The X-ray crystal structure of the heterodimeric MBH from C. necator (MBH; PDB code: 3RGW (Fritsch et al., 2011; Figure 1) was used to extract coordinates of the large subunit (HoxG) as a starting model. During the natural maturation of MBH, the HoxG subunit forms a heterodimer with HoxK only after insertion of the [NiFe] cofactor and cleavage of its C-terminal tail (Hartmann et al., 2018, 2020; Caserta et al., 2022a).

Therefore, this C-terminal peptide is not found in the heterodimeric crystal structure, and it is not present in our model targeting a fully mature HoxG subunit.

The HoxG protein is equipped with a Strep-tag II at its N terminus (Caserta et al., 2022a); therefore, we have used CHARMM software (Brooks et al., 2009) and CHARMM 36 Force-Field (MacKerell, 1998; Best et al., 2012) to model the affinity tag. After prediction using i-Tasser (Zhang, 2008; Roy et al., 2010; Yang et al., 2014), the Strep-tag II sequence has been modelled in a random coil conformation, followed by a geometry optimization in the dielectric medium (ε = 4) for 4,000 steps. Additionally, a short geometry optimization of Strep-tag II while keeping the coordinates of HoxG fixed was done with NAMD (Phillips et al., 2005) in vacuum for 10 ns. The protonation pattern has been computed with Karlsberg2⁺ software (Rabenstein and Knapp, 2001; Kieseritzky and Knapp, 2008; Meyer and Knapp, 2015) at pH 7.0 and applied to titratable groups in all MD simulations (see SI; Supplementary Table S1). The resulting structure with modelled cofactors has been used as a basis for all further MD simulations, both conventional (cMD) and Gaussian accelerated (GaMD; Miao et al., 2015). Amino-acid numbering has been kept following that of the MBH crystal structure. Where necessary, the amino acid numbering of the Strep-tag II sequence has been included with the prefix “tag.”

The [NiFe] active site of MBH has been trapped in several redox states of the catalytic cycle (Sickerman and Hu, 2019; Apfel et al., 2020). For simplicity, we have chosen an EPR silent state, the Ni_a-S state (Ilina et al., 2019; Sickerman and Hu, 2019), which represents the presumed state accepting and releasing H₂. In the Ni_a-S state, both Ni and Fe are in a + 2 redox state, coordinated to four cysteine residues. The atomic partial charges have been taken from early works on similar enzymes (Teixeira et al., 2006), where electrostatic potential was computed quantumchemically with Gaussian (Frisch, 2016) and fitted with RESP (Cornell et al., 1993). During preparatory steps and MD simulations, the [NiFe] active site has been kept rigid with coordinates as in the crystal structure but with corrected ligand orientation [Rippers et al., 2012; PDB code: 3RGW (Fritsch et al., 2011)] by applying conformational restraints on bonds (force constant of 500 kcal), angles (force constant of 100 kcal) and dihedral angles (barrier of 1.0 kcal) with NAMD (Phillips et al., 2005), as in previous applications (Utesch et al., 2013; Heidary et al., 2015). Considering that the main goal of this study is to investigate the global dynamic properties of the large subunit, constraints on the [NiFe] active site will have little to no influence.

The prepared structure of the large subunit with the N-terminal Strep-tag II and crystal water molecules were solvated using a TIP3P (Jorgensen et al., 1983) water box with periodic boundary conditions via a PSFGEN VMD plugin (Humphrey et al., 1996) with an addition of NaCl ions (150 mM). A water box with slightly larger dimensions 102 Å x 121 Å x 93 Å was used to accommodate conformational changes stemming from Strep-tag II flexibility or other protein regions. All MD simulations (conventional and Gaussian accelerated) were performed with software NAMD (Phillips et al., 2005) with 2 fs time step using SHAKE (Ryckaert et al., 1977) fixing bond lengths of hydrogen atoms, Langevin dynamics at 300 K with small friction constant of β = 1 ps⁻¹ avoiding slowing down of dynamics (Blumhagen et al., 1996) and particle-mesh Ewald method (Darden et al., 1993) for electrostatic interactions.

The solvated model system described above was energy minimized, then heated to 300 K for 20 ps while restraining the positions of all protein heavy atoms. During the 60 ps-long preequilibration step, the conformational restraints on the protein atoms were gradually lifted except for the atoms constituting the [NiFe] active site including the three inorganic ligands and the side chains atoms (up to Cβ) of the coordinating cysteine residues. Finally, the entire system was thermally equilibrated for 45 ns via cMD. The preparatory steps before GaMD included a 2 ns extension of the cMD in order to collect potential statistics for the estimation of the GaMD (double boost) acceleration parameters and a 2 ns GaMD equilibration considering the previously calculated boost potential (Miao et al., 2015). GaMDs were run for 250 ns. During these simulations the upper limits of the standard deviation of the total boost potential and the dihedral boost potential were set to 5 kcal/mol. To enlarge the statistical sampling, both cMD and GaMD were repeated twice. Further in text, these simulations are referred to as simulation 1 and simulation 2.

The initial protonation pattern (Supplementary Table S1; SI) was determined based on the initial model [HoxG from the crystal structure 3RGW (Fritsch et al., 2011) with modelled N-terminal Strep-tag II as a random coil, Figure 1], by computing pK_A values with Karlsberg2⁺ (KB2⁺; Rabenstein and Knapp, 2001; Kieseritzky and Knapp, 2008; Meyer and Knapp, 2015), as in previous applications (Wolf et al., 2020; Dragelj et al., 2021a,b; Batebi et al., 2018). The pK_A values have been computed in the range from −10.00 to 20.00, not only for the starting model but also for the structures from GaMD trajectories obtained every 2 ns. The conformational stability of predicted structures was evaluated considering conformational energies from the time frames of GaMDs. Conformational energies were computed by solving the linearized PoissonBoltzmann (LPB) equation with the program “Adaptive Poisson-Boltzmann Solver” (APBS) (Baker et al., 2001). Water molecules and electrolytes were removed and replaced by an implicit ion concentration of 150 mM. The region out of the solvent accessible surface area was treated as a dielectric continuum with ε = 80 and protein volume was treated as a dielectric continuum with ε = 4. These conformational energies were computed using a grid resolution of 0.3 Å for frames taken every 4 ns from the resulting trajectories.

Coarse-grained Brownian Dynamics (BD) simulations were run using the ProPHet (Probing Protein Heterogeneity, available online at https://bioserv.rpbs.univ-paris-diderot.fr/services/ProPHet/) program (Sacquin-Mora, 2014, 2016, 2018). In this approach, the protein is represented using an elastic network model (ENM). Unlike most common coarse-grained models where each residue is described by a single pseudo atom (Tozzini, 2005) ProPHet uses a more detailed representation (Zacharias, 2003) that involves up to 3 pseudo atoms per residue and enables different amino acids to be distinguished. Pseudo atoms closer than the cutoff parameter R_c = 9 Å are joined by Gaussian springs which all have identical spring constants of γ_struct = 0. 42 N.m⁻¹ (0.6 kcal.mol^-1.Å⁻²). The springs are taken to be relaxed for the initial conformation of the protein. The simulations use an implicit solvent representation via the diffusion and random displacement terms in the equation of motion, (Ermak and McCammon, 1978) and hydrodynamic interactions are included through the diffusion tensor (Pastor et al., 1988).

Mechanical properties are obtained from 200,000 BD steps at an interval of 10 fs and a temperature of 300 K. The simulations lead to deformations of roughly 1.5 Å root-mean-square deviation with respect to the protein starting conformation (which by construction corresponds to the system’s equilibrium state). The trajectories are analyzed in terms of the fluctuations of the mean distance between each pseudo atom belonging to a given amino acid and the pseudo atoms belonging to the remaining protein residues. The inverse of these fluctuations yields an effective force constant k_i describing the ease of moving a pseudo atom with respect to the overall protein structure:

where 〈〉 denotes an average taken over the whole simulation and d_i = 〈d_ij〉j* is the average distance from particle i to the other particles j in the protein (the sum over j* implies the exclusion of the pseudo atoms belonging to residue i). The distance between the C_α pseudo atom of residue i and the C_α pseudo atoms of the adjacent residues i-1 and i + 1 are excluded since the corresponding distances are virtually constant. The force constant for each residue is the average of the force constants for all its constituent pseudo atoms i. We will use the term rigidity profile to describe the ordered set of force constants for all the residues of the protein.

Despite the experimental evidence that isolated preHoxG (i.e., a HoxG protein precursor equipped with an [NiFe] active site but still containing its C-terminal extension tail) can be found in different oligomerization states with a prevalent homodimeric fraction, no detailed structural information was available (Hartmann et al., 2018). The three-dimensional structure of the HoxG homodimer was constructed using the cartesian coordinates of the monomeric structure predicted with the lowest conformational energy during 20 ns of GaMD simulation. We used SymmDock Webserver (Schneidman-Duhovny et al., 2005a,b) to investigate the relative orientation of two HoxG_m subunits. The SymmDock algorithm predicts protein complexes by geometry-based rigid docking, whereby the scoring function considers both geometric fit and atomic desolvation energy of resulting constructs. In our protocol we requested the docking of two isolated HoxG units without distance constraints, nor the prior definition of a binding interface. Interestingly, out of the resulting top 10 suggested complexes more than half of them form dimer structures in which both HoxG monomers interact with each other via the HoxG-HoxK interface of the functional MBH heterodimer as seen in Figure 2. For simplicity, we have chosen a HoxG_d model with the highest overall score as a starting model, as the prediction of reliable dimer poses can be computationally very demanding, and it is not the main objective of this work. The HoxG_d model was prepared for MD simulations following the same protocols and conditions as described above for the HoxG_m with an unchanged protonation pattern [determined by Karlsberg2⁺ software (Rabenstein and Knapp, 2001; Kieseritzky and Knapp, 2008; Meyer and Knapp, 2015). HoxG_d was solvated in a water box with dimensions of 121 Å x 114 Å x 114 Å. After 45 ns thermal equilibration at 300 K with cMD, boost parameters for GaMD were estimated for a 150 ns long GaMD of the HoxG_d model.

For the size exclusion chromatography experiments, the as-purified HoxG protein was investigated in various concentration ranges (0.5, 2.5, 12.5 and 60 mg/ml) in 50 mM K_iPO₄, 150 mM NaCl pH 7.4 (purification buffer). Measurements were run on an ÄKTA pure 25 using a Superdex 200 Increase 10/300 GL (Cytiva) column equilibrated with the purification buffer at 4 °C. A calibration curve was made by measuring six protein standards with known molecular weights between 12 and 670 kDa: Thyroglobulin (669 kDa, 9.34 ml), Apoferritin (443 kDa, 10.49 ml), β-Amylase (200 kDa, 11.8 ml), Bovine serum albumin (66 kDa, 14.1 ml), Carbonic anhydrase (29 kDa, 16.8 ml) and cytochrome C (12.3 kDa, 18.24 ml). Additionally, HoxC (the large subunit of the regulatory hydrogenase from C. necator) was included in the calibration series as its oligomerization profile was recently elucidated (Caserta et al., 2020a).

HoxG protein solutions were transferred into a homemade, gas-tight, and temperaturecontrolled (10°C) transmission cell equipped with two sandwiched CaF₂ windows separated by a Teflon spacer with an optical pathlength of 50 μm. Spectra with a resolution of 2 cm⁻¹ were recorded on a Tensor 27 Fourier-Transform spectrometer (Bruker) equipped with an MCT (liquid nitrogen-cooled mercury-cadmium-telluride) detector. The cell compartment was purged with dried air. For a single spectrum 200 individual scans were averaged. A buffer spectrum was used as reference for calculating the corresponding absorbance spectra. OPUS software version 7.5 from Bruker was used for data analysis. For the cofactor-stability experiments, IR spectra of as-isolated HoxG protein solution 35 mg/ml (500 μM) and a threefold higher concentrated sample were recorded consecutively for 7 h.

The structures obtained from Gaussian accelerated molecular dynamics simulations were analyzed in order to get a deeper insight into the dynamics and stability of the HoxG_m subunit. Special attention has been given to the quaternary structure, as well as the preservation of secondary structure elements. Root-meansquare deviation (RMSD), root-mean-square fluctuation (RMSF) of backbone atoms of protein, electrostatic conformational energy, and rigidity profile of the structure were computed.

The RMSF values of backbone Cα atoms from two GaMD simulations (shown in Supplementary Figure S1) are mostly low (below 2 Å), indicating a stable structure with some flexible loops. The highest flexibility is observed for the Strep-tag II affinity tag (Supplementary Figure S1), in line with the fact that this peptide fragment is seldom resolved in X-ray protein structures. In all simulations, no secondary structure elements could be assigned for Step-tag II, which exhibited multiple conformations relative to the HoxG structure (Supplementary Figure S2). Furthermore, the flexibility of the affinity tag may cause destabilization of the terminal β-sheet as seen in Figure 3 (residues 213, yellow). A reliable indicator of the structural stability is the root-mean-square deviation or RMSD value of the protein backbone, which was computed in all cases by excluding the mobile.

Strep-tag II (both from the alignment and the calculations). The RMSD values in cMD (Supplementary Figure S3) show a steady rise to about 2.0 Å. This was also observed for GaMD simulations, where RMSD values reached 3.0 Å (Figure 3B1). Moreover, when several flexible loops, identified based on RMSF values (residues 24–28, 171–191, 373–386, and 244–250), were excluded, the RMSD value was reduced to about 0.5 Å. The flexibility of loop regions is commonly observed in molecular dynamics simulations but it is noteworthy that residues from these loop regions form interactions with the small subunit (HoxK) in the heterodimeric MBH enzyme (Fritsch et al., 2011; Albareda et al., 2019). In contrast, these loops are solvent exposed in the HoxG_m. Interestingly, the loop containing residues 24–28 (Figure 3A3) hosts Glu27, which plays an important proton gate role and it is located in the center of the interface of the two subunits in MBH (Tombolelli and Mroginski, 2019).

Analysis with VMD (Humphrey et al., 1996) showed that the majority of the secondary structures are preserved during both GaMD simulations, supported by a relatively low reduced RMSD, hinting at the stability of HoxG even in the absence of the small subunit. Furthermore, the relative conformational energies of time frames extracted from GaMD simulations yield average values of 345 kJ/mol and 331 kJ/mol above the 28,000 kJ/mol baseline (Supplementary Figure S4) for simulations 1 and 2, respectively, with fluctuations up to 400 kJ/mol. Despite these large fluctuations of the conformational energy, there is no indication of either protein unfolding or large conformational changes in the structures derived from GaMD simulations.

HoxG_d structure, predicted by SymmDock Webserver (Schneidman-Duhovny et al., 2005a,b) was subjected to cMD and GaMD simulations (see Materials and methods for details) in order to obtain a thermally equilibrated moiety for the analysis of structural features. RMSD and RMSF values (Supplementary Figures S5, S6) computed for backbone atoms indicate a high structural stability comprising a “closed” conformation and a preserved orientation of the two subunits. Computed relative conformational energies of time frames extracted from GaMD simulations yield and average value of 382 kJ/mol above the baseline value (Supplementary Figure S7), with fluctuations of around 350 kJ/mol, furthermore confirming the structural integrity of the predicted HoxG homodimer. The homodimer interface (Supplementary Figure S8) contains 52 residues (3,382 Ų) and 49 residues (3,366 Ų) belonging to HoxG-I and HoxG-II, respectively via 6 salt-bridges, 28 hydrogen bonds and 260 non-bonded contacts, as identified with PDBSum (Laskowski, 2009). Salt-bridges found in the thermally equilibrated HoxG_d after 150 ns are: Arg62(I)-Glu21(II), Arg73(I)Glu27(II), Glu21(I)-Arg62(II), Glu27(I)-Arg73(II), Arg267(I)-Asp211(II) and Glu371(I)Arg384(II). Several important residues have been observed to play a role at the interface between two HoxG subunits. Residue Glu27 that plays an important role in proton transfer in MBH is involved in the formation of a stable salt-bridge with Arg62 in both HoxG subunits in the HoxG_d structure (Dementin et al., 2004; Tombolelli and Mroginski, 2019). Cysteines 597 and 75, coordinating the Ni ion, are involved in interface interactions as well, albeit only observed in one subunit (HoxG-I; Supplementary Figure S8) in the HoxG_d.

Computation of mechanical properties and rigidity profiles may help identify the location of the active sites in proteins (Lavery and Sacquin-Mora, 2007; Sacquin-Mora, 2016) and were utilized in this work in order to identify stable regions of the catalytic (large) subunit of MBH (HoxG) in the absence of its small counterpart HoxK. Rigidity profiles were computed for both HoxG_m and HoxG_d models based on time frames from GaMD simulations, as well as, for the entire MBH structure [PDB code: 3RGW (Fritsch et al., 2011)] and its HoxG subunit extracted from 3RGW without further processing (HoxG_c, HoxG_MBH; Figure 4; Supplementary Figures S9, S10).

The rigidity profile obtained for HoxG_MBH (Supplementary Figure S10) is qualitatively similar to those reported in an earlier study (Oteri et al., 2014a) for the soluble and membrane-bound [NiFe]-hydrogenases from D. fructosovorans and A. aeolicus, respectively. The highest force constant peaks obtained for the MBH structure are observed for residues coordinating the active site, i.e., Cys75, Cys78, Cys597 and Cys600 (Figure 4, red line), which is a common feature in proteins with cofactors (Yang and Bahar, 2005; Sacquin-Mora and Lavery, 2006; Sacquin-Mora et al., 2007). Despite the absence of HoxK, a similar trend is observed for HoxG_c (Figure 4, black line) as well as in structures extracted from GaMD simulations (Figure 4, purple line), where peaks have different intensities. In the region of Cys75 and Cys78, the intensities of peaks decrease from about 700 kcal·mol⁻¹·Å⁻² to just below 100 kcal·mol⁻¹·Å⁻² and in the region of Cys597 and Cys600 the magnitude of the force constants is significantly reduced from ca. 300 kcal·mol⁻¹·Å⁻² to half when comparing HoxG_m to the HoxG_MBH. The loss of rigidity around active site cysteines is compensated by an increase in peak intensities around residues Gly412, Arg476 and Arg530. The highest peak (at 373 kcal·mol⁻¹·Å⁻² in Figure 4, purple line) from GaMD simulations is assigned to Arg530, which is also very pronounced in the rigidity profile of HoxG_MBH. Arg530 forms a stable salt-bridge with Asp117 in several structures of CnMBH [PDB codes: 3RGW (Fritsch et al., 2011), 4IUC, 4IUB, 4IUD (Frielingsdorf et al., 2014)] and it is maintained in cMD and GaMD simulations (Supplementary Figure S11).

Generally, residues featured as peak-representatives are within 12 Å of the [NiFe] active site (Figure 4), thereby reflecting the stability of the protein core in the isolated HoxG (Figure 2). However, the [NiFe] active site in the HoxG_m is somewhat less rigid than that in HoxG_MBH with force constants of all cysteines below 100 kcal·mol⁻¹·Å⁻². Loss of rigidity around the active site may be the consequence of the solvent exposure of Ni²⁺, Cys597 and Cys75 and the increased flexibility of the adjacent loop with residues 24–28 (Figure 3A3). These sites are located at the HoxG-HoxK interface in MBH that becomes solvent exposed when isolating the HoxG unit.

Interestingly, the rigidity encompassing the [NiFe] active site, which was lost in the HoxG_m is partially regained in the HoxG_d model (Figures 2, 4). Furthermore, the rigidity of Glu27 seems to be recovered as reflected by a force constant peak of approx. 108 kcal·mol⁻¹·Å⁻², which is similar to that observed in HoxG_MBH. His229, which interacts with the HoxK subunit in MBH, also contributes to the rigidity of the HoxG_d with a force constant peak of approx. 209 kcal·mol⁻¹·Å⁻². Notably, the region encompassing the Arg530 and Phe594 appears less rigid. The change of mechanical properties in this region may also have consequences on the efficiency of substrate channeling and binding to the active site.

The electrostatic properties of proteins are crucial for understanding their interactions with other molecules and/or surfaces, e.g., of electrodes (Oteri et al., 2014b; Heidary et al., 2015). After separation from the HoxK subunit of MBH, the electrostatic properties of the HoxG_m exhibit some differences compared to the HoxG_MBH. The pK_A values (protonation states) initially determined with Karlsberg2⁺ (Rabenstein and Knapp, 2001; Kieseritzky and Knapp, 2008; Meyer and Knapp, 2015), remain unchanged in all MD simulations (Supplementary Table S1). Remarkable is the significant change in the pK_A value of Glu27 upon isolation of the HoxG subunit. This residue is located in the proximity of the [NiFe] center and is immersed in the protein matrix at the HoxG-HoxK interface. A very high pK_A value (> 20) is predicted, indicating a neutral charge state. Upon removal of HoxK, Glu27 becomes solvent exposed and its pK_A drops significantly down to ca. 4, suggesting the prevalence of its anionic form. Therefore, we propose that the mechanism by which protons are supplied to the active site is altered when the HoxK unit is detached from the MBH.

In order to better identify changes in the electrostatic properties arising from the removal of the small HoxK subunit, we also computed the electrostatic potential surface (EPS) of the HoxG_c and that of HoxG_m using APBS (Baker et al., 2001; Figure 5). The EPS of thermally equilibrated HoxG_m (Figure 5B) shows a considerable number of positive charges at the solvent exposed interface (occupied by the HoxK subunit in MBH), where several Arg, Lys and His residues are present. A comparison of the computed EPS for CnMBH and its HoxG subunit is shown in Supplementary Figure S12. In regions different from the former heterodimer interface we observed similar charge localizations patterns, with important exceptions in regions containing flexible loops and the Strep-tag II tag (Figure 6; Supplementary Figure S12). Here, the differences in charge distribution significantly depend on their local conformations and interactions.

The direction and magnitude of the dipole moment in HoxG_m were evaluated over the course of GaMD simulations using VMD (Humphrey et al., 1996), with charges and atomic radii from CHARMM ForceField (MacKerell, 1998; Best et al., 2012). The dipole moment of the HoxG_MBH oriented along the y axis, which was defined from Cα of Asn311 to Cα of Pro22 as reference (Figure 5A). In the course of the GaMD simulation, the total dipole moment of HoxG_m rotates 25° with respect to its initial position. This is shown in Figure 5B as representative conformation, taken from simulation 2 after 100 ns. Thereby, HoxG_m adopts an orientation similar to that in the heterodimer MBH (Supplementary Figure S12; Utesch et al., 2013; Heidary et al., 2015). Furthermore, moderate fluctuations of the magnitude and orientation of the dipole moment are predicted during the simulations (Supplementary Figure S14) like previously reported for other [NiFe]-hydrogenases (Oteri et al., 2014b).

During the GaMD simulations, the magnitude of the dipole moment of the HoxG_m fluctuates moderately around 800 ± 250 Debye (Supplementary Figure S14) and decreases to around 677 ± 100 Debye upon exclusion of the Strep-tag II (Supplementary Table S2). Moreover, also the orientation of the dipole moment in HoxG shows some fluctuations, as the formed angle of the dipole moment computed for HoxG_c in 3RGW (Fritsch et al., 2011) oscillates between 5° and 40° when the Strep Tag-II is taken into account and between 0° and 30° if the tag is absent (the related HoxG_c does not contain a Strep-tag II; Fritsch et al., 2011). Thus, our calculations demonstrate that the presence of a flexible tag region, carrying charged Glu, His⁺ and Lys residues, has a significant influence on the orientation and strength of the total dipole moment of HoxG protein and consequently, on the stability of the electrostatic interactions with potential reactions partners (e.g., proteins and surfaces).

The oligomerization state of HoxG was investigated utilizing size exclusion chromatography. The protein was measured in various concentrations between 0.5 and 60 mg/ml. We observed that at concentrations below 12.5 mg/ml, the chromatographic profile of HoxG comprises mainly a monomeric form (R_V 15.34 ml). At higher protein concentrations, a significant amount of a dimeric form (R_V 14.1 ml) was detected (Figure 7).

The computational work predicted a stable HoxG_d homodimer at the former HoxK interface. This structural arrangement is supposed to contribute to the cofactor stability, protecting the solvent exposed NiFe(CN)₂(CO) site from degradation. To elucidate whether the dimeric arrangement might confer stability to the active site, we used IR spectroscopy. Given that the CO and CN-diatomic ligands of the [NiFe] cofactor have specific spectroscopic signatures that vary with respect to changes in electron density at the active site, IR spectroscopy can provide detailed information on the hydrogenase cofactor monitoring redox changes, hydrogen bonding, protonation state of neighboring residues as well as stability of the [NiFe] site in the protein scaffold (Ash et al., 2017; Tai et al., 2021).

For this purpose, IR spectra of two HoxG samples (ca 30 mg/ml and 100 mg/ml, respectively) were recorded continuously for 7 h. These two protein concentrations were chosen such that, the first sample (between 12.5 and 60 mg/ml) has a higher monomer content, while the 100 mg/ml sample has a higher dimer content. Furthermore, the comparable high protein concentrations enable a clear detection of the active site absorption bands suitable for cofactor quantitation. The spectra were first normalized to the intensity of the amide II band (at ca. 1,551 cm⁻¹) and then the integral of the CO stretching bands region was calculated and plotted as a function of time (Figure 8A). The fully mature HoxG protein has been recently characterized (Caserta et al., 2022a) and its spectrum shows two main CO absorptions at 1,929 and 1,939 cm⁻¹, and a weaker band at 1953 cm⁻¹ that was assigned to an unmatured portion of the protein still containing its C-terminal peptide extension (Hartmann et al., 2018; Caserta et al., 2022a).

By comparing the first (t₀) and last IR spectrum (t_7h) of the two HoxG samples (Figures 8A,B) and their time dependence (Figure 8C), we observed a more pronounced loss of the active site signals in the sample with higher monomer content. Given a strong correlation between the amount of dimeric HoxG and the increased stability of the active site signals, we propose that the dimerization occurs at the interface freed by the HoxK subunit. This is in line with the computational predictions, suggesting a recovery of the active site rigidity for the HoxG protein upon (homo)-dimerization.