (Figure 6A, mutations highlighted in lime green). This variant was recently demonstrated to have enhanced specificity in genome-editing (Eggers et al., 2024). Intriguingly, iGeoCas9 exhibited increased dynamics in the HNH and Rec domains, along with a reduction in gRNA binding free energy similar to the K267E and R332A mutants. These results suggest a distinct allosteric pathway involving additional residues that enables iGeoCas9 to maintain improved DNA cleavage activity despite reduced gRNA binding affinity. In fact, iGeoCas9 samples the greatest conformational space of any variant tested (Figure 6—figure supplement 1), suggesting a high level of flexibility is critical to enhanced specificity in GeoCas9. Collectively, our simulations reveal that mutations K267E and R332A destabilize the GeoCas9 interaction with gRNA, consistent with NMR observations. Furthermore, the enhanced dynamics and altered binding affinities observed in iGeoCas9 indicate potential allosteric mechanisms that optimize its genome-editing functionality, where K267E and R332A evoke a similar, but lesser degree of biophysical change.

The dynamic impact of the GeoRec mutations and their altered gRNA interactions at the biophysical level led us to speculate that either mutation incorporated into full-length GeoCas9 would also alter its DNA cleavage function, especially at elevated temperatures where WT GeoCas9 is most active. Temperature-dependent functional alterations were previously observed for single-point mutations within GeoHNH (Belato et al., 2022b). Although the K267E and R332A mutations slightly diminished on-target DNA cleavage by GeoCas9, the effect was very subtle and these overall cleavage activities followed the temperature dependence of WT GeoCas9 quite closely (Figure 7—figure supplement 1).

To assess the impact of the K267E and R332A mutations on GeoCas9 specificity, we assayed the propensity for off-target cleavage using DNA substrates with mismatches 5–6 or 19–20 base pairs from the PAM site (Figure 7—figure supplement 2, Supplementary files 3 and 4). As a control for on- and off-target activity, we assayed WT SpyCas9 alongside the widely used high-specificity HiFi-SpyCas9 variant (Vakulskas et al., 2018; Figure 7—figure supplement 2, Supplementary file 4) and found a lower percent of digested off-target (mismatched) DNA sequences when compared to WT SpyCas9. As expected, WT GeoCas9 was increasingly sensitive to mismatched target sequences closer to the seed site, which has been demonstrated with SpyCas9 and other Cas systems (Harrington et al., 2017; Lee et al., 2016; Jinek et al., 2012; Hou et al., 2013). No significant differences in activity were observed with digestion durations ranging from 1 to 60 min (Harrington et al., 2017), implying that a 1-min digestion is sufficient for in vitro activity of GeoCas9 with the target DNA template. While these findings generally align with prior investigations of off-target DNA cleavage (Harrington et al., 2017; Lee et al., 2016; Jinek et al., 2012), there are nuanced differences. Specifically, a previous study reported ~10% cleavage of off-target DNA with a mismatch 5–6 base pairs from the PAM by WT GeoCas9 (Harrington et al., 2017). Our results showed nearly 50% cleavage for the same off-target mismatch, but still a significant decrease in cleavage from on-target or 19–20 base pair distal mismatches. This could be due to the relatively high RNP concentrations (600–900 nM) in our assay (for clear visibility on the gel), compared to prior studies with RNP concentrations ≤500 nM (Harrington et al., 2017). Our results corresponded closely to those of prior studies with a 19–20 base pair mismatch, where off-target cleavage is tolerated by WT GeoCas9 (Harrington et al., 2017). Single-point mutants K267E and R332A GeoCas9 have negligible impact on GeoCas9 specificity (both variants follow the trend of WT GeoCas9, Figure 7—figure supplement 2), which contrasts prior work with SpyCas9 that demonstrated robust specificity enhancement with single-point mutations in Rec (Vakulskas et al., 2018; Casini et al., 2018). Additionally, a GeoCas9 double mutant K267E/R332A exhibits decreased on-target cleavage efficiency, which has been noted in high-specificity Cas systems (Chen et al., 2017; Slaymaker et al., 2016; Vakulskas et al., 2018; Casini et al., 2018; Kleinstiver et al., 2016; Hu et al., 2018; Lee et al., 2018). However, the additive effect of the K267E/R332A double mutant still does not enhance GeoCas9 specificity in our assay.

MST-derived binding affinities using full-length Tnnt2 gRNA and full-length WT, K267E, or R332A GeoCas9 indicate that all three proteins have similar affinities for the gRNA used in the functional assays (Figure 7). Thus, mutations do not substantially alter full-length GeoCas9 binding to Tnnt2 gRNA, supporting similar cleavage activities for these proteins. We repeated the MST experiments using a different gRNA sequence derived from the new cryo-EM structure of GeoCas9 (PDB:8UZA), which has a different spacer sequence. We observed a similar trend, as all GeoCas9 variants exhibited comparable affinities for this gRNA. Our functional studies illustrate an apparent resilience of GeoCas9 to major functional changes at the level of Rec, despite comparable mutations having profound functional impacts in mesophilic Cas9s.

The GeoRec1 (residues 90–225) and GeoRec2 (residues 245–456) subdomains, as well as the entire GeoRec lobe (residues 90–456) of G. stearothermophilus Cas9 were engineered into a pET28a vector with a N-terminal His₆-tag and a TEV protease cleavage site. The K267E and R332A mutations were separately introduced into the GeoRec2 plasmid. Plasmids were transformed into BL21 (DE3) cells (New England Biolabs). Protein samples for CD spectroscopy, MST, and functional assays were grown in Lysogeny Broth (LB, Fisher), while isotopically labeled samples for NMR were grown in M9 minimal media (deuterated for GeoRec2 and GeoRec) containing CaCl₂, MgSO₄, MEM vitamins, and 1.0 g/L ¹⁵N ammonium chloride and 2.0 g/L ¹³C glucose (Cambridge Isotope Laboratories), as the sole nitrogen and carbon sources, respectively. Cells were induced with 1 mM IPTG after reaching an OD₆₀₀ of 0.8–1.0 and grown for 4 hr at 37 °C post induction. The cells were harvested by centrifugation, resuspended in a buffer of 50 mM Tris-HCl, 250 mM NaCl, 5 mM imidazole, and 1 mM PMSF at pH 7.4, lysed by ultrasonication, and purified by Ni−NTA affinity chromatography. Following TEV proteolysis of the terminal His-tag, the samples were further purified on a Superdex75 size exclusion column. NMR samples were dialyzed into a buffer containing 20 mM NaPi, 80 mM KCl, 1 mM DTT, and 1 mM EDTA at pH 7.4.

The full-length GeoCas9 plasmid was acquired from Addgene (#87700), expressed in TB media and was expressed and purified as previously described (Harrington et al., 2017). The K267E, R332A, and K267E/R332A variants were introduced into full-length GeoCas9 by modifying the original plasmid acquired from Addgene.

Backbone resonance assignments of GeoRec1 and GeoRec2 were carried out on a Bruker Avance NEO 600 MHz spectrometer at 25 °C. The following triple resonance experiments were collected for each sample: ¹H-¹⁵N TROSY-HSQC, HNCA, HN(CO)CA, HN(CA)CB, HN(COCA)CB, HN(CA)CO and HNCO. All spectra were processed in NMRPipe (Delaglio et al., 1995) and analyzed in Sparky (Lee et al., 2015). Three-dimensional correlations and assignments were made in CARA (Keller, 2005) and GeoRec1 and GeoRec2 backbone assignments were deposited in the BMRB under accession numbers 52363 and 51197, respectively. Backbone resonance assignments of GeoRec were completed by transferring assignments from the individually assigned spectra of GeoRec1 and GeoRec2, as done previously for other large Cas9 fragments (Nerli et al., 2021; De Paula et al., 2025).

NMR spin relaxation experiments were carried out in a temperature-compensated manner at 600 and 850 MHz on Bruker Avance NEO and Avance III HD spectrometers, respectively. CPMG experiments were adapted from the report of Palmer and coworkers (Loria et al., 1999) with a constant relaxation period of 20ms and ν_CPMG values of 0, 25, 50, 75, 100, 150, 250, 500, 750, 800, 900, and 1000 Hz. Exchange parameters were obtained from global fits of the data carried out with RELAX (Bieri et al., 2011) using the R2eff, NoRex, and CR72 models, as well as in-house fitting in GraphPad Prism with the following models:

Model 1: No exchange(1)R2eff=R20

Model 2: Two-state, fast exchange (Meiboom equation Luz and Meiboom, 1963)(2)R2eff=R20+ϕkex[1−4CPMGkextan⁡h(kex4CPMG)]

Global fitting of CPMG profiles was determined to be superior to individual fits based on the Akaike Information Criterion (Cavanaugh and Neath, 2019). Uncertainties in these rates were determined from replicate spectra with duplicate relaxation delays of 0, 25, 50 (×2), 75, 100, 150, 250, 500 (×2), 750, 800 (×2), 900, and 1000 Hz.

Longitudinal and transverse relaxation rates were measured with randomized T₁ delays of 0, 20, 60, 100, 200, 600, 800, and 1200ms and T₂ delays of 0, 16.9, 33.9, 50.9, 67.8, 84.8, and 101.8ms. Peak intensities were quantified in Sparky and the resulting decay profiles were analyzed in Sparky with errors determined from the fitted parameters. Uncertainties in these rates were determined from replicate spectra with duplicate relaxation delays of 20 (x2), 60 (x2), 100, 200, 600 (x2), 800, and 1200ms for T₁ and 16.9, 33.9 (x2), 50.9 (x2), 67.8 (x2), 84.8, 101.8 (x2) ms for T₂. Steady-state ¹H-[¹⁵N] NOE were measured with a 6 s relaxation delay followed by a 3 s saturation (delay) for the saturated (unsaturated) experiments and calculated by I_sat/I_ref. All relaxation experiments were carried out in a temperature-compensated interleaved manner.

Model-free analysis was carried out by fitting relaxation rates to five different forms of the spectral density function with local τ_m, spherical, prolate spheroid, oblate spheroid, or ellipsoid diffusion tensors (Brüschweiler et al., 1995; Mandel et al., 1995; Fushman et al., 1997; Orekhov et al., 1999; Korzhnev et al., 2001; Zhuravleva et al., 2004). The criteria for inclusion of resonances in the diffusion tensor estimate was based on the method of Bax and coworkers (Tjandra et al., 1995). N-H bond lengths were assumed to be 1.02 Å and the ¹⁵N chemical shift anisotropy tensor was –160 ppm. Diffusion tensor parameters were optimized simultaneously in RELAX under the full automated protocol (Bieri et al., 2011). Model selection was iterated until tensor and order parameters did not deviate from the prior iteration.

NMR titrations were performed on a Bruker Avance NEO 600 MHz spectrometer at 25 °C by collecting a series of ¹H-¹⁵N TROSY HSQC spectra with increasing ligand (i.e. RNA) concentration. The ¹H and ¹⁵N carrier frequencies were set to the water resonance and 120 ppm, respectively. Samples of WT, K267E, and R332A GeoRec were titrated with a 39nt portion of a gRNA until no further spectral perturbations were detected. NMR chemical shift perturbations were calculated as:Δδ=(ΔδHN2+ΔδNH2/25)/2

MST experiments were performed on a Monolith X instrument (NanoTemper Technologies), quantifying WT, K267E, R332A, and K267E/R332A GeoRec binding to a 39-nt Cy5-labeled RNA at a concentration of 20 nM in a buffer containing 20 mM sodium phosphate, 150 mM KCl, 5 mM MgCl₂, and 0.1% Triton X-100 at pH 7.6. The GeoRec proteins were serially diluted from a 200 µM stock into 16 microcentrifuge tubes and combined in a 1:1 molar ratio with serially diluted gRNA from a 40 nM stock. After incubation for 5 min at 37 °C in the dark, each sample was loaded into a capillary for measurement. K_d values for the various complexes were calculated using the MO Control software (NanoTemper Technologies). Statistical significance was calculated using a two-tailed T-test.

All GeoCas9 and GeoRec proteins were buffer exchanged into a 20 mM sodium phosphate buffer at pH 7.5, diluted to 1 μM, and loaded into a 2 mm quartz cuvette (JASCO instruments). A CD spectrum was first measured between 200 and 250 nm, after which the sample was progressively heated from 20–90 °C in 1.0 °C increments while ellipticity was monitored at 222 and 208 nm. Phosphate buffer baseline spectra were subtracted from the sample measurements. Prior to CD measurements, GeoCas9-RNP was formed by incubating 3 μM GeoCas9 with its gRNA at a 1:1.5 molar ratio at 37 °C for 10 min. The unfolding CD data was fit in GraphPad Prism to:Ellipticity(T)=[(mfT+bf)+(muT+bu)]exp[(−ΔHD,vHR)(1T−1Tm)]1+exp[(−ΔHD,vHR)(1T−1Tm)]

GeoRec protein purified as described above was crystallized by sitting drop vapor diffusion at room temperature by mixing 1.0 µL of 15 mg/mL GeoRec in a buffer of 20 mM HEPES and 100 mM KCl at pH 7.5 with 2.0 µL of crystallizing condition: 0.15 M calcium chloride, 15% polyethylene glycol 6000, 0.1 M HEPES at pH 7.0. Crystals were cryoprotected in crystallizing condition supplemented with 30% ethylene glycol. Diffraction images were collected at the NSLS-II AMX beamline at Brookhaven National Laboratory under cryogenic conditions. Images were processed using XDS (Kabsch, 2010) and Aimless in CCP4 (Winn et al., 2011). Chain A of the N. meningitidis Cas9 X-ray structure (residues 249–445 only, PDB ID: 6JDQ) was used for molecular replacement with Phaser followed by AutoBuild in Phenix (Liebschner et al., 2019). Electron density was only observed for the GeoRec2 subdomain. The GeoRec2 structure was finalized through manual building in Coot (Emsley et al., 2010) and refinement in Phenix.

Molecular Dynamics (MD) simulations were based on the cryo-EM structure of full-length GeoCas9 (PDB: 8UZA, resolution 3.17 Å) in complex with gRNA and target DNA with two mutations in GeoCas9 (at residues 8 and 582). Four systems were considered for the MD studies: WT, K267E, R332A, K267E/R332A and iGeoCas9. We generated the WT GeoCas9 by back-mutating A8D and A582H from the cryo-EM structure (PDB: 8UZA), followed by introducing the mutations K267E, R332A, or a double mutation (with both K267E and R332A) for the variant systems. Subsequently, we performed MD simulation of iGeoCas9 (PDB: 8UZB, resolution 2.63 Å) consisting of 10 mutations (D8A, E149G, T182I, N206D, P466Q, H582A, Q817R, E843K, E854G, K908R). All systems were solvated with explicit water in a periodic box of ~134 Å x~154 Å x~151 Å resulting in ~276,000 atoms. Counter ions were added to neutralize the systems. MD simulations were performed using a protocol tailored for protein-nucleic acid complexes (Sinha et al., 2023), previously applied in studies of CRISPR-Cas systems (Saha et al., 2024; Arantes et al., 2024; Sinha et al., 2024). All the simulations were performed by using Amber ff19SB force field for protein (Tian et al., 2020), ff99bsc1 corrections and χOL3 corrections for DNA and RNA, respectively (Galindo-Murillo et al., 2016; Zgarbová et al., 2011). Water molecules were described by TIP3P model (Jorgensen et al., 1983). All bonds involving hydrogens were constrained using the LINCS algorithm. A particle mesh Ewald method (PME) with a 10 Å cutoff was used to calculate electrostatics. Energy minimization was performed to relax the water molecules and counterions, keeping the protein-nucleic acid complex fixed with harmonic potential restraints of 100 kcal/mol Ų. Equilibration was performed by gradually increasing the temperature from 0 to 100 K and then to 200 K in canonical NVT ensemble and isothermal-isobaric NPT ensemble. A final temperature of 300 K was maintained via Langevin dynamics with a collision frequency γ=1/ps and a reference pressure of 1 atm was achieved through Berendsen barostat. Production runs were carried out in NVT ensemble for 2 µs for each system in three replicates, resulting in 6 µs per system (totaling 30 µs for all systems). The equations of motion were integrated with the leapfrog Verlet algorithm with a time step of 2 fs. All simulations were conducted using the GPU-empowered version of AMBER 22 (Case et al., 2005). Analysis was performed on the aggregate ensemble (i.e. ~6 μs per system).

To characterize the protein-nucleic acid interactions in all the systems under investigation, we performed contact analysis. A contact was considered to form between two atoms within a cutoff distance of ≤4.5 Å. The binding free energy of gRNA and DNA with GeoCas9 was calculated using the Molecular Mechanics Generalized Born Surface Area (MM-GBSA) method (Mongan et al., 2007; Nguyen et al., 2015; Nguyen et al., 2013) This approach was used to compare the Rec-gRNA binding affinity of WT GeoCas9 with its mutants. For each system, the binding energies were calculated over the ~200 ns ensemble of the stable trajectories at an interval of ~20 ns.

GeoCas9 gRNA templates containing 21-nt spacers targeting the mouse Tnnt2 gene locus were introduced into EcoRI and BamHI sites in pUC57 (Genscript). The plasmid was transformed into BL21(DE3) cells (New England BioLabs) and subsequent restriction digest of the plasmid DNA was carried out using the BamHI restriction enzyme (New England BioLabs) according to the manufacturer’s instructions. Linearized plasmid DNA was immediately purified using the DNA Clean and Concentrator-5 kit (Zymo Research) according to the manufacturer’s instructions. RNA transcription was performed in vitro with the HiScribe T7 High Yield RNA Synthesis Kit (New England BioLabs). DNA substrates containing the target cleavage site (479 base pairs) were produced by polymerase chain reaction (PCR) using mouse genomic DNA as a template and primer pairs 5’-CAAAGAGCTCCTCGTCCAGT-3’ and 5’-ATGGACTCCAGGACCCAAGA-3’ followed by a column purification using the NucleoSpinⓇ Gel and PCR Clean-up Kit (Macherey-Nagel). For the in vitro activity assay, RNP formation was achieved by incubating 3 µM GeoCas9 (WT, K267E, R332A, or K267E/R332A mutant) and 3 µM gRNA at either 37 °C, 60 °C, 75 °C, or 85 °C for 30 min in a reaction buffer of 20 mM Tris, 100 mM KCl, 5 mM MgCl₂, 1 mM DTT, and 5% glycerol at pH 7.5. The 10 µL cleavage reactions were set up by mixing RNP at varying concentrations with 149 nanograms of PCR products on ice followed by incubation at 37 °C for 30 min. The reaction was quenched with 1 µL of proteinase K (20 mg/mL) and subsequent incubation at 56 °C for 10 min. 6 x DNA loading buffer was added to each reaction and 10 µL of reaction mixture per lane was loaded onto an agarose gel. DNA band intensity measurements were carried out with ImageJ.

For in vitro off-target activity assays, RNP formation was achieved by incubating 10 µM GeoCas9 (WT, K267E, R332A, or K267E/R332A mutant) and 10 µM gRNA at 37 °C for 30 min in the reaction buffer described above. The 10 µL cleavage reactions were set up by mixing 1 µM RNP with 150 nanograms of PCR products (off-target DNA sequences listed in Supplementary file 3) on ice followed by incubation at 37 °C for varying time points. The reaction was quenched with 1 µL of proteinase K (20 mg/mL) and subsequent incubation at 56 °C for 10 min. 6 x DNA loading buffer was added to each reaction and 10 µL of reaction mixture per lane was loaded onto an agarose gel. DNA band intensity measurements were carried out with ImageJ. WT and HiFi SpyCas9 control proteins were purchased from Integrated DNA Technologies (IDT, cat. No. 108158 and No. 108160, respectively), as was the associated SpyCas9 gRNA, Alt-R CRISPR-Cas9 gRNA, with an RNA spacer sequence complementing 5’-TGGACAGAGCCTTCTTCTTC-3’. The on-target and off-target DNA sequences used for the SpyCas9 in vitro cleavage assay can be found in Supplementary file 4.