Crystal structure of human mitochondrial trifunctional protein, a fatty acid β-oxidation metabolon

Significance

Fatty acid β-oxidation is the major energy-producing process in all tissues and is performed by four consecutive reactions that cleave fatty acids. Mitochondrial trifunctional protein (TFP) performs the last three of these four reactions. Herein, we report the crystal structure of human TFP and compare and contrast this with the recently reported cryo-EM structure. The crystal structure reveals the channel through which the substrate/product can pass from the first reaction site to the second and onto the third, thereby minimizing leakage of the intermediates/product to the outside of the protein. Our findings provide a better understanding of how this enzyme functions and reveal insight into the development of inhibitors or agonists for the regulation of fatty acid degradation.

Abstract

Membrane-bound mitochondrial trifunctional protein (TFP) catalyzes β-oxidation of long chain fatty acyl-CoAs, employing 2-enoyl-CoA hydratase (ECH), 3-hydroxyl-CoA dehydrogenase (HAD), and 3-ketothiolase (KT) activities consecutively. Inherited deficiency of TFP is a recessive genetic disease, manifesting in hypoketotic hypoglycemia, cardiomyopathy, and sudden death. We have determined the crystal structure of human TFP at 3.6-Å resolution. The biological unit of the protein is α₂β₂. The overall structure of the heterotetramer is the same as that observed by cryo-EM methods. The two β-subunits make a tightly bound homodimer at the center, and two α-subunits are bound to each side of the β₂ dimer, creating an arc, which binds on its concave side to the mitochondrial innermembrane. The catalytic residues in all three active sites are arranged similarly to those of the corresponding, soluble monofunctional enzymes. A structure-based, substrate channeling pathway from the ECH active site to the HAD and KT sites is proposed. The passage from the ECH site to the HAD site is similar to those found in the two bacterial TFPs. However, the passage from the HAD site to the KT site is unique in that the acyl-CoA intermediate can be transferred between the two sites by passing along the mitochondrial inner membrane using the hydrophobic nature of the acyl chain. The 3′-AMP-PPi moiety is guided by the positively charged residues located along the “ceiling” of the channel, suggesting that membrane integrity is an essential part of the channel and is required for the activity of the enzyme.

Fatty acid β-oxidation is the principal energy-yielding process in organisms ranging from bacteria to humans. It is carried out by a series of enzymes that successively cleave acetyl-CoA fragments from fatty acyl-CoA thioesters until the fatty acyl-CoA is completely degraded. The resulting acetyl-CoA is further oxidized in the tricarboxylic acid cycle. Each cycle of β-oxidation pathway is composed of four reactions: acyl-CoA dehydrogenase (ACAD), 2-enoyl-CoA hydratase (ECH), 3-hydroxyacyl-CoA dehydrogenase (HAD), and 3-ketothiolase (KT). In mammalian mitochondria, four separate soluble enzymes carry out each of the four reactions during the processing of short and medium chain fatty acids (for review; ref. 1). However, degradation of very long chain fatty acids is carried out by two proteins that are bound to the mitochondrial inner membrane; the first reaction is carried out by very long chain acyl-CoA dehydrogenase (VLCAD), and the next three reactions are carried out by a single protein, trifunctional protein (TFP) (2). TFP is composed of two subunits: the α-subunit contains the first two of the remaining three activities (ECH and HAD), while the β-subunit bears the KT activity (Fig. 1).

Fig. 1.

Schematic diagram of fatty acid β-oxidation in mitochondria. TFP and VLCAD are associated with the mitochondrial inner membrane. LCAD, MCAD, and SCAD represent long-, medium-, and short-chain acyl-CoA dehydrogenase, respectively.

Mutations in either subunit of TFP can result in reduced activity of all three TFP enzyme activities, leading to disease. In TFP-deficient patients, at least 32 mutations in the α-gene and 30 mutations in the β-gene have been reported. Clinical phenotypes include hypoketotic hypoglycemia associated with metabolic acidosis, cardiomyopathy, and retinopathy (3, 4). Deficiency of the long chain HAD (LCHAD) activity in the α-subunit has also been found in children of women who develop HELLP (hemolysis, elevated liver enzymes, and low platelets) syndrome and AFLP (acute fatty liver pregnancy) syndrome (5), both of which are life-threatening obstetric conditions.

Structures of soluble multifunctional proteins have been determined, including two bacterial TFPs from Pseudomonas fragi (PfTFP, ref. 6) and from Micobacterium tuberculosis (MtTFP, ref. 7) and rat peroxisomal multifunctional enzyme 1 (PMFE-1, ref. 8), the latter of which is a bifunctional protein with ECH and HAD activities, similar to the α-subunit of mitochondrial TFP. However, no membrane-bound mitochondrial TFP or peroxisomal multifunctional enzyme structures have been determined, despite intensive biochemical and structural studies. Only very recently (while we were preparing this manuscript), the structure of human TFP determined by cryo-EM was published (9). Here, we report the crystal structure of human TFP determined at 3.6-Å resolution and compare and contrast the two human TFP structures. While the α₂β₂ structure and membrane orientation are nearly identical, there are notable differences between the models determined by cryo-EM and crystallography, presumably due to the different methodology and resolution. One major difference is the structure of α₆β₆, a different oligomeric state of human TFP found in the crystal structure. Also included in this report is a detailed description of putative substrate channeling pathways between the three active sites. Furthermore, the structure allows us to propose a mechanism of cardiolipin remodeling activity (monolysoacyltransferase) of the TFP α-subunit.

Results and Discussion

Biochemical Characterization of the Escherichia coli-Expressed Human TFP.

Human TFP (hTFP) was expressed in an E. coli system and purified to homogeneity (SI Appendix, Fig. S1). The recombinant enzyme has overall activity (0.7 unit/mg of purified protein; Fig. 2A) comparable to the native enzyme purified from pig heart mitochondria (1.0 unit/mg of purified protein; ref. 10). Although the soluble bacterial TFPs exist as α₂β₂ heterotetramers, the oligomeric state of mitochondrial TFP has been generally accepted as α₄β₄. Native rat liver TFP was reported to exist as α₄β₄ as determined by size-exclusion chromatography (11), and recombinant human TFP was found by sedimentation velocity analysis to be a mixture of α₂β_2, α₄β₄, and α₆β₆, with the α₄β₄ form in the majority. Our size-exclusion chromatographic results (Fig. 2B) show that the protein exists in solution as a mixture of α₆β₆ and α₄β_4, with the latter in the majority, consistent with previously published results of recombinant human TFP (12). Interestingly, in the presence of 24 mM β-octyl glucoside (OBG), the protein was eluted as a sharp peak, corresponding to an apparent molecular mass of ∼320 kDa (close to the calculated value for α₂β₂, 260 kDa). However, in the presence of 0.4 mM n-dodecyl-β-maltoside (DDM), the protein was eluted as a mixture of α₂β₂ and α₄β₄, which is again consistent with the oligomer distribution observed in the cryo-EM studies (9).

Fig. 2.

(A) Enzymatic activity of various TFP proteins. The overall reaction activity is in milliunit/mg (purified TFP), Numbers in parentheses are the numbers of repeated measurements. (B) Chromatographic elution profile of wild type (green), ∆β(170-209) (blue), and F277K (cyan). Purified proteins were applied to a size-exclusion column (Bio-Rad Enrich SEC 650) running buffer containing 25 mM Tris·HCl, pH 7.5, 150 mM NaCl and 5% glycerol at a flow rate of 0.4 mL/min using a Shimadzu Prominence HPLC system. Wild-type TFP in the same running buffer plus 24 mM octyl-β-d-glucopyranoside (black) or plus 0.4 mM n-dodecyl-β-d-maltoside (red) is also shown. The apparent molecular mass of the major peaks 1 and 2; and the shoulder peak 3 are calculated at about 317, 554, and 813 kDa, respectively, corresponding to TFP α₂β₂ (258 kDa), α₄β₄ (516 kDa), and α₆β₆ (774 kDa) oligomeric states.

Crystal Symmetry, Oligomeric States, and Membrane-Binding of Human TFP.

The crystal structure of hTFP has been determined at 3.6-Å resolution (13). The asymmetric unit of the hTFP crystal contains three α₂β₂ units forming a large ball with a diameter of ∼150 Å and has a 32 symmetry (three twofold axes perpendicular to the threefold axis) (Fig. 3 and Movie S1). The inside of the ball is occupied by six helix (H4)–loop–helix (H5) segments (see SI Appendix, Fig. S2 for assignment of the secondary structures, hereafter referred to as H4-H5) corresponding to residues Asp170-Val214, from each of the six β-subunits, and another six helix (H10) segments from the α-subunits. In addition, at the north and south poles of the ball, where three α-subunits meet, lie two hydrophobic residues, Phe277 and Ile275, from the end of H10 of the α-subunit, making hydrophobic interactions (Fig. 3C). Similar hydrophobic interactions were also observed at the dimer interface in the α₄β₄ species found in the recent cryo-EM studies of hTFP (9). In the cryo-EM α₄β₄ structure, the hydrophobic residues (275-IleProPhe-277) of one α₂β₂ heterotetramer interact with the corresponding part of the other tetramer making a “skewed clam shell” like structure. However, size-exclusion chromatographic results (Fig. 2B) showed that the mutant protein (F277K) was still mainly α₄β₄, suggesting that, in the absence of detergent or lipid bilayer, the TFP protein exists as α₄β₄ or α₆β_6, not α₂β₂.

Fig. 3.

Surface representation of the structure of the α₆β₆ form. One asymmetric unit of the human TFP crystal contains three α₂β₂ units, which form a ball with a 32 symmetry (three twofold axes are perpendicular to the threefold axis as marked). All α-subunits are shown in green or orange/yellow shades, and β-subunits are shown in blue or magenta. (A) View down a twofold axis located at the middle of a α₂β₂ unit. (B) View down the threefold axis, where three α-subunits meet. (C) Enlarged view of the hydrophobic interactions among Ile275 and Phe277 of the three α-subunits at the threefold axis.

The Structure of the α₂β₂ Heterotetramer.

Fig. 4 shows two different views of the α₂β₂ heterotetramer. The overall structure of the α₂β₂ heterotetramer is the same as that observed by the cryo-EM method (9). The two β-subunits make a tightly bound homodimer as found in other known thiolase structures. Two α-subunits are bound to each side of the β₂ dimer. The overall folds of each α- and β-subunit alone are essentially the same as the corresponding subunits of two known bacterial TFP structures, one from P. fragi (Pf TFP) (6) and the other from M. tuberculosis (MtTFP) (7). However, the domain/subunit arrangement in human TFP (hTFP) is very different from those of bacterial TFPs. Comparison of the three structures and sequence alignments reveal that there are a few major insertions and some structural divergences in each subunit. These minor structural differences in the hTFP structure compared with the bacterial TFPs play important roles in defining the subunit interfaces, resulting in the protein’s membrane-binding property, which, in turn, leads to the formation of various quaternary structural assemblies of hTFP that have been observed previously. It is likely that the biological assembly of hTFP on the mitochondrial inner membrane is an α₂β_2, not α₄β₄ nor α₆β₆, because in the latter two structures, the membrane-binding side of the molecule is secluded and would not be exposed to bind to the mitochondrial membrane.

Fig. 4.

Cartoon representation of the α₂β₂ structure. The ECH domain (Thr37-Thr333) and HAD domain (Lys334-Gln763) of α1 are shown in light and dark green, respectively. The corresponding domains of α2 are in light and dark orange, respectively. β1 is in blue and β2 in magenta. The shaded area in Left represents possible hydrophobic interactions with a curved membrane, and four red dashed ellipses in Right indicate interfaces between the α- and β-subunits.

Membrane-Binding Affinity.

Comparison of the structures and sequences of three TFPs (i.e., membrane-bound hTFP and two soluble bacterial TFPs) clearly revealed a major insertion in the β-subunit of the hTFP structure, corresponding to residues Met179-Leu207, which include the hydrophobic H4-H5 (Insertion 1 in SI Appendix, Fig. S2A). A closer inspection of the three TFP structures showed that, starting from Ser169, the main chain of the human TFP-β structure differs from the two bacterial TFP structures merging back at Glu221. In this 53-residue divergent region, about 40 residues from Asp170 to Pro209 form the putative membrane binding helices, H4-H5 in the hTFP structure (SI Appendix, Figs. S3A and S4A), while in both bacterial TFPs, this region is shorter, has a different secondary structure, and is involved in making an interface with the ECH domain of their corresponding α-subunits (SI Appendix, Fig. S4A).

To confirm the membrane affinity of H4-H5, we made a mutant, in which the 40 residues between Asp170 and Pro209 in the β-subunit were deleted [∆β(170-209)], and the resulting TFP mutant was expressed in the same manner as the wild-type protein. Compared with the wild-type protein, which was expressed nearly 100% in the membrane fraction, the mutant protein was expressed in and purified from both soluble and membrane fractions with a 2:3 ratio, respectively. Presumably, the remaining membrane-binding affinity of the deletion mutant must lie in the α-subunit. Unlike the β-subunit, for the α-subunit, there is no apparent amino acids insertion between Asp253 and Phe277. However, this region’s secondary structure and spatial arrangement are very different from the bacterial TFPs. In hTFP, this part extrudes out of the main body of the ECH domain of the α-subunit, forming another membrane-binding helix, H10 (SI Appendix, Figs. S3A and S4B), consistent with the membrane-binding property of the α-subunit.

The spatial arrangement of these membrane-binding regions is consistent with the ball-shaped structure of the α₆β₆ form described above. These membrane-binding regions are situated inside the “ball” and sequestered from the outside aqueous solution. Furthermore, these membrane regions are likely interacting with each other directly or indirectly via detergent molecules to further stabilize the α₄β₄ and α₆β₆ forms. There are some patches of electron density inside the ball that are scattered in between the membrane-binding helices, especially between the α-subunit helix H10 and the β-subunit helix H5. These are most likely either from detergent molecules that remain bound to the protein during the protein purification (Tween 20) or C₈E₅ detergent molecules included in the crystallization media. As with the F277K mutant, the amount of the α₆β₆ form in the β-subunit deletion mutant [∆β(170-209)] was decreased, but the α₄β₄ amount remained the same as the wild type (Fig. 2B). This is due probably to the fact that an exquisite arrangement of the three Ile275/Phe277 interactions among the three α-subunits is required to assemble the α₆β₆ ball structure. The deletion of H4-H5 of the β-subunit would eliminate the interactions between the membrane-binding helices in both subunits, resulting in making the H10 of the α-subunit more mobile and weakening the pivotal Ile275/Phe277 interactions among the three α-subunits.

Importance of the Membrane Integrity.

Judging from the distances between the membrane-binding helices in the α₂β₂ structure, the curvature of the bend is ∼50 Å (i.e., a circle of 50-Å radius). This suggests that, in the absence of membrane, large micelles (diameter of ∼100 Å) would be required to stabilize the curved shape of the α₂β₂ structure. Interestingly, in the presence of 24 mM OBG [1× critical micelle concentration (CMC)], only the α₂β₂ form exists, while in the presence of 0.4 mM DDM (3× CMC), about two-thirds exists as α₄β₄ and the remaining one-third as α₂β₂ (Fig. 2B). OBG forms only very small size micelles, while DDM can form stable 16 × 28 Å oblate-shaped micelles (14). This is consistent with the EM results that in the presence of OBG, no stable hTFP image can be recorded since OBG does not form the large stable micelles required to stabilize the membrane curvature to fit the α₂β₂ structure. However, in the presence of DDM, both α₂β₂ and α₄β₄ forms were observed in the EM images (9).

Catalytic Residues in All Three Active Sites Are Conserved.

In the hTFP structure, as in the two bacterial TFPs, the α-subunit is composed of two large domains. The N-terminal domain contains enoyl-CoA hydratase activity (ECH) and the C-terminal domain contains 3-hydroxyacyl-CoA dehydrogenase activity (HAD). The β-subunit contains the keto thiolase activity (KT). The structures of these three active sites are well conserved from those of the corresponding mitochondrial soluble enzymes. For each of the three active sites in the hTFP structure, the corresponding substrate can be modeled in, demonstrating that each active site has a large/long enough cavity to accommodate long fatty acyl chains, and that the catalytic residues for each active site are all preserved (SI Appendix, Fig. S5).

Subunit Interfaces.

In hTFP, each α-subunit makes two interfaces with the β₂ dimer: one between the ECH domain of α1 and β1 and the other between the HAD domain of α1 and β2 (Fig. 4, Right; note the assignment of α1, β1, α2, and β2). There are short-sequence insertions in both α-and β-subunits: one with residues Leu225-Ile237 in the α-subunit (loop between S7 and H9 and extending into part of H9 in SI Appendix, Fig. S2B) and Ala392-Lys408 in the β-subunit forming a short helix–loop–helix structure (SI Appendix, Fig. S2A). These two regions together with helix H6 in the β-subunit (three red-highlighted regions in SI Appendix, Fig. S2) form the first subunit interface between the ECH domain of α1- and the β1-subunit (Fig. 4, Right and SI Appendix, Fig. S3B, Left) with an interface area of 820 Ų, containing a significant amount of hydrophobic interaction as characterized by the PISA interface calculation (www.ebi.ac.uk/pdbe/pisa). In the bacterial TFPs, there is only one interface between the two subunits, i.e., between the ECH domain of the α-subunit and the β-subunit. The HAD domain of the α-subunit makes no other contact with the β₂-dimer, allowing some flexibility between the β₂ dimer and the HAD domain (6). This flexibility in the PfTFP is necessary for the substrate channeling between the HAD active site in the α-subunit to the KT active site in the β-subunit (6, 15). However, in hTFP there is a second interface between the α1-subunit and the β₂ dimer, which is formed between helices H16 and H17 of the α1 HAD domain and H1, H2, S6, and S7 of the β2 subunit (area, 607 Ų) (six pink-highlighted regions in SI Appendix, Fig. S2, also Fig. 4, Right and SI Appendix, Fig. S3B, Right). With both the ECH and HAD domains fixed to the β₂ dimer, the hTFP α₂β₂ structure is more rigid than the bacterial TFPs. This also infers that the curved membrane-binding domain arrangement observed in the α₂β₂ structure is indeed that of the hTFP structure in mitochondria.

Substrate Channeling Between Active Sites and the Concept of a Metabolon.

The concept of a metabolon was first proposed by Welch (16) over 40 y ago and heavily promoted by Srere (17). The concept is that many groups of enzymes within common metabolic pathways are physically associated together into “metabolons,” e.g., TCA cycle enzymes, glycolytic enzymes, or fatty acid oxidation enzymes. The idea makes good sense, but no stable such complexes have been observed in vitro, indicating that these complexes are weakly associated and perhaps their assembly needs to be regulated. Recently, there have been many observations that support the existence of metabolons. For example, Benkovic and coworkers (18) demonstrated by fluorescent labeling techniques that the enzymes in purine biosynthetic enzymes colocalize in vivo (“purinosome”). The large separations between the ECH, HAD, and KT active sites strongly suggests that a sequential substrate channeling between the three sites would be beneficial for TFP enzyme efficiency. Indeed, Schulz and coworkers (10) and Eaton et al. (2) have provided evidence for channeling between the active sites of the TFP protein. Since TFP contains three enzyme activities utilizing a channeling mechanism, we consider that TFP is a mini metabolon of the fatty acid β-oxidation (2). Here, we propose a structure-based channeling pathway from the ECH active site to HAD to KT active sites.

From the ECH Site to the HAD Site.

Ishikawa et al. (6) suggested that in PfTFP, the substrate channeling from the ECH to HAD active sites occurs by flipping the hydrophobic acyl group of the substrate, while fixing the AMP-PPi moiety of the acyl-CoA substrate to the common binding pocket, thus avoiding the substrate diffusing into the solvent. Although the substrate binding modes that have been modeled in the ECH and HAD sites of the hTFP structure (SI Appendix, Figs. S5 and S6) do not share a common binding pocket for the AMP-PPi moiety, the substrate channeling mechanism between these two active sites appears similar to that suggested for PfTFP. In hTFP, the channel from the ECH to HAD active sites is also located in the cleft between the ECH and HAD domains, as shown in Fig. 5. Along both edges of the cleft, there are arrays of positively charged residues stretching out to the middle of the cleft, and of particular interest is a cluster of positively charged residues, including Arg165, Arg205, Arg208, Arg211, Lys411, Lys414, and Lys415, that are situated about halfway between the two active sites (Fig. 5 and SI Appendix, Fig. S6). In addition, a series of hydrophobic residues are lining the “bottom” of the cleft, which could make hydrophobic interactions with the fatty acyl group of the substrate during the substrate channeling. The distance between the two active sites, ECH and HAD, is about 27 Å. The distance was measured between the pyrophosphate oxygen atoms (-P-O-P-) of each substrate bound to the active sites, since the highly negatively charged pyrophosphate group plays an important role in the electrostatic retention of the substrate, preventing it from diffusing into the bulk aqueous solvent. The depth of the cleft, measured from the hydrophobic residues lying at the bottom to the cluster of the positive residues covering the cleft is about 20 Å. Therefore, when the negatively charged AMP-PPi moiety of the substrate is fixed by the cluster of positively charged residues at the ceiling of the channel/tunnel, there is enough room for the acyl group to change its orientation with respect to the AMP-PPi group from that found in the ECH active site to that observed at the HAD active site. Thus, the landscape and the topology of the ECH-to-HAD pathway appears to be well-suited for channeling the fattyacyl-CoA intermediate, 3-hydroxyacyl-CoA, from the ECH site to the HAD site.

Fig. 5.

Proposed substrate channels between the three active sites. The same color scheme is used as in Fig. 4, unless stated otherwise. (A) Proposed channel from the ECH to HAD binding sites in the α-subunit (pink) and from the HAD to KT binding sites in the β-subunit (gray). The channels were created using the program HOLLOW (hollow.sourceforge.net). When creating the channel from HAD/α1 to KT/β1, a set of lipid bilayer molecules was also included to mimic the bound membrane. The dashed line represents the other possible substrate pathway from HAD/α1 to KT/β2, that has a large portion open to the solvent (see text for details). Substrates/intermediates are shown with cyan sticks. For clarity, the substrate acyl binding cavities are not included. (B) A different view from A with a 90° rotation along the x axis. For clarity, the α2-subunit is not shown. The alternative channel from HAD/α1 to KT/β2 (dashed line) clearly shows the solvent-exposed passage. (C) Schematic drawing of the substrate channeling between the three active sites. The negatively charged 3′- AMP-PPi moiety (shown as red octagons labeled as AMP-PPi) and the fattyacyl group (black zigzags) are linked by a pantetheine group. The channel between the ECH and HAD sites (pink) has a length of about 27 Å, a width of ∼12 Å at the widest section, and a depth of ∼20 Å (receding into the plane of the diagram). Its volume is ∼2,900 ų. The channel between the HAD and KT sites (gray) is ∼49 Å long with a cross-section of ∼10 × 10 Å, excluding the hydrophobic membrane bilayer (beneath the plane of the diagram). Its volume is ∼6,000 ų. The overall substrate/product channeling events, after 2-enoyl-CoA binds to the ECH active site and 3-hydroxyacyl-CoA is formed, are as follows: (1) The negatively charged AMP-PPi is relocated from the ECH active site to the positively charged area situated at the interface between the ECH/HAD domains (dotted red octagon). (2) The hydrophobic acyl group is released from the ECH active site and relocated to the HAD active site. (3) The AMP-PPi then binds to the HAD active site. The above three steps complete the channeling from the ECH to HAD sites. (4) Then, the negatively charged AMP-PPi moves from the HAD binding site back to the positively charged area (dotted red octagon). (5) The hydrophobic ketoacyl group is released from the HAD active site and moves into the hydrophobic membrane bilayer. (6) The 3-ketoacyl-CoA intermediate wades through the HAD-to-KT channel and finally binds to the KT active site.

From the HAD Site to the KT Site.

Although the structures of each individual domain/subunit in hTFP are very similar to the corresponding domains in two known bacterial TFP structures [MtTFP (PDB ID code 4B3I) and PfTFP (PDB ID code 1WDK)], the arrangements of the two α-subunits relative to the β₂ dimer are very different. Therefore, the distances between the HAD active site of the α-subunit and the KT active site in the β-subunit are very different among the three TFPs. There are two possible pathways from the HAD active site of the α-subunit to the KT active site of the β-subunit: α1 to β1 and α1 to β2. (For bacterial TFPs, α1, β1, α2, and β2 are defined such that α1 to β1 and α2 to β2 are closer than the distance between α1to β2 and α2 to β1. For hTFP, they are defined as shown in Fig. 4.) For both bacterial TFPs, one pathway is obviously preferred to the other. In PfTFP, the distance from the HAD active site of the α1 subunit to the KT active site of β1 is 21 Å, while the distance from α1 to β2 is 52 Å (6). In MtTFP, the distance from the HAD active site of the α1 subunit to the KT active site of the β1 subunit is 31 Å, while the distance for α1 to β2 is 58 Å (7). In addition, this α1-to-β2 passage is also blocked by the ECH domain. However, in hTFP, the situation is very different. First, the distances between the two active sites (for both α1 to β1 and α1 to β2) are much longer than those of the preferred pathways in both MtTFP and PfTFP; thus, they would require a more efficient, tighter channeling. Second, at first glance, there is no preferable pathway: The distance from the HAD/α1 active site to the KT/β1 active site is 49 Å, while that to the β2-subunit is 52 Å, not significantly different. However, a careful inspection of the environment near the two pathways revealed that they are very different. As shown in Fig. 5A, considering the presence of a membrane bilayer at the bottom, a relatively well-protected channel can be formed between the HAD/α1 and the KT/β1 active sites. However, the pathway from the HAD/α1 to the KT/β2 active sites is almost completely exposed to the bulk solvent on one side of the pathway (dashed line in Fig. 5 A and B). Another significant feature for the α1 to β1 pathway is the abundance of positively charged residues along this pathway (SI Appendix, Fig. S6). Most noticeably, there are two clusters of positively charged residues next to each other, Arg205, Arg208, Arg211, Lys213, and Lys214 of α1 in one cluster and Lys411, Lys413, Lys414, and Lys415 in the other, located at about one-third of the way into the channel (SI Appendix, Fig. S6C). The first three residues in each cluster are shared with the ECH-HAD channel. Similar to the channel between the ECH and HAD active sites, the negatively charged AMP-PPi can be attracted by the clusters of positively charged residues at the ceiling (opposite side of the membrane) of the channel. Because the mitochondrial membrane is composed mainly of neutral lipids (44% phosphatidyl choline and 34% phosphatidyl ethanolamine) (19), it is unlikely that the negatively charged AMP-PPi group of the substrate would be attracted to membrane head groups. More importantly, the hydrophobic fatty acyl group can be easily directed to the hydrophobic membrane bilayer and reorient itself from the HAD active site toward the KT active site. For these reasons, we conclude that the pathway from the α1 HAD active site to the β1 KT active site is preferred to the one from the α1 HAD active site to the β2 KT active site, contrary to the conclusions made from the cryo-EM studies (9). A schematic model of substrate channeling is shown in Fig. 5C (see legend for description of the model). Further studies, including computational analysis of simulation of substrate transfer, will be required to establish how the substrates/intermediates are transferred from one active site to the other.

Membrane Is Required for Channel Formation.

Since the membrane is an essential part of substrate channeling, any disruption of interactions between the membrane and the membrane-binding regions of the protein would affect the channel integrity as well as the overall enzyme activity. Our activity data (Fig. 2A) show that in the presence of OBG, there is no activity, while only 66% activity was observed in the presence of DDM. These activity data appear to coincide with the percentage amount of the α₄β₄ form determined by our size-exclusion chromatographic results (Fig. 2B), suggesting that the α₂β₂ form has no activity. At first glance, this contradicts the fact that in mitochondria, hTFP exists as the α₂β₂ form and must be active. There are two possible explanations for this inconsistency: (i) OBG is an inhibitor by binding to the active site and (ii) OBG disrupts membrane integrity. For the following reasons, the latter is a more plausible explanation. When hTFP exists in the α₂β₂ form in the presence of OBG or DDM, their micelles are not large enough to form stable interactions with the TFP’s membrane-binding regions, resulting in the formation of an unstable (OBG) or partially stable (DDM) curved membrane, not with the curvature required to form a stable α₂β₂ heterotetramer. In the case of the α₄β₄ form, the curved membrane is conserved by combining two α₂β₂ tetramers that results in a skewed clam shell (see Importance of the Membrane Integrity above). Apparently, DDM micelles are capable of stabilizing the curved membrane structure required for the formation of the clam shell structure. Thus, the α₄β₄ form in the presence of DDM is active. Similarly, when the enzyme assay was performed in the presence of dimyristoyl-sn-glycero-3-phosphocholine (DMPC) liposomes, which can provide the curved membrane, an increased overall activity was observed (Fig. 2A), again consistent with the second explanation.

Pathogenic Mutations.

Inherited deficiency of TFP or LCHAD is a recessive genetic disorder. Among over 60 missense mutations found in patients, the most common TFP mutation is the E510Q mutation in the α-subunit (c.1528G > C), accounting for about 87% of the alleles in LCHAD deficiency and a cause of AFLP (20). When we expressed the E510Q mutant in E. coli, the enzyme is stable but has drastically decreased activity (14% of wild type, Fig. 2A and SI Appendix, Fig. S1), consistent with the results from IJlst et al. (20), which showed diminished HAD activity (∼25% of wild type) when the mutant was expressed in Saccharomyces cerevisiae. This result is not surprising, since Glu510 is a part of the catalytic dyad for the HAD activity of TFP (see section on Catalytic Residues in All Three Active Sites Are Conserved, also SI Appendix, Fig. S5). Other mutations in both the α- and β-subunits have been shown to result in the loss of all three enzyme activities, resulting in complete TFP deficiency. A mapping on the α₂β₂ structure of the 46 mutation sites currently identified in patients of TFP deficiency has been reported (9).

The Linkage Between hTFP α-Subunit and Cardiolipin Remodeling.

Cardiolipin is a major phospholipid component of mammalian mitochondrial inner membrane and supports the mitochondrial respiratory chain complexes (21). It has a characteristic acyl chain composition that depends on the function of phospholipid-lysophospholipid transacylases, including tafazzin and the newly identified monolysocardiolipin acyltransferase-1 (MLCLAT-1). MLCLAT-1 is identical to the TFP α-subunit (TFPα) that lacks the first N-terminal 191 residues (not counting the mitochondrial transit peptide) (22). Based on the structure of TFPα, we propose a plausible catalytic mechanism of MLCL acyltransferase (SI Appendix, Fig. S7). From the inspection of the HAD active site and binding cavities of the substrate (3-hydroxyacyl-CoA) and cofactor NAD⁺, we concluded that the MLCL molecule would bind at the NAD⁺-binding groove in the HAD active site of TFPα. For the HAD activity in TFPα, residues His498 and Glu510 act as a catalytic dyad to abstract the 3-hydroxyl proton from the substrate 3-hydroxyacyl-CoA (23). In MLCLAT-1, this pair of residues would activate the hydroxyl group of MLCL by forming an H bond between MLCL-OH and Nε2 of His498. The activated MLCL-OH then undergoes a nucleophilic attack of the C1 of linoleoyl-CoA, and the tetrahedral intermediate is stabilized by an oxyanion hole consisting of the hydroxyl sidechains of Thr548 and/or Ser477, in a similar manner found in a squash glycerol-3-phosphate (1)-acyltransferase (G3PAT, ref. 24). Although TFPα does not contain a conserved HisX₄Asp motif as observed in classical acylglycerophosphate transferases, the constellation of His-Asp/Glu in TFPα, together with the C1 atom of modeled linoleoyl-CoA, is identical to that observed in the squash G3PAT structure. Ser477 lies in close proximity to the C1-carbonyl group of linoleoyl-CoA and is conserved in all monofunctional HADs, bifunctional enzymes, and TFPs. However, for the MLCLAT-1 activity, Thr548 might be in a better position to stabilize the tetrahedral intermediate. Thr548 is not conserved in most HADs, bifunctional enzymes, and soluble bacterial TFPs, but is conserved in all known mitochondrial TFP sequences, including human, pig, bovine, rat, mouse, and zebrafish. Further studies including site-directed mutagenesis are necessary to confirm the exact role of each residue in the active site.

Concluding Remarks.

The structure of human TFP has been determined by X-ray crystallography at 3.6-Å resolution. The structure reveals how the substrate can travel through the three active sites. The pathway between the ECH to the HAD sites is similar to those found in two bacterial TFPs. However, the pathway from the HAD to the KT sites is novel in that the acyl-CoA substrate can be transferred between the two active sites by wading along the mitochondrial inner membrane using the hydrophobic nature of the acyl chain, while the AMP-PPi moiety is guided by the positively charged residues located along the ceiling of the channel. Thus, the membrane integrity is an essential part of the channel and is required for the overall activity of the enzyme. It is likely that VLCAD and TFP would associate closely on the mitochondrial membrane and form a metabolon of fatty acid β-oxidation.

Materials and Methods

Protein Cloning, Expression, and Purification.

Two cDNAs each encoding the α- (TFP_α/pcDNA3.3) or β-subunit (TFP_β/pcDNA3.3) of human TFP protein were the generous gift of Suresh Kumar at the Medical College of Wisconsin, Milwaukee. These cDNAs were subcloned into E. coli expression vectors, pET21b and pET28a, resulting in TFP_α/pET21b and TFP_β/pET28a, respectively. Both expression plasmids were cotransformed with groEL/ES plasmid into E. coli BL21(DE3) competent cells. The protein was purified using Ni-NTA affinity chromatography. Detailed description of these experimental procedures is available in SI Appendix.

Enzyme Activity Assay.

The TFP-specific activity for the overall reaction (i.e., three reactions coupled) was assayed as described (10), Briefly, 5 μL of TFP solution containing about 1–10 μg of TFP protein was added to 0.5 mL of assay solution, containing 0.1 M potassium phosphate buffer (pH 7.6), 1 mM NAD⁺, 0.2 mM CoA, and 20 µM 2-transhexadecenoyl-CoA, to initiate the overall enzyme reaction at 23 °C. The product NADH was monitored by the absorbance increase at 340 nm. One unit of enzyme activity is defined as 1 µmole of NADH produced/min. To obtain the TFP activities in the presence of detergents, 0.2 mM DMPC, 24 mM OBG, or 1 mM DDM was included in the assay solution.

Crystallization, Data Collection, and Structure Determination.

Crystals were obtained by hanging drop vapor-diffusion method, with crystallization dips composed of 2 μL of the protein solution (0.1 mM hTFP, 20 mM NAD⁺, 1 mM acetoacetyl-CoA, and 0.5% C8E5) and 2 μL of reservoir solution (0.1 M Hepes buffer pH 7.0, 12% PEG3350, and 0.2 M MgCl₂). Diffraction data were collected at Beamline IMCA-CAT 17-ID-B at the Advanced Photon Source, Argonne National Laboratory, Argonne, IL. Data were processed by programs Mosflm and Scala in the CCP4 program package (25). The initial structure was determined using the Phaser program (26), as detailed in SI Appendix, Methods. Refinement was carried out using iterative cycles of CNS refinement followed by manual fitting and rebuilding using the COOT graphics software (27). Chains A and B (β₂) and G and H (α₂) have the most residues modeled in and, therefore, unless otherwise stated, Chains A, B, G, and H corresponding to one α₂β₂ heterotetramer were used for structural interpretations. Data collection and processing statistics and the final refinement statistics are given in SI Appendix, Table S1.