In silico Insights into the Dimer Structure and Deiodinase Activity of Type III Iodothyronine Deiodinase from Bioinformatics, Molecular Dynamics Simulations, and ONIOM Calculations

Abstract

The homodimeric family of iodothyronine deiodinases (Dios) regioselectively remove iodine from thyroid hormones. Currently, structural data has only been reported for monomer of the monomeric mus type III catalytic thioredoxin (Trx) fold domain (Dio3^Trx), but the mode of dimerization has not yet been determined. Various groups have proposed dimer interfaces that are similar in structure to the A-type and B-type dimerization descriptions of peroxiredoxins. Computational methods are used to compare the sequence of Dio3^Trx to related proteins known to form A-type and B-type dimers. Sequence analysis and in silico protein-protein docking methods suggest that Dio3^Trx is more consistent with proteins that adopt B-type dimerization. Molecular dynamics (MD) simulations of the refined Dio3^Trx dimer constructed using the SymmDock and GalaxyRefineComplex databases indicate that stable dimer formation along the β₄α₃ interface consistent with other Trx fold B-type dimers. MMGBSA calculations show that the dimer is stabilized by interdimer interactions between the β-sheets and α-helices. A comparison of MD simulations of the apo and thyroxine-bound dimers suggests that the active site binding pocket is not affected by dimerization. Determination of the transition state for deiodination of thyroxine from the monomer structure using ONIOM methods provides an activation barrier consistent with previous small model DFT studies.

1. INTRODUCTION

The iodothyronine deiodinase (Dio) family of selenoproteins controls thyroid hormone (TH) levels through selective deiodination.^1–14 The crystal structure of the mouse Dio3 catalytic domain (PDB = 4TR4)¹⁵ confirms the earlier predictions of thioredoxin (Trx) fold of five β-strands flanked by four α-helices (Figure 1A).^16–20 Unlike other Trx proteins, which contain a CXXC motif in the active site, Dios incorporate a SCTU motif (U = Sec) replacing cysteine with the rare selenocysteine residue. Dio Type 1 can deiodinate at either the inner or outer ring of the thyroid hormones, but Dio 2 (Dio2) and 3 (Dio3) are regioselective for outer ring deiodination (ORD) and inner ring deiodination (IRD), respectively. ORD by Dio2 activates thyroxine (T₄) while IRD by Dio3 produces inactive TH metabolites.^{3,6,12,21–24}

Figure 1.

Potential dimer interfaces for Dio based upon Prx. (A) Scheme of Dio3’ thioredoxin fold domain indicating potential dimerization interfaces, (B) Representative structure of an oligomeric Prx: plasmodium vivax (PDB = 4L0U), (C) Close-up of the A-type and B-type dimer interfaces in the Prx1a decamer highlighting the A- and B-type interfaces.

The Trx domain of Dio3 bears some structural similarity to the peroxiredoxin family through the insertion of a β_N1β_N2θ motif at the N-terminus and a α_Dβ_D motif in the β₂α₂ turn. Dio differs from peroxiredoxins (Prx) in the length of the β₂α_D loop which is short in Prx, but consists of a flexible 20-residue Ω-loop adjacent to the active site.²⁵ In molecular dynamics (MD) simulations, a key loop conformation exposes a pocket to the active site where T₄ can undergo deiodination by Sec. Halogen bonding to the Asp211 carboxylate of the Ω-loop contributes to selectivity for IRD.²⁶ Deiodinase activity can be inhibited by endocrine disruptors such as polybrominated diphenyl ethers, polychlorinated biphenyls, dibenzodioxins, iodoacetic acid, and gold thioglucose.^27–30

Direct structural data is challenging for the active membrane-bound Dio homodimer, because both the Trx fold and transmembrane (TM) domain is necessary for deiodinase activity.^31,32 Sagar et al. superimposed Dio2 onto the X-ray structure of oxidized human thioredoxin (PDB = 1ERU) and assumed that the Ω-loop folds into an inter-strand β-sheet to create an active site cavity.^31,33,34 More recently, Schweizer et al. proposed that Dio3 dimerizes at the α₃β₄ interface¹ analogous to 2-Cys peroxiredoxin (i.e., HBP23, PDB = 1QQ2).³⁵ Although the transmembrane domain is essential for dimerization, an active Dio, partially active proteins can be formed from a full-length monomer paired with one lacking the membrane-bound domain.³¹ These results suggest that dimerization of the Trx domains could be sufficient for catalysis to occur, but that dimer formation is challenging in the absence of the TM domain.

The Dio3 structure is comparable to Trx fold proteins that form dimers or oligomers (e.g., glutaredoxins (Grxs), protein disulfide bond isomerases, and peroxiredoxins (Prxs).³⁶ Prx form (α₂)₅ decamers (e.g., plasmodium vivax Prx1a (PDB = 4L0U).³⁷ Oxidized Prx can dissociate to homodimers along either the A- or B-type decamer interface, which produces either B-type or A-type dimers, respectively (Figure 1B, 1C).³⁸ The A-type (or “alternate”) dimer interface occurs near the α₂ and β₁α₁ turn surface. Human Trx (1ERU) homodimerizes at a similar surface. B-type (for β-strand) dimer interfaces consist of interactions of α₃ and β₄, the latter forming an intermonomer β-sheet.³⁹ Prx are further classified into different subfamilies (Prx1, Prx5, Prx6, Tpx) with each preferring A-type (Prx5 and Tpx) or B-type (Prx1 and Prx6) dimerization.³⁹

Bioinformatic tools and molecular dynamics (MD) simulations are used to predict a homodimer structure for the Dio3 Trx catalytic domain. Sequence and structure comparisons are used to narrow the potential interface based upon similar Trx fold proteins.⁴⁰ A refined symmetric dimer was subjected to MD simulations to test for dimer stability. Hydrogen bonding analysis and MMGBSA calculations were performed to determine the key residues and interactions at the proposed dimer interface. Finally, hybrid quantum-mechanical/molecular mechanical (QM/MM) calculations of the activation barrier for T₄ deiodination were performed and found to be consistent with previous small molecule DFT studies.

2. METHODS

Sequence alignments of selected Trx fold proteins were performed with the Clustal Omega server using FASTA sequences obtained from the Protein Databank (1ERU,³⁴ 4L0U,³⁷ 1XIY,³⁸ 1OC3,⁴¹ 3I43,⁴² 1Q98,⁴³ 2P5R,⁴⁴ 4KCE,⁴⁵ 1QQ2,⁴⁶ 2C0D⁴⁷), corrected for mutations, and truncated prior to the β_N1β_N2θ insertion motif (Figures 2 and 3).⁴⁸ Online databases for protein-protein docking have been developed over the past few decades with various algorithms for predicting dimer structures.^49–51 Monomers of the Dio3 truncated at Asn137 and capped at the N-terminus with an acetyl group (Dio3^Trx) were submitted to template-based (GalaxyGemini) and template-free (HawkDock) servers.^26,52 Both the X-ray structure and the T₄-bound structure containing the stable Ω-loop conformation determined in a previous study²⁶ were used for dimerization attempts. Template-based modeling assumes protein-protein complexes based upon known protein structures in the Protein Data Bank if the interacting pairs share over 30% sequence identity.⁵³ Template-free docking databases generates sets of structures that are subsequently scored and ranked based upon known interfacial interactions.⁵⁴ Preliminary dimers suitable for production runs were predicted using SymmDock, which uses a geometry-based algorithm to produce symmetric multimers with a given order n.⁴⁰ Residues within the β₄-sheet (Thr268 to Gln272) were treated as part of the interface and the distance between the midpoint residue Met270 of the β-sheets, was constrained to 0–7 Å. GalaxyRefineComplex was used to resolve the close contacts at the interface of the best SymmDock-produced dimer by symmetric repacking the sidechain residues.⁵⁵ GalaxyRefineComplex uses a sampling method in which short MD simulations are run to facilitate protein-protein dimerization by allowing the sidechains to dictate inter-protein orientations and intra-protein backbone conformations.⁵⁵ This refining process has been shown to produce the most favorable structures in comparative studies.^50,56,57 Refined structures are evaluated according to the critical assessment of predicted interactions (CAPRI), a widely accepted measure of structure quality that inspects generated structures for unfavorable interactions, such as close positively charged sidechains.⁵⁷

Figure 2.

Multiple sequence alignment of mus Dio3 relative to A-type dimer proteins. Key secondary structures of the Trx fold are indicated with the dimer interface residues highlighted in yellow. The locations of the conserved Phe residues in ‘ball-and-socket’ interactions are indicated with stars.

Figure 3.

Multiple sequence alignment of mus Dio3 relative to A-type dimer proteins. Key secondary structures of the Trx fold are indicated with the dimer interface residues highlighted in yellow. The location of the resolving Cys for Prx is indicated with a star.

MD simulations were performed using the ff99sb force field⁵⁸ in AMBER 16 and the PMEMD GPU routines.^59–62 The initial Dio3^Trx monomer structure was taken from the previous study,²⁶ specifically the relaxed T₄-bound conformation C (see Figs 3 and 4 of reference 26). Parameters for Sec selenolate and the Sec-T₄ complex were used as described in a previous study.²⁶ Specifically, the strong I···Se XB interaction between Sec and an inner ring iodine was represented as a covalent bond.^63,64 Additionally, charged dummy atoms on the ancillary iodines were used to account for weaker XB electrostatic interactions with regions of positive electrostatic potential (σ-holes).⁶⁵ The protonation state of the monomers were determined by H++^66,67 as previously described.²⁶ The initial dimer models obtained from GalaxyRefineComplex were solvated in a 10 Å octahedral box of TIP3P water.⁶⁸ Six Na⁺ counterions were required to neutralize the system. The solvent was allowed to relax around the protein in stages that incrementally released a constraint on the protein structure over 12000 steps. The relaxed dimer models were warmed to 300 K with the protein constrained followed by equilibration in the NVT ensemble using Langevin dynamics in stages that incrementally released constraints upon the dimer interface region. Systems were further equilibrated in the NPT ensemble prior to 10 successive 100ns production runs (timestep = 1 fs, total production simulation time = 1μS). The AmberTools suite was used to evaluate the energetics, the hydrophobic contacts, and key hydrogen bonding interactions within the dimer.⁶⁹ Bond lengths involving hydrogen were constrained using SHAKE.⁷⁰ Molecular mechanics generalized Born surface area (MMGBSA) and Poisson-Boltzmann surface area (MMPBSA) calculations were performed using the MMPBSA.py Python script to quantify the free energy of dimerization and the energetic contributions of dimer interface pairwise interactions in the MD simulations.^71,72 These methods have been shown to have high accuracy in predicting dimerization energies and are able to identify energetic contributions of individual residues through decomposition analysis.^73,74 The free energy of formation of the solvated dimer can be calculated as

$$\mathrm{\Delta }{G}_{dimer,solv}^0=\mathrm{\Delta }{G}_{dimer,vacuum}^0+\mathrm{\Delta }{G}_{solv,dimer}^0-2\mathrm{\Delta }{G}_{solv,monomer}^0$$ (1)-

where $\mathrm{\Delta }{G}_{dimer,solv}^0$ is the free energy of the solvated dimer; $\mathrm{\Delta }{G}_{dimer,vacuum}^0$ is the free energy of the dimer in vacuum; $\mathrm{\Delta }{G}_{solv,dimer}^0$ is the solvation free energy of the dimer; and $\mathrm{\Delta }{G}_{solv,monomer}^0$ is the solvation free energy of solvation of the monomer. The MMGBSA and MMPBSA methods calculate free energies of solvation by estimating their electrostatic and nonpolar contributions using the generalized Born or Poisson-Boltzmann equations and an empirically derived correction, respectively.

Two-layer ONIOM⁷⁵ QM/MM method within Gaussian09⁷⁶ was used to determine the deiodination transition state from the monomeric Dio3^Trx-T₄ complex from previous MD simulations.^26,75,77 The ONIOM job preparation, monitoring and result analysis were conducted following the procedure implemented in the TAO package.⁷⁸ Sidechain residues within 5.0 Å of T₄ or within the cryptic pocket that accommodates substrate binding were unconstrained (Ser167-Met174, Arg176, Glu200-Pro203, Trp207, Thr209, Asp211, Ser212-Tyr214, Tyr257-Arg260, Arg275-Asp280, Tyr280-Gln281).²⁶ The sidechain of the active site Sec residue, T₄ and two water molecules participating in a hydrogen bonding network with the T₄ ammonium group were included in the QM region. The AMBER force field implemented in Gaussian 09 was used to describe the MM portion.⁷⁹ The QM portion was represented using density functional theory with the mPW1PW91 functional.⁸⁰ Selenium and iodine were represented by the Wadt-Hay effective core potential basis set augmented with diffuse and polarization functions.⁸¹ All other atoms used a valence triple-ζ basis set with polarization functions.⁸² The initial atomic charges were adapted from a previous study.²⁶ The geometry optimizations using a mechanical embedding scheme and the transition state (TS) search used the quadratic coupled algorithm implemented in the Gaussian package.⁷⁶ Unrefined TS and product structures were re-optimized with ONIOM(mPW1PW91:AMBER).

3. RESULTS AND DISCUSSION

3.1. Comparative sequence analysis of potential dimer interface.

Given the structural similarity of Dio^Trx to Prx, the sequences of a selection of A- and B-type dimers were compared using multiple sequence alignment (Figures 2 and 3).⁴⁸ Dio3 aligns poorly with both A-type and B-type dimers due to the presence of the Ω-loop insertion in the Dio3 sequence. While Sec is conserved with the peroxidic Cys of B-type dimers, this final residue of the SCTU active site motif aligns with the C-terminal residue of the CXXC motif of A-type Trx-type proteins. This result could suggest that the UP(P/S)F Dio sequence at the beginning of α₂ corresponds to the Trx fold CXXC motif and that the up-sequence location of the S(C/A)TU active site in the β₁α₂ turn is to accommodate reaction with the large iodine center. The sequence identity between Dio3 and A-type proteins is low (10–15%) with only 3 conservative substitutions located in β₁ and the turn following β₂ in multiple sequence alignment. In contrast, the Dio3 Trx fold sequence identity with the B-type dimers is slightly higher (15–20%) and close to the ‘twilight zone’ of 20–25%, suggesting that some conserved regions could have evolved from a common ancestor.⁸³

A-type dimers typically conserve a Phe residue at either the β₁α₁ (Prx5) or β_Dα₂ (Tpx) turns to form interdimer “ball-and-socket” hydrophobic interactions, mutation of which destabilizes the dimer.^84,85 ‘Ball-and-socket’ interactions have also been observed in proteins that do not have A-type dimerization such as glutathione peroxidase (Gpx) and λ Cro.^37,84,85 Although the human Trx dimer (PDB: 1ERT, 1ERU) is supported by an interstrand disulfide bond, Trp31 within the β₁α₁ turn could play a similar role to Phe in A-type interfaces by interacting with Val71′/Met74′ on the opposite monomer.³⁴ In Dio3, Thr169 and His219 occupy the positions analogous to Phe in Prx5 or Tpx, respectively, suggesting ‘ball-and-socket’ interactions in Dio3 are unlikely (Figure 3). Phe258 in the α₂β₃ turn is the only Phe located at the Dio3 surface. However, this residue covers the active site in the monomer X-ray structure and forms part of the wall of the active site pocket in MD simulations.^15,26 In addition, the sterics and positioning of the unique Dio Ω-loop near the β₁α₁ and α₂ regions could prohibit A-type dimerization. Structural overlay of Dio3^Trx monomers on A-type dimers places α_D in close contact with the Ω-loop on opposite sides of the dimer interface. Therefore, A-type dimer formation would preclude the Sagar et al. proposal, based upon 1ERU, that the Ω-loop could fold into a β-sheet. The proposed β-sheet regions (²⁰⁷WVTT²¹⁰ and ²¹⁴YII²¹⁷) which contain ‘hotspot’ residues such as Tyr and Trp have been shown to stabilize dimer interfaces as well as residues commonly found in exterior β-sheets.^83,86 However, these sequences never organize into β-strands during the 20 μs simulation of the Dio3 Trx monomer suggesting that β-sheet formation for an 1ERU-type dimer is unlikely.

The sequence of β₄, α₃, and the β₄α₃ turn, which make up the B-type interface, varies between Dio3 and Prx proteins (Figure 3). Dio3 has a slightly shorter β₄-sheet compared to 2-Cys Prxs with the terminal MYQ triad in the β₄-sheet of Dio3 overlapping with the initial (V/L)(V/R)(D/H/Q) triad in 2P5R, 4KCE, 1QQ2, and 2C0D.^35,44,47 Poplar Gpx5 (PDB: 2P5Q, 2P5R) also forms a B-type-like dimer along this interface with a β₄ sequence similar in length to Dio3 but with low overall sequence identity.⁴⁴ A Gly-Arg dyad is conserved within the β₄α₃-turn of 2-Cys peroxiredoxin proteins 2C0D, 1QQ2, and 4KCE. This dyad is up-sequence in Dio3 as part of a flexible Gly-rich α₃β₄-turn that accommodates the amino acid group of the TH substrate. The Arg275 sidechain oscillates between open and closed over a polar cleft that binds of the T₄ amino acid group.²⁶ The ²⁸¹QVSEL²⁸⁵ pentad of Dio3 is moderately similar to an (S/N)V(D/Q)E(V/I) sequence in Prx, both occurring in the N-terminal segment of α₃. Dio3 has Arg286 one residue downstream of a conserved Arg in 2-Cys Prxs, consistent with stabilizing intramonomer Glu-Arg salt bridges observed in B-type dimers.

3.2. Prediction of the Dio3 dimer structure.

Preliminary protein-protein docking of monomers of Dio3^Trx (X-ray structure and the T₄-bound structure) was attempted using GalaxyGemini and HawkDock.²⁶ GalaxyGemini predicted only two B-type dimers (Table S1) based on a thioredoxin-like protein from Aeropyrum pernix (PDB: 3HA9) and poplar Gpx5 (PDB: 2P5Q).⁴⁴ HawkDock produced primarily asymmetric hits from the X-ray structure, but found one model B-type dimer using T₄-bound Dio3^Trx as the most energetically favorable of this set of predicted dimers (Figure S1). A symmetric dimer with -interfaces at α₂ and the Ω-loop bore some similarities to A-type dimers but was unstable in trial MD simulations. These preliminary results support the evidence from sequence analysis that Dio is more similar to proteins that adopt a B-type dimer. However, the predicted B-type structures contained clashes between interfacial sidechains which led to unstable dimers in MD simulations. Additionally, formation of the interstrand extended β-sheet as observed in B-type dimers was inhibited by a slight twist of β₄ in the monomers.

SymmDock was used to focus structural prediction on symmetric B-type dimers. The monomer structures were modified to impose β-strand backbone dihedral angles on β₄ to prevent structural clashes that caused dimers to quickly fall apart during MD simulations. The highest scoring dimer structure with anti-parallel interactions of the α₃β₄ regions was adjusted to remove close contacts between the sidechains of Arg286 in the α₃-helix and the Thr268/Gly273′ pairings in the β₄α₃ turn prior to refinement of the interface sidechains by GalaxyRefineComplex. Three high-quality structures were obtained with similar orientations of Arg291 forming a close interaction to the carbonyl oxygen of Gln295′ on the opposite α-helix (Figure 4A). The guess with the lowest 1-RMSD orients the sidechains of Tyr271 and Trp288 for stabilizing π-stacking interactions (Figure 4B) consistent with these residues serving as “hotspot” residues to stabilize the dimer interface.⁸⁷ This model dimer was the most stable in preliminary MD simulations and selected for longer simulations.

Figure 4.

Overlays of the three highest quality models generated by GalaxyRefineComplex emphasizing the orientations of (A) the sidechains of Arg291 and Gln295 and (B) the sidechains of Tyr271 and Trp288. The blue model with a Tyr271-Trp288 π-stacking interaction was the most stable in test simulations and used for production runs of the dimer.

3.3. MD Simulations of the dimer structure

MD simulations were carried out for dimer models both with T₄-bound to both monomers and in the apo state. The apo- and holo-proteins are generally stable over the course of triplicate 1 μs simulations (Figure 5). The T₄ residues remain bound to the active site pocket supported by an outer-ring O···Se XB interaction to the Asp211 carboxylate in both monomer subunits. In the apo simulations, the Ω-loops and β₄α₃ turns are more free to change conformation in the absence of the T₄ substrate, contributing to the increased variance in the RMSD of the apo dimer (Figure 5A).²⁶ The monomer interface fluctuates slightly over the course of the simulations (Figure 5B) due to opening and closing of Arg275 over the polar binding pocket for the amino acid group of the TH substrate. The key β₄-β₄ interaction, a Met270-Gln272′ is largely intact over the six simulations with the exception of one apo run where the interaction dissociates (Figure 5C), but reforms and remains intact through the remainder of the run. Full dissociation of the monomers is prevented by strong electrostatic interactions between the charged residues of α₃. This observation is consistent with Dio favoring a B-type dimer as predicted by bioinformatic analysis.

Figure 5.

(A) RMSD plot of the apo and holo states of the dimer model. The protein is slightly more mobile in the apo state due to the absence of the large T₄ substrates. (B) RMSD plot of the β₄α₃ interface residues of the apo and holo dimer states. The flexibility in this region is attributed to the α₃-α₃ interface, which contains itinerant salt bridges, and the β₄α₃ turn, which contributes to the binding pocked for the amino acid end of T₄. (C) Distance plots of the two stable hydrogen bonds between Met270-Gln272′ backbone atoms for the simulations of the holo and apo proteins. (D) Representative snapshot of the Dio3 dimer. The β₄-β₄ interaction stabilizes the intermonomer β-sheet consistent with a B-type Prx interface. (E) Snapshots of the Met270-Gln272′ and Met270′-Gln272 hydrogen bonding interactions.

MMGBSA has been shown to have high accuracy in predicting dimerization energies and is able to identify energetic contributions of individual residues through decomposition analysis.^73,74 MMGBSA interaction energies indicate that dimer formation is more stable in the absence of T₄ (ΔG_holo = −46.24 ± 16.29 kcal mol⁻¹; ΔG_apo = −56.26 ± 15.96 kcal mol⁻¹). The time evolution of interaction energies shows the electrostatic contributions far outweigh van der Waals interactions within the interface region (Figure S2).

The contributions of per-residue interactions at the dimer interface were calculated using the pairwise energy decomposition analysis in MMPBSA for all possible pairings between interfacial residues 267–301 (Table 1). These interaction energies account for all instances of hydrogen bonding between residues, meaning that more than one potential hydrogen bonding interaction between residue pairs could persist during the simulation.⁸⁸ Hydrogen bonds can form or be lost rapidly over the course of an MD simulation, which could lead to large variance in per-residue hydrogen bonding energies.⁸⁸ The magnitude of the pairwise standard deviations (SDs) can identify whether hydrogen bonds are dynamic (high SDs) or static (low SDs).

Table 1.

MMPSBA energy contributions of key per-residue interaction along the dimer interface region of Dio3 during the apo- and holo-protein simulation

	Interaction	apo ΔG (kcal mol−1)	holo ΔG (kcal mol−1)
β-β Interdimer (BB-BB)	Met270-Gln272′	−3.84 ± 0.07	−4.13 ± 0.38
	Thr268-Gln272′	−0.54 ± 0.50	−1.86 ± 0.65
	Met270-Tyr271′	−2.16 ± 0.30	−2.35 ± 0.29
	Ile269-Gln272′	−1.90 ± 0.31	−2.17 ± 0.19
α-α Interdimer (SC-SC)	Arg291-Glu294′	−3.96 ± 1.29	−1.17 ± 0.44
	Glu294-Gln295′	N/A	−1.24 ± 0.17
	Arg291-Gln295′	−1.99 ± 1.00	−1.87 ± 0.56
α-α Intramonomer (SC-SC)	Arg291-Glu290	−2.08 ± 0.83	−0.23 ± 0.14
	Arg291-Glu294	−1.11 ± 0.34	−2.66 ± 0.59

An extended interstrand β-sheet interactions is common to B-type proteins (e.g., 2C0D and 4KCE) except of 2P5R. The strongest and most persistent hydrogen bonding interactions in the Dio3 dimer are the interstrand β-β Met270-Gln272′ BB-BB interactions (ΔG_holo = −4.13 ± 0.38 kcal mol⁻¹, ΔG_apo = −3.84 ± 0.07 kcal mol⁻¹, Figure 5C,E). Intermittent interstrand β-β BB-BB interactions are also found between the Thr268-Gln272′, Met270-Tyr271′, and Ile269-Gln272′ pairs. These interactions are relatively static but become slightly destabilized in the apo dimers (ΔΔG = +1.32, +0.19, and +0.26 kcal mol⁻¹ respectively, where ΔΔG represents the difference between the holo and apo free energies).

Interstrand α₃-α₃ interactions are slightly less favorable than those that form the interdimer β-sheets (Table 1) although some B-type dimers (e.g., 3HA9 and 2C0D) lack α₃-α₃ interactions. Intramonomer salt bridges involving an α₃ Arg and nearby acidic sidechains (Glu/Asp) are found in B-type proteins 1QQ2, 2P5R, and 4KCE (Figure 6) and similar interactions may stabilize the Dio3 dimer. The orientations of Arg156 in 1QQ2 and Arg152 in 4KCE slightly overlay with Arg291 of Dio3, as all form SC-SC interactions with nearby Glu residues. In comparison, Arg161 of 2P5R assumes an extended conformation with the inward-facing Asp162. Interdimer α-α SC-SC interactions between Arg291, Glu294′ and Gln295′ account for the largest contributions to the Dio3^Trx dimerization energy (Table 1). In the apo simulation, the interdimer Arg291-Glu290′ and Arg291-Gln295′ SC-SC interactions become more favorable (ΔΔG = −2.79 and -0.12 kcal mol⁻¹ respectively) as the interdimer Arg291-Glu294′ and Arg291-Gln295′ SC-SC interactions become more dynamic. Additionally, the interdimer Glu294-Gln295′ SC-SC interactions and intramonomer Arg291–294 SC-SC interactions become destabilized in the apo dimer. In addition, the intramonomer Arg291-Glu290 interactions strengthen by 1.85 kcal mol⁻¹ and become more dynamic after removal of the T₄ residues.

Figure 6.

Comparison of the orientation of Arg291 in Dio3 to A) Arg156 of 1QQ2; B) Arg161 of 2P5R; C) Arg187 of 4KCE. Arg291 forms salt bridges with acidic residues similar to B-type dimer interfaces in Prx.

3.4. ONIOM Modeling of Deiodination Transition State.

Given that MD simulations of the holo B-type Dio3^Trx dimers do not indicate significant changes to the T₄ binding pocket, the monomer was used as the basis for determining the transition state for deiodination. Two-layer ONIOM (mPW1PW91:AMBER) calculations were initiated from the previously determined Dio3^Trx-T₄ monomer complex.²⁶ The binding pocket, T₄ and waters within the binding pocket, and residues with 5 Å of T₄ with the remaining protein, solvent and counterions to be fully constrained. Deiodination has been shown to occurs through a halogen-bonded intermediate with Dio donating a proton to complete the nucleophilic attack of Sec on the C-I bond.⁶³ Although it was previously proposed Dio3 provided a proton shuttle through the His219-Glu200-Ser167 triad,¹⁵ MD simulations indicated that these residues did not face the interior of the binding pocket.²⁶ Instead, it was recommended that water molecules found within the T₄-bound active site and close to the site of protonation may act as proton donors for the nucleophilic deiodination.²⁶ Working from this hypothesis, a deiodination reaction pathway was mapped from an initial state in which T₄ interacts with Sec170 through an inner-ring I⋯Se XB, supported by an outer-ring T₄-Asp211 SC I⋯O XB.²⁶ Two water molecules that form a hydrogen bonded network to the T₄ amino group were included in the QM region with T₄ and the Sec sidechain.

In the optimized reactant complex, the I⋯Se XB is slightly weaker than that found in the simple methyl selenolate complex with T₄ (Δd(C–I)=0.03 Å and d(Se–I)=3.17 Å versus Δd(C–I)=0.20 Å and d(Se–I)=2.92 Å).²⁷ due to the effects of the protein environment. The TS (Figure 7) was found by following the C–H_wat bond-forming coordinate followed by optimization to the TS structure in which the C-I bond-breaking motion at the TS is coupled to proton transfer along the two-water network from the T₄ ammonium group. The second water migrates to form a two-water bridge between the forming carbanion and T₄ amine to stabilize proton transfer. The ONIOM(mPW1PW91:AMBER) activation barrier (E_a = 18.5 kcal mol⁻¹) is comparable to that for small DFT models of XB-based deiodination (17.6 kcal mol^-1.⁶³ as well as other computational deiodination models.⁸⁹ As the water proton is transferred to the carbon, the C–I bond elongates by 1.07 Å, and the Se–I distance reduces to 2.66 Å, later along the reaction coordinate than found for a small DFT model using imidazoles as proton donors (Δd(C-I)= 0.31 Å and d(Se-I)=2.733 Å).⁴⁴ The optimized product complex (P) of reverse T₃ (rT₃) and Sec selenenyl iodide (d(Se-I) = 2.61 Å) lies −13.5 kcal mol⁻¹ below the reactant complex. In this structure, rT₃ is repelled by the selenenyl iodide group, but is held in place by the T₄-Asp211 SC I⋯O XB interaction and the hydrogen bond interactions between the sidechains of Tyr257 and Arg275 and the carboxylic group of T₄.

Figure 7.

ONIOM transition state for the deiodination of T₄ in the Dio3^Trx monomer active site. Waters in the active site facilitate proton transfer from the ammonium group of T₄ to the carbanion formed by nucleophilic attack of Sec on the C–I bond. The activation barrier from the I···Se intermediate is comparable to previous DFT models.

4. CONCLUSIONS

Understanding the structure of Dio proteins are important for designing drug targets for treating thyroid hormone-related illnesses. However, the mechanism of dimerization of the Dio family has not yet been characterized experimentally. in silico methods have been used to propose a potential dimerization interface, an important step toward understanding the relationship between the tertiary structure of Dio proteins and their activity. Previously proposed dimers are roughly comparable to the A-type and B-type dimers formed in the oxidation of oligomeric Prx-type proteins. Dio lacks conserved features of the A-type interface (i.e., the “ball-and-socket” residues) while limited similarity in its β₄α₃ sequence B-type proteins provides a strong hint as to the Dio dimerization interface. Building upon this analysis, docking algorithms produce more hits for B-type proteins further suggesting that this type of dimerization is more likely for the Dio family.

MD simulations of dimers generated using protein-protein docking databases provide stable B-type dimers in both the apo and holo forms. Since dimerization did not significantly alter the T₄ binding site, less expensive QM/MM calculations on the monomer were used to model the deiodination reaction pathway. A transition state for deiodination in which active site waters facilitate proton shuttling from the T₄ amino group to the phenyl anion formed during nucleophilic attack has an activation barrier in agreement with previous small model DFT studies. MMPBSA pairwise decomposition analysis identified key interactions within the β₄α₃ region of the dimerization interface. Backbone-backbone interactions between Met270-Gln272′ are primarily responsible for stabilizing the intermonomer β-sheet, while salt bridges between Arg291 and acidic residues hold the α-helices together. Mutation studies targeting the β₄α₃ interface residues, particularly those shown to stabilize the interaction, could test their effects on dimer formation and catalytic function. Note, however, the N-terminal transmembrane domain also makes significant contributions to dimerization. Further studies of the full protein dimer mounted in a lipid bilayer are needed for a complete picture of the structure and activity of this important class of selenoproteins.