Crystal structure and interaction studies of human DHTKD1 provide insight into a mitochondrial megacomplex in lysine catabolism

DHTKD1 is a lesser-studied E1 enzyme among the family of 2-oxoacid de­hydrogenases. In complex with E2 (di­hydro­lipo­amide succinyltransferase, DLST) and E3 (dihydrolipo­amide de­hydrogenase, DLD) components, DHTKD1 is involved in lysine and tryptophan catabolism by catalysing the oxidative de­carboxyl­ation of 2-oxoadipate (2OA) in mitochondria. Here, the 1.9 Å resolution crystal structure of human DHTKD1 is solved in complex with the thi­amine diphosphate co-factor. The structure reveals how the DHTKD1 active site is modelled upon the well characterized homologue 2-oxoglutarate (2OG) de­hydrogenase but engineered specifically to accommodate its preference for the longer substrate of 2OA over 2OG. A 4.7 Å resolution reconstruction of the human DLST catalytic core is also generated by single-particle electron microscopy, revealing a 24-mer cubic scaffold for assembling DHTKD1 and DLD protomers into a megacomplex. It is further demonstrated that missense DHTKD1 variants causing the inborn error of 2-amino­adipic and 2-oxoadipic aciduria impact on the complex formation, either directly by disrupting the interaction with DLST, or indirectly through destabilizing the DHTKD1 protein. This study provides the starting framework for developing DHTKD1 modulators to probe the intricate mitochondrial energy metabolism.

1. Introduction

The family of multi-component 2-oxoacid de­hydrogenase complexes, of which pyruvate de­hydrogenase (PDHc), branched chain α-ketoacid de­hydrogenase (BCKDHc) and 2-oxoglutarate de­hydrogenase (OGDHc) complexes are canonical members, catalyse the oxidative de­carboxyl­ation of 2-oxoacids (e.g. pyruvate, 2-oxoisovalerate and 2-oxoglutar­ate) into their corresponding acyl-CoA thio­esters, generating the reducing equivalent NADH (nicotinamide adenine di­nucleotide in a reduced form). These biochemical reactions play crucial roles in intermediary metabolism, and are tightly regulated by phospho­rylation and allosteric effectors (Yeaman, 1989; Reed, 2001).

The overall reaction catalysed by 2-oxoacid de­hydrogenases is dissected into three sequential steps each catalysed by an individual enzyme (Perham, 1991; Jordan, 2003). In the first step, rate limiting for the overall reaction, the E1 enzyme (a 2-oxoacid de­carboxyl­ase; EC 1.2.4.2) catalyses the irreversible de­carboxyl­ation of 2-oxoacids via the thi­amine diphosphate (ThDP) co-factor and subsequent transfer of the de­carboxyl­ated acyl intermediate on an oxidized di­hydro­lipoyl group that is covalently amidated to the E2 enzyme (a di­hydro­lipoyl acyl­transferase; EC 2.3.1.61). In the second step, E2 transfers the acyl moiety from the di­hydro­lipoyl group onto a CoA-SH acceptor, generating acyl-CoA and a reduced di­hydro­lipoyl group. In the final step, one FAD-dependent (flavin adenine dinucleotide) E3 enzyme universal to all complexes (di­hydro­lipo­amide de­hydrogenase, DLD; EC 1.8.1.4) re-oxidizes the di­hydro­lipoyl group by transferring one reducing equivalent of NAD⁺ to yield NADH.

To achieve the overall oxidative de­carboxyl­ation reaction, multiple copies of the E1, E2 and E3 components classically assemble into a supramolecular complex reaching 4–10 MDa in weight (Marrott et al., 2014). Structural studies have shown E2 enzymes from various organisms to exist in a high-order cubic 24-mer or dodecahedral 60-mer (Izard et al., 1999), acting as a scaffold onto which copies of E1 and E3 are assembled. Such a quaternary arrangement, a classic example of a metabolon, provides a means by which products of one reaction are funnelled into the catalytic centres of the next reactions to enhance enzymatic efficiency and avoid undesirable side reactions (Cohen & Pielak, 2017). For example, the E2-attached di­hydro­lipoyl co-factor is expected to shuttle catalytic intermediate substrates between E1 and E3 enzymes by means of a `swinging-arm' mechanism (Zhou et al., 2001; Reed & Hackert, 1990; Perham et al., 2002).

The human genome encodes five E1-type de­carboxyl­ases (PDH, BCKDH, OGDH, DHTKD1 and OGDHL), among which OGDH, OGDHL and DHTKD1 form a more evolutionarily related subgroup with respect to the E1 architecture and the E2 enzyme employed (Bunik & Degtyarev, 2008). The OGDHc complex, composed of OGDH as E1, di­hydro­lipo­amide succinyltransferase (DLST) as E2 and DLD as E3, converts the metabolite 2-oxoglutarate (2OG) to succinyl-CoA and serves as a rate-limiting step in the Krebs cycle (Araújo et al., 2013). A close homologue of OGDH, the OGDH-like protein (OGDHL) is expressed in the brain (Bunik et al., 2008) and implicated in brain pathways of glutamate and Ca²⁺ sensing. While its precise physiological role is not defined, OGDHL is considered as a tissue-specific isoenzyme of OGDH. A second analogue of OGDH, the enzyme DHTKD1 (de­hydrogenase E1 and transketolase domain-containing protein 1) is positioned in the last step of lysine and tryptophan catabolism with the common product being 2-oxoadipate (2OA), one methyl­ene group longer than 2OG. To catalyse the oxidative de­carboxyl­ation of 2OA to glutaryl-CoA (Nemeria, Gerfen, Yang et al., 2018), DHTKD1 recruits the same E2 (DLST) and E3 (DLD) as OGDH to form the 2-oxoadipate de­hydrogenase complex (OADHc) (Goncalves et al., 2016; Nemeria, Gerfen, Nareddy et al., 2018), implying that DLST also acts as a di­hydro­lipo­amide glutaryl­transferase. Both DHTKD1 (Quinlan et al., 2014; Bunik & Brand, 2018) and OGDH (Xu et al., 2013; Sherrill et al., 2018) are emerging as contributors of reactive oxygen species in mitochondria, through catalysing a side reaction in the forward reaction that results in superoxide/H₂O₂ formation (Goncalves et al., 2016; Bunik & Sievers, 2002). The identification of DHTKD1 as an additional reactive oxygen species (ROS) source in mitochondria implies a contribution to oxidative stress under pathophysiological conditions such as those associated with mitochondrial abnormalities and neurodegeneration (Jordan et al., 2019). To this end, DHTKD1 is increasingly recognized as essential for mitochondrial function and energy production, whereby loss of DHTKD1 function is associated with decreased adenosine triphosphate (ATP) production, increased ROS production and impaired mitochondrial biogenesis in cultured cells (Xu et al., 2013; Sherrill et al., 2018).

In support of this role, inherited DHTKD1 mutations are identified as the molecular cause of two rare Mendelian disorders. 2-Amino­adipic and 2-oxoadipic aciduria (OMIM 204750) is an inborn error of metabolism with questionable clinical consequence (Fischer et al., 1974), characterized biochemically by increased urinary excretion of 2-oxoadipate and its transamination product 2-amino­adipate (Danhauser et al., 2012; Duran et al., 1984). Among <30 reported cases caused by autosomal recessive missense and nonsense mutations (Hagen et al., 2015), p.G729R and p.R455Q are common variants. Additionally, a nonsense DHTKD1 mutation causes Charcot–Marie–Tooth disease type 2Q (CMT2Q, OMIM 615025), an autosomal dominant neurodegenerative disorder characterized by motor and sensory neuropathies (Xu et al., 2012).

While structural studies have been carried out for PDH and BCKDH E1 enzymes from various organisms across the phyla, including human (Ævarsson et al., 2000; Ciszak et al., 2003), only prokaryotic OGDHs have been crystallized. These include the apo structure of Escherichia coli OGDH (ecOGDH) (Frank et al., 2007), as well as various structures of Mycobacterium smegmatis OGDH (msOGDH) complexed with active-site catalytic intermediates (Wagner et al., 2011, 2014, 2019). In this study, we report the crystal structure of human DHTKD1 (hDHTKD1) and a cryo-EM reconstruction of the human DLST (hDLST) catalytic core. We also characterize disease-causing variants of DHTKD1 for protein thermostability and interaction with DLST.

2. Materials and methods

2.1. Expression and purification of hDHTKD1 and hDLST

Site-directed mutations were constructed using the QuikChange mutagenesis kit (Stratagene) and confirmed by sequencing. All primers are available upon request. Wild-type (WT) and variant DHTKD1 proteins, as well as all DLST proteins, were expressed in E. coli BL21(DE3)R3-Rosetta cells from 1–6 l of Terrific Broth culture. Cultures were grown at 37°C until an optical density (OD₆₀₀) of 1.0, when they were cooled to 18°C and induced with 0.1 mM IPTG overnight. Cultures were harvested at 4000g for 30 min. Cell pellets were lysed by sonication at 35% amplitude, 5 s on 10 s off, and centrifuged at 35 000g. The clarified cell extract was incubated with Ni-NTA resin pre-equilibrated with lysis buffer (50 mM HEPES pH 7.5, 500 mM NaCl, 20 mM imidazole, 5% glycerol, 0.5 mM TCEP). The column was washed with 80 ml binding buffer (50 mM HEPES pH 7.5, 500 mM NaCl, 5% glycerol, 20 mM imidazole, 0.5 mM TCEP) and 80 ml wash buffer (50 mM HEPES pH 7.5, 500 mM NaCl, 5% glycerol, 40 mM imidazole, 0.5 mM TCEP), and eluted with 15 ml of elution buffer (50 mM HEPES pH 7.5, 500 mM NaCl, 5% glycerol, 250 mM imidazole, 0.5 mM TCEP). The eluant fractions were concentrated to 5 ml and applied to a Superdex 200 16/60 column pre-equilibrated in GF buffer (50 mM HEPES pH 7.5, 500 mM NaCl, 0.5 mM TCEP, 5% glycerol). Eluted protein fractions were concentrated to 10–15 mg ml⁻¹. Lipoylation of DLST proteins was verified by intact mass spectrometry.

2.2. Co-expression of DHTKD1 and DLST

The DHTKD1–DLST complex used in this study was co-expressed in both E. coli and insect Sf9 cells. For E. coli co-expression, hDHTKD1_45–919 was subcloned into the pCDF-LIC vector (incorporating a His-tag) and the resultant plasmid was co-transformed with the plasmid encoding untagged hDLST_68–453 in the pNIC-CT10HStII vector. Co-transformed cultures were grown and protein purification was performed, as described above for DHTKD1 alone. For co-expression in insect cells, baculoviruses were produced by transformation of DH10Bac cells. Viruses were amplified by infecting Sf9 insect cells in 250 ml of sf900II serum free protein-free insect-cell medium (Thermo Fisher Scientific) and grown for 65 h at 27°C in 1 l shakers. Sf9 culture was co-infected in a 1:1 ratio with two of third-generation viruses each at 1.5 ml l⁻¹. One baculovirus pFB-Bio5 vector expresses His-tagged hDHTKD1_45–919 and the other baculovirus pFB-LIC-Bse vector expresses His-tagged hDLST_68–453. The cultures were grown at 27°C for 72 h in 3 l flasks before harvesting at 900g for 30 min. The purification of Sf9 expressed proteins was carried out mostly as above. It only differed in adding 1:1000 benzonase to the lysis buffer and a gentler sonication cycle of 4 s on, 12 s off.

2.3. Crystallization and structure determination of DHTKD1

Crystals were grown by the vapour-diffusion method. To crystallize hDHTKD1_45–919, concentrated protein was incubated for 30 min on ice with 3 mM MgCl₂ and 3 mM ThDP before being centrifuged for 10 min at 13 500g to remove any precipitation. Sitting drops containing 75 nl of protein (10 mg ml⁻¹) and 75 nl of well solution containing 20%(w/v) PEG 3350, 0.1 M bis-tris-propane pH 8.5, 0.2 M sodium formate and 10%(v/v) ethyl­ene glycol were equilibrated at 4°C. Crystals were mounted and frozen without additional cryo-protectant, as the crystallization condition contains 10%(v/v) ethyl­ene glycol. Diffraction data were collected at the Diamond Light Source beamline I03 and processed using the CCP4 program suite (Winn et al., 2011). hDHTKD1_45–919 crystallized in the primitive space group P1 with two molecules in the asymmetric unit. The structure was solved by molecular replacement using the program Phaser (McCoy et al., 2005) and the E. coli OGDH structure (PDB code 2jgd; Frank et al., 2007) as the search model. The structure was refined using Phenix (Adams et al., 2010), followed by iterative cycles of model building in Coot (Emsley & Cowtan, 2004). Statistics for data collection and refinement are summarized in Table 1. Protein interfaces were analysed with the software PISA (Krissinel & Henrick, 2007).

Table 1
Summary of diffraction and refinement statistics

The numbers in parentheses represent data in the highest-resolution shell.

Data collection
Synchrotron radiation facility	Diamond Light Source
Beamline	I03
Detector	Dectris PILATUS3 6M
Wavelength (Å)	0.9762
Reflections (measured/unique)	145107 (13870)
Space group	P1
Unit-cell parameters
a, b, c (Å)	78.55, 81.22, 86.90
α, β, γ (°)	63.43, 76.96, 72.06
Resolution (Å)	46.01–1.87 (1.937–1.87)
Rmerge	0.10 (0.44)
〈I/σ(I)〉	4.96 (1.04)
CC 1/2	0.99 (0.64)
Completeness (%)	96.45 (91.94)
Multiplicity	1.8 (1.5)
Refinement
Resolution (Å)	1.9
Rwork/Rfree (%)	0.19/0.22
No. of reflections
Working set	144961 (13819)
Test set	7088 (698)
Total no. of atoms	13331
Ligands at active site	55
No. of water molecules	1066
Wilson B factor (Å^2)	21.81
RMS deviations
Bond lengths (Å)	0.007
Bond angles (°)	0.84
Ramachandran plot
Favoured (%)	96.9
Allowed (%)	2.87
Outliers (%)	0.23
Molecules in asymmetric unit	2
PDB code	6sy1

2.4. DHTKD1 enzyme assay

The enzymatic activity assay was performed in triplicates, in a buffer containing 35 mM potassium phosphate (KH₂PO₄), 0.5 mM EDTA, 0.5 mM MgSO₄, 2 mM 2OA or 2OG, 1 mM ThDP, 5 mM sodium azide (NaN₃) and 60 µM 2,6-dichloro­phenolindophenol (DCPIP), pH 7.4. The activity was determined as a reduction of DCPIP at λ = 610–750 nm, 30°C (Sauer et al., 2005), with and without 2OA or 2OG. The dye DCPIP changes colour from blue to colourless when being reduced (VanderJagt et al., 1986). To obtain K_m and V_max, different concentrations of 2OA (0.1, 0.05, 0.1, 0.25, 0.5, 0.75, 1 and 2 mM) and no substrate were measured in a 96-well microtitre plate (total well volume = 300 µl). The ensuing OD values were plotted on a graph (slope = 1/V_max; Y intercept = K_m/V_max) to calculate K_m and V_max using the Hanes Woolf plot:

where V₀ = initial velocity, [S] = substrate concentration and V_max = maximum velocity.

2.5. Small-angle X-ray scattering

Small-angle X-ray scattering (SAXS) experiments were performed at a wavelength of 0.99 Å at the Diamond Light Source beamline B21 coupled to the appropriate size-exclusion column (Harwell, UK) and equipped with a PILATUS 2M 2D detector at a distance of 4.014 m from the sample, 0.005 < q < 0.4 Å⁻¹ (q = 4π sin θ/λ, 2θ is the scattering angle). hDHTKD1_45–919 at 20 mg ml⁻¹ in 10 mM HEPES-NaOH pH 7.5, 200 mM NaCl, 0.5 mM TCEP and 2% glycerol was applied onto a Shodex KW404-4F column. hDHTKD1_45–919 co-expressed with hDLST_68–453 in baculo Sf9 cells at 5 mg ml⁻¹ in 25 mM HEPES pH 7.5, 200 mM NaCl, 0.5 mM TCEP, 2% glycerol and 1% sucrose was applied onto the Shodex KW404-4F column. hDLST at 10 mg ml⁻¹ in 25 mM HEPES pH 7.5, 200 mM NaCl, 0.5 mM TCEP, 2% glycerol and 1% sucrose was applied onto a Shodex 405-4F column.

SAXS measurements were performed at 20°C using an exposure time of 3 s frame⁻¹. SAXS data were processed and analyzed using the ATSAS program package (Franke et al., 2017) and Scatter (https://www.bioisis.net/scatter). The radius of gyration R_g and forward scattering I(0) were calculated by Guinier approximation. The maximum particle dimension D_max and P(r) function were evaluated using the program GNOM (Svergun, 1992).

2.6. Solution analysis

Analytical gel filtration was performed on a Superdex 200 Increase 10/300 GL column or Superose 6 Increase 10/300 GL (GE Healthcare) pre-equilibrated with 20 mM HEPES pH 7.5, 150 mM NaCl and 0.5 mM TCEP.

2.7. Differential scanning fluorimetry (DSF)

DSF was performed in a 96-well plate using an Mx3005P RT PCR machine (Stratagene) with excitation and emission filters of 492 and 610 nm, respectively. Each well (20 µl) consisted of protein (2 mg ml⁻¹ in 100 mM HEPES pH 7.5, 150 mM NaCl, 5% glycerol), SYPRO Orange (Invitrogen, diluted 1000-fold of the manufacturer's stock). Fluorescence intensities were measured from 25 to 96°C with a ramp rate of 1°C min⁻¹. T_m was determined by plotting the intensity as a function of temperature and fitting the curve to a Boltzmann equation. Temperature shifts, ΔT_m, were determined as described (Niesen et al., 2007) and final graphs were generated using GraphPad Prism (v.7; GraphPad software). Assays were carried out in technical triplicate.

2.8. MIDAS protein–metabolite screening

Protein–metabolite interaction screening using an updated MIDAS platform was performed similar to Orsak et al. (2012). Briefly, a flow-injection analysis mass-spectrometry (FIA-MS) validated library of 412 metabolite standards was combined into four defined screening pools in 150 mM ammonium acetate pH 7.4. For each metabolite pool, 5 µl of target protein was arrayed in triplicate across a SWISSCI 10 MWC 96-well microdialysis plate (protein chambers). To the trans side of each dialysis well, 300 µl of a 50 µM metabolite pool supplemented with 1 mM ThDP and 1 mM MgCl₂ was arrayed in triplicate per hDHTKD1_45–919 protein (metabolite chambers). Dialysis plates were placed in the dark at 4°C on a rotating shaker (120 rev min⁻¹) and incubated for 40 h. Post-dialysis, protein- and metabolite-chamber dialysates were retrieved, normalized and diluted 1:10 in 80% methanol, incubated for 30 min on ice, and centrifuged at 3200g RCF for 15 min to remove precipitated protein. Analytes were aliquoted across a 384-well microvolume plate and placed at 4°C in a Shimadzu SIL-20ACXR autosampler for FIA-MS analysis. Then, 2 µl of each sample was analysed in technical triplicate by FIA-MS on a SCIEX X500R QTOF MS with interspersed injections of blanks.

2.9. MIDAS data analysis

FIA-MS spectra collected from MIDAS protein–metabolite screening was qualitatively and quantitatively processed in SCIEX OS 1.5 software to determine relative metabolite abundance by integrating the mean area under the curve across technical triplicates. Log2(fold change) for each metabolite was calculated from the relative metabolite abundance in the protein chamber (numerator) and metabolite chamber (denominator) from dialysis triplicates. For each technical triplicate, up to one outlier was removed using a z-score cutoff of five (<0.1% of observations). The corrected technical replicates were collapsed to one mean fold-change summary per protein–metabolite pair. To remove fold-change variation that was not specific to a given metabolite–protein pair, the first three principal components of the cumulative screening dataset were removed (∼75% of observed variance) creating Log2(corrected fold change). Protein–metabolite z scores were determined by comparing the target protein–metabolite Log2(corrected fold change) to a no-signal model for that metabolite using measures of the central tendency (median) and standard deviation (extrapolated from the 25–75% quantiles) which are not biased by the signals in the tails of a metabolite's fold-change distribution. z scores were false-discovery rate controlled using Storey's q value (https://github.com/jdstorey/qvalue). Protein–metabolite interactions with p values < 0.05 and q values < 0.1 were considered significant.

2.10. Grid preparation and EM data collection

3 µl of 0.4 mg ml⁻¹ purified complex from E. coli or Sf9 cells were applied to the glow-discharged Quantifoil Au R1.2/1.3 grid (Structure Probe). Blotting and vitrification in liquid ethane was carried out using a Vitrobot Mark IV (FEI Company) at 4°C and 95% humidity with a 9 s wait and a 3 s blot at zero blotting force from both sides. Cryo grids were loaded into a Glacios transmission electron microscope (ThermoFisher Scientific) operating at 200 keV with a Falcon3 camera. For the E. coli sample, three screening images were recorded in linear mode with a pixel size of 0.96 Å and a defocus set to −3 µm. The Sf9 sample was also recorded in linear mode with a pixel size of 0.96 Å and a defocus range of −1 to −3.1 µm (steps of 0.3 µm). Data were collected with a total dose of 32.52 e Å⁻² and images were recorded with a 1 s exposure over 19 frames. Details are summarized in Table S1 in the Supporting information. Representative micrographs of both datasets are shown in Figs. S9(a) and S9(b) in the Supporting information.

2.11. EM data processing

The three single-frame micrographs from the E. coli dataset were used for manual picking after contrast transfer function (CTF) parameters were determined by CTFFIND4.1 (Rohou & Grigorieff, 2015). A total of 572 particles were extracted with a box size of 344 pixels, and 2D classification was performed. A total of five classes containing 272 particles were compared with an equivalent set of particles derived from the Sf9 dataset [Fig. S9(c)]. The full data-processing workflow for the Sf9 derived complex is illustrated in Fig. S10. A total of 619 dose-fractioned movies were corrected for drift using RELION's MotionCor2 (Zheng et al., 2017) with the dose-weighting option. CTF parameters were determined by CTFFIND4.1 (Rohou & Grigorieff, 2015). A subset of 1244 particles were manually picked and extracted with a box size of 344 pixels rescaled to 172 pixels. Eight classes were selected from one round of 2D classification for reference-based autopicking using RELION 3.0 (Scheres & Chen, 2012). 165 739 particles were extracted with a box size of 344 pixels rescaled to 172 pixels. All downstream particle classification, refinement and post-processing steps were performed in RELION 3.0 (Scheres & Chen, 2012). Junk particles were removed using 2D classification. Parallel rounds of 3D classification with (O) and without octahedral symmetry (C1) revealed no classes with additional density visible beyond the C-terminal catalytic domain. Reclassification with a soft mask also resulted in classes with no apparent density for the DLST N terminus. Subsequently, masked 3D refinement with the highest-resolution class from the symmetry-imposed masked 3D classification resulted in a 5 Å map. Finally, a further round of masked 3D classification without alignment and a regularization parameter T value of 20 was used to identify classes with the highest-resolution features. The resulting 3356 particles were refined to a global resolution of 4.7 Å based on the Fourier shell correlation (FSC) 0.143 threshold criterion. The orientation distribution of the final map was visualized with Chimera. Local resolution was calculated with RELION 3.0 (Scheres & Chen, 2012).

2.12. EM model building and refinement

Model building and refinement was carried out using the suite of programs in CCP-EM (Burnley et al., 2017). To fit a template to the final map the E. coli DLST orthologue structure (PDB code 1scz; Schormann et al., unpublished work) was used. The sequence was humanized and residues truncated to the alpha carbon using CHAINSAW (Stein, 2008). The oligomeric structure was docked into the density map, sharpened with a B factor of −281 Ų, using MOLREP. One round of refinement using REFMAC5 was carried out with ProSMART restraints generated from the E. coli DLST orthologue to avoid overfitting. Figures displaying model fit to density were made using Chimera (Pettersen et al., 2004). Overfitting was monitored through simultaneous refinement against the two half maps from the final 3D refinement. FSC between map and model was calculated using model validation in CCP-EM (Burnley et al., 2017).

3. Results and discussion

3.1. Overall structure of hDHTKD1 homodimer

hDHTKD1 is a 919-amino acid (aa) polypeptide [Fig. 1(a)], with the N-terminal 22 aa predicted to form the mitochondrial targeting signal peptide (Bunik & Degtyarev, 2008). We expressed in E. coli the soluble proteins for the precursor hDHTKD1_1–919, the predicted mature protein (hDHTKD1_23–919), as well as a further truncated construct (hDHTKD1_45–919) removing the putatively disordered aa 24–44 (Fig. S1). Despite various attempts, only the construct hDHTKD1_45–919 yielded crystals, upon pre-incubation with ThDP and Mg²⁺ prior to crystallization trials. This protein construct is active in vitro, exhibiting E1 de­carboxyl­ase activity with 2OA as substrate in a colorimetric assay using 2,6-dichlorphenol indophenol as reductant (V_max = 14.2 µmol min⁻¹ mg⁻¹ protein, K_m, 2OA = 0.2 mM; see Section 2.4).

Figure 1 Overall structure of hDHTKD1. (a) A schematic of hDHTKD1 domain organization, indicating disease-causing missense mutations with arrows. (b) Structure of the hDHTKD1 homodimer, where one subunit is coloured according to the domain organization from panel (a) and the other subunit is coloured grey. ThDP co-factors and Mg^2+ ions are shown as sticks and spheres, respectively. (c) Sites of missense mutations are shown as black spheres on one hDHTKD1 protomer [same view as panel (b)]. (d) Structural superposition of hDHTKD1, ecOGDH (PDB code 2jgd) and msOGDH (PDB code 6r2b; Wagner et al., 2019) highlighting their different inter-domain linkers (coloured purple, blue and light brown, respectively). (e) A magnified view of the dotted box in panel (d), showing how the hDHTKD1 inter-domain linker (purple) packs against two loop regions (black). Binding positions for the allosteric effectors AMP in the ecOGDH structure (PDB code 2jgd) and acetyl-CoA in the msOGDH structure (PDB code 2y0p; Wagner et al., 2011) are shown as blue and brown meshes, respectively. (f) Overall architectures of the hDHTKD1 homodimer (left), the human (PDHA–PDHB)2 heterotetramer (middle, PDB code 1ni4; Ciszak et al., 2003) and the human (BCKDHA–BCKDHB)2 heterotetramer (right, PDB code 1dtw; Ævarsson et al., 2000) are shown. hDHTKD1 homodimer has a larger volume because of the insertions coloured red.

The crystal structure of hDHTKD1_45–919 is determined to 1.9 Å resolution by molecular replacement, using the E. coli OGDH structure (PDB code 2jgd, 38% sequence identity; Frank et al., 2007) as the search template (Table 1). The asymmetric unit contains two DHTKD1 protomers [A and B; Fig. 1(b)] arranged as an intertwined obligate homodimer in a similar manner to OGDH, burying a large 5600 Ų (18%) area of monomeric accessible surface at the dimer interface. This crystal homodimer is consistent with SAXS analysis of hDHTKD1_45–919 protein in solution, with the theoretical scattering curve of the dimer displaying a good fit to experimental data (χ² of 3.8, Fig. S2).

Our DHTKD1 structural model [Fig. 1(c)] consists of residues 53–915 from both chains, with the exception that no electron density was observed for two surface exposed loop regions (aa 274–275_{chain A}/274–277_{chain B} and aa 502–508_{chain A}/505–508_{chain B}). DHTKD1 is structurally composed of an N-terminal helical bundle (aa 53–127) followed by three α/β domains (α/β1, aa 129–496, Pfam PF00676; α/β2, aa 528–788, PF02779; and α/β3, aa 789–915, PF16870). These four structural regions assemble into two halves, inter-connected by an extended linker (aa 497–527) that threads along the protein surface [Figs. 1(b) and 1(c)].

3.2. Structural comparison of DHTKD1 with other E1 enzymes

As expected, a DALI search (Holm & Sander, 1995) reveals that the closest structural homologue to hDHTKD1 is msOGDH [z score 55.7, root-mean-square deviation (RMSD) of 1.9 Å and 38% sequence identity] and ecOGDH (z score 50.3, RMSD of 1.8 Å and 40% sequence identity). The main structural divergence is found in their interdomain linkers, which traverse the respective protein surface via different trajectories [Fig. 1(d)]. Importantly, the hDHTKD1 linker (498–527) packs against two loop regions that are longer than the equivalents in ecOGDH and msOGDH [Fig. 1(e)]. These include the `active site loop' (aa 247–258), and a DHTKD1-unique region (aa 720–733) identified as `Δ3' in the work of Bunik & Degtyarev (2008) (Fig. S3). The path traversed by the hDHTKD1 linker, not an artefact from crystal packing (Fig. S4), also overlaps with the binding sites for the allosteric activators of ecOGDH [acetyl-CoA, (Frank et al., 2007)] and msOGDH [AMP, (Wagner et al., 2011)] revealed from their structures [Fig. 1(e), meshes]. These allosteric sites are probably not present in DHTKD1 structure because of low sequence conservation in the neighbourhood (Fig. S3).

To a lesser degree, hDHTKD1 is also structurally homologous to the E1 enzymes of human PDH and BCKDH (also known as 2-oxoisovalerate de­hydrogenase) [z score 30, RMSD of 3.0–3.5 Å and 16% sequence identity (Ævarsson et al., 2000; Ciszak et al., 2003)], which are heterotetramers built from two copies of two subunits [Fig. 1(f)]. This contrasts with DHTKD1, OGDH and presumably OGDHL, which are homodimers. PDH and BCKDH form a more compact shape, lacking several surface insertions to the α/β core that are unique to the DHTKD1/OGDH/OGDHL subgroup. These include the helical bundle at the N terminus, and the β- and helical hairpins (aa 527–565, 606–630) within the α/β2 domain [Fig. 1(f), red ribbons]. PDH and BCKDH structures also contain K⁺ binding sites that play a role in enzymatic regulation [Fig. 1(f), green spheres]. We did not observe any difference density that suggests metal binding in the equivalent region of DHTKD1. Metal-dependent regulation is also featured in mammalian OGDH enzymes (Rutter et al., 1989; Lawlis & Roche, 1981), mediated by Ca²⁺ binding motifs unique to the OGDH N terminus and a region equivalent to the DHTKD1 Δ3 (Rigden & Galperin, 2004). Again, these motifs are not present in prokaryotic OGDHs or DHTKD1.

3.3. The DHTKD1 active site favours 2OA as substrate

Each DHTKD1 subunit in the crystal homodimer is bound with a ThDP co-factor [Figs. 1(b) and 1(c)], at a site formed from both subunits (Fig. S5). The ThDP pyrophosphate moiety binds to the α/β1 domain of one subunit, the pyrimidine ring binds to the α/β2 of the other subunit, while the central thia­zolium ring sits between the subunits. The ThDP binding residues are highly conserved among OGDH and E1 homologues (Fig. S3). These include Asp333 and Asn366 which bridge the ThDP pyrophosphates with the Mg²⁺ ion. Also, Leu290 acts as a hydro­phobic wedge to form the characteristic V-shaped conformation of ThDP, bringing the N4′ amino group of the pyrimidine ring into close proximity (3.08 Å) with the C2 proton of the thia­zolium ring. Essential for catalysis, Glu640 triggers a proton relay to activate the co-factor into a reactive ylide (Fig. S5). Reaction then ensues via a nucleophilic attack by the ThDP ylide on the keto carbon of the substrate, forming a pre-de­carboxyl­ation intermediate that is in turn de­carboxyl­ated into an enamine-like ThDP adduct.

Compared with an apo E1 structure such as that of ecOGDH, our ThDP-bound DHTKD1 structure highlights four loop segments in the active site that undergo disorder-to-order transition during co-factor binding [Fig. 2(a)]. Using nomenclature from the work of Bunik & Degtyarev (2008) (Fig. S3), these include: `Region 1' (aa 187–195), which contributes Tyr190 to the substrate binding site; the active site loop (aa 247–258), with different length and sequence from OGDHs; `loop 1' (aa 366–383), which contributes the L₃₆₈GY₃₇₀ motif to bind the ThDP pyrophosphate; and `loop 2' (aa 434–445), which contributes residues to engage with the E2 enzyme for acyl­transfer (e.g. His435). The conformations seen in our holo structure are similar to those of msOGDH structures bound with the post-de­carboxyl­ation co-factor conjugates (Wagner et al., 2014) [Fig. 2(b)].

Figure 2 Substrate binding pocket of hDHTKD1. (a) A view of the active site from the hDHTKD1 structure (pink ribbon) bound with ThDP (sticks, white carbon atoms) and Mg^2+ (yellow), overlaid with the ecOGDH structure in the apo form (light blue ribbon; PDB code 2jgd), to highlight structural differences in various loop regions. (b) The same hDHTKD1 active-site view as in panel (a), overlaid with msOGDH structures in complex with the first post-de­carboxyl­ation intermediate (cyan ribbons for proteins, sticks with cyan carbon atoms for ligands; PDB code 3zht) and second post-de­carboxyl­ation intermediate (blue ribbons for proteins, sticks with blue carbon atoms for ligands; PDB code 3zhu). (c) The hDHTKD1 substrate pocket is lined by residues from both subunits of the homodimer (pink and blue lines). Overlaid in this pocket is the putative 2OA ligand (pink sticks) modelled in our density (Fig. S6) and the post-de­carboxyl­ation intermediates (cyan sticks) bound to the msOGDH structures. (d) The hDHTKD1–metabolite interactome as determined by MIDAS. hDHTKD1 significantly depleted 2OA and α-oxoisovalerate, and enriched dAMP and dGMP. The cutoff for significance was p < 0.05 and q < 0.1.

In one X-ray dataset, we observed OMIT map electron density at one active site of the homodimer that is not accounted for by the co-factor or any component of the crystallization condition [Fig. S6(a)]. This density is adjacent to but disjointed from the ThDP co-factor [Fig. S6(b)], at a location partly overlapping the two conformations of post-de­carboxyl­ation intermediate seen in the msOGDH structures (PDB codes 3zht and 3zhu; Wagner et al., 2014) [Fig. 2(b)]. The size of this density feature can accommodate a C6 ligand such as 2OA without covalent linkage to ThDP. Although the observed ligand, likely co-purified with the protein, did not undergo enzymatic turnover, the keto carbon can be placed at 3.5 Å from the ThDP thia­zolium C2 and hence be compatible with the nucleophilic attack and subsequent de­carboxyl­ation [Fig. S6(c)]. While in good agreement with OMIT map, the 2OA model was not included in the deposited structure, in light of no further experimental evidence of its presence.

DHTKD1 and OGDH overlap to some extent in their in vitro reactivity towards 2OG and 2OA (Nemeria, Gerfen, Yang et al., 2018; Leandro et al., 2019). For example, soaking msOGDH crystals with 2OA and 2OG both yielded similar post-de­carboxyl­ation intermediates (Wagner et al., 2014). Nevertheless, hDHTKD1 turns over 2OA with 40-fold higher catalytic efficiency than over 2OG (Nemeria, Gerfen, Yang et al., 2018). The hDHTKD1 active site reveals several amino acids poised to interact with the substrate, which are not conserved with OGDH orthologues. Two of them involve substitution to more polar residues i.e. Tyr190 (from Phe_OGDH) and Tyr370 (Phe_OGDH), while the other two involve substitution to less bulky residues i.e Ser263 (Tyr_OGDH) and Asp707 (Glu_OGDH) (Fig. S3). Overlaying the two msOGDH post-de­carboxyl­ation intermediates [first and second conformers, sticks in Fig. 2(b); Wagner et al. (2014)] onto the hDHTKD1 substrate pocket clearly explained how these substituted amino acids can stabilize catalytic intermediates generated from the longer 2OA substrate [Fig. 2(c)]. The 2OA terminal carboxyl group from the first conformer in msOGDH (PDB code 3zht) can be sandwiched between hDHTKD1 Tyr190 (Phe_OGDH) and Tyr370 (Phe_OGDH) to form polar interactions, while hDHTKD1 Ser263 (Tyr_OGDH) and Asp707 (Glu_OGDH) increase pocket volume to accommodate the terminal carboxyl group from the second conformer in msOGDH (PDB code 3zhu). It remains to be determined whether the post-de­carboxyl­ation intermediate of hDHTKD1 also exists in dual conformation. One can rationalize that the DHTKD1 substrate pocket is engineered to accommodate the slightly larger and more polar 2OA substrate, providing a structural basis for its superior catalytic efficiency over 2OG.

3.4. DHTKD1 preferentially interacts with 2OA in solution

We further explored the substrate preference of hDHTKD1 by mapping its metabolite interactome using MIDAS, a mass spectrometry-based equilibrium dialysis approach (Orsak et al., 2012). From a screening library of 412 human metabolites, 2OA was observed as the most significant (p < 4.33 × 10⁻⁵⁴, q < 2.59 × 10⁻⁵¹) interaction with hDHTKD1_45–919 in the presence of ThDP and Mg²⁺ [Fig. 2(d), see Supplementary File S1 in the Supporting information]. Furthermore, 2OA had the most negative Log2(corrected fold change) value (−1.17), suggesting that hDHTKD1 enzymatically processed 2OA during the MIDAS screening. Relative to 2OA, the 2OG interaction with hDHTKD1 was not significant (p < 0.17, q < 0.74) and had a relatively small negative Log2(corrected fold change) value (−0.28). The higher confidence and fold change observed for 2OA, relative to 2OG, are in complete agreement with the substrate preference of hDHTKD1.

α-Ketoisovalerate (also known as α-oxoisovalerate), the primary product of valine degradation by branched-chain-amino-acid amino­transferases, was the second most significant metabolite (p < 8.00 × 10⁻⁴, q < 2.06 × 10⁻²) and had the second most negative Log2(corrected fold change) value (−0.25), suggesting hDHTKD1 could interact with and may also enzymatically process α-oxoisovalerate. The de­oxy­purine monophosphates, dAMP and dGMP, had significant (p < 4.47 × 10⁻¹¹, q < 4.84 × 10⁻⁹ and p < 5.13 × 10⁻¹⁸, q < 1.07 × 10⁻¹⁵, respectively) positive Log2(corrected fold change) values (0.92 and 1.27), suggesting binding to hDHTKD1. These results support observations that purine nucleotides functionally regulate eukaryotic OGDHc (Lawlis & Roche, 1981; Craig & Wedding, 1980) and perhaps OADHc. Further experiments are warranted to understand the functional relevance of α-oxoisovalerate and nucleotide mono­phosphates on DHTKD1 activity.

3.5. DHTKD1 and DLST form direct interactions

There is literature evidence that the E1 and E2 components of 2-oxoacid de­hydrogenase complexes interact directly as a binary subcomplex, in the absence of E3 (Zhou et al., 2018; Park et al., 2004; Patel et al., 2009). For some E1 enzymes such as OGDH and PDH, the N terminus is known to be important for the direct interaction with E2 (Zhou et al., 2018; Park et al., 2004) and E3 (McCartney et al., 1998), although this region is notably different for DHTKD1. For example, the hDHTKD1 precursor encodes a mere 50-aa segment before the first α-helix of the structure, while the hOGDH equivalent region is longer (121 aa) and contains two DLST-binding motifs (Zhou et al., 2018) not preserved in DHTKD1 [Fig. S1(b)]. This suggests that the manner in which DHTKD1 and OGDH (E1) interact with DLST (E2) could be different.

hDLST as a precursor protein is structurally composed of [Fig. 3(a)]: the mitochondrial target sequence (aa 1–67), the N-terminal single lipoyl domain (aa 68–154) to which a di­hydro­lipoyl moiety is covalently attached through a lysine residue (Lys110), the C-terminal catalytic domain responsible for the multimeric assembly and harbouring the acyl-transferase active site (aa 211–453), and the flexible inter-domain linker (aa 155–210). We opted to reconstitute the DHTKD1–DLST binary complex by co-expressing His-tagged hDHTKD1_45–919 and hDLST_68–453 in E. coli followed by affinity chromatography. Untagged hDLST_68–453 was found to co-purify with His-tagged hDHTKD1_45–919 immobilized on Ni affinity resin [Fig. 3(b)], and to a similar extent with hDHTKD1_1–919 [Fig. 3(c)] and hDHTKD1_23–919 [Figs. 3(d), S7(a), S7(b) and S7(c)]. Hence the hDHTKD1 N-terminal 45 aa, not present in our structural model and replacing the DLST-binding motifs mapped for hOGDH, does not play a role in the DHTKD1–DLST interaction. Size-exclusion chromatography (SEC) using an analytical Superose 6 Increase column eluted the complex at V_e = 10.4 ml (V_e = elution volume), as compared with hDHTKD1_45–919 protein alone which eluted later at V_e = 16.3 ml [Fig. 3(d)]. Our attempts to mix the binary DHTKD1–DLST complex with purified DLD did not yield a stable three-way complex in SEC [Fig. S7(d)], as was the case shown for OGDHc previously (Zhou et al., 2018).

Figure 3 Interaction studies of hDHTKD1 and hDLST. (a) Constructs of hDHTKD1 and hDLST (black bars) used in the affinity pulldown experiments. (b)–(d) SDS–PAGE showing affinity pulldown of untagged hDLST68–453 from (b) His6-tagged hDHTKD145–919, (c) hDHTKD11–919 and (d) hDHTKD123–919. The original uncropped SDS–PAGE gels are shown in Fig. S7. For panel (b), the lanes loaded are: 1, flow-through; 2–6, wash fractions of increasing imidazole concentration; and 7–11, elution fractions with 250 mM imidazole. SDS–PAGE gel fragments from panels (c) and (d) show only the first three elution fractions i.e. lanes 7, 8 and 9 [equivalent to the lanes marked in red in panel (b)]. (e) Chromatogram and SDS–PAGE from SEC runs of hDHTKD145–919 protein alone (dashed line) and of hDHTKD145–919 co-expressed with hDLST68–453. Elution volumes for the complex peak and hDHTKD1-alone peak are shown. (f) An EM map of the hDLST 24-mer catalytic core overlaid with a humanized model of E. coli DLST (PDB code 1scz). The three views show how the trimer building block (monomers shown as blue, orange and green ribbons) is assembled into the 24-mer core via fourfold (left), twofold (middle) and threefold (right) symmetry axes, respectively. Inset of the left view shows how the first residue (aa 219, red spheres) from each of the 24 hDLST catalytic cores is distributed at the surface of the cube structure.

Similar DHTKD1–DLST complex can also be formed by co-expression in the baculo Sf9 cells. When expressed alone in Sf9, the hDLST_68–453 protein is highly prone to degradation, with a significant proportion fragmenting into two halves [Figs. S8(a) and S8(b)]. When the DHTKD1 and DLST proteins are co-expressed, hDHTKD1_45–919 co-purified in SEC together with both the hDLST_68–453 intact protein and the C-terminal fragment (containing the catalytic core), while the N-terminal fragment (containing the lipoyl domain and linker) was not part of this complex [Fig. S8(c)]. This suggests that the DLST N-terminal fragment alone is not sufficient to interact with DHTKD1, although this DLST region was previously mapped to be interacting with the binding motifs at the hOGDH-unique N terminus (Zhou et al., 2018).

Altogether, our data reinforce the notion that DHTKD1 and OGDH interact with DLST differently despite the structural conservation. This difference is not surprising considering these enzymes are present in distinct cellular contexts. While DHTKD1 is responsible for the last step of lysine and tryptophan catabolism, OGDHc operates as a rate-limiting step in the Krebs cycle. Therefore, they are expected to be regulated by different mechanisms, both spatially (e.g. by employing distinct binding partners, co-factors and post-translational modifications) and temporally (e.g. by displaying different affinity/kinetics towards binding partners/co-factors).

3.6. Insight into complex assembly from cryo-EM and SAXS studies

To provide a structural context for the DHTKD1–DLST interactions, we attempted single-particle cryo-EM on the reconstituted binary complexes co-expressed in E. coli and Sf9 cells. Electron micrographs displayed the characteristic cubic cage structures of approximate dimensions 130 × 130 × 130 Å (Fig. S9), as observed for E. coli DLST (Knapp et al., 1998) and other E2 enzymes such as Azotobacter vinelandii PDH E2 (Mattevi et al., 1992) and bovine BCKDH E2 (Kato et al., 2006).

We collected a dataset from the Sf9 co-expressed complex and generated a 3D reconstruction at 4.7 Å global resolution derived from 3356 particles (Figs. S10, S11 and Table S1). Local resolution analysis reveals a range between 4.7 and 6.3 Å [Figs. S11(a) and S11(c)]. The low resolution may in part be explained by orientation bias along the fourfold symmetry axis [Fig. S11(b)]. The EM reconstruction shows 24 DLST C-terminal catalytic domains assembled as eight trimer building blocks into a cubic cage with octahedral symmetry [Fig. 3(f)] and allows tracing of a humanized DLST model (aa 219–453 of hDLST) based on the E. coli structures [PDB codes 1e2o and 1scz; 60% identity; Albert et al. (2000); Knapp et al. (1998)] [Fig. S11(f)].

Considering the sequence conservation, the catalytic cores of E. coli and hDLST display essentially identical topology and symmetry along two-, three- and fourfold axes [Fig. 3(f)]. In this assembly, all 24 C-terminal catalytic domains have their first residue (aa 219) exposed to the surface of the core [Fig. 3(f), inset], presumably projecting the adjacent inter-domain linker outwards from the core in order to deliver the N-terminal lipoyl domain for engagement with E1 and E3. After processing the dataset with extensive 3D classification comparisons with and without imposing symmetry (Fig. S10), there is unfortunately no discernible density for further regions of DLST (e.g. N-terminal lipoyl domain) or for the DHTKD1 protein. It is likely that the DLST lipoyl domain and much of the linker region are highly flexible and dynamic, in agreement with previous attempts to structurally characterize other full-length E2 enzymes such as human PDH E2 (Yu et al., 2008). Additionally, the DHTKD1–DLST inter­action could be short lived, as shown for other E1–E2 complexes. It is also possible that this region could be partially denatured by the air–water interface.

We also subjected the E. coli co-expressed complex to cryo-EM and observed more heterogeneous particles on the micrograph, some of which reveal extra density emanating from the cubic core to ∼10–20 Å [Fig. S9(a)]. This contrasts with the aforementioned Sf9 co-expressed complex, where particles are predominantly homogenous and contain only cubic cages [Fig. S9(b)]. Five 2D classes comprising of 272 particles from the E. coli co-expressed complex (manually picked across three screening micrographs and representing 48% of the maximum particles available) were compared with 2D classes with an equivalent number of particles picked from the large dataset of the Sf9 co-expressed complex [Fig. S9(c)]. Again, loosely defined and heterogeneous density is visible at the corners of the cubic core within the E. coli co-expressed complex, whereas this extra density is missing or lost through 2D classification of the equivalent Sf9 co-expressed complex. We reasoned that the additional density protruding from the core represents an ordered segment of the linker region and perhaps in some instances engages with the interaction partner DHTKD1. However, a larger high-resolution dataset of the E. coli complex would be required to discern the exact contribution of this additional density.

The positioning of 24 lipoyl domains at the exterior of the DLST core implies that they are all potentially available for engagement with E1 and E3. To explore the underlying stoichiometry for the DHTKD1–DLST interaction, we characterized the Sf9 co-expressed binary complex using SEC-SAXS (Fig. S12). The molecular weight (MW) of hDHTKD1_45–919:hDLST_68–453 derived from SAXS porod volume is 2.45 MDa, which is in close agreement with a MW of 2.7 MDa calculated for 24 × DLST and 16 × DHTKD1 protomers, assuming a stoichiometry of one DHTKD1 dimer per DLST trimer building block as suggested previously for hOGDHc (Zhou et al., 2018). It remains unknown whether the two active sites within a DHTKD1 dimer are engaged by one or two lipoyl domains. With either possibility, it is apparent that not all 24 lipoyl domains from one DLST core were engaged with DHTKD1 at the same time.

3.7. Structural mapping of disease-causing DHTKD1 mutations

To date, seven DHTKD1 missense mutations have been reported as the molecular cause for 2-amino­adipic and 2-oxoadipic aciduria. At the protein level, three (p.L234G, p.Q305H and p.R455Q) are located within the α/β1 domain, while the other four (p.R715C, p.G729R, p.P773L and p.S777P) are clustered in the α/β2 domain [Figs. 1(a) and 1(c)]. From over 100 DHTKD1 and OGDH orthologues surveyed, the aa 715 position is invariantly Arg, while aa positions 305 and 777 are also highly conserved (82% and 93%, respectively) [Fig. 4(a)]. None of these residues directly affect the conserved catalytic machinery common to the 2-oxoacid de­hydrogenase family.

Figure 4 Characterization of hDHTKD1 disease-causing variants. (a) Weblogo diagrams generated from an alignment of >100 DHTKD1 orthologues and homologues, showing sequence conservation for the regions surrounding the seven missense mutation sites. (b) Small-scale expression and purification for hDHTKD1 WT and variants. SDS–PAGE gel slices showing lysate (L) and eluant (E) samples from affinity purification of hDHTKD145–919 WT, p.S777P and p.L234G. The position of the hDHTKD1 bands is marked by an arrow. Original uncropped gels are shown in Fig. S13 (lanes i–v were cropped from one gel and lanes vi–vii were cropped from another). (c) DSF melting curves for hDHTKD145–919 WT and p.P773L, with an inset showing the derived melting temperature (Tm) values for WT and five variants. (d) Affinity pulldown of hDLST68–453 by immobilized His-tagged hDHTKD145–919 p.R715C variant. The SDS–PAGE shown is one of two biological replicates (Fig. S15). The lanes are loaded with the following samples: 1, flow through; 2–6, wash fractions of increasing imidazole concentration; and 7–11, elution fractions with 250 mM imidazole. (e) hDHTKD1 homodimer showing the sites of the seven missense mutations (black spheres) on one subunit (yellow ribbon). (f)–(h) Atomic environments surrounding the mutation sites (f) p.L234G, (g) p.P773L and p.S777P, and (h) p.R715C. All amino acids shown, including sites of missense mutations, are of the WT protein.

To understand their putative biochemical defects, we reconstructed the seven DHTKD1 missense mutations recombinantly. All hDHTKD1_45–919 variants were expressed as soluble protein to a similar level as WT, with the exception of p.L234G and p.S777P which showed significantly lower yields and propensity to degradation, suggesting these variant proteins are misfolded compared with WT [Figs. 4(b) and Fig. S13]. Leu234 is located at the protein centre ∼20 Å from the co-factor site [Fig. 4(e)] and the p.L234G change introduces a smaller side chain thereby leaving a cavity at the hydro­phobic core [Fig. 4(f)]. Ser777 is partially exposed to the protein surface [Fig. 4(e)] and the p.S777P change introduces a proline side chain that probably disrupts hydrogen bonds with neighbouring residues [Fig. 4(g)].

The five remaining variant proteins were isolated, purified and subjected to thermostability analysis by DSF. Four of them (p.Q305H, p.R455Q, p.R715C and p.G729R) demonstrated similar single unfolding–folding transition as hDHTKD1 WT [Fig. 4(c), inset]. However, p.P773L exhibited a significantly reduced melting temperature (ΔT_m = −5.2°C), suggesting that while expressed as soluble protein this variant is thermally more labile than WT [Fig. 4(c)]. Pro773 forms a bend for the surface-exposed loop which connects the α/β2 and α/β3 domains and also harbours the above-mentioned Ser777. Replacing Pro773 with Leu probably alters the structural integrity of this loop [Fig. 4(h)] and could affect protein folding. The observation of two destabilizing mutations within this one loop region strongly implies its importance in the functioning of DHTKD1.

Arg715 is located at the twofold axis of the homodimer and together with Arg712 forms a salt-bridge network with Asp677 of the opposite subunit [Fig. 4(h)]. Arg712, Arg715 and Asp677 are invariant amino acid positions across DHTKD1 and OGDH orthologues, indicating the importance of this salt-bridge network. Arg715 is positioned immediately after `loop 3', a signature motif conserved across all E1 enzymes, including the invariant His708 that is involved in the reductive acyl transfer to E2 (Fig. S3) (Wynn et al., 2003). SEC of the p.R715C variant revealed a similar chromatogram profile to WT hDHTKD1_45–919, indicating intact dimer formation (Fig. S14). Nevertheless, when assayed in our co-expression and affinity pulldown, the hDHTKD1_45–919 p.R715C variant has significantly reduced ability to bind hDLST_68–453 directly [Figs. 4(d) and S15], compared with WT [Fig. 3(b)]. Therefore, the effect of p.R715C substitution could be transmitted from the dimer interface to engagement with E2 through loop 3. As a control, hDHTKD1_45–919 bearing the p.R455Q or p.P773L substitution (both located at the protein exterior) interacts with hDLST_68–453 to a similar extent as WT.

Our data did not reveal any discernible defects on protein stability or interaction with DLST in vitro for the p.Q305H, p.R455Q and p.G729R variants, the latter two being found in the majority of reported cases of 2-amino­adipic and 2-oxo­adipic aciduria. These results imply that additional functions or unknown binding partners could be involved in the OADHc. Future efforts can be focused on studying their in vivo impact using patient-relevant cells.

4. Conclusions

DHTKD1 is emerging as a key player in mitochondrial metabolism through its influence in lysine metabolism, energy production and ROS balance. The structure of DHTKD1 presented here provides the first template for the rational design of DHTKD1 small-molecule modulators to probe the enzyme's role in these mitochondrial functions and the associated disease states. DHTKD1 exhibits key structural differences from other E1 enzymes, particularly OGDH, with the key finding being that the active-site substrate pocket in DHTKD1 is larger in size and of more polar character than OGDH. These subtle modifications would favour the 2OA substrate, providing a molecular basis for the subtle difference in substrate specificity that was reported previously. The additional sequence and structural variations at the protein exterior would also impact on protein–protein interaction and explain how the two enzymes could, to some extent, engage with the cognate E2 and E3 components differently despite their close homology. These features are likely to be exploited by future chemistry efforts for the development of DHTKD1-specific modulators.

We have reconstituted the DHTKD1–DLST complex in vitro and demonstrated for the first time that complex formation is disrupted in some disease-causing variants, probably via indirect (e.g. destabilizing DHTKD1 to reduce its steady-state level) or direct (e.g. altering the binding interface of DHTKD1) mechanisms. These data underscore the importance of DHTKD1 functioning within the context of the OADHc complex. Our cryo-EM and complementary studies provide insight into how DLST forms a multimeric core to recruit multiple DHTKD1 protomers into this mega assembly. Considering that both DHTKD1 and OGDH recruit DLST as their E2 component, future studies are warranted to explore the existence of a `hybrid' complex in which lipoyl domains from one DLST multimeric core could engage with both DHTKD1 and OGDH at the same time. This could allow crosstalk and regulation between the OADHc and OGDHc complexes for their concerted mitochondrial functions.

5. Related literature

The following reference is cited in the supporting information for this article: Svergun et al. (1995).

6. Data availability

X-ray coordinates and structure factors have been deposited to the Protein Data Bank (PDB) (accession code 6sy1). Cryo-EM data have been deposited to the Electron Microscopy Data Bank (EMDB) (accession code EMD-11014).

Footnotes

‡These authors contributed equally to this work.