All chemicals were purchased from Sigma-Aldrich (Merck) unless specified otherwise.

The coding sequence for wild-type human PLOD3 gene (GenBank accession number BC011674.2) was obtained from Source Bioscience. Oligonucleotides containing in-frame 5′-BamHI and 3′-NotI were designed and used to sub-clone the coding sequence devoid of the N-terminal signal peptide into a pCR8 vector, that was also used as a template for subsequent experiments. Single-point mutations were generated using Phusion Site Directed Mutagenesis (Invitrogen) with the oligonucleotides listed in Table 1. The linear mutagenized plasmids were phosphorylated using T4 polynucleotide kinase (Invitrogen) prior to ligation using T4 DNA ligase (Invitrogen). All plasmids were checked by Sanger sequencing prior to cloning into the pUPE.106.08 expression vector. This expression vector, kindly provided by U-protein Express, BV (U-PE, Netherlands) provides the N-terminal cystatin signal peptide, followed by a N-terminal 6xHis-tag and a recognition site for Tobacco Etch Virus (TEV) protease prior to the in-frame BamHI-NotI restriction cassette, followed by an in-frame stop codon.

Recombinant tagged LH3/PLOD3 were produced using suspension cultures of HEK293F (Invitrogen) cells, maintained and transfected according to (Faravelli et al., 2021). Cells were not authenticated and not tested for mycoplasma contamination. Briefly, cells were transfected at cell densities of 1 million/ml using 3 μg of polyethyleneimine (PEI; Polysciences) for 1 μg of pUPE.106.08-LH3/PLOD3 plasmid DNA per mL of cells. Cultures were supplemented with 0.6% Primatone RL 4 h after transfection. The cell medium containing secreted LH3/PLOD3 was collected 6 days after transfection by centrifugation at 1,000 × g for 15 min.

The LH3/PLOD3-containing medium from HEK293F cell cultures was filtered through a syringe 0.8 μm filter (Sartorius). The pH and ionic strength of the filtered medium were adjusted using a 5X concentrated buffer stock to reach a final concentration of 25 mM 4-(2-hydroxyethyl)−1-piperazineethanesulfonic acid (HEPES)/NaOH, 500 mM NaCl, 30 mM imidazole, pH 8.0. Recombinant LH3/PLOD3 was purified using a combination of affinity and size-exclusion chromatography on Äkta systems (GE Healthcare) according to (Scietti et al., 2018). The filtered supernatant was first loaded onto a 20 ml His-Prep FF column (GE Healthcare) and eluted using 250 mM imidazole. The eluate was then loaded onto a 5 ml HiPrep desalting FF column (GE Healthcare) equilibrated in 25 mM HEPES/NaOH, 500 mM NaCl, pH 8.0. The N-terminal His-tag was cleaved using overnight His-tagged TEV protease digestion at 4°C followed by affinity-based removal of TEV protease and the cleaved His-tag using a 5 ml HisTrap FF (GE Healthcare). Recombinant LH3 was concentrated to 5 mg/ml using 30,000 MWCO Vivaspin Turbo centrifugal filters (Sartorius), then loaded onto a Superdex 200 10/300 GL (preparative scale) or onto a Superdex 200 5/150 GL (analytical scale) columns (GE Healthcare) equilibrated with 25 mM HEPES/NaOH, 200 mM NaCl, pH 8.0. LH3/PLOD3-containing fractions as assessed from SDS-PAGE analysis were pooled, concentrated, and stored at −80°C until further usage.

Synthetic collagen peptides were purchased from China Peptides. Peptides tested were ARGIKGIRGFS and GIKGIKGIKGIK sequences (Scietti et al., 2018). 5 μM LH3/PLOD3 was incubated with 50 μM FeCl₂, 100 μM 2-OG, 500 μM ascorbate, 1 mM peptide substrate and 0–500 μM 2,2′-bipyridine. Reactions were allowed to proceed for 3 h at 37°C. 10 μl of each sample were supplemented with 38 μl of Milli-Q water and acidified by addition of 2 μl of formic acid (FA) to reach a final volume of 50 μl, then analyzed on an UHPLC-HRMS/MS system (AB Sciex, United States). LC unit (ExionLC AD) consists of a column oven thermostated at 40°C, an autosampler cooled at 10°C and a binary gradient pump system. MS instrument consists of a high resolution QTOF mass spectrometer (AB Sciex X500B) equipped with a Turbo V Ion source and a Twin Sprayer ESI (electrospray ionization) probe, controlled by SCIEX OS 2.1 software. Peptides were separated by reverse phase (RP) HPLC on a Hypersil Gold C18 column (150 × 2.1 mm, 3 μm particle size, 175 Å pore size, Thermo Fisher Scientific) using a linear gradient (2–50% solvent B in 15 min) in which solvent A consisted of 0.1% aqueous FA and solvent B of acetonitrile (CAN) containing 0.1% FA. Flow rate was 0.2 ml/min. Mass spectra were generated in positive polarity under constant instrumental conditions: ion spray voltage 4,500 V, declustering potential 100 V, curtain gas 30 psi, ion source gas 1 40 psi, ion source gas 2 45 psi, temperature 350°C, collision energy 10 V. Spectra analyses were performed using SCIEX OS 2.1 software. Statistical evaluations based on pair sample comparisons between uncoupled and coupled assay values using Student’s t-test in Prism 7 (Graphpad software).

Reaction mixtures (5 μl total volume) were prepared according to (Scietti et al., 2018) by sequentially adding LH3/PLOD3 at 0.2 mg/ml, 0–1 mM peptide substrate or 4 mg/ml gelatin in water (solubilized through heating denaturation at 95°C for 10 min), 500 μM ascorbate, 100 μM 2-OG, variable concentrations of FeCl₂ (0–200 μM) and let incubate for 1 h at 37°C. Reactions were stopped by heating samples at 95°C for 2 min prior to transfer into Proxiplate white 384-well plates (Perkin-Elmer), then 5 μl of the Succinate-Glo reagent I (Promega) were added and let incubate 1 h at 25°C, after that 10 μl of the Succinate-Glo reagent II (Promega) were added and let incubate 10 min at 25°C. The plates were then transferred into a GloMax Discovery plate reader (Promega) configured according to manufacturer’s instructions for luminescence detection. All experiments were performed in triplicates. Control experiments were performed using identical conditions by selectively removing LH3/PLOD3, 2-OG or peptide substrates. Data were analyzed and plotted using Prism 7 (Graphpad Software). Statistical evaluations based on pair sample comparisons between uncoupled and coupled assay values using Student’s t-test in Prism 7 (Graphpad software).

DSF assays were performed on LH3/PLOD3 wild-type using a Tycho NT.6 instrument (NanoTemper Technologies). LH3/PLOD3 samples at a concentration of 1 mg/ml in a buffer composed of 25 mM HEPES/NaOH, 200 mM NaCl, pH 8. Binding assays were performed by incubating LH3/PLOD3 with variable FeCl₂ and 2,2′-bipyridil concentrations. Data were analyzed and plotted using GraphPad Prism 7 (Graphpad Software).

Recombinant LH3/PLOD3 was subject to labeling using the NHS-RED kit (NanoTemper Technologies) according to manufacturer’s instructions. Labeled LH3/PLOD3 at a concentration of 50 nM was incubated in a buffer composed of 25 mM TRIS/HCl, 100 mM NaCl, pH 7.5 with variable concentrations of FeCl₂ for 40 min. The samples were then transferred into Dianthus 384-well plates for Temperature-Related Intensity Change (TRIC) (NanoTemper Technologies), and centrifuged at 1,000 g for 2 min. TRIC measurements were performed immediately after centrifugation using a Dianthus NT.23 instrument (NanoTemper Technologies). The samples were first measured for 1 s without heating and for 5 s with the IR-laser turned on. Normalized fluorescence values (F _norm ), described as ratios between fluorescence values after and prior to infrared laser activation were collected and plotted as a function of ligand concentration (Schulte et al., 2021). Determination of binding affinities was carried out using the DI.Screening Analysis software (NanoTemper Technologies). Data were then exported and plotted using GraphPad Prism 7 (Graphpad Software).

Simulations were started from the extrapolation of the dimeric LH domain of human LH3/PLOD3 from its experimental crystal structure (PDB 6FXR) (Scietti et al., 2018) using COOT (Emsley et al., 2010); two systems were prepared: in one system both Fe²⁺ cations were left (LH3_Fe2), while in the other one only the catalytic Fe²⁺ was left in (LH3_Fe1). Residues were modeled in their standard protonation states at physiological pH, as predicted by PROPKA, version 3.1 (Sondergaard et al., 2011): this resulted in one disulfide bridge (between Cys 563 and 698), histidines 546, 586, 643, 681, 717 being protonated on Nε2, and histidines 595, 613, 667, 711, 719 protonated on Nδ1 (comprising histidines in both Fe²⁺ binding sites). Crystallographic waters were taken from the published crystal structure of LH3/PLOD3 (PDB: 6FXR) (Scietti et al., 2018). Hydrogen atoms were introduced using the tleap utility in AmberTools (version 19) (Case et al., 2005), as well as –NH₃ ⁺ and –COO^– caps at the N- and C-termini, respectively. All molecular dynamics simulations (MD) were carried out with the AMBER software package (version 18) (Case et al., 2005; Case et al., 2018), using its GPU-accelerated (Salomon-Ferrer et al., 2013) pmemd.cuda utility during equilibration and production, and sander otherwise; three independent MD replicas (different random seeds) were carried out for LH3_Fe1 and LH3_Fe2 alike. A 8.0 Å cutoff was applied for the calculation of Lennard-Jones and Coulomb interactions between nonbonded atoms; beyond this limit, only Coulomb interactions were computed, using the particle mesh Ewald approach (Darden et al., 1993). Each replica’s production stage was 1 μs in length, and conducted with a 2 fs time-step in the NpT ensemble (with a temperature of 300 K enforced via Langevin’s thermostat (Loncharich et al., 1992); collision frequency 1 ps^–1, and a 1 atm pressure enforced by Berendsen’s barostat) (Berendsen et al., 1984). Preproduction stages for each replica consisted in minimization (10 steps of steepest descent + 290 steps of conjugate gradient); heating (20 ps; NpT; 25–300 K; increasingly softer harmonic restraints on Cα atoms; k = 5.0 kcal mol^–1 Å, collision frequency 0.75 ps^–1 , with 2 fs time-step); and equilibration (1.0 ns; NpT; 300 K collision frequency 1 ps^–1 , with 2 fs time-step). Analyses of MD trajectories were carried out with the CPPTRAJ program distributed within the AmberTools suite (version 19) (Case et al., 2005) or with code written in-house.

To characterize the impact of the second Fe²⁺ on the internal dynamics of LH domain of LH3, we made use of the previously introduced distance fluctuation (DF) analysis (Morra et al., 2012; Moroni et al., 2018). For each MD trajectory of the two systems, we computed on the combined meta trajectory the matrix of distance fluctuations, in which each element of the matrix corresponds to the DF parameter. DF is defined, for a couple of amino acids i and j, as the variance of the time-dependent distance d _ij of the C α atoms: D F i j   = 〈 ( d i j   −   〈 d i j   〉 ) 2 〉 where the brackets indicate the time-average over the trajectory. This parameter is invariant under translations and rotations of the molecules and, unlike the covariance matrix, does not depend on the choice of a particular protein reference structure. The resulting DF matrix can be used to assess the intrinsic flexibility of proteins. This parameter characterizes residues that move in a highly coordinated fashion, and it is actually able to reflect the presence of specific coordination patterns and quasi-rigid domains motion in the protein of interest. In particular, pairs of amino acids belonging to the same quasi-rigid domain are associated with small distance fluctuations and vice versa.

Lennard-Jones and intramolecular bonded parameters for the 2-oxoglutaric acid cofactor (2-OG) in the catalytic site of LH3/PLOD3 were assigned according to the generalized Amber forcefield (GAFF) (Wang et al., 2004), using AmberTools’ antechamber and parmchk2 utilities (Case et al., 2005), after adding methyl hydrogens using the reduce tool (Case et al., 2005). Assignment of charges and (intermolecular) 2-OG-Fe²⁺ bonded parameters were performed as discussed below. Parametrization of both Fe²⁺ binding sites was carried out using the MCPB.py utility (Li and Merz, 2016) in conjunction with density functional theory calculations (DFT) using the Gaussian09 suite (Frisch et al., 2009). Intermolecular bonded parameters for all residues in both Fe²⁺-binding sites and for the 2-OG cofactor were derived by MCPB.py (Li and Merz, 2016) applying the Seminario method (Seminario, 1996) on a DFT-derived Hessian matrix. More specifically, this Hessian matrix was calculated at the B3LYP (Lee et al., 1988; Becke, 1993)/6-31G(d) level of theory after optimizing a “small” model of both binding sites at the same level to a confirmed minimum (no imaginary frequencies). Such “small” models (generated by MCPB.py (Li and Merz, 2016)) included: 1) the Fe²⁺ cation in that particular binding site; 2) binding site residue sidechains up to their Cβ, with a shorter Cβ-H bond replacing Cα-Cβ; and 3) in the catalytic site, the entire 2-OG cofactor. Residues’ Cβ atoms and 2-OG’s carboxylate oxygens farthest from Fe²⁺ were frozen during optimization and excluded from frequency calculations. Atomic point charges on both Fe²⁺ cations on individual binding site residues and on the entire 2-OG cofactor were fitted by MCPB.py (Li and Merz, 2016) using the RESP method (Bayly et al., 1993) based on the outcome of (single-point) ESP charge fitting calculations (Besler et al., 1990) at a higher DFT level, on a “large” version of each Fe²⁺ binding site. More specifically, the chosen level of DFT for ESP charge fitting (Besler et al., 1990) was B3LYP (Lee et al., 1988; Becke, 1993)/6-31G(d)/def2-SV(P), (Weigend and Ahlrichs, 2005) with the def2-SV(P) basis set specifically applied to Fe²⁺. “Large” binding site models (generated by MCPB.py (Li and Merz, 2016)) comprised Fe²⁺, 2-OG, binding site residues in their entirety (i.e., with their backbone) as well as backbones of Glu596 and His668, contiguous to His595/Asp597 and His667/Asp669, respectively. All contiguous backbone fragments in the large models are capped by acetyl and N-methyl moieties at their N- and C-termini, respectively. Gaussian09 (Frisch et al., 2009) was programmed to perform ESP charge fitting (Besler et al., 1990) over 10 spherical shells around each atom, with 17 grid points per square Bohr. As per MCPB.py’s default (Li and Merz, 2016), the atomic radius of both Fe²⁺ cations was taken to be 1.409 Å. In all DFT calculations, Fe²⁺ centers in both the catalytic and noncatalytic binding sites were modeled in their quintet state, after comparative optimizations of each site with Fe²⁺ in the triplet and singlet state confirmed—at the B3LYP (Lee et al., 1988; Becke, 1993)/6-31G(d) level of theory—that the quintet state is the most energetically stable in both cases (data not published). All remaining LH3/PLOD3 residues—including intra-residue bonded parameters for residues in both Fe²⁺-binding sites—were treated with the ff14SB forcefield (Maier et al., 2015), whereas Na⁺ countercations were modeled with parameters by Joung and Cheatham (Joung and Cheatham, 2008): these are compatible with the TIP3P model in use for water (Jorgensen et al., 1983).

The presence of amino acid networks shared by Fe²⁺, 2-OG-dependent dioxygenase enzymes in their catalytic sites provides a general template for structure-based design of potential inhibitors. The analysis of the residues surrounding the catalytic Fe²⁺ and the 2-OG indeed supported the possibility that 2-OG analogs may act as competitive inhibitors of the co-substrate molecule (Rose et al., 2011). In human LH3/PLOD3, the catalytic Fe²⁺ is strongly coordinated by the side chains of His667, Asp669, and His719 of the DSBH fold, and by the 2-OG co-substrate, capped towards the outer solvent by a flexible capping loop defined by residues 590-610 (Figure 2). Using a combination of nano-differential scanning fluorimetry (nanoDSF) and luminescence-based activity assays, we screened a small, focused library of 2-OG analogs (Table 2) searching for compounds capable of inhibiting LH activity. After thorough testing, none of the compounds tested showed binding/folding stabilization in nanoDSF, nor inhibition of enzymatic conversion of 2-OG into succinate. Likewise, a custom-designed library of compounds selected through in silico virtual screening of specific candidate binders of the LH3/PLOD3 catalytic site did not provide suitable hints for inhibitors of LH activity using these assays.

Intrigued by the recalcitrance to inhibition of the LH catalytic site by 2-OG analogs, we focused our attention to the distinguishing features displayed by this domain when compared to homologous Fe²⁺, 2-OG-dependent dioxygenases, and in particular to the capping loop and the stable conformation adopted in the presence of excess [Fe²⁺]. This interlocked state may indeed constitute an obstacle when dealing with inhibition of the LH catalytic site, and the relative positioning of the capping loop is crucial for inhibitor accessibility to the active pocket. We decided to perform a thorough investigation of the flexibility of this loop in silico and with site-directed mutagenesis in vitro to validate the possible significance of the second Fe²⁺ binding site prior to attempting to quantitatively probe the specific metal ion binding to the two distinct LH3/PLOD sites within the LH domain.

To investigate the impact of the second Fe²⁺ on the structural conformation of the capping loop 590-610, the C-terminal LH domain involved in LH/PLOD dimerization underwent all-atom MD simulation in explicit solvent, with and without the second Fe²⁺ ion. In the following, we refer to the simulated system with the second Fe²⁺ as LH3_Fe2, while the system simulated without it has been named LH3_Fe1. Three independent replicate simulations were carried out for the two systems, each 1 μs long. In each independent replicate we used identical simulation parameters (see Material and Methods), varying only the initial velocities of atoms via random assignments from a Maxwell distribution consistent with the required temperature. Visual inspection of MD simulations shows that both systems are characterized by minimal atom fluctuations, suggesting that this fragment of the enzyme is stiff and allows for minimal protein motions away from the starting (crystal) structure, except for the gate loop and some protein region at the dimer interface.

The first step in the analysis of the dynamics of LH3/PLOD3 C-terminus and its potential variation as a function of the presence/absence of the second Fe²⁺, was the identification of protein regions displaying higher levels of local flexibility during dynamics, through the computation of the local fluctuations (LF). This parameter detects the flexibility of a given residue with respect to the neighboring amino acids; the comparison of this calculation for the two systems (i.e., LH3_Fe1 and LH3_Fe2), allows to extrapolate variations in protein regions more affected by the presence/absence of the second Fe²⁺. LF is calculated for residue i, as the mean of the variances of the distance d _ij of the C_α (i) and C_α (j) atoms, for j = i-2, i-2, i+1, i+2. In the LH3_Fe1 system residues 593-611 of the gate loop of both monomers displayed higher flexibility as compared to the LH3_Fe2 system (Figures 3A,B). The peak at residues 640-646, constituted by residues of one monomer which contact the gate loop of the other monomer, is more defined in the LH3_Fe1 system (Figures 3A,B). To quantify the observed differences in the dynamics of the enzyme during MD in the two simulated systems, we calculate the root-mean-square deviation (RMSD) distributions of MD trajectories, after optimal rigid body superposition of backbone atoms of each snapshot with the corresponding atoms of the crystal structure, where the gate loop is in the closed conformation.

The RMSD distributions of the loop 590-610 in the two monomers of the single trajectories are reported in Figures 3C,D, showing the different behavior of the two systems and the variations in the single trajectories. A threshold of 3 Å was used to distinguish the closed from the open conformation, based on visual inspection of MD simulations and superposition of snapshots representing the two states. These plots show that the gate loop is stabilized in the closed conformation in the LH3_Fe2, while the absence of second Fe²⁺ impacts the stability of the loop, shifting the population of LH3_Fe1 towards the open conformation, even though both systems can visit the two states. To understand the impact of the second Fe²⁺ on the internal dynamics of LH3/PLOD3, we performed the distance fluctuation analysis on the meta trajectories of the two systems, obtained concatenating the three MD replicas of each system. The DF parameter, which is the variance of the inter-residue distance d_ij, was calculated for any pair of residues during the trajectory (see Material and Methods). The resulting DF matrices calculated for LH3_Fe1 and LH3_Fe2 shown in Figures 3E,F, report on the fluctuation of the inter-residue distance in the corresponding residue pairs and matrix regions, describing the intrinsic flexibility and coordination of the LH domains. Relatively low DF values identify protein regions that move together, in coordination. The comparison of these matrices can be used to evaluate possible changes of the internal dynamics and coordination due to the presence/absence of the second Fe²⁺.

Overall, the matrices for both systems turned out to be similar, with largely overlapping patterns of small and large fluctuations of inter-residue distances. In both systems, the internal dynamics of the two domains is characterized by small atomic fluctuations, confirming that protein structure is stiff. Indeed, the highest DF parameters in both systems correspond to protein regions that do not adopt well-ordered 3D structures, which are intrinsically more flexible. Generally, the inter-domain motion is characterized by higher fluctuations than the internal domain motion. In particular, in the LH3_Fe1 system, residues 650-738 of one monomer, corresponding to the beta-sheet core, move in a more coordinated way with the other monomer than the rest of the protein (Figure 3F).

The significance of the differences observed between the two DF matrices was evaluated with a statistical analysis based on F-test. We used LH3_Fe2 as a reference state for comparing the two matrices. Red dots in the LH3_Fe1 system (Figure 3F) correspond to residues which show significant difference in DF values in respect to the corresponding DF values in LH3_Fe2, according to this test. This analysis highlights that relevant differences in protein motion for the two systems concern residues of the gate loop 590-610 and residues 640-650 forming the loop that connects the alpha helix 618-639 with the beta-sheet core of the protein. This loop partially forms the dimerization interface of the two domains, as well as residues 694-702, which displays higher DF values in the LH3_Fe1 system (Figures 3F,G).

To further investigate the role of the relative positioning of the capping loop for the accessibility to the active pocket depending on the second Fe²⁺ ion, we computed the part of the van der Waals surface of residues forming the catalytic site (protein residues within 6 Å of 2-OG and Fe²⁺) that are accessible to solvent, that is the solvent accessible surface area, across each frame of MD trajectories, using the VMD program (Humphrey et al., 1996). This analysis gives a measure of the capability of the loop in regulating the access of the substrate to the catalytic site, depending on the presence of the second Fe2+. Figure 4 shows the distribution of the solvent accessible surface area calculated for the three replicas, for each system and monomer. In both monomers the distributions of the LH3_Fe1 system are shifted towards slightly higher values compared to the LH3_Fe2, suggesting that the presence of the Fe²⁺ has an impact on the access to the active site by modulating the opening/closure of the loop, even though both systems can visit the two distinct states, in agreement with the RMSD analysis discussed above.

To inspect the effect of the second Fe²⁺ on the dynamics of the capping loop at a finer level, we evaluated the persistence over the simulation time of the native and new interactions established by this loop with the rest of the protein. The results of this analysis show that most of the native contacts are maintained during the simulation time in both systems (Supplementary Figure S1A,B), while in LH3_Fe1 some residues of the loop tend to weaken some of these interactions compared to LH3_Fe2. As for new stable contacts, we observed that in LH3_Fe1 the loop can’t form new interactions with the rest of the protein due to its higher mobility, while in the LH3_Fe2 the loop establishes new steady interactions with residues 643-648 of the other monomer (Supplementary Figure S1C).

As mentioned above, in the crystal structure of the LH domain obtained in the presence of excess [Fe²⁺], residue Arg599 belonging to the loop 590-610 forms a salt bridge with the 2-OG co-substrate, yielding a conformation that mimics the collagen lysine substrate (Scietti et al., 2018) (Figures 2C,D). To check the impact of the presence/absence of the second Fe²⁺ on the stabilization of this bond, we tracked the hydrogen bonds between Arg599 and 2-OG co-substrate over the course of the trajectories. This analysis showed that in most of the conformations visited during the dynamics, Arg599 forms a salt-bridge with 2-OG in both systems, with a in slight prevalence in LH3_Fe2 (Supplementary Figure S1D–F). Taken together, the results from the in silico analysis corroborate the observation that the second Fe²⁺ constrains the LH domain in a tightly interlocked conformation with reduced conformational motility in proximity to the enzyme’s catalytic site.

To better examine the impact of the second Fe²⁺-binding site on enzymatic activity and substrate accessibility, we generated the Asp597Ala and Asp611Ala variants of human LH3/PLOD3 using site-directed mutagenesis. In particular, we focused on Asp597 as a critical capping loop residue involved in the coordination of the second Fe²⁺, whereas Asp611 could represent its counterpart as central to the Fe²⁺ coordination platform on the surface of the LH domain (Figures 2C,D). Both mutants could be expressed and purified to homogeneity, and showed yields, folding stability and oligomerization states comparable to wild-type LH3/PLOD3 (Supplementary Figure S2). We probed their Fe²⁺ binding affinity using Temperature-Related Intensity Change (TRIC) and found that both mutants showed 4-times lower affinity compared to wild-type LH3/PLOD3 (Supplementary Figure S3). We were not surprised of the weak binding observed: the catalytic Fe²⁺ is known to have structural roles within the LH domain and cannot be removed to probe its binding (Guo et al., 2018; Scietti et al., 2018), therefore we assumed that the binding data collected exclusively refer to possible additional Fe²⁺ binding sites. When tested for their enzymatic activity in the presence of acceptor substrates such as gelatin (i.e., coupled activity), the mutants showed slightly higher (i.e., 1.5X) activity than their wild-type counterpart (Figure 5A, right). Nevertheless, we consistently observed a much higher (i.e., 3-5X) degree of 2-OG conversion into succinate without the need of an acceptor substrate (i.e., uncoupled activity, Figure 5A, left). Prompted by this observation, we decided to investigate the modulatory effect of [Fe²⁺] on both uncoupled and coupled enzymatic activities of these mutants. We found that the uncoupled activities of both LH3/PLOD3 Asp597Ala and Asp611Ala mutants were less inhibited by high [Fe²⁺] compared to wild-type LH3/PLOD3 (Figure 5B, left), whereas the modulation of the coupled enzymatic activities was almost unaffected by the presence of the point mutations (Figure 5B, right). Based on the combined results obtained from the MD simulations and mutagenesis, we reasoned that binding of a second metal ion on the surface of the LH domain induces a conformationally stable, self-inhibited state which in turn reduces the ability of LH/PLOD enzymes to process 2-OG into succinate via uncoupled activity.

Given the unique presence of two distinct Fe²⁺ binding sites in close proximity, but with opposite effects on enzymatic activity, we wondered whether small-molecule inhibitors acting through metal ion chelation could efficiently modulate substrate processing in LH/PLOD enzymes. To date, despite large scale screenings identified potential hits (Devkota et al., 2019), specific inhibitors of the lysyl hydroxylase activity are missing. The only known inhibitor of LH/PLOD enzymes is the 2,2’-bipyridine (BPY), a non-specific inhibitor that act as chelating agent. BPY chemical structure is characterized by two pyridyl rings, heterocyclic chemical moieties containing nitrogens (Figure 6A). Although the mechanism of action of BPY on LH/PLOD enzymes is not well characterized, being a metal chelator, it likely acts on the Fe²⁺ ion present in the lysyl hydroxylase domain (Ikeda et al., 1994; Rose et al., 2011; Vasta and Raines, 2015; Jover et al., 2018). We therefore decided to explore the actual impact of BPY inhibition on human LH3/PLOD3 enzymatic activity. Firstly, we attempted to evaluate concentration-dependent inhibition using luminescence-based assays, but we realized that usage of BPY was not compatible with the assay setup, likely due to inhibition of the Mg²⁺-dependent luciferase reaction by the reagent. We therefore focused on a strategy to directly investigate hydroxylysine formation on synthetic collagen peptides using mass spectrometry (MS) by evaluating the relative ratios between non-hydroxylated and hydroxylated lysine side chains (Figure 6B). Using this method, we could perform accurate quantitation of the dose-dependent effects of BPY on LH3/PLOD3 enzymatic activity.

In absence of BPY, collagen peptide substrate processing was consistent with previously reported data (Scietti et al., 2018). Given a BPY:Fe²⁺ chelation stoichiometry of 3:1, and a half-maximal BPY concentration required to form complexes with 20 μM Fe²⁺(Fe₂₀–EC₅₀) experimentally determined in about 40 μM (Vasta and Raines, 2015), we would have expected an initial reduction of LH activity at a [Fe^{2 +}]/[BPY] ratio of 2:1. Surprisingly, assays performed with concentrations of BPY up to 25 μM unexpectedly revealed increasing Hyl/Lys ratios, suggesting enhancement of enzymatic activity, whereas only higher BPY concentrations caused the expected dramatic decay in substrate processing (Figures 6B,C). Considering that standard assays are carried out by supplementing a fixed concentration of Fe²⁺ (i.e., 50 μM), we wondered whether the activity boost observed at relatively low BPY concentrations could be associated to Fe²⁺ sequestration from the non-catalytic site on capping loop, whereas inhibition could be caused by chelation of the catalytic Fe²⁺.

We therefore carried out additional assays, in which we simultaneously varied [Fe²⁺] and [BPY]. Our results (Figure 6D, Supplementary Figure S4) showed that the LH3/PLOD3 catalytic activity always reaches its maximum at [Fe²⁺]/[BPY] ratio of 2:1, supporting a fine balance between sequestration of excess inhibitory metal ions bound to the capping loop and inhibition through chelation of Fe²⁺ in the catalytic site. These results suggest that usage of BPY as a lead compound for the development of inhibitors of LH/PLOD enzymatic activity demands particular care, as unexpected concentration-dependent opposite effects may be produced through release of the self-inhibitory, Fe²⁺-dependent capping loop.