Mycobacterium tuberculosis assembles a unique hexameric E2p core of the pyruvate dehydrogenase complex

TL;DR

Mycobacterium tuberculosis has a unique structure for a key metabolic complex, which helps it manage energy and stress.

Contribution

Discovery of a noncanonical hexameric E2p core in Mtb's pyruvate dehydrogenase complex.

Findings

DlaT assembles into hexamers and dodecamers at micromolar concentrations.

Hexamer is the functional E2p core of Mtb's PDHc.

Unique interfaces prevent formation of classic 24- or 60-mer structures.

Abstract

The pyruvate dehydrogenase complex (PDHc) is a universally conserved multienzyme system that converts pyruvate into acetyl-CoA for entry into the tricarboxylic acid cycle and for NADH production. Its central scaffold, the dihydrolipoyl transacetylase (E2p), forms an oligomeric inner core that recruits pyruvate dehydrogenase (E1p) and dihydrolipoyl dehydrogenase (E3). All previously characterized PDHc assemblies adopt either an octahedral 24-mer or an icosahedral 60-mer E2p core, each constructed from trimeric building blocks. We recently showed that the Mycobacterium tuberculosis (Mtb) E2p protein DlaT also functions as the core of the pathogen’s peroxynitrite reductase/peroxidase complex. Here, using cryo-EM, we demonstrate that DlaT assembles into discrete hexamers and dodecamers at micromolar concentrations, which approximate intracellular DlaT concentrations in Mtb. Structure-guided mutagenesis combined with in vitro activity assays indicates that the hexamer represents the functional E2p core of the Mtb PDHc. This noncanonical architecture arises from unique interfaces between DlaT trimers that preclude formation of the classic spherical 24- or 60-mer structures. We propose that this specialized E2p organization enables Mtb to regulate metabolic activities and to remodel the E2p core for engagement in the peroxynitrite reductase/peroxidase antioxidant pathway under stress. Our findings reveal an unexpected diversity in PDHc architecture and uncover a distinct organization principle for the core metabolic complex in mycobacteria.

Figures

Full text

The pyruvate dehydrogenase complex (PDHc) maintains cellular glucose homeostasis by linking glycolysis to the tricarboxylic acid cycle. Deficiency in PDHc function impairs carbohydrate utilization in both prokaryotes and eukaryotes and in humans manifests as lactic acidosis and neurological disorders (1, 2, 3). PDHc belongs to the structurally and functionally related family of two-oxoacid dehydrogenase complexes, which also includes the 2-oxoglutarate dehydrogenase complex (OGDHc) and the branched-chain α-ketoacid dehydrogenase complex (BCKDHc) (4, 5, 6, 7, 8). These complexes share a common tripartite organization consisting of E1 decarboxylase, an E2 acyl transferase, and the universally shared E3 dihydrolipoyl dehydrogenase (8). Each complex contains its specific E1 and E2 components (E1o/E2o for OGDHc; E1b/E2b for branched-chain α-ketoacid dehydrogenase complex), whereas E3 is common to all. Genes encoding these complexes are often clustered in operons and cotranscribed under a single promoter. PDHc comprises pyruvate dehydrogenase (E1p), dihydrolipoyl transacetylase (E2p), and E3. Together, they catalyze three sequential reactions that convert pyruvate into acetyl-CoA (6, 9). E1p decarboxylates pyruvate to form a C2α-hydroxyethylidene–thiamine pyrophosphate intermediate, which then attacks the oxidized lipoate at the E2p lipoyl domain (LD) to form S8-acetyldihydrolipoyl-E2p. E2p transfers the resulting acetyl group from its swinging lipoyl arm to CoA to generate acetyl-CoA and dihydrolipoate (10). Finally, E3 reoxidizes dihydrolipoate through reduction of its active site Cys–Cys pair and subsequent electron transfer to FAD cofactor and NAD+, yielding NADH.

PDHc is highly organized around a symmetric inner core formed by the E2p catalytic domain (CD) (11). In Gram-negative bacteria, such as Azotobacter vinelandii and Escherichia coli, the E2p CD assembles as a 24-mer cubic octahedron (12, 13, 14, 15), whereas Gram-positive bacteria and eukaryotes form a 60-mer icosahedral core (11, 16, 17). In higher eukaryotes, the E3 binding protein (protein X) may replace or supplement some E2p within this core (18, 19, 20, 21, 22). E2p consists of a multidomain N-terminal LD, a peripheral subunit–binding domain (PSBD), and a C-terminal CD, connected by flexible linkers that allow the LDs to access E1p and E3 active sites (10, 23, 24). E1p and E3 bind the E2p PSBD in a mutually exclusive but selective manner (25).

All structurally characterized PDHc inner cores adopt either a 24-mer or 60-mer architecture, with one exception: Corynebacterium glutamicum (Cglu) forms a minimal trimeric E2p inner core without further oligomerization (26). Because Mycobacterium tuberculosis (Mtb) and Cglu belong to the same order (Mycobacteriales), the same study proposed that the Mtb PDHc may also form similar trimeric cores. Our long-standing interest in the Mtb metabolic and antioxidant pathways revealed their intersection at E2p (DlaT) (27, 28, 29, 30, 31). DlaT supports classical PDHc-mediated carbon utilization as well as the mycobacteria-specific peroxynitrite reductase/peroxidase system. In the latter, reduced lipoate on E2p powers the thioredoxin-like protein AhpD, which in turn reduces the peroxiredoxin AhpC to repair oxidatively damaged proteins (29, 30). We previously showed that reduced E2p could be generated either through E1p/E1o-dependent ketoacid decarboxylation and subsequent acyl transfer to CoA or directly by NADH-dependent reduction by E3, providing metabolic flexibility in supporting antioxidant activities. This multifunctionality likely contributes to the role of DlaT in Mtb virulence (32). Moreover, small-molecule inhibitors targeting DlaT selectively kill nonreplicating Mtb, a major reservoir of latent infection (33).

These observations prompted us to ask whether the structural features of the Mtb PDHc core and its interactions with partner enzymes could inform further inhibitor development and explain how E2p supports both metabolic and antioxidant functions. In particular, we sought to determine whether Mtb E2p architecture may confer enhanced accessibility of the AhpD trimer to the E2p LDs. Here, we report that the Mtb E2p core adopts a previously unrecognized hexameric organization with concentration-dependent oligomeric flexibility, likely reflecting adaptation to the metabolic and stress environment encountered by the pathogen. Through structural and biochemical analyses, we show that DlaT assembles a dimer-of-trimers hexamer, that this hexamer is the probable functional PDHc core, and that its formation is driven by a distinct β-propeller element. Thus, the Mtb PDHc features a unique E2p inner core architecture that is distinct from previously known models.

Results

The hexamer is the basic unit of the Mtb E2p core

Mtb DlaT contains two N-terminal LDs (LD1 and LD2), a PSBD, and a C-terminal CD (Fig. 1A). We expressed DlaT in E. coli, purified it to homogeneity, confirmed its in vitro activity, and verified lipoylation at Lys43 and Lys162 by Western blot and mass spectrometry (Figs. S1–S3). Cryo-EM analysis of full-length DlaT yielded three EM maps. The first, resolved to 2.65 Å with D3 symmetry, represents a hexamer composed of two back-to-back trimers (Figs. 1B, S4 and S5). The second, resolved to 2.51 Å with D2 symmetry, corresponds to a dodecamer formed by two side-by-side associated hexamers (Figs. S4 and S6, A–E). The third map, obtained from heterogeneous refinement, contained three hexamers and accounted for ∼28% of particles originally classified as dodecamers (Fig. S6F).Figure 1**Mtb DlaT assembles into a hexamer and a dodecamer in vitro.**A, domain organization of Mtb DlaT. B and C, cryo-EM map (B) and corresponding atomic model (C) of the DlaT hexamer. D and E, DlaT remains hexameric in the presence of CoA. The EM density was shown with the fitted atomic model (D), and an enlarged view highlights the bound CoA (cyan) and coordinating residues (E). Mtb, Mycobacterium tuberculosis.

An AlphaFold model (UniProt ID: P9WIS7) was used as the starting template for building the hexamer structure. For clarity, residues from neighboring subunits within a DlaT trimer are labeled with a prime (′); residues from the second trimer of the same hexamer are denoted by a double prime (″); and residues from the second hexamer in a dodecamer are labeled with a triple prime (‴). Consistent with previous E2 structures from prokaryotic and eukaryotic 2-oxoacid dehydrogenase complexes, only the CD (from His313 to the C terminus) was resolved, whereas LD1/LD2 and the PSBD were not visible, reflecting their high mobility (16, 34, 35, 36) (Fig. 1, A and C). The CD assembles hierarchically: individual monomers first form trimers, and two trimers then stack back to back to create a dumbbell-shaped hexamer (Fig. 1C). Although the trimer is the universal building block of all known 2-oxoacid dehydrogenase complexes (37), the Mtb DlaT trimer does not further assemble into the canonical octahedral (24-mer) or icosahedral (60-mer) architectures. Instead, DlaT forms discrete hexamers and higher oligomers, built from hexamers rather than from trimers.

Because DlaT catalyzes transfers of an acetyl group from S-acetyl dihydrolipoamide to CoA, and the active site lies near the trimer edge where the hexamers assemble, we asked whether the absence of substrate might have biased the oligomeric state. A medium-resolution cryo-EM structure of CoA-bound DlaT showed that substrate binding does not alter the dimer-of-trimers architecture (Figs. 1, D and E and S7 and S8).

Subunit–subunit and trimer–trimer interactions in the Mtb hexameric E2p inner core

Within each DlaT trimer, adjacent subunits form two β-strand–mediated interfaces (Fig. 2, A–C). The first involves pairing of the Gly316–Lys320 strand with the opposing Lys401′–Tyr405′ strand; the second pairs Leu474–Thr477 with Thr343′–Thr345′. Additional hydrogen bonds—Arg315–Glu409 and Asp476–Tyr521—further stabilize these antiparallel interactions. Overall, the trimer structure closely resembles that of other E2p acyl transferases (38, 39).Figure 2**Subunit–subunit and trimer–trimer interfaces in the Mtb DlaT hexamer.**A–C, polar interactions at the subunit–subunit interface within the trimer. Local EM density (B and C) shows how each subunit contributes one β-strand to form a two-stranded antiparallel β-sheet. D, the trimer–trimer interface of the hexamer. The middle zoom shows the two opposing antiparallel β-sheets; the right zoom highlights key polar interactions. Hexamer formation occurs through back-to-back association of the β-propeller regions of two trimers. E, structural alignment of the Mtb DlaT antiparallel β-sheet with corresponding elements from E2p proteins assembles octahedral (Escherichia coli E2p, PDB code: 4N72; Mtb E2b, PDB code: 6ZZN) or icosahedral (human E2p, PDB code: 6CT0) cores. Cglu E2p (PDB code: 6ZZI) is the only E2p trimer that does not assemble into higher-order structures. F, mass photometry of Cglu E2p at 20 nM and 10 μM, showing exclusively trimeric species. G, sequence alignment of the antiparallel β-sheet region involved in trimer–trimer interactions in Mtb DlaT and Cglu E2p. H, structural comparison of the trimer–trimer interface in Mtb DlaT (green and brown) and in Cglu E2p (dark and light gray). The absence of the Mtb DlaT Arg498-equivalent residue in Cglu E2p causes rotation of the β-sheet toward the catalytic domain, shortening and rigidifying the loop, and weakening potential intertrimer interactions. The Cglu E2p trimer was docked based on secondary structure alignment with Mtb DlaT. Cglu, Corynebacterium glutamicum; Mtb, Mycobacterium tuberculosis; Ec, Escherichia coli; Hs, Homo sapiens; PDB, Protein Data Bank.

Hexamer formation occurs through a back-to-back association of two trimers (Fig. 2D). Each trimer presents three two-stranded antiparallel β-sheets (residues 494–513) that project from its base to form a three-bladed β-propeller (Fig. 2D). The two opposing β-propellers interact through three antiparallel contacts, each stabilized by three hydrogen bonds: two between the main-chain atoms of Ile509 and Ile509″ and one between the side-chain atoms of Ser508 and Ser508″ (Fig. 2D).

To understand why Mtb DlaT adopts this unique dimer-of-trimers architecture, we compared its antiparallel β-sheets with those of homologous E2p structures (Fig. 2E). In Mtb DlaT, the β-sheet and its connecting loop are extended and bent backward, positioning the Ser508–Ile509 dipeptide to pair with the same motif from the opposing trimer. By contrast, the corresponding β-sheets in human E2p (code: 6CT0), E. coli E2p (PDB code: 4N72), and Mtb E2b (PDB code: 6ZZN) are shorter, placing the analogous dipeptides too close to the loop to permit intertrimer pairing. Moreover, the residue equivalent to Mtb Ser508 is glycine in human E2p and glutamate in E. coli E2p and Mtb E2b—substitutions that either eliminate the stabilizing hydrogen bonds or introduce charge repulsion, preventing hexamer formation.

Structural basis for trimer assembly in Cglu E2p versus hexamer formation in Mtb DlaT

Cglu E2p also contains an antiparallel β-sheet structure with a Ser–Ile motif, yet crystallography suggests a trimeric assembly (26). In the Cglu structure (PDB code: 6ZZK), a crystallographic twofold axis links two trimers in a configuration reminiscent of the Mtb DlaT hexamer. To determine whether Cglu E2p forms hexamers in solution, we expressed and purified the protein and analyzed it by mass photometry and cryo-EM (Figs. S9 and S10). At both high (10 μM) and very low (20 nM) concentrations, Cglu E2p behaved exclusively as a trimer (Fig. 2F), and cryo-EM reconstruction confirmed a trimeric state under conditions identical to those used for Mtb DlaT (Fig. S10E), consistent with prior crystallographic observations (26).

Although Cglu and Mtb E2p share high sequence homology across LD, PSBD, and CD, the antiparallel β-sheet responsible for hexamer formation in Mtb is less conserved (Fig. 2G). Notably, Cglu E2p lacks the residue corresponding to Mtb DlaT Arg498, resulting in a slight rotation of the β-sheet toward the CD and a shorter, more rigid loop. In Mtb DlaT, in addition to the Ser508–Ile509 pairing, weaker (3.5–3.6 Å) interactions occur between the loop and Arg512 from the other trimer (Fig. 2H). In Cglu E2p, the equivalent loop remains ∼7 Å away, preventing these interactions and disrupting the Ser–Ser stabilization observed in Mtb. These structural differences likely underlie the distinct oligomeric preferences of the two enzymes—hexameric for Mtb DlaT and trimeric for Cglu E2p.

DlaT hexamers further assemble into higher-order oligomers

As described above, Mtb DlaT hexamers can further associate to form larger assemblies (Fig 3, A and B). Approximately 60% of particles correspond to a two-hexamer dodecamer (Fig. 3A), whereas ∼28% of particles initially classified as dodecamers contain an additional hexamer, forming a three-hexamer complex (Fig. 3B). A fourth DlaT hexamer can also be accommodated, although this larger assembly may only form at elevated protein concentrations (Fig. 3C). These observations demonstrate that the hexamer—not the trimer—is the fundamental building block of the Mtb E2p core, in contrast to the octahedral and icosahedral cores of other PDHc systems, which use trimers as their basic unit.Figure 3**The Mtb DlaT hexamer is the fundamental building block for higher-order oligomers.**A, cryo-EM map and atomic model of the two-hexamer (dodecameric) assembly. B, cryo-EM map overlaid with the atomic model of three-hexamer assembly. C, schematic illustrating the stepwise assembly of higher-order DlaT oligomers. The green hexamer in the four-hexamer model was computationally docked using the same hexamer–hexamer interface observed in the experimentally determined two- and three-hexamer structures. Mtb, Mycobacterium tuberculosis.

Across the 2-, 3-, or 4-hexamer assemblies, the interfaces are structurally identical. Hexamers associate side by side with their long axis tilted ∼60° relative to one another. A conserved 10-residue peptide (Gly430–Arg439) from each hexamer interacts antiparallelly with the same segment from the neighboring hexamer (Fig. 4, A and B). The dodecameric interface is supported by hydrogen bonds—between Asp431 and Ser433‴ and between Gly430 and Ser433‴—as well as a salt bridge linking Asp431 to Arg439‴ (Fig. 4B).Figure 4**Structural basis of hexamer–hexamer interactions in higher-order DlaT oligomers.**A, EM density at the interface between two hexamers. B, enlargement of the interface, highlighting the polar contacts between hexamers; interacting residues are shown in sticks. C, interaction mediating the octahedral inner core of Mtb E2b (PDB code: 6ZZN). A C-terminal α-helical turn (amino acids 387–393, salmon) from one trimer inserts into a pocket formed by residues 277 to 280 and the C-terminal helix (amino acids 387–393, light blue) of an adjacent trimer. D, interaction mediating the icosahedral inner core of human E2p (PDB code: 6CT0). A C-terminal α-helix (amino acids 642–647, brown) inserts into a pocket formed by residues 531 to 534 and the partner C-terminal helix (amino acids 642–647, blue). E, in contrast, Mtb DlaT cannot form octahedral or icosahedral assemblies because its C-terminal α-helix (green) occupies the position that would form the canonical pocket. Instead, this helix interacts directly with the C-terminal helix of a neighboring trimer (golden), mediating hexamer–hexamer association. F, alignment of C-terminal helices from E2p/E2b proteins. The terminal helices of Mtb DlaT (hexamer/dodecamer) and Cglu E2p (trimer, PDB code: 6ZZI) project in orientations distinct from those in octahedral and icosahedral cores. G, comparison of trimer arrangement in E2p cores. Trimers associate with a bending angle in octahedral (cyan) and icosahedral (blue) assemblies but align nearly parallel in the Mtb dodecamer (green). Mtb, Mycobacterium tuberculosis; Hs, Homo sapiens; Bt, Bos taurus; PDB, Protein Data Bank.

Interestingly, the homologous sequences of the hexamer–hexamer interface in Mtb E2b (Asp277–Thr280; PDB code: 6ZZN) and human E2p (Ile531–Val534; PDB code: 6CT0) participate in a different structural role: together with the C-terminal helix, they form a pocket that receives the terminal helix of an adjacent trimer, thereby creating the concatenated trimers that build octahedral or icosahedral scaffolds (Fig. 4, C and D). In Mtb DlaT, however, this pocket is preoccupied by its own final helical turn (Gly545–Leu553), preventing insertion of a partner helix (Fig. 4E). Alignment of C-terminal helices from E2p and E2b subunits further shows that the terminal helices of Mtb DlaT and Cglu E2p (PDB code: 6ZZI) point away from the orientation required for classical octahedral or icosahedral assemblies (Fig. 4F).

Thus, the unique configuration of the Mtb DlaT C-terminal helix blocks formation of the canonical spherical E2p interface and instead enforces the use of the decapeptide (residues 430–439) to mediate hexamer–hexamer interactions. This results in a fundamentally different arrangement of neighboring trimers compared with the architectures of octahedral and icosahedral E2p cores (Fig. 4G).

Cellular DlaT concentrations support the hexameric assembly

Enzyme oligomerization can be sensitive to protein concentration. Because the hexameric state of Mtb DlaT was observed by cryo-EM at 14 μM, we sought to determine whether intracellular concentrations in mycobacteria are compatible with this assembly state. To this end, we quantified DlaT levels in Mtb or Mycobacterium smegmatis (Msm) by Western blot densitometry of cell-free soluble extracts, using purified recombinant Mtb DlaT as a calibration standard. We estimated intracellular DlaT concentrations to be at the lower end of micromolar concentrations (∼1–5 μM) in Mtb and 20 to 40 μM in Msm (Figs. 5, A and B and S11). The Mtb value is likely an underestimation because lysates prepared under Biosafety Level 3 conditions required filtration for decontamination, which could result in partial material loss prior to analysis. Even so, our measurement aligns with previous proteomic datasets reporting that DlaT is among the most abundant proteins in Mtb, ranking in the top 5% of the proteome (∼2000 ppm; ∼5 μM; https://pax-db.org).Figure 5**Cell abundance of DlaT.**A, quantification of intracellular DlaT in Mtb using purified Mtb DlaT as a standard. B, quantification of intracellular DlaT in Msm was also calibrated using purified Mtb DlaT as a standard. Numbers above the blots refer to the micrograms of cellular lysates or purified recombinant DlaT per lane. A, shows a representative experiment, with additional repeats provided in Fig. S11. B, each lane on the blot represents an independent preparation of cell-free soluble extracts from Msm. Msm, Mycobacterium smegmatis; Mtb, Mycobacterium tuberculosis.

Together, these data indicate that endogenous DlaT concentrations in mycobacteria are well within the range that favors hexamer formation, supporting the physiological relevance of the hexameric assembly observed in vitro.

The hexamer is the likely functional unit of the Mtb E2p core

To assess how oligomerization influences DlaT function, we introduced mutations at (i) the trimer–trimer interface that forms the hexamer and (ii) the hexamer–hexamer interface that mediates higher-order assemblies. At the trimer–trimer interface, we generated five variants: a conservative substitution (S508A) that alters side-chain hydrogen bonding, three mutations predicted to disrupt main-chain pairing (I509P, S508A/I509P, and S508P/I509P), and an E2br chimera in which the Mtb DlaT antiparallel β-sheet (residues 497–511) was replaced with the corresponding region from Mtb E2b (residues 343–352). All constructs were expressed in E. coli. Two variants (S508A/I509P and E2br) were either degraded or insoluble (Fig. S12). The remaining three (S508A, I509P, and S508P/I509P) were purified and analyzed by gel filtration, mass photometry, and activity assays (Fig. 6).Figure 6Functional importance of the Mtb DlaT hexamer. A, size-exclusion chromatography profiles of WT and trimer–trimer interface mutant proteins (S508A and I509P) on a Superose 6 Increase 10x300 mm column. S508A shifts toward higher mass, whereas I509P shifts toward lower mass. B, mass photometry–based molecular mass distributions of WT and mutant DlaT proteins. Measurements nominally labeled “10 μM” were acquired using the MassFluidix HC add-on, which performs rapid (∼50 ms) on-chip dilution of the samples by ∼5000-fold before detection. Accordingly, 10 μM reflects the predilution loading concentration rather than the concentration at which complexes were measured. C, SDS-PAGE (left) and native PAGE (right) analyses of WT and mutant DlaT variants. D, PDH activity assay of WT and mutant Mtb DlaT proteins. E, concentration-dependent PDH activity of WT and S508A. S508A variant enriched for higher-order oligomers shows increased enzymatic activity. Mtb, Mycobacterium tuberculosis; PDH, pyruvate dehydrogenase.

The I509P protein eluted at a smaller apparent mass than WT DlaT (Fig. 6A). Mass photometry at 20 nM revealed a mixture of monomer (∼67%) and trimer (∼33%) (Fig. 6B). Cryo-EM 2D classification confirmed the trimer as the dominant species (Fig. S13) for I509P. The functional PDH assay revealed that the I509P protein retained minimal (∼9%) activity in comparison to the WT PDH activity. In contrast, S508A displayed a larger apparent mass in gel filtration and in native PAGE (Fig. 6, A and C). Mass photometry indicated that S508A remained predominantly hexameric at 20 nM (Fig. 6B), implying enhanced trimer–trimer affinity. PDH assays showed that S508A was ∼50% more active than WT DlaT (Fig. 6D). A head-to-head comparison between the WT and S508A DlaT in the PDH assay at variable concentrations revealed that S508A consistently showed greater activity, which could be attributed to its predominantly hexameric oligomeric state even at lower concentrations (Fig. 6E). Together, these results indicate that the hexamer is a more catalytically competent state than the trimer. The double mutant S508P/I509P was largely monomeric by mass photometry (Fig. S14) and displayed nearly no enzymatic activity (Fig. 6D).

We next probed the importance of the hexamer–hexamer interface, mediated principally by Asp431, Leu432, and Ser433 (Fig. 4, A and B). A triple mutation (D431P/L432A/S433P) resulted in little soluble protein (Fig. S12), but single or double mutations (S433P or D431P/S433P) yielded proteins that migrated as sharper, faster-moving bands on native gels (Fig. 6C), consistent with disruption of higher-order oligomers and enrichment of single hexamers. These variant proteins retained the WT level of enzymatic activity (Fig. 6D) consistent with hexamers acting as minimal functional units of the Mtb E2p core. A C-terminal truncation removing the final helix (N545) resulted in aggregation and was not further characterized (Fig. S12). Overall, mutational analysis supports that the DlaT hexamer is the functional PDHc core, and higher oligomers are dispensable for catalysis.

In vitro PDHc assembly supports the DlaT hexamer as the functional core

We next asked if the peripheral PDHc enzymes—E1p (AceE) and E3 (Lpd)—preferentially associate with a particular DlaT oligomeric state during complex assembly. Individually purified AceE and Lpd were mixed with either WT DlaT or the hexamer-disrupting double mutant DlaT protein (S508P/I509P) at a molar ratio of 2:1:2, and the resulting complexes were purified by gel filtration and analyzed by SDS-PAGE and cryo-EM (Fig. S15 and S16).

Both WT and variant DlaT proteins successfully assembled ternary PDHcs with AceE and Lpd. As expected, the complex containing the WT DlaT eluted earlier (12.1 ml) and therefore had a higher apparent molecular mass than the complex assembled with the S508P/I509P protein (14.9 ml) (Fig. S15).

Attempts to visualize the intact PDHc by cryo-EM yielded 2D class averages corresponding only to individual components—DlaT oligomers, AceE or Lpd, bound to the DlaT PSBD—but not the full complex (Fig. S16). This likely reflects the highly flexible linker between the DlaT PSBD and CD, which hampers visualization of their combined architecture.

Nevertheless, assembly with AceE and Lpd produced a notable shift in the oligomeric distribution of DlaT. Whereas purified WT DlaT alone contains both hexamers and dodecamers, the PDHc mixture contained predominantly hexamers with a very minor population of trimers. The disappearance of dodecamers suggests that AceE and Lpd binding destabilizes the hexamer–hexamer interfaces that support higher-order oligomers. These findings further support the DlaT hexamer—not dodecamer or trimer—as the functional E2p core within the assembled Mtb PDHc.

Discussion

The oligomeric E2p inner core nucleates assembly of the PDHc, and all previously characterized PDHc architectures contain either an octahedral (24-mer) or an icosahedral (60-mer) E2p oligomer. Here, we show that Mtb instead uses a distinct E2p core composed of a DlaT hexamer. High-resolution cryo-EM structures of the Mtb DlaT hexamer and dodecamer revealed that approximately one-third of particles exist as single hexamers, whereas two-thirds form complexes of two or three hexamers. Although the canonical E2p building block is a trimer—including in Mtb, where DlaT first assembles a trimer—the unique architecture of the Mtb DlaT CD prevents trimers from assembling into classical spherical 24- or 60-mer structures. Instead, two trimers dimerize through a distinct β-propeller element to form a hexamer (Fig. 7). And these hexamers can further associate side by side into higher oligomers that are neither octahedral nor icosahedral.Figure 7**Unique stacking interactions give rise to the hexamer-based Mtb E2p core.**A, two DlaT trimers associate back to back through their β-propellers to form a dimer-of-trimers hexamer. These hexamers can further assemble into dodecamers or higher-order oligomers through side-by-side interactions between trimers belonging to adjacent hexamers. B, in contrast, classic spherical E2p cores form octahedral or icosahedral assemblies in which trimers interact side by side, rather than back to back as observed for Mtb DlaT. A representative icosahedral E2p core is shown in surface view (PDB code: 7UOM). Mtb, Mycobacterium tuberculosis; PDB, Protein Data Bank.

DlaT oligomerization is also dynamic and concentration dependent. At low nanomolar concentrations, DlaT behaves primarily as a trimer; at micromolar concentrations, it transitions into hexamers and dodecamers. This behavior contrasts sharply with that of the homologous Cglu E2p, which remains trimeric across the same concentration range (26). Our data show that the hexamer is the functional unit of the Mtb E2p core that supports enzymatic activity in PDHc. Moreover, PDH assembly converts dodecamers into hexamers, suggesting that the oligomeric transition may serve as a regulatory mechanism. The high cellular DlaT concentrations strongly favor hexamer formation in vivo, supporting the hexamer as the functional PDHc core.

We propose that Mtb exploits concentration-dependent oligomerization to regulate carbon flow and stress responses. Unlike the eukaryotes, where phosphorylation modulates PDHc activity (40), prokaryotic metabolic complexes are primarily regulated transcriptionally. Yet, Mtb lacks operon organization for most central metabolic enzymes—including PDHc and OGDHc components—except for the lowly expressed BCKDH (bkdABC) operon (41), and genes encoding E1p, E1o, E2p, and E3 are dispersed throughout the genome (42). Furthermore, Mtb lacks catabolite repression and can cocatabolize multiple carbon sources simultaneously (43). In such a context, oligomerization-dependent regulation could provide a rapid, energy-efficient means to store E2p in higher-order assemblies that can readily dissociate into active hexamers (or less active trimers) upon binding substrate-loaded E1p/E1o. This mechanism could also explain how a single E2p supports both the pyruvate and α-ketoglutarate dehydrogenase reactions (31, 42) and how DlaT interfaces with AhpD to power peroxynitrite reductase activity driven from both carbon-derived (via E1p/E1o) and NADH-derived (via E3) intracellular pools.

Mtb experiences extreme environmental pressures within the host—acidic pH, nutrient limitation, hypoxia, and continuous exposure to reactive oxygen and nitrogen species. Long-term survival in this environment requires metabolic flexibility and rapid adaptability: Mtb has evolved a unique, oligomerization-driven architecture for its PDHc inner core, distinct from any other bacterial or eukaryotic system. Combined with prior work identifying DlaT as a promising antituberculosis target (33), the high-resolution structures presented here may guide the development of new inhibitors targeting DlaT or its partner enzymes. Future work is needed to determine the full architecture of Mtb PDHc, a challenge likely amplified by the inherent high flexibility of the complex.

Experimental procedures

Protein expression and purification (Mtb DlaT, AceE, Lpd, and Cglu E2p)

Recombinant Mtb proteins (DlaT, AceE, and Lpd) were overexpressed in E. coli and purified as described before (30, 42, 44). Except for AceE, which carries an N-terminal His-tagged, all recombinant proteins were expressed without affinity tags and purified in their native form to >95% purity. Selected DlaT variants (E2br and N545) were additionally tested as N-terminal His-tagged constructs to facilitate purification and increase protein recovery.

The Cglu E2p (UniProt ID: Q8NNJ2) open reading frame was synthesized (GenScript) and cloned into pET-11c at the NdeI/NheI sites. Overexpression in E. coli followed the Mtb DlaT protocol. Briefly, cultures grown in LB were supplemented with 200 μM lipoic acid (from a 200 mM methanol stock) at an absorbance of ∼0.7 at 600 nm, followed by induction with 1 mM IPTG (35°C, 3.5 h). Cglu E2p was purified by ammonium sulfate precipitation (40% saturation) and Q Sepharose chromatography in 25 mM potassium phosphate buffer (pH 7.0) with a 0 to 1 M NaCl linear gradient.

Mutations in Mtb DlaT were generated using the QuikChange II XL site-directed mutagenesis kit (Agilent). Mutagenic primers were designed using the QuikChange online tool (https://www.agilent.com/store/primerDesignProgram.jsp), and all constructs were sequence verified. Native lipoylation of recombinant DlaT (WT and variants) and Cglu E2p was confirmed by Western blotting using anti–lipoic acid antibody (Sigma, 437695, 1:5000 dilution) (Fig. S1).

For in vitro PDHc reconstitution, AceE, DlaT, and Lpd were mixed at a molar ratio of 2:1:2 and incubated for 15 min at room temperature in 20 mM Tris (pH 7.5), 5 mM MgCl_2_, and 100 mM KCl. Complexes were resolved by size-exclusion chromatography on a Superose 6 Increase 10 × 300 mm column (Cytiva).

PDH activity assay

Recombinant PDHc activity was measured using an NADH production assay. AceE, DlaT, and Lpd were mixed in 50 mM potassium phosphate buffer (pH 7.0) at a molar ratio of 2:1:2 and incubated for 15 min at room temperature. The assay mixture was prepared at 2× concentration, and reactions were initiated by mixing equal volumes (40 μl) of protein mixture and assay mixture in a 384-well plate. NADH production was monitored at 340 nm for 10 min at 37 °C. Control reactions lacked CoA. Final reaction component concentrations were 200 nM AceE, 100 nM DlaT, 200 nM Lpd, 0.2 mM thiamine pyrophosphate, 1 mM MgCl_2_, 1 mM NAD^+^, 1 mM pyruvate, and 250 μM CoA in 50 mM potassium phosphate (pH 7.0). Rates of NAD^+^ reduction were calculated from the linear portion of the progress curves using the NADH extinction coefficient (ε340 = 6223 M^-1^ cm^-1^).

Quantification of intracellular DlaT concentration

Intracellular DlaT levels were estimated by quantifying DlaT in cell-free soluble extracts from Mtb or Msm using Western blot densitometry. A single anti–DlaT-reactive band was observed for both recombinant DlaT protein and native cell extracts; this band migrated at a higher apparent molecular weight (57–61 kDa) because of the double lipoylation post-translational modification, which slows migration in SDS-PAGE. Purified recombinant Mtb DlaT served as a standard. Known amounts (10–1000 ng) were loaded onto 4% to 12% gradient SDS-PAGE gels, resolved, transferred to nitrocellulose, probed with anti-DlaT antibody (1:5000 dilution; overnight, 4 °C), and detected with IRDye800 or IRDye680 secondary antibodies using an Azure 600 imager. Mtb and Msm lysate samples were run in the same gel with the pure recombinant DlaT standard curve to ensure identical exposure conditions. Band intensities were quantified in Fiji-5/ImageJ. The resulting standard curve showed linear detection across 10 to 1000 ng and was used to extrapolate the DlaT content of lysate samples.

Molar concentrations in cell-free extracts were calculated using the molecular masses of DlaT: 57.5 kDa (Mtb) and 61.5 kDa (Msm).

Intracellular concentrations were then estimated using two approaches to bracket the range of plausible mycobacterial cell volumes. In method 1, following the geometric assumption in the study by Tian et al. (45), cells were modeled as cylinders with widths of 0.3 to 0.6 μm and lengths of 1 to 4 μm, yielding an average volume of 0.4 μm^3^. A similar estimate (0.5 μm^3^) was reported by Sarathy et al. (46). In method 2, based on radiolabeling measurements in E. coli by Mengin-Lecreulx et al. (47), we assumed that a bacterial cell volume is 2 × 10^-15^ l (2 μm^3^) with a water content of 1.5 × 10^-15^ l. To account for the higher lipid content of mycobacteria, we assumed that 40% to 60% of the total volume corresponds to lipids, yielding an estimated cellular volume of 0.6 to 0.9 μm^3^. Total cell numbers were inferred from an absorbance at 580 nm measurement (1 absorbance at 580 nm = 5 × 10^8^ colony-forming unit/ml), with each bacillus assumed to represent 1 colony-forming unit.

Mass photometry

Mass photometry was performed on a Refeyn TwoMP instrument following the manufacturer’s recommended protocol. Briefly, 18 μl of imaging buffer (20 mM Tris, pH 7.5, 5 mM MgCl_2_, and 100 mM KCl) was added to a 6-well cassette mounted on a precleaned Refeyn carrier slide to establish focus. Two microliters of protein sample were then added, mixed, and imaged for 60 s using Refeyn AcquireMP. Contrast and mass distributions were analyzed with Refeyn DiscoverMP software.

For Mtb DlaT, a working concentration of 20 nM provided optimal signal to noise. For measurements at higher concentrations (up to 10 μM), the MassFluidix HC add-on was used to achieve rapid on-chip dilution and enable accurate detection of multiple oligomeric species.

Cryo-EM sample preparation and data collection

Cryo-EM data were collected at the Cryo-EM Facility of the Van Andel Institute. Mtb DlaT samples were prepared at 0.8 mg/ml in 20 mM Tris (pH 7.5), 5 mM MgCl_2_, and 100 mM KCl. Cryo-EM grids were prepared using a Vitrobot Mark IV (FEI) set to 6°C and 100% humidity. Quantifoil R1.2/1.3 Au 300-mesh ‬holey carbon grids were glow-discharged for 30 s at 30 W in a Gatan Model 950 Advanced Plasma System immediately before use. Three microliters of protein sample were applied to each grid, incubated for 5 s, and blotted with Whatman 595 filter paper (blot force 3, blot time 3 s). Grids were vitrified by plunging into liquid ethane cooled by liquid nitrogen. ‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬

Data were acquired on a Titan Krios (ThermoFisher) equipped with a Quantum 967 energy filter and post-GIF K3 summit direct electron detector (BioQuantum). Movies were collected in the super-resolution counting mode using SerialEM at a calibrated pixel size of 0.414 Å. Defocus values ranged from −1.3 to −1.8 μm. The electron dose rate was 40 e^–^/Å^2^/s with a 1.5 s exposure, yielding a total dose of 60 e^–^/Å^2^ per movie. A total of 17,053 movies were collected for the Mtb DlaT dataset.‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬

Cryo-EM grid preparation and data collection for Cglu E2p, the Mtb DlaT–CoA complex, DlaT(I509P), DlaT(S508P/I509P), and PDHc followed the same procedure. Protein concentrations used were 0.8 mg/ml (PDHc), 2 mg/ml (DlaT–CoA complex), 0.8 mg/ml (I509P), 0.7 mg/ml (S508P/I509P), and 1.2 mg/ml (Cglu E2p). The PDHc dataset (2670 movies) was collected on a Talos Arctica equipped with a K2 direct detector at a total dose of 50 e^–^/Å^2^.‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬‬ Dataset for the DlaT–CoA complex (3144 movies), DlaT(I509P) (6224 movies), DlaT(S508P/I509P) (1218 movies), and Cglu E2p (1039 movies) were collected on the Talos Arctica equipped with a K3 detector with a total dose of 55 e^–^/Å^2^ per movie.

Cryo-EM data processing and 3D reconstruction

For the Mtb DlaT dataset, image processing was carried out using RELION 3.1.1 and CryoSPARC 3.3.2 (48, 49) (Fig. S4). Beam-induced motion was corrected in MotionCor2 (50) with a twofold binning factor. Motion-corrected micrographs were imported into CryoSPARC, and contrast transfer function parameters were estimated with CTFFIND4 (51) (Fig. S5A). Particle selection combined blob picking, template picking, and Topaz. Blob-picked particles from 1000 randomly chosen micrographs were averaged in 2D and used to train Topaz. In total, 4,178,622 particles were extracted from 16,459 micrographs and subjected to reference-free 2D classification (Fig. S5B). A subset of 2,041,423 particles displaying clear structural features was retained for ab initio 3D reconstruction, which yielded two distinct map classes. Each class underwent an additional round of ab initio reconstruction to remove misaligned particles. Particles belonging to well-defined classes were pooled for multireference heterogeneous refinement.

During heterogeneous refinement of the dodecamer class, one subset displayed additional density adjacent to the main structure. The highest-quality subsets were combined—379,608 particles for the hexamer and 714,313 particles for the dodecamer—and refined using homogeneous refinement followed by nonuniform refinement in CryoSPARC. Final resolutions were 2.65 Å for the hexamer (D3 symmetry) and 2.51 Å for the dodecamer (D2 symmetry), as determined by gold-standard Fourier shell correlation (0.143) using masked maps (Figs. S5 and S6). Data processing for the Mtb DlaT–CoA complex, Cglu E2p, DlaT(I509P), DlaT(S508P/I509P), and the PDHc followed the same workflow and is detailed in Figs. S7–S10, S13, and S15.

Model building

An AlphaFold-predicted model of Mtb DlaT (UniProt ID: P9WIS7) was used as the starting template. Three CD monomers were aligned to the Cglu E2p structure (PDB code: 6ZZK) (26) to generate a trimeric building block, which was then fitted to the DlaT hexamer EM map using UCSF Chimera. Manual model adjustment was performed in Coot (MRC-LMB) followed by real-space refinement in Phenix (Lawrence Berkeley National Laboratory) (52, 53, 54). The refined model was subsequently fitted into the dodecamer map to build the full dodecameric assembly. Model quality and stereochemistry were assessed using MolProbity (55).

Data availability

The cryo-EM maps have been deposited in the Electron Microscopy Databank under accession codes EMD-72630 (DlaT hexamer), EMD-72639 (DlaT dodecamer), EMDB-72666 (DlaT–CoA hexamer), and EMDB-72667 (Cglu E2p trimer). Corresponding atomic coordinates have been deposited in the PDB with accession codes 9Y6T (DlaT hexamer), 9Y72 (DlaT dodecamer), and 9Y7V (DlaT-CoA hexamer).

Supporting information

This article contains supporting information.

Conflict of interest

The authors declare that they have no conflicts of interest with the contents of this article.