This work describes results obtained from five experimental structures of MtHISN5, including three crystal structures at 1.55, 1.69, and 2.2 Å resolutions ( Supplementary Figures S1A–C ) and two cryoEM structures resolved at 2.4 Å (MtHISN5-unliganded, PDB ID: 7OJ5) and 2.2 Å (complex MtHISN5 with 2S,3S-IGP (referred to as IG2), PDB ID: 8QAV) ( Figure 2 ). The latter was obtained using the commercially available IGP (cIGP).

The crystal structure at 1.55 Å was obtained in space group R3 and comprises eight chains in the asymmetric unit (ASU). The crystal structure at 2.2 Å is isomorphous to the 1.55 Å structure but contains different ligands, e.g., citrate (CIT, Supplementary Figure S1D , see below). We also obtained the structure in space group I4 at 1.69 Å resolution with six chains in the ASU. The crystal structures cover the MtHISN5 sequence starting from Gly77 or Ala78 through Arg260/261 (chain and structure-dependent). Both cryoEM structures (unliganded and IG2 complex) have been reconstructed from EM maps with octahedral symmetry, and therefore contain 24 identical protein chains, spanning from Ala78 to Arg261 (IG2 complex) and Arg262 (unliganded).

The MtHISN5 subunit contains a four α-helix bundle sandwiched between two β-sheets whose four strands are almost perpendicular to each other ( Figure 3A ). In solution, MtHISN5 forms a homo 24-mer with octahedral (432) symmetry ( Figures 2A, F , 3B ) of approx. 110 Å diameter and a total mass of 540 kDa. It remains unclear why natural selection has promoted 24-merization for IGPDs/HISN5s. Possible drivers may be (i) minimization of energy cost and amino acid usage (Akashi and Gojobori, 2002; Seligmann, 2003), (ii) proteome stability and efficiency of translation (Kepp, 2020), and (iii) cost of gene expression (Frumkin et al., 2017). Another evolutionary driver of oligomerization may be the benefit of cooperative regulation of the enzyme activity. However, to the best of our knowledge, such a property has never been reported for any IGPD enzyme. The dimensions of the MtHISN5 oligomer did not change significantly after the binding of IG2, indicating a lack of major conformational rearrangements. The active site of MtHISN5 (see below) contains two Mn²⁺ cations (Mn1 and Mn2, Figure 3C ) bound by residues belonging to different subunits. In other species, it has been shown that withdrawal of Mn²⁺ causes HISN5 dissociation to inactive trimers while re-addition of Mn²⁺ or other divalent metal cations (e.g., Co²⁺, Cd²⁺, Ni²⁺, Fe²⁺, and Zn²⁺) reassembles the enzyme (Sinha et al., 2004; Glynn et al., 2005b). However, we did not observe MtHISN5 trimers at any stage of the purification process, even when 40 mM EDTA was used (not shown).

As mentioned above, each subunit of MtHISN5 contains two Mn²⁺ cations (Mn1 and Mn2). The cryoEM and crystal structures share the same pattern of Mn²⁺ coordination ( Figures 3C, D ). Mn1 is coordinated octahedrally by Nϵ of His141, Nϵ of His213, carboxyl of Glu145, Nϵ of His238* from a neighboring subunit (*), and a water molecule. The Mn2 cations are complexed by Nϵ of His142, Nϵ of His115*, Nϵ of His237*, carboxyl of Glu241*, and a water molecule. The Mn2 coordination sphere is incomplete in the MtHISN5-unliganded cryoEM structure, owing to the lower map resolution. The crystal structures were obtained in the presence of formate (FMT) between Mn1 and Mn2, which completes their coordination spheres ( Figure 3C ). The corresponding position is occupied by a water molecule in the unliganded cryoEM structure ( Figure 2E ) or by the imidazole moiety in the MtHISN5-IG2 complex ( Figure 2J ).

The cryoEM MtHISN5-IG2 complex structure reveals detailed information about the most likely substrate positioning before catalysis. The imidazole ring is trapped between manganese ions, facing its Nδ towards Mn2 and Nϵ towards Mn1. While the formation of the active site with the bi-Mn²⁺ cluster requires a contribution of residues from two subunits, a third subunit (**) participates in substrate/product binding by contributing guanidines of Arg167** and Arg189** that bind the phosphate group ( Figure 3D ). However, based on our complex with IG2, these polar H-bonds are rather weak, with the donor-acceptor distances of 3.5 – 3.8 Å. Participation of the corresponding Arg121 of Mycobacterium tuberculosis HisB in closing the active site has been pointed out recently by Kumar and coworkers (Kumar et al., 2022).

It must be emphasized that our MtHISN5-IG2 complex structure is a spectacular example where the cryoEM maps are of such a high quality that they permit resolving stereoisomers of ligands bound to a protein. In fact, we expected the reaction product, IAP as the enzyme was incubated with cIGP (at 2 mM concentration) for 2 days prior to the cryoEM grid preparation. We clearly recognized IG2 based on EM maps ( Figure 3D , inset). The improvement in map resolution is an important advancement compared to AtHISN5, which yielded a 3.1-Å EM map (PDB ID: 6EZJ), making it difficult to determine the positioning of ligands and water molecules as well as distinguishing between the R- and S-isomers (IGP vs IG2) (Rawson et al., 2018). In this context, using MtHISN5 as a model and our pipeline for the cryoEM structure-based development of novel herbicides will be a significant improvement.

In addition to the IG2 observed in our cryoEM complex, we identified several types of molecules bound to MtHISN5, suggesting hot-spots prone to bind certain chemical moieties. The crystal structure at 1.55 Å resolution contains imidazole (IMD), sodium ion, chloride ion, formate (FMT), 1,2-ethanediol (EDO), glycerol (GOL), and tris(hydroxymethyl) aminomethane (TRS). The structure at 1.69 contains FMT, GOL, and TRS. The structure at 2.2 Å contains a chloride ion, FMT, EDO, acetate (ACT), sulfate ion, and CIT. The most abundant binders among these structures are EDOs (n = 41) and FMTs (n = 49). EDOs bind mostly to the inner surface and interfaces between subunits ( Figure 4A ), but a few are also found on the outer surface and in the vicinity of the active site (4-6 Å, Figure 4B ). FMTs primary location was in the active sites, between Mn1 and Mn2 ( Figures 3C , 4B ). However, it was very interesting to find that IMD (IMD1, Figure 4B ) bound not between the Mn²⁺ ions in the active site, but instead was positioned approximately 3.8 Å from the C atom of FMT. This is in contrast to other reported structures (AtHISN5, PDB ID: 4MU1 (Bisson et al., 2015)), where imidazole mimicked part of the substrate/product (between Mn1 and Mn2). In MtHISN5, IMD1 forms hydrogen bonds with FMT and water. We postulate that the carboxylate and imidazole binding sites can be used as pharmacophores for the design of selective inhibitors of plant HISN5. The fact that we see FMT between Mn²⁺ and imidazole positioned differently suggests that the bi-Mn²⁺ cluster has a high affinity for carboxylate, in addition to the imidazole of IGP or IAP.

To search for small molecules representing a broader chemical space, we performed virtual screening (VS) by in silico docking 3.3 mln lead-like molecules from the ZINC database (Sterling and Irwin, 2015) in the neighborhood of the MtHISN5 active site. Six molecules scoring the highest binding energy gain (between -10.0 and -9.4 kcal/mol) are shown in Figure 5 . The top-scoring molecules satisfy the following criteria: (i) the content of heteroatoms that improve water solubility and can potentially ensure specific binding to the protein, and (ii) the potential for parallel or T-shaped π-stacking with surrounding residues.

Intrigued by the highly symmetric and porous structure of MtHISN5, we analyzed the molecular tunnels (in addition to the one presented in Figure 4A ) that could let small molecules penetrate inside the enzyme. To this end, we studied the cryoEM MtHISN5-unliganded structure in the Caver Analyst 2.0 and Caver 3.0.3 PyMol plugin (Chovancova et al., 2012; Jurcik et al., 2018). Most of the tunnels are distributed along the 2-, 3-, and 4-fold axes ( Figure 6 ). The average diameter of the tunnels along the 2-fold axes is large enough to allow infiltration only by very small molecules (< 60 Da), such as EDO, FMT, and water. Larger molecules, however, could permeate the inner cavity of MtHISN5 through the tunnels along the 3-fold and 4-fold axes, which are approximately 2.5 times wider in diameter. Consistently, we identified TRS (121 Da) and CIT (192 Da, Supplementary Figure S1D ) in the 3- and 4-fold tunnels, respectively.

MtHISN5 (Uniprot ID: I3SDM5) is encoded by the HISN5 gene (Ensembl: MTR_1g103820, Gene database: LOC25485215) located on chromosome 1. According to the Gene Database (Benson et al., 2013) and the TargetP 2.0 webserver (Emanuelsson et al., 2000), MtHISN5 contains an exon corresponding to a 76 amino acid residues long chloroplast transit peptide. The following intron separates the sequence of the genuine MtHISN5 enzyme, which complies with previous observations for a canonical isoform in A. thaliana, HISN5B (Gene database: AT4G14910). This pattern suggests the fusion of the exon encoding the transit peptide with the exon encoding the enzyme sequence during evolution, which has been observed for several nuclear genes encoding chloroplast proteins (Wolter et al., 1988; Gantt et al., 1991; Stepansky and Leustek, 2006). A likely reason for the compartmentalization of the HBP is the interconnection with de novo purine metabolism that occurs in chloroplasts (and mitochondria). The HBP shares a precursor, 5-phosphoribosyl-1-pyrophosphate (PRPP) and an intermediate, aminoimidazolecarboximide ribonucleotide (AICAR), with de novo purine metabolism (Smith and Atkins, 2002; Witte and Herde, 2020). The obtained MtHISN5 structures also allowed us to investigate the two transcript isoforms, X1 which is 1300 nt long, and X2 (1227 nt) in M. truncatula. The isoform X2 lacks the 5^th exon, which would result in a protein missing the loop-β7-loop fragment (residues Asp185 to Gln209). Therefore, it is very unlikely that the isoform X2 is expressed as a functional enzyme.

To assess the similarity between prokaryotic and eukaryotic IGPD/HISN5 enzymes, we analyzed 12 710 sequences from the InterPro family IPR000807 by calculating a sequence similarity network (SSN, Figure 7 ). The result showed a close relationship between plants (Viridiplantae), green algae (Chlorophyta), and Cyanobacteria suggesting that plant HISN5s derive from cyanobacterial IGPDs. This is consistent with the endosymbiotic theory. To verify this observation, we generated a phylogenetic tree based on homologous sequences of MtHISN5 ( Supplementary Figure S2 ). The tree also showed a close relationship between the plant and cyanobacterial IGPDs, supporting the cyanobacterial origin of plant HISN5 enzymes. In contrast, we have recently shown that plant HISN2 and HISN6 are distant homologs of their cyanobacterial counterparts and are likely to have been acquired by HGT (Witek et al., 2021; Rutkiewicz et al., 2023).

The analysis also revealed a relatively large group of bacterial bifunctional enzymes presenting IGPD and HPP activity ( Figure 7 ). Previous analyses of the phylogenetic origin of bacterial bifunctional hisB, i.e., hisNB, genes were limited to the classes of γ- and ϵ-Proteobacteria (Brilli and Fani, 2004; Kinateder et al., 2023). Our results suggest that the presence of fused genes occurs also in other bacterial phyla, providing new models for his genes’ evolution, especially interesting when combined with novel approaches for enzyme functional annotation (Kinateder et al., 2024). These bifunctional enzymes exist in Acidobacteriota (Acidobacteria), Actinomycetota (Actinobacteria), Bacteroidota (Bacterioidetes), Calditrichota (Calditrichaeota), Pseudomonadota (γ-Proteobacteria) and Spirochaetota (Spirochaetes); named according to the recent nomenclature update by the International Code of Nomenclature of Prokaryotes (Oren and Garrity, 2021) (former names are in parentheses).

In general, the majority of bacterial sequences present high sequence variability between phyla, which is much more significant than, for example, within plants. Moreover, fungal IGPD sequences are disconnected from other groups, suggesting that they differentiated early and have evolved in parallel. The conservation of HISN5 sequences in plant species, clearly distant from other kingdoms, suggests the possibility of reaching general-purpose herbicidal activity by potent HISN5 inhibitors. Such a potential is discussed in detail in the next chapter.

The overall fold of IGPD enzymes, including their 24-meric assembly, is highly conserved between kingdoms of life, even though sequence conservation varies strongly. More precisely, MtHISN5 sequence shares 88% identity with A. thaliana HISN5B (AtHISN5B; 89% with AtHISN5A), 45% with Acanthamoeba castellani (AcIGPD), 40% with Saccharomyces cerevisiae (ScIGPD), and 40% with Staphylococcus aureus (SaIGPD). Therefore, we decided to perform a detailed analysis to pinpoint differences that could be exploited to ensure the selectivity of future inhibitors of plant HISN5 versus bacterial IGPD homologs. To obtain a perspective on the conserved and variable regions, we analyzed residue conservation using the ConSurf web server (Ashkenazy et al., 2016). Results of the Multiple Sequence Alignment (MSA) amongst all kingdoms reveal that the highest conservation score was assigned to residues forming HISN5 active sites and coordinating manganese ions: (i) Mn1 by His141, Glu145, His213, His238*, and (ii) Mn2 by His115*, His237*, Glu241*, His142 ( Figure 8A ). There were other residues located close to the active site area that were assigned the highest conservation rank and are also conserved in the aforementioned species, e.g., Asp146 or Lys245 that take part in ligand binding ( Figure 8B ) or Arg167 and Arg189 that stabilize the substrate’s phosphate group by weak hydrogen bonding ( Figure 3D ). Interestingly, residues comprising both β-sheets that are located on the outer and inner surfaces of the oligomer are rather variable across sequences from all kingdoms ( Figure 8C ). It is possible that this variability stems from the evolutionary pressure caused by operating in various environments, such as the bacterial and fungal cytosol and chloroplast stroma of plants.

When only protein sequences within the plant kingdom are considered, residues forming β-sheets on the outer and inner HISN5 surfaces are more conserved ( Figure 8D ). Still, more variability is observed at the outer surface compared to the inner ( Figure 8E ), suggesting some evolutionary pressure to maintain the environment in the hollow core of HISN5. In addition to the active site residues, the highest conservation was observed at the inter-subunit interfaces. Residues forming the central α-helical bundle are also highly conserved in plants ( Figure 8E ). Interestingly, several loops are also conserved, which is common not only for loops involved in substrate recognition, but also for those shaping internal tunnels, channels, or voids (Kress et al., 2018).

We then compared the ConSurf data obtained for all kingdoms and exclusively for plants to identify a surface region in the vicinity of the active site that would be conserved in plants but vary in other kingdoms ( Figures 9A, B ). The goal was to propose a development pathway that would ensure both selectivity for plant HISN5 sequences and high potency, which will be crucial for designing HISN5 inhibitors and the subsequent development of herbicides. In this context, the cleft near Thr153, Ser187, His191, Asn220, and Thr221 ( Figure 9C ) is the most interesting, being variable in other species ( Figure 9A ) and highly conserved in plants ( Figure 9B ). This cleft connects active sites of two MtHISN5 subunits (the Mn1-Mn1* distance is ~27 Å). The cleft is long and intrinsically symmetric, as it is crossed by one of the 2-fold axes. It is also rather hydrophilic and negatively charged ( Figures 9C, D , respectively).

A common compound optimization method involves linking molecules that bind to a target at separate sites (Kirsch et al., 2019). The most versatile linkers are oligoethylene glycol chains. However, neither ethylene glycol molecules nor polyethylene glycol (PEG) fragments were identified in our structures, despite their use at high concentrations. This is consistent with the negative charge in the cleft center, which makes it predisposed to bind H-bond donors. However, PEG is an H-bond acceptor, except for the terminal hydroxyl groups. Therefore, we conducted another VS campaign focused on the cleft to find more suitable linkers between the two active sites. This time, however, we narrowed the screening library to include only more polar compounds (logP ≤ 2, ~1.3 mln molecules). The estimated binding energies were weaker than those obtained by VS in the active site (-8.1 vs. -10 kcal/mol, Supplementary Figure S3 ). Nonetheless, interesting common features became apparent in the top-scoring molecules. First, the fragment near the 2-fold axis between Pro190 and Pro190* appears prone to bind hydrophilic, preferably aliphatic, and six-membered-ring moieties, whereby secondary amines are H-bond donors. In contrast, more hydrophobic moieties tend to bind between the apolar parts of the Arg189 and Thr221* side chains, which are closer to the active sites. It is very important to note that targeting this cleft will enable the design of symmetric inhibitors that reach two active sites simultaneously. Such a multivalency can enhance the overall binding affinity and selectivity, reducing the likelihood of interactions with off-target proteins and minimizing side effects. Based on our experimental, comparative, and computational approaches, the scaffold which would best correspond to the HISN5 pharmacophore, would include competitive inhibitor (Comp. Inh.) moieties, at the poles of the molecule ( Figure 9D ). Next, aromatic moieties would connect to the molecule center that would be created by a symmetric and polar moiety with H-bond donors. Importantly, the competitive inhibitor moiety does not need to be a substrate/product or a transition state analog but could also contain carboxylate and imidazole moieties, as based on our crystal structures.

The supply of IGP is limited, and only 1-mg packages are currently available. Previously used absorption-based methods require large quantities of IGP as one needs to prepare 8 to 12 separate solutions, each containing a different substrate concentration, to obtain a single set of experimental data (Hawkes et al., 1995). Therefore, we decided to adopt and adjust the isothermal titration calorimetry single-injection method (ITC-SIM (Wang et al., 2020)). ITC-SIM consumes only ~200 µL of substrate at saturating concentration (instead of 8-12 mL) to obtain the enzyme kinetics graph showing rate of reaction as a function of substrate concentration. Our first ITC trials were performed on the same cIGP that was used for the cryoEM experiments. Although other IGPD enzymes were assayed using cIGP from the same source, we were not able to determine kinetic parameters. We could only confirm that the enzyme was active, as the curves clearly showed an exothermic event when compared to blank experiments ( Figures 10A, B ). Nonetheless, these measurements provided interesting insights into the MtHISN5 behavior with cIGP. The exothermic reaction occurred after a lag phase, whose length was directly proportional to the cIGP concentration. During the lag phase, the heat production was very low, whereas theoretically, one should observe saturation with the substrate at the beginning of the experiment ( Figure 10A ). When using 24, 65, 253, and 800 µM of cIGP, this lag lasted for about 5, 7, 12, and 25 minutes, respectively. Subsequently, the highest heat production was observed, reflecting the achievement of V_max under the given conditions. Interestingly, with 2 mM cIGP, the initial lag reached the maximum time that the PEAQ-ITC apparatus offers (170 min). This peculiar behavior can likely be attributed to the contamination of cIGP with IG2, as revealed by our cryoEM experiments. The presence of IG2 (which cannot be enzymatically converted to IAP) in cIGP was also observed by others (Saika et al., 1993; Bisson et al., 2015). Hence, it was impossible to saturate MtHISN5 with the cIGP substrate because an increase of the 2R,3S diastereoisomer concentration always elevated the amount of 2S,3S (IG2), which is a competitive binder.

To eliminate IG2 in the substrate sample, we synthesized IGP enzymatically (hereafter referred to as eIGP). We utilized MtHISN1-4 and E. coli inorganic pyrophosphatase, each purified separately by Ni²⁺ affinity chromatography. Each reaction was monitored by absorbance ( Supplementary Figure S4 ). After the reaction with MtHISN4 was completed, the mixture was run through the Ni²⁺ resin to eliminate the proteins. This approach yields stereochemically pure IGP owing to the stereoselectivity of the HBP enzymes. The obtained yield of eIGP synthesis was 80%.

The ITC-SIM data obtained with eIGP were significantly more informative than those obtained with cIGP ( Figure 10C ). Despite the strong buffer mismatch, deriving from the high salt content in the vacuum-concentrated eIGP sample (in SpeedVac) and a likely heat effect of protein-substrate initial interaction, it was possible to observe the V_max plateau and the substrate depletion curve. Those data allowed us to fit the Michaelis-Menten equation and calculate the kinetic parameters: K_M = 227 ± 45 µM, k _cat = 3.4 ± 0.3 s^-1 ( Figure 10D ). The K_M value reported for AtHISN5 was 170 µM (Tada et al., 1995; Bisson et al., 2015), while HISN5 enzymes in crops exhibit a wide K_M range of 49 and 83 µM in Triticum aestivum (germ wheat and mature wheat, respectively), 600 µM in barley (Hordeum vulgare), up to 1.7 mM in oat, Avenia sativa (Wiater et al., 1971c). As for the k _cat value, according to our best knowledge and the BRENDA Enzymes database (Chang et al., 2021), only k _cat = 1400 for M. tuberculosis IGPD has been reported (Ahangar et al., 2013). Importantly, when comparing kinetic parameters from different studies, one should bear in mind different measurement conditions. Concentration of eIGP in the reaction buffer (optimal for enzymes synthesizing the substrate) results in relatively high salt content in the MtHISN5 reaction (200 mM KCl and 100 mM NaCl), which may lower k _cat. Nonetheless, the presented ITC-SIM enabled us to measure the kinetics for MtHISN5 and obtain K_M and k _cat values consuming ~200 µL of saturating (3 mM) eIGP per one replicate. Such amounts are significantly lower than those required by former methods (Hawkes et al., 1995).

Sequence similarity networks were calculated using the EFI-EST web server (Zallot et al., 2019). Input data consisted of 36 356 sequences belonging to the InterPro Family IPR000807 and were later reduced to 12 710 UniRef90 sequences. The calculations were based on sequences of 180 – 390 residues long and an alignment score of 80. Output results were visualized in CytoScape 3.3 (Shannon et al., 2003).

The coding sequence (CDS) of the MtHISN5 gene was retrieved from the NCBI database (entry XP_013469848.1). The CDS was PCR-amplified using the primers MtHISN5-Nt77-F (TACTTCCAATCCAATGCCGGTGCTAGAATTGGAGAGATGAAAAGG) and MtHISN5-CtFL-R (TTATCCACTTCCAATGTTAACTACGCGACAGAACCCCTTTTGAA). The PCR product was purified and cloned into the expression plasmid pMCSG68 using the ligase-independent cloning (LIC) method (Kim et al., 2011). The final MtHISN5 construct in this work was N-terminally truncated at Ser70 to obtain a higher yield of expression and enzyme stability compared to other tested constructs, truncated at residues 15, 30, 45, and 56. Because of the problematic purification of the enzyme fused with the His-tag, we deleted the tag entirely and instead inserted the 70-76 region of the original MtHISN5 sequence using the polymerase incomplete primer extension (PIPE) method (Klock et al., 2008). The primers used for the PIPE method were MtHISN5-Nt70-delHT-F (GAAGGAGATATACATATGCAACTTTCCCATATTGACTCAGGTGC) and MtHISN5-Nt70-delHT-R (CATATTGTTATATCTCCTTCTTAAAGTTAAACAAATATTATTTTCTAGAGGGG). The cloning correctness was confirmed by DNA sequencing. Overexpression was carried out in BL21 Gold E. coli cells (Agilent Technologies) in an LB medium containing 150 μg/mL ampicillin. The cultures were grown at 37°C and shaken at 180 rpm. When OD₆₀₀ reached 1.0, the temperature was lowered to 18°C, and the overexpression was induced using 0.5 mM isopropyl-D-thiogalactopyranoside (IPTG) and went on for 18 h; MnCl₂ at 10 mM final concentration was added after IPTG. The cultures were centrifuged at 5000 × g for 15 min at 4°C and the sediment was suspended in 30-35 mL of purification buffer (40mM Tris–HCl pH 8.0, 40 mM NaCl, 4 mM Mn²⁺, 0.4 mM EDTA) and frozen at -75°C for purification.

The cells were disrupted by sonication (5 min with intervals for cooling), and cell debris was removed by centrifugation at 25,000 × g for 30 min at 4°C. We then followed a method of purification with DEAE-cellulose and NaCl gradient described by Glynn (Glynn et al., 2005b), however, MtHISN5 was not present in the increasing NaCl gradient but in the first flow-through. Therefore, we decided to skip the gradient procedure and modify the method. After centrifugation at 25,000 × g, MtHISN5 was precipitated by the addition of 1.7 M ammonium sulfate. The precipitate was collected by centrifugation at 20,000 × g for 15 min at 4°C and dissolved in 2 mL of purification buffer. The protein solution was filtered through a 0.45 µm syringe filter and 2.0 mL were loaded onto a Superose 6 column previously equilibrated with the purification buffer for size-exclusion chromatography (SEC). After elution, fractions containing MtHISN5 were combined, concentrated to ~ 2.0 mL using Amicon Ultra Centrifugal Filters (Merck), and loaded onto a Superdex 200 column, also equilibrated with the purification buffer. Eluted fractions containing pure MtHISN5 (based on SDS-PAGE) were pooled and concentrated.

The concentration was assessed by two methods because of the low extinction coefficient (ϵ = 4470 M^-1 cm^-1) and possible contamination with small molecules absorbing at 280 nm, thus creating a false-positive result of a higher than actual protein concentration (>50 mg/mL). The initial measurements were conducted at λ = 280 nm. The second spectrophotometric measurement was performed using the Bradford method (Bradford, 1976). Crystal structures were obtained from concentrations of 11-15 mg/mL, whereas cryoEM structures were obtained from 1 mg/mL.

MtHISN1 and MtHISN2 for enzymatic synthesis of IGP were produced as described previously (Ruszkowski, 2018; Witek et al., 2021), omitting TEV cleavage and dialysis. MtHISN3 (N-terminally truncated at residue 42), MtHISN4 (truncated at residue 48), and E. coli inorganic pyrophosphatase were obtained following the procedure described for MtHISN1, with SEC directly after elution from Ni-NTA resin. The following primers were used to amplify MtHISN3 and MtHISN4 CDSs: MtHISN3-Nt42-F TACTTCCAATCCAATGCCTCTCCACCTTCAATTTCAATGCTCCGTTCAATTC, MtHISN3-CtFL-R TTATCCACTTTCCAATGTTAAGCCACTGAGACCTTTTGCTGGTTATGC, MtHISN4-Nt48-F TACTTCCAATCCAATGCCACTTCTATATGATTCTGTTGTGACTTTGCTTGATTATGGTG, and MtHISN4-CtFL-R TTATCCACTTCCAATGTTAGATTCGGACTTCTATGCCTTCATTCAACAAATGTTCTTT.

MtHISN5 was crystallized using the vapor diffusion method (hanging drop) and all crystallizations were set up manually at 20°C. The structure at 1.55 Å was obtained from the Morpheus screen (Gorrec, 2009) in Molecular Dimensions (MD1-46), condition 2-27 (G3) containing 0.1 M carboxylic acids (0.2 M sodium formate, 0.2 M ammonium acetate, 0.2 M sodium citrate tribasic dihydrate, 0.2 M potassium sodium tartrate tetrahydrate, 0.2 M sodium oxamate), 0.1 M imidazole, 4-morpholine-ethane-sulfonic acid (MES) adjusted by ratio to pH 6.5 and 30% v/v precipitant mix (40% v/v glycerol and 20% v/v PEG 4000).

The ShotGun Screen (SG-1 MD1-89-ECO), from Molecular Dimensions (Fazio et al., 2014) supplemented with 15% glycerol yielded the structure at 1.69 Å resolution. The measured crystals were obtained by mixing 2.0 µL of the MtHISN5 solution with 2.0 µL of the condition 2-11 (E11) containing 2.0 M sodium formate, 0.1 M sodium acetate pH 4.6.

The structure at 2.2 Å was obtained based on the Morpheus screen (Gorrec, 2009) (MD1-46) in condition 2-31 (G7) containing 0.1 M carboxylic acids (0.2 M sodium formate, 0.2 M ammonium acetate, 0.2 M sodium citrate tribasic dihydrate, 0.2 M potassium sodium tartrate tetrahydrate, 0.2 M sodium oxamate), sodium 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid (HEPES), 3-(N-morpholino) propane sulfonic acid (MOPS) at pH 7.5, and 30% v/v precipitant mix (40% v/v glycerol and 20% v/v PEG 4000). The crystals were cryoprotected with 15% glycerol, vitrified in liquid nitrogen, and stored for data measurement. Diffraction data for structures at 1.55 and 1.69 Å were measured at the P13 beamline at the PETRA III synchrotron in Hamburg, Germany. The diffraction data for the structure at 2.2 Å were measured using an in-house X-ray diffractometer, Rigaku XtaLAB Synergy-R. All datasets were processed using the XDS Package (Kabsch, 2010). Data statistics are summarized in Table 1 .

The crystal structure of MtHISN5 was solved using molecular replacement based on the structure of A. thaliana HISN5 (PDB ID: 4MU0) in PHASER (Mccoy et al., 2007). The initial model was built using Phenix.Autobuild (Terwilliger et al., 2008). The ACHESYM server was used to rearrange the model within the unit cell (Kowiel et al., 2014). Automatic model refinement was performed in Phenix.Refine (Afonine et al., 2018) and manual corrections were conducted in COOT (Emsley et al., 2010). For generation of ligand restraints, Phenix.eLBOW was used (Moriarty et al., 2009). The structure at 1.55 Å was refined anisotropically, and structures at 1.69 Å and 2.2 Å were refined isotropically with translation-libration-screw (TLS) parameters. The refinement statistics are included in Table 1 .

Preparation of the cryoEM samples and the data collection were performed at the SOLARIS CryoEM Facility (Kraków, Poland). Quantifoil TEM grids (300 mesh R1.2/1.3 copper) were glow-discharged on EM ACE200 (Leica Microsystems). The grids were then placed inside the FEI Vitrobot Mark IV chamber set to 100% humidity at 4°C and a total of 2.5 μl of the protein solution was applied (blotting parameters: blot time, 4 s; wait time, 10 s; drain time, 0 s; blot force, 0; blot total, 1). The grids were plunge-vitrified in liquid ethane and clipped in liquid nitrogen. The data were collected on a Titan Krios microscope (Thermo Fisher Scientific) operated at 300 kV, equipped with an FEI Falcon III (4k x 4k) direct electron detector. The detector operated in counting mode at 96 000× magnification, resulting in a calibrated physical pixel size of 0.86 Å px⁻¹. The micrographs were acquired as 40-frame movies (the total dose of 40 e⁻ Å⁻²) at under-focus with a defocus range of −3.0 to −0.9 μm and 0.3 μm defocus step. A total of 954 and 3330 micrographs were collected for the MtHISN5-unliganded and MtHISN5-IG2 structures, respectively.

For the MtHISN5-unliganded dataset, reconstruction of the map was performed in Relion 3.1 within the CCP-EM package (Scheres, 2012; Wood et al., 2015). The contrast transfer function (CTF) was estimated in CTFFIND4 (Rohou and Grigorieff, 2015); 871 micrographs passed the curation step. 2D references for particle picking were obtained after automatic picking of 8707 particles from 31 micrographs. Template-based picking identified 456759 particles which were extracted as 352-pixel boxes. Based on the 2D classification, 424472 particles (41 classes out of 50) were selected and subjected to 3D classification. Finally, 229366 good particles were used for high-resolution refinement with the O symmetry imposed. Per-particle CTF refinement and Bayesian particle polishing after the first refinement improved the resolution from 3.2 to 2.4 Å (gold-standard Fourier-shell correlation, GSFSC; without masking and postprocessing). The “shiny” particles were then imported into Cryosparc 4.4 and used in non-uniform refinement which further improved the resolution to 2.25 Å.

The MtHISN5-IG2 cryoEM data were processed in Cryosparc 4.1 (Punjani et al., 2017). 2990 micrographs were selected based on the CTF fit resolution, ice thickness, and accumulated motion. Blob picking, followed by 2D classification was used to generate four templates for automatic picking. After inspection of the picks, 661163 particles were retained and extracted in 300-pixel boxes for 2D classification (100 classes). The good 25 classes contained 605614 particles which were used to build the initial map and for high-resolution refinement. The best resolution (2.2 Å, GSFSC) was obtained using the non-uniform refinement protocol.

The molecular model of AtHISN5 was placed into the map in UCSF Chimera 1.15 (Pettersen et al., 2004) and the sequence was fitted using Phenix.Autobuild (Terwilliger et al., 2008). Manual corrections to the models were done in Coot (Emsley et al., 2010), between iterative rounds of automatic real-space model refinements in Phenix.Refine (Afonine et al., 2012). The latter also validated the model geometry and model-to-map correlation; details are listed in Table 2 .

All docking experiments were performed in AutoDock Vina (Trott and Olson, 2010) with the exhaustiveness = 8, using Python scripts to automate and parallelize the work. The receptor files were prepared with the UCSF Chimera DockPrep tool (Pettersen et al., 2004). VS in the active site was run using the library of lead-like molecules (3,344,603 in-stock compounds; 300-350 Da, logP ≤ 3.5) downloaded from the ZINC15 database (Sterling and Irwin, 2015) in December 2021. The structure in the I4 space group (1.69 Å resolution) was used as the receptor. The search box was centered at -2, 44, 63 Å (x, y, z) and measured 31 × 23 × 27 Å. To propose potential binders in the cleft conserved in plants but variable in other kingdoms, we selected a subset from the lead-like library in the ZINC15 database (Sterling and Irwin, 2015) containing more soluble molecules of logP ≤2 (1,355,624 docking-ready files downloaded in March 2022). The unliganded cryoEM structure (PDB ID: 7OJ5) was used as the receptor. The search box with the dimensions of 20 × 20 × 23 Å was centered at 114, 115, 155 Å (x, y, z). All results were scored based on the calculated binding energy gain.

Multiple sequence alignment and the analysis of residue variability were performed using the ConSurf Server (Ashkenazy et al., 2016). The multiple sequence alignment for both high (> 95%) and low (> 35%) sequence identity for homologs was built using the MAFFT algorithm. The homologs were collected from UNIREF90 by the CS-BLAST search algorithm. The calculations resulted in 54 unique sequences of high percent identity and 150 unique sequences of low identity. The conservation scores were assigned using the Bayesian method of calculation and the best-fit model of substitution for proteins.

The phylogenetic tree was constructed using BLAST (Altschul et al., 1997) pairwise alignments. Distances were calculated using Kimura’s method (Kimura, 1983), and the tree was built using the Neighbor-Joining method (Saitou and Nei, 1987). BLASTP search using MtHISN5-Nt70 sequence as a query, resulted in 477 Swissprot records with sequence identity between 30 and 90%. Sequences with the lowest percent identity were chosen as an outgroup root (Thermotoga sp.). The tree was visualized in IcyTree (Vaughan, 2017).

Molecular tunnel analysis was conducted using CAVER Analyst 2.0 (Jurcik et al., 2018) and CAVER 3.0.3 (Chovancova et al., 2012) plugin for PyMOL 2.4.0 software. To detect the molecular tunnels, we used the following starting point coordinates 129.834, 136.402, 130.468 (x, y, z), probe radius 1.4, shell radius 6.0, shell depth 4, clustering threshold 3.5, frame weighting coefficient = 1.0, frame clustering threshold = 1.0 and number of iterations = 12. Received tunnels and their distance from water molecules within active sites were depicted and calculated using UCSF Chimera 1.15 (Pettersen et al., 2004) and PyMOL (Schrodinger).

All protein models and structural alignment using the Needleman – Wunsch algorithm and the BLOSUM-62 matrix were visualized using UCSF Chimera 1.15 (Pettersen et al., 2004) and UCSF ChimeraX 1.5 (Pettersen et al., 2021).

Enzymatic synthesis required in vitro reconstitution of five steps of the HBP leading to the formation of 2R,3S-IGP. The reaction was conducted in a total volume of 2.75 mL at 295 K and absorbance was measured at λ = 290 nm. All enzymes used in the synthesis were His-tagged at the N-termini. All steps were conducted in the kinetic buffer (Tris-HCl 50 mM, pH 8.0; MgCl₂ 4 mM; KCl 100 mM; NaCl 50 mM, TCEP 1 mM). The first step required 1 µM MtHISN1, 40 µM E. coli pyrophosphatase, and 2 mM ATP. The mixture was blanked and the reaction was initiated using 2 mM PRPP. The reaction was conducted until a plateau was reached (approximately 50 min, absorbance reached 1.1, Supplementary Figure S4A ). The mixture was blanked once again, and 1 µM MtHISN2 was added. After 20 min when A = 2.0 ( Supplementary Figure S4B ), the mixture was blanked and 1 µM MtHISN3, 5 µM MtHISN4, and 5 mM L-glutamine were added simultaneously to start the last steps of enzymatic synthesis. After approximately 150 s, a decrease in absorbance was observed at 300 nm, indicating that PR-FAR was converted by MtHISN4 into 2R,3S-IGP ( Supplementary Figure S4C ). The mixture was incubated for 5 minutes on ice with 300 µL of Ni-NTA resin (GE Healthcare) and centrifuged at 2000 × g for 3 minutes. Notably, we tried purification using membrane filters to remove enzymes and potential intermediates, but we did not detect eIGP in the flowthrough, suggesting that eIGP was captured by the membrane. The resin was pre-equilibrated in binding buffer (Tris-HCl 50 mM, pH 8.0; NaCl 500 mM, imidazole 20 mM, TCEP 1 mM, 10% glycerol). The supernatant containing IGP was aspirated, transferred into a fresh Eppendorf tube, and kept on ice. The concentration of 2R,3S-IGP was assessed using a glutamate assay kit (Sigma Aldrich). Glutamate was formed from glutamine by MtHISN4, therefore its concentration of 1.6 mM stoichiometrically corresponded to a concentration of 2R,3S-IGP, indicating 80% yield of eIGP synthesis.

The kinetics of MtHISN5 were monitored using microcalorimetry (MicroCal PEAQ-ITC and MicroCal iTC200, Malvern). ITC-SIM has been used for this purpose (Wang et al., 2020). The reaction was conducted in SEC buffer (Tris-HCl 40 mM pH 7.8, NaCl 40 mM, MnCl₂ 4 mM, EDTA 0.4 mM) at 30°C. The differential power (DP) of all ITC-SIM experiments was set to 10 µcal/mol and the stirring speed at 650 rpm. For experiments with cIGP (Santa Cruz Biotechnology, sc-218019), the protein in the cell was maintained at a low nanomolar concentration (5-10 nM), and the substrate in the syringe was added in one injection to the final concentration varying from 24 to 235 µM. Measurements were stopped when the baseline returned to the initial state, signaling substrate depletion. For one of the tested setups of the ITC-SIM experiment, a blank experiment was performed consisting of one injection of 30 µL of 0.5 mM cIGP into the cell containing the buffer, and the observation time was set to the maximum for the apparatus (10 000 s) to check the possibility of heat effect derived from the non-enzymatic substrate degradation. Additionally, an inverted ITC-SIM system was introduced to achieve a higher cIGP concentration (by circumventing the problem of the high heat of substrate dilution), where the substrate (in the cell) was kept at 0.8- and 2-mM concentrations and concentrated enzyme in the syringe (capped with 3 µL of the buffer to prevent the early leakage of the enzyme) was injected in 1 portion into the cell at a final concentration of 12 and 15 nM, respectively.

For the ITC-SIM experiments on eIGP, the substrate was concentrated using a SpeedVac. We maintained the concentration of eIGP in the cell in the range of 1.9-3.4 mM and injected the concentrated MtHISN5 to its final concentration of 792-901 nM (a 3.8 µL aliquot of 42-48 µM enzyme, subunit concentration). Measurements were stopped ~10 min after the baseline returned to near the initial state, signaling total substrate depletion. The DP baseline was analyzed after the initial DP drop (coming from the buffer mismatch and initial protein-substrate interaction) and fitted to the ‘Enzyme Kinetics – Single Injection’ model within the MicroCal PEAQ-ITC analysis software. Briefly, the raw data were transformed into reaction rates and IGP concentrations and fitted to the Michaelis-Menten equation. The final kinetic parameters were calculated by averaging the values obtained from the three separate experiments.