Theoretical study of ArcB and its dimerization, interaction with anaerobic metabolites, and activation of ArcA

ArcA/B models

Protein models were either downloaded from the AlphaFold database hosted at EBI (https://alphafold.ebi.ac.uk/) or generated in the collaborative resource on line (https://colab.research.google.com/github/sokrypton/ColabFold/blob/main/AlphaFold2.ipynb) (Mirdita et al., 2022), without using template mode and concatenating the sequences by using “:” for each chain modeled as dimers (for ArcB or ArcA separately) or higher order oligomers (homodimers or heterooligomers using ArcA) (Evans et al., 2021). All the remaining settings were used in the default mode. GPU was used with extended RAM for oligomeric models. All models generated are provided as Supplemental Files (Supplemental Zip File). Chimera was used to measure the distance between Cys¹⁸⁰ in the PAS domain and the two monomers in the dimer model. Model visualization was conducted with PyMol (Schrödinger & DeLano, 2020). Reference structures were obtained from the Protein Data Bank (https://www.rcsb.org/) as PDB files, and the PDB ID numbers are indicated on the figure legends.

ArcB docking with anaerobic metabolites

ArcB protein and ligand binding prediction was conducted with CB-Dock (Liu et al., 2020), a cavity-detection blind docking algorithm, that predicts the binding of ligands in a protein without knowing information about the binding sites and their properties. One advantage of this algorithm is that it predicts a curvature-based detection approach and performs docking with Autodock Vina achieving a ~70% success rate. Also, this algorithm performs a blind docking, providing a non-biased result. The CB-Dock server was used with the default settings. Docking data is presented in Table 1 showing the Vina score (the weighted sum of atomic interactions; McNutt et al., 2021) and cavity size (in Å). Images were obtained on the same web server. The structures of metabolites were downloaded as sdf files from PubChem and prepared for docking in UCSF Chimera as previously described (Butt et al., 2020; Pettersen et al., 2004) using the Dock Prep tool to add hydrogens and determine the charge and are included in the Supplemental File. The binding of metabolites with the isolated PAS domain was conducted. The PAS domain was modeled, including part of the connecting alpha helix region (residues 138–280 of the full-length protein) with AlphaFold2 as described in the previous section. Ligand physicochemical properties were calculated in MolCalc (Jensen & Kromann, 2013) by text mining the target molecule and using the default settings to calculate the properties of the ligands; images of dipole charge distribution were also generated in the same server and obtained as PNG files.

Leucine zipper analysis

For assessing putative leucine zipper sequences, amino acid sequences of 30 HK of E. coli (obtained from KEGG; Kanehisa & Goto, 2000) were analyzed with GLAM2 (Frith et al., 2008) with the default settings for motif discovery using gapped alignments. The best score predicted motif (e value) for each HK was mapped to the whole sequence of each HK and identified in the protein sequences using MAST (Frith et al., 2008) with the default settings. The predicted motif was indicated also in the multimer model of ArcB by manually highlighting the residues in the protein using PyMol (Fig. S2).

ArcB homologs analysis

ArcB HK homologs analysis was conducted based on the report by Jung et al. (2008), where two types of ArcB HK have been identified. Protein structural models were downloaded from the AlphaFold database using the accession numbers reported by Jung et al. (2008) or modeled for those not available, for Type 1: Photobacterium profundum and Type 2: Mannheimia succiniciproducens ArcB homologs. Protein structure comparisons were carried out with mTM-align using the default options to highlight common core regions (Dong et al., 2018). US-align was also used with the default settings for accurate global alignments and to determine if the topology is shared among the two types of ArcB kinases (Zhang et al., 2022).

Protein models were visualized with PyMOL or Chimera (Schrödinger & DeLano, 2020; Pettersen et al., 2004).

The full-length ArcB model

The ArcB kinase is comprised of a short cytoplasmic domain (residues 1 to 22), two transmembrane helices (TM) (residues 22 to 41 for TM helix 1 and 58 to 77 for TM helix 2), a short periplasmic domain between the two transmembrane domains (residues 42 to 57), and a PAS domain (residues 177–267), where the two cysteine residues are located (180 and 241). The catalytic domain is located between residues 290 to 520 (primary transmitter domain or H1), where the ATP binding box is located (spanning residues 399 to 473) and the primary transmitter residue (His²⁹²). The primary receiver domain (or D1) is located between residues 520 to 640, where the conserved Asp residue is situated (576). The phosphotransfer domain (Hpt) is in residues 640 to 778, with a conserved histidine residue (717) located within this region (Alvarez & Georgellis, 2010). The most relevant features are indicated in Fig. 1A.

To gain insight into the accuracy of the AlphaFold2 models, a series of comparisons were carried out to regions of the kinase that have been determined experimentally. The ArcB AlphaFold2 model showed that it is highly accurate in the full-length protein (Fig. 1B). In Fig. 1B, is shown in pLDDT color scheme. This score evaluates distance differences of all atoms in a model, including validation of stereochemical plausibility (Mariani et al., 2013), indicating expected distance error in Å. The less confident sections of the protein are the transmembrane domain and the histidine-containing phosphotransfer (Hpt) domain connecting helix with the PAD domain and the primary receiver domain.

The model was compared with the backbone structure of the membrane domain (PDB ID: 2KSD, Maslennikov et al., 2010), which NMR determined. As shown in Fig. 1C, the structural alignment of the PAS domain showed an RMSD of 1.91 but a TM-score of 0.528. The alignment may be explained by the loose helical packing of this kinase’s TM domain. In this case, a more flexible transmembrane domain 2 is found in the experimentally determined structure, perhaps due to the technique used (solid-state NMR, recorded at 45 °C) and the analysis method that lacked detergents, so the structure may be affected by hydrophobic hindrance. Additionally, the secondary transmitter domain or Hpt domain (Fig. 1D) was compared with the three available structures of this domain (PDB 1FR0, 2A0B, and 1A0B) (Kato et al., 1997, 1999; Ikegami et al., 2001). The comparison revealed an RMSD of 0.87 Å and a TM-score of 0.94, which suggests a nearly identical structure, even when including PDB 1FR0, which was the Hpt domain determined by NMR and shown to be a highly dynamic structure. The overall structural alignment with experimentally determined structures suggests that the prediction made by AlphaFold2 agrees with experimentally determined structures and is useful for further analysis.

Functional analysis of the ArcB model

The next structural feature to analyze is the leucine zipper, which was characterized previously by Matsushika & Mizuno (2000). For such analysis, the dimer model was generated using AlphaFold2. In Fig. 2A, the dimer is shown, each monomer in a different color. The leucine residues involved in the leucine zipper are indicated in Fig. 2B.

As shown previously, Nuñez Oreza et al. (2012) identified key residues in the helix spanning the cytoplasmic domain after the second transmembrane domain (residues 70–121) that are needed for the signal transduction mechanism. In Fig. 2B, the multimer model shows that all the leucine residues face each other. The model presented by Nuñez Oreza et al. (2012) suggests that specific leucine residues play a crucial role in maintaining the stability of the dimer and facilitating the signal detection and transduction mechanism. Experimental evidence supports this model, particularly regarding these key leucine residues. When these residues are mutated, either the regulation by the quinone redox state is eliminated (cyan box in Fig. 2B) or the activity is completely lost (magenta box in Fig. 2B) (Matsushika & Mizuno, 2000; Nuñez Oreza et al., 2012). Also, the other leucine residues that in vivo render a dominant negative phenotype are located near the second TM domain (Fig. 2B, black arrows). The mechanism that generates a dominant negative phenotype may be related to taking further apart the PAS domain from each monomer. The conformation leads to the in vitro and in vivo phenotype of the mutants in these residues or a null kinase activity as shown in vitro for the Leu⁸⁷→Val mutant (Nuñez Oreza et al., 2012). The Leu¹⁰²→Val mutant has higher activity in anaerobic conditions, suggesting that the kink found in the helix connecting the transmembrane domain with the PAS domain may acquire more flexibility, as seen in the multimer model (Fig. 2B). Also, the structural data are consistent with the predicted positions of the leucine residues proposed by Matsushika & Mizuno (2000), and the data with the report by Nuñez Oreza et al. (2012) is consistent with the regulatory effect of these leucine residues.

In Fig. S2A, the presence of leucine zippers found in 21 out of 30 sensor kinases from E. coli, indicates that ArcB leucine zipper is located spanning residues 20 to 41 (black arrow), which is in the second TM helix (Fig. S2B). The result suggests that the inner membrane leucine zipper may not be functional for establishing the kinase dimer, but the rest that has been experimentally validated is the real leucine zipper. Still, it may be related to stabilizing the TM domains only or interacting with specific lipids in the membrane.

Next, the PAS domain was analyzed for proper folding prediction compared to the structure of PAS domains determined experimentally. In Fig. 3A, the independent ArcB PAS domain was modeled with high accuracy spanning residues 139–279 (for numbering, refer to Fig. S3). In Fig. 3A, the structural comparison between the model (grey structure) and PAS domain alone (cyan) indicates that the two structures have identical TM-score 0.99119. The two cysteine residues needed for kinase regulation (Malpica et al., 2004) are located near the connecting helix between TM domain 2 and the rest of the kinase, Cys¹⁸⁰ in the upper and Cys²⁴¹ in the lower part of the PAS domain (Fig. 3B).

Further, to assess the structural conservation of the PAS domain, the 139–279 model was aligned with eight PAS domains whose structures had been experimentally determined. As shown in Fig. 3C, the PAS domain of ArcB exhibits a conserved core domain with other PAS domains from bacteria and eukaryotic organisms, suggesting that this domain may also be involved in the fine-tuning of the activation or inhibition of the kinase (see below).

Previously, Matsushika & Mizuno (2000) reported that Asn¹⁸¹→ Ala mutant showed a similar phenotype to the one displayed by an ArcB lacking the complete PAS domain (ΔPAS). In Fig. 3D, we show that the PAS domain mutant and wild type in this residue are identical in folding, and the location of the Cys¹⁸⁰ residue is identical in both predicted structures. The resulting null phenotype found by Matsushika & Mizuno (2000) is unclear by both the biochemical and genetic data as well as for the predicted structural folding of the PAS domain bearing this mutation and the resulting position of the cysteine residues.

The activation and silencing of the kinase depend on two cysteine residues (180 and 241, Malpica et al., 2004). To date, no structural data has been provided to link the disulfide bridge formation between Cys¹⁸⁰ pairs in the kinase dimer, except functional analysis using a hybrid kinase with the membrane domain of the Tar chemoreceptor showing that the kinase has a rotational on/off mechanism (Kwon, Georgellis & Lin, 2003). The AlphaFold2 multimer shows that Cys¹⁸⁰ residues are oriented apart in the predicted model (Fig. 4A), and the linear distance between the two residues is over 25 Å. Biochemical studies have estimated that the optimal distance between two cysteine residues is in the range of 2.4 and 4 Å (Bhattacharyya, Pal & Chakrabarti, 2004) and is dependent on the environment surrounding the cysteine residue (Bhattacharyya, Pal & Chakrabarti, 2004). As shown in Fig. 4B, the kinase shows a 25.811 Å in a linear distance. However, as demonstrated by Kwon, Georgellis & Lin (2003), the rotation of ArcB can be up to 100°. Thus, these two residues are likely to be in proximity to form a disulfide bond.

To gain further insight into whether the model is positioned in an active or inactive state. In the AlphaFold2 dimer model, the first residues of the cytosolic domain are stripped of the rest of the protein, analyzed from the cell membrane point of view, and tilted towards the front to have, in perspective, the Cys¹⁸⁰ residues (Fig. 4B). The leucine residues are positioned as Kwon, Georgellis & Lin (2003) suggested in the regulable form of the kinase (residues Gln⁷⁹, Leu^80, and Glu⁸¹), with the two cysteine residues pointing apart. The structural model suggests that at least the rotation must bring up to 2–4 Å apart each monomer to form the disulfide bond between the two Cys¹⁸⁰. Once assembled, this will bring the two catalytic domains apart and stimulate the phosphatase activity (which has been previously shown in Malpica et al. (2004)), and the model is in the off state but orientated in the regulatable form of the kinase (Kwon, Georgellis & Lin, 2003). Therefore, the model is consistent with a rotational on/off mechanism, and this is required for the disulfide bond formation between Cys¹⁸⁰ on each monomer and a 100° rotation to render the kinase active. Cys²⁴¹ is on the same surface and partially involved in the signal-sensing mechanism (Fig. 4A). Figure 4B shows that Cys²⁴¹ is buried deeper in the PAS domain, explaining its lesser role in kinase inactivation.

Figure 4C shows that the Cys¹⁸⁰ residue is located near Tyr¹⁷⁰ and Phe¹⁷⁷. Aromatic residues are relevant for the disulfide bond formation since the presence of an aromatic ring favors its formation (Bhattacharyya, Pal & Chakrabarti, 2004). In Fig. 4C, the Cys¹⁸⁰ (cyan) and Tyr¹⁷⁰ (magenta) are highlighted in the dimeric structure of ArcB, indicating that the Tyr¹⁷⁰ residue delimits the orientation of the disulfide bond formation and the stability since its located towards the Cys¹⁸⁰ residue on each ArcB monomer. Phe¹⁷⁷ may have a lesser role since its orientation is farther apart from the cysteine residue (Bhattacharyya, Pal & Chakrabarti, 2004).

The binding of anaerobic metabolites and the relevance of D-lactate

Docking methods allow the detection of putative binding sites of molecules relevant to the function of a protein or to screen for drug discovery (Tripathi & Bankaitis, 2017). CB-Dock, an algorithm that predicts the binding of ligands in a protein without knowing the binding sites’ properties, was used to detect putative cavities where anaerobic metabolites might bind (Liu et al., 2020). The approach used for this analysis was using this method with quinone and menaquinone molecules as recognized regulators of the kinase activity in the full-length kinase or the ArcB PAS domain (Georgellis, Kwon & Lin, 2001; Liu et al., 2020; Cao & Li, 2014) and then evaluate other anaerobic metabolites for binding. This approach will allow us to analyze whether the PAS domain, with a highly conserved structure (Fig. 3C) or in the full-length kinase, there are high probability binding sites for anaerobic metabolites. Also, the exact binding of D-lactate, for instance, has not been determined. The catalytic domain may be active for binding other molecules or perhaps an allosteric pocket that regulates kinase ATP binding or phosphotransfer reaction efficiency present. We began by determining the binding of Q₀, which has been frequently utilized in vitro assays and causes reduced hindrance due to its lack of a long aliphatic chain. Figure 5A indicates that Q₀ and menaquinone can bind in the catalytic domain, while Fig. 6A shows that it can also bind in the PAS domain.

Second, in the case of ubiquinone, 2-octa prenyl-3-methyl-5-hydroxy-6-methoxy-1,4-benzoquinone was used and is the corresponding ubiquinone synthesized in E. coli (Kwon, Kotsakis & Meganathan, 2000). This molecule can bind to the catalytic and PAS domains according to CB-docking (Figs. 5A and 6B). Both Q₀ and ubiquinone bind to a similar pocket in the PAS domain, with the aromatic ring directed towards the Cys¹⁸⁰ location (indicated by the red asterisk in Fig. 6). The same observation was found for menaquinone (Fig. S4 for the full-length protein and Fig. S5 for the PAS domain). According to the biochemical data, the binding domain is the PAS domain (Malpica et al., 2004). However, the CB-Dock tool data indicates ubiquinone is the most probable binding molecule (Table 1) in the full-length and PAS domain.

With the above data, then the binding of anaerobic metabolites was explored. Caution must be taken since the binding of quinone molecules was predicted to bind in the catalytic and PAS domains. In the case of anaerobic metabolites, one possibility is that the metabolite D-lactate is sensed in the catalytic domain since it has been shown to enhance the kinase activity both in vitro and in vivo (Georgellis, Kwon & Lin, 1999; Rodriguez, Kwon & Georgellis, 2004). CB-Dock tool prediction with the full kinase structure, D-lactate is predicted to bind in the deep pocket found in the ATP-binding domain (residues in the N, G1, and G2 boxes) (Fig. 5B). However, this is also predicted in the isolated PAS domain (Fig. 6C, for the residues involved in the binding, refer to Fig. S3). In the PAS domain case, the prediction suggests that binding is in a cavity with charged residues. Thus far, no experimental evidence of the binding sites of D-lactate has been provided. These two cavities in the sensor kinase indicate the possibility of utilizing two mutagenesis strategies to investigate the regulation mechanism. We suggest the following approaches: (1) conducting site-directed mutagenesis of the residues in the ATP binding boxes that do not affect the in vitro phosphorylation rate, and (2) performing mutagenesis targeting the 78–778 region of the kinase that demonstrates quinone-mediated regulation in vitro. Directing mutations to the PAS domain may uncover further evidence favoring or discarding this analysis and ultimately showing the regulation exerted by these anaerobic metabolites.

When the same analysis is conducted with L-lactate and other anaerobic metabolites (Figs. 5, 6, S4 and S5), the kinase can bind in the same predicted regions as D-lactate. L-lactate is capable of binding in both the ATP-binding region and the PAS domain (Figs. 5 and 6). Table 1 shows the differences in the binding score and cavity size, where other anaerobic metabolites (butyrate, pyruvate, succinate, formate, and ethanol) are bound either in the ATP-binding region or in the PAS domain (Figs. S5 and 6, respectively). The results show that D-lactate has a high probability cavity, and binding for D-lactate and L-lactate can also bind to the same pocket in the same region but form a less strongly bound due to a reduced number of contacts. Butyrate and pyruvate also show a probability like D-lactate (Table 1). However, in the PAS domain, the results are inverted; L-lactate has an equal binding probability and a bigger cavity to bind. The results from CB-Dock suggest that the two regions may regulate kinase differently; an interesting experimental approach could be to analyze the autophosphorylation rate of mutants lacking the residues involved in binding either metabolite and the effect of oxidized and reduced quinone.

As shown in Table 1, the same cavity has a high probability of binding in the full-length kinase. Butyrate, pyruvate, and succinate occupy the same cavity size. Formate and ethanol showed a low likelihood of binding. These results are consistent with the PAS domain alone (Table 1). Overall, exploring the ATP binding box and the PAS domain involved in binding anaerobic metabolism-derived metabolites can further provide evidence of other regulatory conditions for the ArcB sensor kinase that are worth exploring experimentally.

The calculation of the overall characteristics of each metabolite is a useful approach to determining why the binding is probable in the molecules tested. In Table 2, the properties of these molecules are shown, and in Fig. S6, a graphical representation of the predicted charge distribution for each anaerobic metabolite tested is shown.

The reduced binding of L-lactate can be explained by a stronger dipole (Table 2 and Fig. S6), which explains the position in the full-length kinase as bound outward from the identified cavity (Fig. 5), in an inverse manner to the PAS domain (Fig. 6). Despite the results presented here, a question remains regarding whether L-lactate can modulate the disulfide bond formation in Cys¹⁸⁰ allosterically, like the experimental evidence shown for D-lactate (Georgellis, Kwon & Lin, 1999; Rodriguez, Kwon & Georgellis, 2004). The available evidence has not provided proof of L-lactate binding in vitro or in vivo due to the higher affinity for D-lactate. Also, the cavity size and Vina score for succinate and pyruvate is similar to D and L-lactate binding sites. However, both metabolites show a stronger dipole moment and higher total entropy (Table 2), suggesting that the binding of these two relevant anaerobic metabolites may depend on the cytoplasmic pH; perhaps the determination of ArcB activity under acidic conditions may render regulation by any of the tested anaerobic metabolites.

The properties of the rest of the anaerobic metabolites are similar, except for ethanol, which has the lowest dipole and solvation energy. The net charge is distributed along the molecule similarly to D-lactate, with the varying size of each molecule (Fig. S6). From a physicochemical point of view, the rest of the anaerobic metabolites are candidates for regulating ArcB at either the ATP binding box or PAS domain due to their similar chemical features. Putative binding cavities of the same size may, in vivo, provide further regulation of the kinase activity or sensitivity to the quinone pool redox state.

The interaction between ArcB and ArcA

ArcB exerts its action by activating ArcA via a phosphotransfer reaction with the conserved Asp⁵⁴. To date, the only available structure of ArcB with a bound response regulator is to CheY (Yaku et al., 1997; Kato, Mizuno & Hakoshima, 1998). The reported interaction, however, is not between the cognate response regulator of ArcB. In Fig. 7, using the multimer capability of AlphaFold2, the predicted interaction of ArcA and ArcB is shown. In Fig. S7, the ArcA dimer model shows that the conformation predicted is consistent with the available crystal structures (ArcA receiver domain without and with beryllium fluoride to mimic phosphorylation, Toro-Roman, Mack & Stock, 2005). Also, the quality assessment of the five models of the ArcA tetramer is shown in Fig. S8. Unambiguously, biochemical studies have demonstrated that the autophosphorylation and then phosphotransfer route is from His²⁹² to Asp⁵⁷⁶ of the same monomer of ArcB. The phosphotransfer from Asp⁵⁷⁶ to His⁷¹⁷ occurs in an intermolecular reaction between the two monomers of ArcB and then the transfer to ArcA (Teran-Melo et al., 2018).

However, no structural data are available to determine in vitro and in vivo that the geometry and spatial organization of ArcB in the interaction with ArcA can ultimately resolve the phosphotransfer pathway, even though the extensive biochemical and genetic data support the reaction pathway (Kwon, Georgellis & Lin, 2000; Peña-Sandoval & Georgellis, 2010; Teran-Melo et al., 2018).

In order to predict the binding of ArcA to ArcB, the model of full-length ArcA was generated to assess its accuracy against the receiver domain of ArcA that has been experimentally determined (Toro-Roman, Mack & Stock, 2005). In Fig. 7A, the two reported conformations of ArcA were modeled (for model data, refer to Fig. S7A through C) as a dimer and a tetramer (Toro-Roman, Mack & Stock, 2005). The dimer model renders a near identical conformation with the crystal structure of the ArcA receiver domain (Figs. 7A and S7D, TM-score 0.96829). The tetrameric form of ArcA is found by size exclusion chromatography (Toro-Roman, Mack & Stock, 2005). The model also predicts a spatial orientation of each monomer as in the dimeric ArcA (Fig. 7A, model statistics in Fig. S8). The modeled ArcA is consistent with the structural information available (Toro-Roman, Mack & Stock, 2005).

Then, the ArcB interaction with ArcA was modeled. First, the kinase dimer was modeled with a monomeric form of ArcA bound to each subunit of ArcB (Fig. 7B, statistics of the model in Fig. S9). In this model, the receiver domain of ArcA is positioned in the Hpt domain of ArcB on each subunit, which is consistent with the vast experimental evidence on the phosphotransfer mechanism (Teran-Melo et al., 2018).

However, the DNA-binding form of ArcA is either a dimer or a tetramer (Toro-Roman, Mack & Stock, 2005). Therefore, modeling ArcB with either the dimeric (statistics in Fig. S10) or tetrameric (statistics in Fig. S11) form of ArcA may provide insight into the regulation of the response regulator activation.

Firstly, in the model contemplating an ArcA dimer, all the predictions indicate two possible interactions between ArcA and ArcB (Fig. 7C). The ArcA interaction is with the Hpt domain (one monomer) in all cases. The second molecule (Fig. 7B) demonstrates direct binding to ArcA in the catalytic domain. Two conformations of the ArcA monomers are predicted and shown in two shades of orange, with model statistics presented in Fig. S10. The models suggest that ArcA may need to be phosphorylated before dimerization, consistent with previous research (Toro-Roman, Mack & Stock, 2005).

Secondly, Toro-Roman, Mack & Stock (2005) found that upon phosphorylation, ArcA exhibits a higher-order structure as a tetramer. As shown here, the tetrameric ArcA can arrange in the same orientation as in the dimer crystal. This is consistent with the experimental determination of the activation domain and the activated form as a tetramer reported by Toro-Roman, Mack & Stock (2005). In Fig. 7, the interaction between the ArcB dimer and the monomer of ArcA is predicted to bind to the Hpt domain (Fig. 7B).

When the kinase and response regulator were modeled in different stoichiometries, i.e., as a dimer with one ArcA, with the ArcA dimer, and with the ArcA tetramer, the resulting models showed different architectures. Figure 7D shows the model of ArcB with the tetrameric form of ArcA, revealing two spatial distributions. The first is two dimers bound between the Hpt domain and the PAS domain (left image in Fig. 7D), while the second is a dimer bound to the Hpt and another dimer bound to the catalytic domain (right image in Fig. 7D). Taking together the data in Figs. 7C and 7D, the most plausible mechanism involves the phosphorylation of an ArcA monomer. Then it forms higher order complexes, perhaps as shown in Fig. 7C, one ArcA monomer is bound to the kinase awaiting phosphorylation. Further biochemical data is needed to elucidate a stepwise mechanism of ArcA activation.

The current model for phosphotransfer in ArcB is as follows: from H1 to D1, the phosphotransfer occurs in the same subunit, and then, the phosphate group is transferred to the Hpt domain (H2). Using the model presented in Fig. 8A, a specific focus was taken to assess the orientation of the catalytic residues in ArcB in relation to the position of ArcA. The location of key residues and the domains involved in each step of the phosphotransfer mechanism is presented in Fig. 8B. This model suggests that the active kinase interacts with the response regulator (ArcA, in green) to the H2 domain (in blue). In this model, H1 (orange) and D1 (in yellow) are placed in the same region of the dimeric ArcB, whereas the H2 domain is placed afar from D1. The predicted flexible part shown in the model explains the in vitro and in vivo demonstration that the final phosphotransfer pathway can occur by displacement of the Hpt domain (H2) towards D1. It is more flexible in the 290–778 kinase fragment (Fig. S12). In Fig. 1C, the ArcB dimer is a snapshot of an immobile ArcB.

The crystal structure of the Hpt domain contains just the first part of the flexible region (Fig. 1). Therefore, the most significant limitation of protein models and crystals is that these regions’ dynamics are lost. That may also explain why no crystal structures spanning the complete cytosolic part of ArcB is available yet. The model presented here suggests that all the catalytic domains are highly mobile. For the Hpt domain, depicted in Figs. 8B and 8C show the side where the Asp⁵⁷⁶ residue resides and meets the opposite on each monomer. The Hpt domain can move towards this residue of the opposing subunit due to a flexible connector, consistent with the biochemical findings of Teran-Melo et al. (2018). One important feature of the models shown here is the position of the cysteine residues in the PAS domain; all models predict both Cys¹⁸⁰ facing in the opposite direction, consistent with the active form of the kinase. One remaining question is whether ArcA, once bound, the phosphotransfer reaction occurs directly and the kinase is already locked in the conformation needed to be in close proximity to ArcA (as depicted in Fig. 8C) or, first, the Hpt domain is phosphorylated and this induces the displacemnt into the position shown in the model of Fig. 8A, allowing to bind ArcA and then transferring the phosphate group to ArcA Asp⁵⁴ residue. In Fig. S13C, the monomeric form of ArcA with the predicted binding of ArcA shows that the Asp⁵⁴ residue is closer to His⁷¹⁷ than the position shown in the dimeric form of ArcB (Fig. 8B), showing the previously different phosphotransfer rate described previously between the signal decay using in vitro ArcB fragments of H1-D1-H2 (slower signal decay) vs. D1-H2 (faster signal decay) (Georgellis et al., 1998).

The reverse phosphotransfer in ArcB is relevant for signal decay (Georgellis et al., 1998) via the transfer from Asp⁵⁴ of ArcA to the conserved histidine (His⁷¹⁷) in the Hpt domain and then releasing the phosphate group from D1 domain (Georgellis et al., 1998; Peña-Sandoval, Kwon & Georgellis, 2005; Teran-Melo et al., 2018). In Fig. S13, the model of the monomer of ArcB interacting with ArcA shows a more flexed connector between the second transmembrane domain and the PAS domain (Fig. S13B), and the catalytic domains are positioned outwards (Fig. S13B). One relevant feature in this model is that the more open state is consistent with the reverse phosphotransfer seen in signal decay since the phosphor group travels in the revers phosphor relay reaction from D2 to H2 and then in D1 is released to the cytoplasm. This model shows that the position of Cys¹⁸⁰ has shifted away from the second transmembrane domain and is now located opposite to its position in the dimeric kinase, as depicted in Fig. 4B. To confirm this, a model of residues 290–778 was generated, both alone and in complex with ArcA, as shown in Fig. S12. This model shows that the Hpt domain is less compact than in the dimer or the dimer in the complex with ArcA (Fig. 7). In this model, Asp⁵⁷⁶ is in the opposite position of His⁷¹⁷, suggesting that a rotation in the same plane can occur in both forward and reverse phosphotransfer.

With the models presented here, the extensive biochemical and genetic data are consistent with the predicted structure of ArcB. The next step would be to determine the crystal structure in active and inactive states to assess the same transitional states of the Hpt domain in the active or inactive states of the kinase.

Structural features of ArcB homologs

The available protein models in the AlphaFold database and the comparison between the types of ArcB kinases can uncover mechanistic insights into signal sensing. Figure 9 shows a structural comparison between type 1 and 2 ArcB kinases (Jung et al., 2008). The distinction between the two types of ArcB kinase is proposed by Jung et al. (2008) by the lack of almost the entire linker region (residues 93 to 271, E. coli ArcB numbering, Uniprot P0AEC3), which includes the PAS domain. As proposed by Jung et al. (2008), the organisms bearing an ArcB without a PAS domain also lack the presence of ubiquinone, except for Pasteurella (Uniprot A0A126QD99), which is worth studying further to assess the mechanism of signal sensing.

Regarding the possible sensing mechanism, the cysteine residue involved in the activation and deactivation of ArcB in E. coli is absent in type 2 kinases. Genetic data indicate that both Mannheimia (Uniprot A0A378NAL7) and Haemophilus (Uniprot A0A2S9SIC6) ArcB homologs can provide toluidine blue resistance and redox regulation in vivo in an arcB null mutant in E. coli respectively (Jung et al., 2008; Georgellis et al., 2001). However, Mannheimia ArcB in vitro fails to respond to ubiquinone and menadione or the regulation exerted by anaerobic metabolites (Jung et al., 2008). This suggests the need to explore further the residues identified in the CB-Docking analysis shown here in the PAS domain.

ArcB proteins from different bacterial species have variations in their amino acid sequences, including the number and location of cysteine residues. For example, the ArcB protein from Mannheimia has three cysteine residues located in the second transmembrane domain (Cys³⁷), the linker region of D1 (Cys⁴⁷⁰), and the Hpt domain (Cys⁵⁵⁴). On the other hand, the ArcB protein from Haemophilus influenzae has five cysteine residues located in different domains of the protein. Specifically, there is one in the first transmembrane domain (Cys³⁷, as in Mannheimia ArcB), one in the ATP-binding box and H1 domain (Cys²⁶⁸), one in the D1 domain (Cys⁴⁷²), and two in the Hpt domain (Cys⁵⁷⁴ and Cys⁵⁹⁶). Figure 9B shows that cysteine residues in type 2 ArcB proteins are clustered in either the second transmembrane domain or the D1 and Hpt domains (in magenta). Further investigation is required to determine the role of these cysteine residues and how they contribute to the signal transduction mechanism, possibly through cysteine disulfide bonding in the second transmembrane domain or interaction with other proteins that have yet to be identified.

The value of protein modeling

The models presented by AlphaFold2 and AlphaFold2 multimer consistently predict accurate structures that provide meaningful data or confirm robust experimental evidence, for which the only missing aspect is structural confirmation. Based on the available structures for ArcB, we confirmed that the structure contains highly flexible regions, the symmetry and architecture of the dimer are consistent with previous observations of the regulatory mechanism, and the interaction with ArcA is also compatible with genetic and biochemical data regarding the phosphorelay mechanism. The search for binding sites for anaerobic metabolites also uncovered putative target sites that have not been explored, specifically the ATP-binding box domain. This result is also consistent with type 2 kinases lacking the PAS domain, which fails to be regulated by D-lactate but can respond to the effect of external electron acceptors like ArcB from E. coli.

The short periplasmic domain in ArcB remains an important topic. Our analysis shows that the architecture of the membrane-spanning domains is tightly packed, with a kink in the second transmembrane domain (Fig. 1), which is relevant to forming the dimer and is stabilized by the leucine zipper (Fig. 2). The kinase activation may require additional aids for the rotation and avoiding diffusion in the membrane, which cannot be predicted from the generated model. We propose that the regulation of the rotation of the kinase may be related to the cell membrane structure, such as the effect of differential lipid concentration, type, and properties, including lipid raft-like structures. Guzmán-Flores et al. (2019) demonstrated the existence of detergent-resistant membranes in E. coli, but they did not find ArcB in these membrane domains. However, they did find ten proteins with unknown functions that may be associated with these domains and interact (perhaps weakly) with ArcB and other kinases, limiting their diffusion in the membrane plane. The same detergent-resistant domains have been shown to directly affect the kinase activity, as reported for the Staphylococcus aureus SaeRS TCS (Yeo et al., 2023). To test the effect of lipid raft-like structures on kinase activity, experiments using membrane-destabilizing compounds that may alter either the location or the functional dimers formed for signal sensing may shed light on the fundamental role of membrane micro-domains (García-Fernández et al., 2017).

The PAS domain is an ancient protein module found in all kingdoms of life (Nambu et al., 1991; Vogt & Schippers, 2015). Overall, the PAS domain structure is highly conserved among signal transduction proteins. In Fig. 3, we show the structural alignment of the ArcB sensor kinase with other PAS-containing proteins, both prokaryotic and eukaryotic, deposited in the PDB. The structure contains lower sequence conservation but high structural conservation, suggesting that the 3D space of this domain is relevant for its sensory role. Still, the specificity is constrained to the key residues involved in ligand binding. Also, this suggests that the sensory function of the PAS domain is broader than other more divergent signal sensory domains since, here, the regulatory Cys¹⁸⁰ is a key element for signal perception. Also, exploring the Asn¹⁸¹→ Ala mutant in the PAS domain with the provided structural data suggests no major structural changes are observed. Perhaps the local physicochemical environment is needed for the proper oxidoreduction of Cys¹⁸⁰. However, AlphaFold2 models must be taken with caution, specifically for the impact on ΔΔG values, since Pak et al. (2023) found that the impact on pLDDT values has low to no correlation with experimental ΔΔG values. Since no structural variations were found, Asn¹⁸¹ mutants with other residues are suggested to explore the impact of the physicochemical environment.

The effect of D-lactate has been addressed in vivo (Rodriguez, Kwon & Georgellis, 2004) and in vitro (Georgellis, Kwon & Lin, 1999). In both instances, the effect of D-lactate was observed in the full-length protein; thus, the exact pocket where the anaerobic metabolite exerts its influence remains unknown. Here, two putative pockets were identified, one in the catalytic H1 domain and the second associated with the PAS domain with lower probability. The full-length anaerobic metabolites were identified by cavity detection and blind docking to bind to the catalytic domain (Fig. 5). The model of the PAS domain alone reveals a potential cavity that could bind anaerobic metabolites and perhaps disturb the redox mechanism sensing, thereby enhancing kinase activity as previously shown (Malpica et al., 2004; Rodriguez, Kwon & Georgellis, 2004). Site-directed mutagenesis and biochemical analysis can be used to identify key residues for D-lactate binding and other anaerobic metabolites, which may be hidden from the high effect exerted by D-lactate. Ultimately, this can determine the main site for D-lactate binding and additional potential binding pockets both in vitro and in vivo.

Evolution of TCSs with emphasis on protein-protein contacts

One key question in the evolution of TCSs is their specificity and reduced crosstalk. Typically, a new TCS arises through operon duplication since the sensor kinase and response regulator are usually found in the same operon. This raises the question of how specificity is established, which can occur in two ways. The first is the evolution of a unique protein-protein interface that develops a new functional pathway. The second is through the independent evolution of two TCSs (the four proteins) to generate specificity (Nocedal & Laub, 2022), thus establishing a delimited regulatory pathway. In their work, Nocedal & Laub (2022) showed that three mutations are needed to develop the independence of a TCS.

Interestingly, these mutations occurred on the histidine kinase of one system and on the response regulator of the other, resulting in the evolution of specificity rather than changes in phosphotransfer speed between them. Based on the data presented in Figs. 7 and 8, it is less plausible that in vivo CheY can interact with the ArcB kinase (as shown in Yaku et al., 1997 and Kato, Mizuno & Hakoshima, 1998). The exerted effect on swimming behavior is related to the expression of flagellar genes (Kato et al., 2007) rather than a physical interaction between the chemotactic CheY regulator and ArcB. However, Kato et al. (2007) suggested that ArcA may have another pathway that exerts this regulation. The role of acetyl phosphate has been eliminated as a possible regulatory mechanism (Liu et al., 2009). Further analysis of other regulatory mechanisms, particularly regarding the dynamics of ArcA interaction with ArcB in phosphorylation and dephosphorylation reactions, is needed.

The modeling of ArcA in interaction with ArcB in this study shows that either the dimeric or tetrameric forms of ArcA, only one monomer binds to the Hpt domain. According to available data on the multimerization of ArcA, the complex formation between phosphorylated and unphosphorylated ArcA is 1:1 (Jeon et al., 2001). Therefore, the predictions shown here suggest that in the model of ArcA activation where only one ArcA monomer is needed to be phosphorylated, the other ArcA monomers interact weakly with ArcB awaiting to form an active dimer and then interact with a second activated ArcA dimer to form the DNA-binding ArcA tetramer (Jeon et al., 2001).

Signal sensing

The mechanistic role of Cys¹⁸⁰ has been experimentally determined; this residue is needed to form a disulfide bridge that usually is in the range of ~2.4–4.0 Å to form (Chaney & Steinrauf, 1974). Also, the environment near the cysteine residue is involved in disulfide bond formation (Bhattacharyya, Pal & Chakrabarti, 2004). Here, the presence of Tyr¹⁷⁰ suggests that the electrons in the aromatic π-system avoid the lone sulfur pair of electrons, modifying the plane at which the disulfide is located (shown here in Figs. 4B and 4C) (Bhattacharyya, Pal & Chakrabarti, 2004). In type 1 kinases, Tyr¹⁷⁰ is followed by either Phe¹⁶⁹ or Tyr¹⁶⁹ (Georgellis et al., 2001); both residues, in the predicted structures, are facing in the opposite direction than Tyr¹⁷⁰, suggesting that the presence of aromatic residues hampers the formation of Cys¹⁸⁰-Cys²⁴¹ disulfide bonds and orients both subunits to each other. Also, the location of Phe¹⁷⁷ may play a role in the redox state of Cys¹⁸⁰ due to the orientation towards this residue. The experimental assessment of the two aromatic amino acids near Cys¹⁸⁰ may shed light on the fine-tuning of the regulation of ArcB.

The two types of ArcB kinases found to date suggest that in type 2 kinases, the sensing of the signal may require the interaction with another protein rather than by disulfide bond formation or the involvement of other cysteine residues not located in the linker between the transmembrane domains and the catalytic domains. Further research on organisms bearing type 2 kinases will illuminate this subject.

Overall, in this work, we addressed the following. First, AlphaFold2 models are accurate in two regions that have been experimentally determined for the sensor kinase. Second, the multimer algorithm of AlphaFold2 provided a model consistent with the available experimental evidence on the phosphorelay mechanism of ArcB. Third, CB-Dock analysis suggests that two cavities can be involved in the binding of anaerobic metabolites. The binding depends on the ligands’ cavity size and physicochemical properties. Finally, there is still much work to be done regarding the experimental determination of the structure of both type 1 and 2 ArcB kinases, at least the cytosolic portion of the kinase, to assess the oxidation state and the environment surrounding the cysteine residues.