Research Article

Architecture of Mammalian Fatty Acid Synthase at 4.5 Å Resolution

See allHide authors and affiliations

Science  03 Mar 2006:
Vol. 311, Issue 5765, pp. 1258-1262
DOI: 10.1126/science.1123248


The homodimeric mammalian fatty acid synthase is one of the most complex cellular multienzymes, in that each 270-kilodalton polypeptide chain carries all seven functional domains required for fatty acid synthesis. We have calculated a 4.5 angstrom–resolution x-ray crystallographic map of porcine fatty acid synthase, highly homologous to the human multienzyme, and placed homologous template structures of all individual catalytic domains responsible for the cyclic elongation of fatty acid chains into the electron density. The positioning of domains reveals the complex architecture of the multienzyme forming an intertwined dimer with two lateral semicircular reaction chambers, each containing a full set of catalytic domains required for fatty acid elongation. Large distances between active sites and conformational differences between the reaction chambers demonstrate that mobility of the acyl carrier protein and general flexibility of the multienzyme must accompany handover of the reaction intermediates during the reaction cycle.

Fatty acids are central building blocks of life. They are constituents of biological membranes, energy storage compounds, and messenger substances, and they act as post-translational protein modifiers and modulate gene expression. Consequently, the de novo synthesis of fatty acids is essential for all organisms. It involves a conserved set of chemical reactions for the cyclic stepwise elongation of activated precursors by two-carbon units (1, 2) (Fig. 1). The growing fatty acid is attached to a carrier protein, acyl carrier protein (ACP), throughout its synthesis and is, in mammals, released by a thioesterase (TE) once it reaches 16 or 18 carbon atoms in length (3). Although all organisms use variations of this common synthetic scheme, surprisingly, three distinct architectures for fatty acid synthesis have evolved. In bacteria, all reactions are carried out by individual, monofunctional proteins in a dissociated or type II fatty acid synthase (FAS) system (1). In contrast, the eukaryotic type I FAS consists of large, multifunctional polypeptides. Fungal FAS is a 2.6-MD α6β6 dodecamer, in which the catalytic domains are distributed over two distinct subunits (4, 5). The FAS of vertebrates and mammals is an α2 homodimer of a single 270-kD polypeptide. It harbors all catalytic activities required for the synthetic cycle and, in addition, ACP (Fig. 1), making it one of the most complex mammalian enzymes (2).

Fig. 1.

Catalytic cycle and domain organization. The reaction cycle of FAS is initiated by the transfer of the acyl moiety of the starter substrate acetyl-CoA to the acyl carrier protein (ACP, gray) catalyzed by the malonyl-CoA-/acetyl-CoA-ACP-transacylase (MAT, red), which also transacylates the malonyl group of the elongation substrate malonyl-CoA to ACP. The β-ketoacyl synthase (KS, orange) catalyzes the decarboxylative condensation of the acyl intermediate with malonyl-ACP to a β-ketoacyl-ACP intermediate, acetoacetyl-ACP in the first cycle. The β-carbon is processed by nicotinamide adenine dinucleotide phosphate (NADPH)–dependent reduction through β-ketoacyl reductase (KR, yellow). The resulting β-hydroxyacyl-ACP is dehydrated by a dehydratase (DH, light green) to a β-enoyl intermediate, which is reduced by the NADPH-dependent β-enoyl reductase (ER, dark green) to yield a four-carbon acyl substrate for further cyclic elongation with two-carbon units derived from malonyl-CoA until a substrate length of C16 to C18 is reached. Finally, the product is released from the ACP by the thioesterase (TE, blue). The lower panel shows the linear domain organization of mammalian FAS.

Because of its role in fatty acid synthesis, human FAS is a target for drug development against obesity and obesity-related diseases, including diabetes and cardiovascular disorders. FAS inhibitors have shown potential for weight reduction in animal models (6, 7), though their exact mode of action is under discussion (8). FAS is overexpressed in many forms of cancer (9), and FAS inhibitors have demonstrated antitumor activity (10).

Mammalian FAS serves as a paradigm for a class of multifunctional enzymes known as megasynthases. Members of this family use iterative condensations of carboxylic acid (polyketide synthases, PKS) or amino acid (nonribosomal peptide synthetases, NRPS) building blocks to assemble a variety of secondary metabolites with important biological properties, including immunosuppressants and antibiotics (11). Whereas the NRPS are only conceptually related to FAS, modular PKS systems share a common set of catalytic domains with mammalian FAS. Furthermore, the functional domains in mammalian FAS and modular PKS are often arranged in similar order at the sequence level, as exemplified by desoxyerythronolide B synthase (DEBS) (12), which is involved in erythromycin biosynthesis.

Currently, no experimental structural information beyond low-resolution electron microscopic reconstructions (13, 14) is available for complete eukaryotic type I FAS or intact PKS modules. However, high-resolution structures of isolated bacterial FAS enzymes yielded important insights into the general reaction mechanisms of fatty acid synthesis (1), and crystal structures of recombinant isolated FAS and PKS domains, such as FAS TE (3), revealed details about individual active sites of these systems. Here, we present the crystal structure of mammalian FAS at 4.5 Å resolution. It enables accurate placement of all catalytic domains of the fatty acid elongation cycle and provides insight into domain organization in mammalian FAS. ACP and TE domains could not be placed, presumably because of their inherent flexibility.

Structure determination. FAS was purified from porcine mammary gland by established procedures (15). On the basis of amino acid sequence identities between human, bovine or rat, and porcine FAS of more than 70%, the latter is representative for all mammalian FAS systems. FAS crystals in the monoclinic space group P21 with a maximum size of 0.40 mm by 0.07 mm by 0.02 mm were grown by the vapor-diffusion method using polyethylene glycol 3350 as the precipitant at pH 6.7 to 7.3 and diffracted to a maximum resolution of 4.3 Å. Experimental phases to 4.5 Å resolution were determined using multiple isomorphous replacement with anomalous scattering and improved by density modification. Secondary-structure elements are clearly recognizable in most parts of the molecule, as expected for a 4.5 Å–resolution crystallographic map. Based on the identification of secondary-structure elements, all catalytic domains of the fatty acid elongation cycle were placed into the electron density map (Fig. 2). However, it was not possible to unambiguously trace the interdomain linking regions. The ACP and TE domains could not be placed with confidence most likely because of their inherent flexibility or flexible attachment and have not been included in the current model.

Fig. 2.

Electron density fit of domain homologs. (A) KS domain fitted with E. coli FabB (PDB accession code: 1ek4). (B) MAT fitted with Streptomyces coelicolor FabD (PDB: 1nm2). (C) DH pseudo-dimer fitted with two monomers of dimeric E. coli FabA (PDB:1mka).(D) ER fitted with T. thermophilus quinone reductase (PDB: 1iz0). (E) KR fitted with E. coli FabG (PDB: 1i01). In (A) to (E), the right side shows a slab view of the models fitted as rigid bodies into the experimental electron density (contoured at 1σ level) in an orientation similar to that of the fold representation of the respective homologous proteins on the left side.

Overall structure and domain assignment. Mammalian FAS adopts an X-shape with a central body extended at the upper and lower ends by “arms” and “legs,” respectively. The overall dimensions of the complex of 210 Å by 180 Å by 90 Å are in good agreement with earlier low-resolution electron microscopic observations (13, 14, 16) (Fig. 3). An approximate two-fold rotational axis of symmetry relating the two monomers of homodimeric FAS extends vertically through the FAS body, as indicated in Fig. 3. However, an assignment of domains to the two distinct monomers is not yet possible, because the current resolution is insufficient to unambiguously trace interdomain connecting regions.

Fig. 3.

Structural overview. Fitted domains (colored as in Fig. 1) are shown with a semitransparent surface representation of the experimental electron density (contoured at 1σ level) around one dimeric FAS. White stars indicate the pseudosymmetry-related suggested attachment regions for ACP and TE, where only on the right side a large volume of blurred density is visible. (A) Front view: FAS consists of a lower part comprising the KS (lower body) and MAT domains (legs) connected at the waist with an upper part formed by the DH, ER (upper body), and KR domains (arms). (B) Top view of FAS with the ER and KR domains resting on the DH domains. (C) Bottom view showing the arrangement of the KS and MAT domains and the continuous electron density between the KS and MAT domains. In (A) to (C), the approximate position of the pseudo-twofold dimer axis is indicated by an arrow and ellipsoid.

Even though the sequence identity between individual proteins of the bacterial type II FAS system and mammalian FAS is low in some areas, most of the mammalian FAS domains adopt a fold similar to that of their bacterial counterparts (Table 1 and Fig. 2). Starting from the N terminus of mammalian FAS, the β-ketoacyl synthase (KS) domains are located in the lower body (Fig. 3, A and C) and closely resemble the Escherichia coli KS I (FabB) (17) (Fig. 2A). The malonyl-coenzyme A (CoA)-/acetyl-CoA-ACP-transacylase (MAT) domains form the two “legs” of FAS (Fig. 3, A and C) and are homologs of the bacterial malonyl transferase (FabD) (18) (Fig. 2B). The dehydratase (DH) domains comprise the upper body of FAS (Fig. 3A). Despite a lack of sequence homology, each of these domains adopts a “double hot dog” fold (Fig. 2C) closely related to the fold of the dimeric bacterial dehydratases FabA (19) and FabZ (20) and related pseudo-dimeric eukaryotic enzymes (21). The β-enoyl reductase (ER) domain is a member of the medium-chain dehydrogenase family (22). The best structural match was obtained with a zinc-free bacterial quinone reductase (Fig. 2D) (23) with the application of a small rotation of the catalytic relative to the nucleotide-binding domain. Notably, the structure of a PKS ER domain fragment [Protein Data Bank (PDB) accession code: 1pqw] closely resembles that of the cofactor-binding domain. The ER domains sit on top of the DH domains at the upper end of the FAS body (Fig. 3, A and B). The last catalytic domains of the fatty acid elongation cycle, the β-ketoacylreductase (KR) domains, are located adjacent to the ER domains in the FAS arms (Fig. 3, A and B). KR belongs to the short-chain dehydrogenase family (24), comprising bacterial enoyl- and ketoreductases, and was modeled with E. coli KR (FabG) (25) (Fig. 2E).

Table 1.

Domains of the mammalian FAS elongation cycle, and their structural homologs and functional analogs.

Mammalian FAS domainOligomerization statePlaced structural homologsOligomerization stateFunctionally related bacterial FAS proteins
KS Dimeric FabB, E. coli Dimeric FabB, FabF, FabH
MAT Monomeric FabD, S. coelicolor Monomeric FabD
DH Pseudo-dimericView inline FabA, E. coli Dimeric FabA, FabZ
ER Dimeric Quinone reductase, T. thermophilus Dimeric FabI, FabK, FabL
KR Monomeric FabG, E. coli Tetrameric FabG
  • View inline* The DH pseudo-dimer occurs within one polypeptide chain of FAS and is not formed across the dimer interface between the two FAS subunits. The full FAS dimer thus contains two individual pseudo-dimeric DH domains.

  • At the end of one arm (right in Fig. 3, A and B), a blurred volume of electron density is observed, which is nearly completely absent at the end of the other arm. Most likely, it represents a particularly mobile part of FAS, which is only partly stabilized by the observed crystal contacts. It might be interpreted as arising from the C-terminal ACP and TE domains of one monomer based on the size of the density and the close vicinity to the KR domain, which is directly preceding ACP and TE in linear sequence. This assignment agrees with the location of the TE domain at the ends of the long axis of FAS inferred from visualization of antibody complexes of harderian gland FAS (16) and the approximately equidistant location of a labeled ACP phosphopantheteine to both types of reductase centers (26). The high inherent flexibility of the TE domain has already been demonstrated by limited proteolysis (27), fluorescence and mutational studies (28), and the functional interaction of FAS with thioesterase II (29). Furthermore, structures of the isolated human TE (3) and rat ACP domains (30) suggest the presence of considerable intradomain flexibility.

    The central ∼650 residues of mammalian FAS have previously been assigned as the “core” or “interdomain” (31) (Fig. 1), which is characterized by the absence of catalytic centers and lower sequence conservation and has been implicated in FAS dimerization (32). The structure of mammalian FAS reveals that the DH domain that precedes the “core” forms a “double hot dog” fold and has about twice the expected size at sequence level (2). Consequently, the length of the catalytically inactive “core” is reduced to about 450 residues. However, no additional electron density corresponding to a compact domain of such size could be identified, suggesting that it may be disordered or distributed in between other domains serving a structural role (fig. S1, A and B).

    Intersubunit and interdomain connections. In the early, classical model, mammalian FAS was represented as an H-shaped dimer with linear head-to-tail arrangement of subunits, which are centrally connected by the noncatalytic “core” (32). On the basis of structural and functional characterization of recombinant mutant FAS and complementation assays, Smith and co-workers revised the initial model and depicted FAS as an intertwined head-to-head dimer with distinct conformations at various stages of its catalytic cycle (2, 14). The current structure fundamentally agrees with the revised model and demonstrates that mammalian FAS is, indeed, an intertwined dimer with a large dimerization interface running through the body of the molecule, perpendicular to the interface proposed in the classical scheme (13).

    The KS domains dimerize in the same way as the homologous homodimeric E. coli KS I enzyme (FabB) (17) (fig. S1D) with their N termini in close proximity; this is in agreement with cross-links between the N termini of companion KS domains via engineered cysteine residues (33). Another important contribution to the dimer interface comes from the ER domain. Based on the placement of the homologous Thermus thermophilus quinone oxidoreductase monomers into the electron density, also the ER domains of mammalian FAS associate in the same way as the isolated homologous bacterial enzyme (23). The interaction is guided by the formation of a continuous 12-stranded β sheet between the nucleotide-binding domains of the monomers (fig. S1C). Consistent with a role of the ER domain in dimerization, the thermal stability of homodimeric ER active-site mutants of FAS is substantially reduced (34). As estimated from the structures of the homologous bacterial enzymes, the two homophilic interactions between the ER and KS domains contribute ∼5000 Å2 to the total dimer interface of mammalian FAS. Substantial intersubunit contacts are also formed along the pseudo-twofold axis in the waist region by the lower parts of the DH domains (Fig. 3A). The unassigned interspersed regions and interdomain connections could mediate further intersubunit interactions (fig. S1).

    In the current structural model, the KS domains are surrounded by linking regions interconnecting KS/MAT and MAT/DH, which apparently build up a mixed α/β-fold adapter between KS and MAT. At their top, the KS domains are contacting the lower part of the DH domains, connecting the lower and upper part of the body in the waist region. The spatial arrangement of these domains may explain why the shortest recombinant N-terminal FAS construct with KS activity must, in addition to KS, also enclose MAT and part of the DH domain, which are surrounding KS in the current structure, and why this construct shows dimerization properties similar to those of the full-length FAS (33). The example of KR demonstrates that the oligomerization contacts are not transferred from the isolated bacterial homologs to the mammalian FAS domains as a rule (Table 1): Whereas the E. coli KR (FabG) is tetrameric (25), the two KR domains of mammalian FAS do not interact.

    Active sites and reaction chambers. The placement of homologous structures with known catalytic mechanism into the 4.5 Å electron density map accurately defines the positions of active sites of mammalian FAS. During the catalytic cycle of FAS, the growing acyl chain remains attached to the phosphopantetheine arm of ACP, with the exception of the temporal transfer to the KS active site. From the location of the KR domain, to which the ACP is tethered by only a short linker, and the position of the TE domain inferred from earlier work (16), it is possible to establish the approximate position of ACP close to the ends of the FAS “arms.” On the basis of the structural information presented here, the active centers of FAS fall into two groups, according to their accessibility to one or the other ACP domain: one complete set of domains required for productive elongation in each of the two lateral clefts (Fig. 4A). In the observed conformation of FAS, it appears difficult for the ACP to reach any of the active centers of the opposite side cleft—not only because of long distances, but also because the KS and DH domains protrude sideways and block the way around the body to the other cleft (Fig. 4B). Consequently, the lateral clefts define two preferred reaction chambers (Fig. 4A).

    Fig. 4.

    Active sites and reaction chamber. (A) Front view of FAS with ribbon representations of fitted domains colored as in Fig. 1. The overall shape is indicated by the outline of electron density; gray and blue colors of the outline mark the nonmodeled KS/MAT interconnection and suggested ACP/TE location, respectively. The positions of active sites in the two reaction chambers are indicated by solid white and blue spheres. Hollow spheres in domain colors that surround the active sites denote the length of the phosphopantheteine arm, reflecting how close ACP has to approach the individual domains during the catalytic cycle. The active sites are connected in order of the reaction sequence with distances between the active sites indicated for the left reaction chamber. (B) Side view into one reaction chamber as indicated by a white arrow in (A); for clarity, only surface representations of the fitted domains are shown. Active sites of fitted domains of one reaction chamber are indicated by a color gradient to white on the respective surfaces.

    In the order of the FAS reaction cycle, the lower part of the reaction chamber comprises the first two enzymatic domains, with MAT forming the legs and KS the central body. The distance between their deep-set active sites is ∼71 Å; distances between the entrances to the substrate-binding clefts are considerably shorter. The visible connections in electron density between the KS and MAT domains suggest that the connected domains belong to one monomer (Fig. 3C). Based on this assumption, the KS and MAT domains of each reaction chamber are contributed by different monomers. The upper part of the chamber is composed of the three β-carbon processing domains: KR, DH, and ER. The active site of KR is located at a distance of ∼72 Å to the preceding domain in the reaction sequence, KS. The DH active site resides only about 37 Å away from the KR active site, but its substrate-binding cleft points in a slightly different direction. The ER and DH domains are in close proximity, with a distance of ∼32 Å between their active centers. The arrangement of these two catalytic sites would even allow ACP to shuttle the substrates between them without substantially changing its position.

    Conformational variability and reaction mechanism. Considering the overall shape of FAS, crystallized in the absence of cofactors or substrates, it is noticeable that its conformation results in two nonidentical reaction chambers. One chamber (Fig. 3A, left; Fig. 5, A and B) is considerably narrower than the other: The distance between the active centers of the peripheral MAT and KR domains is 72 Å in the narrow chamber, but 87 Å in the wide chamber (Fig. 5B). Surprisingly, also electron microscopic reconstructions of mammalian FAS in the presence of substrates (13, 14) frequently yielded asymmetric structures. Together, these observations might suggest a physiological relevance of the observed asymmetry of FAS, although we cannot exclude that the crystallized FAS was only trapped in one out of multiple possible conformations. A superposition of FAS onto itself based on the twofold relation between the central KS domains reveals hinge regions that cause the observed asymmetry of the clefts (Fig. 5B). The central hinge is located very close to the substrate-binding lids of the KS domain dimer in the “waist” region connecting the lower and upper parts of FAS. Notably, the crystal structure of the homodimeric bacterial homolog of KS, FabB, revealed an asymmetric mode of substrate binding (17). Furthermore, even under saturating substrate conditions, the KS domain of FAS binds single substrates only substoichiometrically (35). Therefore, it is tempting to speculate that asymmetric binding and release of substrates by the KS dimer may affect the conformations of the KS substrate-binding loops at the waist region of the FAS and induce opening and closing of the reaction chambers.

    Fig. 5.

    Interdomain hinges and conformational variability. For structural comparison, the FAS dimer is superimposed onto itself by applying the transformation relating the dimer of KS domains as indicated by an arrow. As a result, the left reaction chamber is transformed onto the right one and vice versa. The original orientation is shown in red, the transformed one in yellow. (A) Only secondary structural elements of the fitted domains are shown. Largest differences are observed for the positions of the KR and MAT domains at the periphery. The approximate position of the pseudo-twofold dimer axis is indicated by an arrow. (B) Experimental electron density is schematically shown as an outline. The positions of active sites are indicated by spheres, hinges by crossed circles. The left reaction chamber is considerably narrower than the right one with a difference of distances between the KR and MAT active sites of about 15 Å as indicated for the original orientation in red.

    Around a second hinge, the MAT domains in the “legs” of FAS undergo a slight up-and-down motion relative to the KS domain (Fig. 5B). The extended interface on both sides of the MAT/KS joint (Fig. 3A), however, appears to preclude large-scale motions of the MAT domain. A third hinge resides at the less solid contact between the KR domains and the pairs of ER and DH domains, which are held together by a substantial interface (Fig. 5B). The phosphopantetheine group of ACP obviously does not serve as a “swinging arm,” as proposed in very early models of type I FAS (36). As indicated in Fig. 4A, its length is just sufficient to reach the deep-set active centers, even assuming that the ACP is in close proximity to the respective domains. In the dissociated bacterial system, substrate-loaded ACP interacts transiently with individual FAS proteins through a proposed common ACP-binding motif in these proteins (37). On the basis of the observed structural homology, such guiding interactions might also facilitate the entry of ACP-bound substrates into deep-set active sites in mammalian FAS.

    From the arrangement of active sites within one chamber, it is obvious that considerable flexibility of ACP, which might result from a combination of its internal and linker flexibility, combined with modest domain motions, as indicated by the observed asymmetry, are required to enable access of ACP to all domains of one cleft, likely involving KS and MAT domains from different FAS subunits. However, biochemical studies have established the existence of redundant alternative routes for fatty acid elongation by FAS (34, 38), which increase the overall efficiency of FAS (2). The interaction between the ACP-bound substrate and the DH, ER, KR, and TE domains is almost exclusively an intrasubunit process. However, loading and condensation may involve either intersubunit or intrasubunit interactions between KS and MAT with ACP (39). On the basis of mutant complementation and cross-linking, 20 to 35% of all elongation cycles proceed via the alternative route, involving intrasubunit interplay between ACP and KS (2, 40). Even a heterodimeric FAS with all catalytic domains of one subunit inactivated by mutation still exhibits 16% of the wild-type activity (38). Considering the large distance between the involved domains—for example, between the suggested position of one ACP and both MAT domains—the presented structure accounts for a major conformation but would not explain a minor alternative synthetic route. The existence of multiple conformations of FAS and their dependence, for example, on the presence of substrates were inferred from early cross-linking experiments (41) and have recently been observed in electron microscopic studies (14).

    Implications for the megasynthase family. Mammalian FAS is a paradigm for the structural organization of modular PKS. These huge homodimeric proteins are assembled from multiple modules, each capable of catalyzing one elongation step, equivalent to a single elongation cycle in FAS, with various extends of β-carbon processing. The minimal PKS module consists of a KS, an acyl transferase (AT), and an ACP domain. Extensions of this minimal set by β-carbon processing domains, together with the substrate preference of the AT domain, determine the product of a particular module. Under physiological conditions, substrate transport through PKS modules is colinear with the arrangement of modules at sequence level, such that the order of distinct modules determines the chemical structure of products, which are released by a terminating thioesterase (11). Homodimeric modular PKS have been envisioned as parallel, interwound supramolecular helices with a structural core formed by KS, AT, and ACP domains and off-axis extensions by varying numbers of β-carbon processing domains (42). This arrangement is represented in the structure of mammalian FAS by the KS/MAT domain blocks in the lower part of mammalian FAS, dimerized via the homophilic interactions of the KS domain and segregated upper segments comprising the β-carbon processing, ACP, and TE domains.

    The peripheral positioning of MAT domains of mammalian FAS in the “leg” region and their attachment through an interface formed by noncatalytic linker regions might indicate that the related AT domains of modular PKS could also be placed off-axis without a role in dimerization. Such positioning agrees well with the existence of modular PKS lacking an internal AT domain, in which the AT functionality is supplied in trans by monofunctional AT proteins (43). In mammalian FAS, the downstream domain block beyond MAT, including β-carbon processing domains, ACP, and TE, is in contact with the substrate-binding region of the KS domain in the waist region, and it might be speculated that a similar arrangement in PKS modules could provide the possibility of cross talk between KS and downstream domains. Of course, the structure of cyclically acting mammalian FAS cannot provide direct insight into intermodular substrate shuttling in PKS. However, the domain architecture revealed here provides a first structural view onto the common scaffold of mammalian FAS and modular PKS.

    Conclusion. The overall architecture of mammalian FAS has been revealed by x-ray crystallography at intermediate resolution. The dimeric synthase adopts an asymmetric X-shaped conformation with two reaction chambers on each side formed by a full set of enzymatic domains required for fatty acid elongation, which are separated by considerable distances. Substantial flexibility of the reaction chamber must accompany the handover of reaction intermediates during the FAS cycle, and further conformational transitions are required to explain the presence of alternative inter- and intrasubunit synthetic routes in FAS. The results presented here provide a new structural basis to further experiments required for a detailed understanding of the complex mechanism of mammalian FAS. Furthermore, continued work on the current crystal system may ultimately provide an atomic model of mammalian FAS.

    Supporting Online Material

    Materials and Methods

    Fig. S1

    Table S1

    References and Notes

    Stay Connected to Science

    Navigate This Article