Crystal structure of FAS thioesterase domainwith polyunsaturated fatty acyl adduct andinhibition by dihomo-γ-linolenic acidWei Zhang1, Bornali Chakravarty1, Fei Zheng1, Ziwei Gu1, Hongmei Wu, Jianqiang Mao1,Salih J. Wakil2, and Florante A. Quiocho2

Verna and Marrs McLean Department of Biochemistry and Molecular Biology, Baylor College of Medicine, One Baylor Plaza, Houston, TX 77030

Contributed by Salih J. Wakil, August 1, 2011 (sent for review March 16, 2011)

Human fatty acid synthase (hFAS) is a homodimeric multidomainenzyme that catalyzes a series of reactions leading to the de novobiosynthesis of long-chain fatty acids, mainly palmitate. The car-boxy-terminal thioesterase (TE) domain determines the length ofthe fatty acyl chain and its ultimate release by hydrolysis. Becauseof the upregulation of hFAS in a variety of cancers, it is a targetfor antiproliferative agent development. Dietary long-chain poly-unsaturated fatty acids (PUFAs) have been known to confer ben-eficial effects on many diseases and health conditions, includingcancers, inflammations, diabetes, and heart diseases, but theprecise molecular mechanisms involved have not been elucidated.We report the 1.48 Å crystal structure of the hFAS TE domain cova-lently modified and inactivated by methyl γ-linolenylfluoropho-sphonate. Whereas the structure confirmed the phosphorylationby the phosphonate head group of the active site serine, it alsounexpectedly revealed the binding of the 18-carbon polyunsatu-rated γ-linolenyl tail in a long groove-tunnel site, which itself isformed mainly by the emergence of an α helix (the “helix flap”).We then found inhibition of the TE domain activity by the PUFAdihomo-γ-linolenic acid; γ- and α-linolenic acids, two popular dietaryPUFAs, were less effective. Dihomo-γ-linolenic acid also inhibitedfatty acid biosynthesis in 3T3-L1 preadipocytes and selective humanbreast cancer cell lines, including SKBR3 andMDAMB231. In additionto revealing a novel mechanism for the molecular recognition of apolyunsaturated fatty acyl chain, our results offer a new frameworkfor developing potent FAS inhibitors as therapeutics against cancersand other diseases.

X-ray crystallography ∣ conformational change ∣ fatty acid selectivity ∣two-subdomain binding site

Fatty acids, the main constituents of fat, biological membranelipids, energy storage, and signal transducer compounds, are

absorbed from diet or synthesized de novo from acetyl-CoA,malonyl-CoA, and NADPH through multiple rounds of a seriesof reactions catalyzed by the homodimeric large multifunctionalfatty acid synthase (FAS) (1). The reactions are carried out byseven different catalytic domains in one catalytic center. Thethioesterase (TE) domain, which is located at the carboxyl end ofthe subunit, plays critical dual roles in regulating the length ofthe fatty acyl chains (C14∶0, C16∶0, C18∶0), with the palmitoyl(C16∶0) being optimal, and in finally hydrolyzing the acyl-S-phos-phopantetheine thioester bound to the preceding acyl-carrierprotein domain to release the fatty acid.

FAS has been shown to be associated with a variety of humandiseases and adverse health conditions such as obesity, cardiovas-cular disease, inflammation, and cancer. Its relationship withcancer pathogenesis has attracted the most attention (reviewedin refs. 2 and 3). In most normal human tissues, FAS expressionis very low, but it is aberrantly elevated in a variety of cancers,including carcinoma of the prostate, breast, ovary, colon, endo-metrium, thyroid and skin. The elevated FAS protein expressionlevel is also correlated with an increase in de novo fatty acid

synthesis in corresponding tissues. The coupling of FAS expres-sion with tumor progression led to the notion that FAS protein isassociated with the growth and survival of cancer cells. Moreover,the identification of the oncogenic antigen-519 (OA-519) fromhuman breast carcinoma cells as FAS has made it an importantdiagnostic and prognostic marker for breast cancer patients (4, 5).The importance of FAS in cancer cells is further underscoredby the observation that inhibitions of FAS catalytic activitiesare selectively cytotoxic to human cancer cells (reviewed in ref. 6).These include inhibitions of the activities of the following cata-lytic domains: ketoacyl synthase by cerulenin (7, 8), C75 (9), andepigallocatechin-3-gallate (a green tea component) (10); enoylreductase by triclosan, a widely used antibacterial in soaps,mouthwashes, etc. (11); and thioesterase by Orlistat, an antiobe-sity drug approved by the Food and Drug Administration (12).Taken together, these observations continue to make FAS anattractive target for developing unique antitumor compounds.

Long-chain polyunsaturated fatty acids (PUFAs) are essentialdietary fatty acids, mainly supplied by vegetables and fishes. Theyhave been shown to have beneficial effects on human health, suchas improved heart disease-related outcomes, antiproliferativeeffects on tumor cells, and anti-inflammatory properties. Likethe association of FAS with cancer pathogenesis, the antiproli-ferative effect of PUFAs, especially those belonging to the (n-3)or ω-3 and (n-6) or ω-3 series, has attracted considerable atten-tion (reviewed in ref. 13). Because the literature on the beneficialeffects of PUFAs is vast and continues to be debated and dis-cussed, we summarize briefly those studies that may be relevantto our structural and biochemical studies of the effects of PUFAs,especially in conjunction with FAS.

Several studies have shown that dietary PUFAs and syntheticPUFAs, particularly γ-linolenic acid (GLA, 18∶3n-6) and α-lino-lenic acid (ALA, 18∶3n-3) (see Scheme S1), suppress hepaticlipogenesis as compared with saturated and monounsaturatedfatty acids, and that this suppression is accompanied by the inhi-bition of the hepatic synthesis of FAS (14–16). (Synthetic PUFAsused in these experiments are often esterified to enhance cell up-take and then presumably hydrolyzed by lipases to their activefree acid form.) GLA and ALA also selectively kill cancer cellswhile causing little or no harm to normal cells (17). Exogenoussupplementation with high concentrations of the methyl esters of

Author contributions: W.Z., B.C., F.Z., S.J.W., and F.A.Q. designed research; W.Z., B.C., F.Z.,Z.G., H.W., and J.M. performed research; W.Z., B.C., F.Z., Z.G., H.W., J.M., S.J.W., and F.A.Q.analyzed data; and F.A.Q. wrote the paper.

The authors declare no conflict of interest.

Data deposition: The atomic coordinates and structure factors have been deposited in theProtein Data Bank, www.pdb.org (PDB ID code 3TJM).1W.Z. and B.C. contributed equally to the structural analysis and Z.G., F.Z., and J.M. to the invitro and in vivo studies.

2To whom correspondence may be addressed. E-mail: faq@bcm.edu or swakil@bcm.tmc.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1112334108/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1112334108 PNAS ∣ September 20, 2011 ∣ vol. 108 ∣ no. 38 ∣ 15757–15762

BIOCH

EMISTR

Y

Dow

nloa

ded

by g

uest

on

June

7, 2

020

GLA and ALA has also been demonstrated to significantly sup-press FAS activity and expression in tumor cells (18). Addition-ally, both PUFAs have been indicated to suppress overexpressionof Her2/neu oncogene (19). Finally, PUFAs have been shown toenhance the cytotoxic activity of several anticancer drugs (e.g.,taxol, paclitaxel) in cell culture and in tumor-bearing mice (13).

In mammals, including man, dihomo-γ-linolenic acid (DGLA)is metabolized from linoleic acid via GLA and further convertedto arachidonic acid (AA), but because of the poor conversionefficiency, DGLA, not AA, accumulates after GLA supplemen-tation (20). Moreover, it has been shown recently that adminis-tration of exogenous DGLA oil (in the form of triglyceride) inrats is more effective than GLA administration for increasingthe DGLA concentration within the body (21).

The precise molecular mechanism(s) by which long-chainPUFAs exert their beneficial effects in a variety of chronic diseasestates, especially cancers, are not completely understood. Thesame is true for the mode of action(s) of PUFAs on FAS activity.A popular but controversial view is that the cytotoxic and anti-proliferative effects of exogenous GLA or ALA are due to for-mation of peroxidation products that could then inhibit FAS(13, 22). On the other hand, experimental data showed thatPUFAs themselves are responsible for FAS suppression (23). Tofurther our molecular understanding of these issues, we carriedout structural and biochemical studies of the 32 kDa TE domainof hFAS covalently modified by a methylfluorophosphonatehead group linked to a γ-linolenyl acyl tail and, prompted by theresults, investigated the inhibitions of related PUFAs on the TEactivity, fatty acid biosynthesis in 3T3-L1 preadipocytes andhuman breast cancer cell lines.

ResultsCrystal Structure of the TE Domain of hFAS with a Covalently BoundMethyl γ-Linolenylphosphonate. Long-chain polyunsaturated fatty

acyl fluorophosphonates have been demonstrated previously toirreversibly inactivate phospholipases through rapid phosphoryla-tion of a reactive serine residue in the active site (24, 25). Ourstudies reported here were set in motion by the determinationof the X-ray structure of the hFAS TE domain treated with two-fold excess of methyl γ-linolenyl fluorophosphonate (MGLFP;Scheme S1) (see Methods and SI Methods). (Unless further spe-cified, henceforth the FAS and its TE domain described in ourwork are those of hFAS.) The structure was determined by mo-lecular replacement technique, using the ligand-free TE domainstructure (26) as the search model, and refined at 1.48 Å resolu-tion to final R-cryst and R-free values of 19.37% and 22.89%,respectively, and excellent geometry (Table S1). The resolution isa considerable improvement over those of previous FAS TE do-main structures in the ligand-free form (2.6 Å resolution) (26)and in complex with the inhibitor Orlistat (2.3 Å resolution) (27).

Four prominent features of the binding of MGLFP, whosewell-defined electron density is shown in Fig. 1A, are as follows.First, a covalent bond is formed between the phosphorus atom ofthe methylphosphonate head group and the hydroxyl side chainof Ser 2308, one of the catalytic triad of residues in TE (26), yield-ing a methyl γ-linolenylphosphonate (MGLP) adduct (Fig. 1 Band C). This finding is reflected by the result of thioesteraseactivity measurement, performed following the structure deter-mination, showing essentially complete loss of activity after treat-ment of TE in solution with fourfold excess of MGLFP (e.g., seeTable 1). Second, the γ-linolenyl (18-carbon) acyl chain is nestledand significantly buried (approximately 75% of its accessiblesurface area) in a long contiguous groove and tunnel (Fig. 1C).Approximately the first eight carbons of the γ-linolenyl chainreside in the groove, and the remaining 10 carbons occupy theentire length of the tunnel. Third, ligand binding causes the emer-gence of an amphiphatic α-helix (residues 2342–2355), called the“helix flap” or “gatekeeper helix,” which forms the major wall of,

Fig. 1. Structure of the hFAS TE domain with a covalently bound methyl γ-linolenylphosphonate (MGLP). (A) Ribbon backbone trace of the structure. TheσA-weighted 1.45 Å Fo–Fc electron density (yellow) of MGLP (green ball-and-stick model) is contoured at 0.8σ above the mean density. The α-helix (the helixflap or gatekeeper helix) colored in red and adjacent to the MGLP density is composed of residues 2342–2355. The TE domain structure is made up of twosubdomains—a much larger subdomain with the α/β hydrolase fold (Right) and a smaller subdomain with an all α-helices motif (α5–α8) (Left) (26). Because thehelix flap, which is missing in the first TE domain structure, precedes α5-helix, it is also identified as α5′ to preserve the original secondary structure identifica-tion. Renumbering all the helices in the TE–MGLP structure makes α5′ equivalent to α5, α5–α6, etc. (B) Schematic representation of the interactions (<4 Ådistance) between MGLP and TE amino acid residue using LIGPLOT diagram. Residues Ser2308, Asp2338, and His2481 (red labels) form the catalytic triadin the active sites. Residues involved in hydrophobic contacts with MGLP are colored in black and demarcated by spoked red arc, and those involved in hydro-gen bonding to MGLP are colored in green along with the values of the distances. The distance of the covalent bond between the phosphonate phosphorusand Ser2308 side chain O is 1.81 Å. Most of the residues interacting with the γ-linolenyl acyl chain reside in three helices: Val 2344, Tyr 2347, Arg 2352, and Tyr2351 in α5′ or helix flap; Glu 2366, Ala 2367, and Phe 2370 in α5; Tyr 2424, Leu 2427, Arg 2428, and Glu 2431 in α8. (C) Transparent surface rendition of the activesite (same view as in panel A) with embedded backbone trace of the TE structure. MGLP in a ball-stick model is covalently linked to Ser2308 sidechain O atom,which is represented as a van der Waals “dot” surface. Even-numbered carbon atoms of the γ-linolenyl acyl chain are labeled. Site 1 label marks the opening ofthe pocket of Site 1.

15758 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1112334108 Zhang et al.

Dow

nloa

ded

by g

uest

on

June

7, 2

020

and, consequently, gives rise to, the groove-tunnel site (Fig. 1 Aand C). Spanning the two TE subdomains (Fig. 1C), the well-defined groove-tunnel commences near Ser 2308 in the large sub-domain and terminates deep in the tunnel in the small subdomainformed by the triangulation of three helices—the helix flap, α5(which follows the helix flap) and α8 (Fig. 1A). This finding un-derscores a unique and important role of the two subdomainscombined. Fourth, a total of 73 contacts (<4 Å distances) aremade between MGLP and 16 TE residues (Fig. 1B). Indicativeof high complementarity, the majority of the contacts (44 total)are between the γ-linolenyl chain and nonpolar TE groups; 16contacts were made with residues deployed by the helix flap.The other contacts were made with residues from the N terminusof α5 helix and the middle of α8 helix.

The long groove-tunnel, formed almost solely by the appear-ance of the gatekeeper helix, has heretofore not been observedin any previous structure of TE, either in a natural form or asan isolated domain. The critical α helix could have arisen by aγ-linolenyl chain-facilitated formation of the helix from a disor-dered polypeptide segment or a stabilization of a very flexible αhelix. (Secondary structure predication indicated a low-probabil-ity α helix for the segment.) Because the ordered helix is approxi-mately 10 Å to the nearest TE domain molecule in the crystallattice, its appearance is unlikely influenced by crystal packing.

With no electron density for the segment of residues 2342–2355 in the structure of the ligand-free TE domain (26), thegroove-tunnel is not obvious. Nevertheless, the ligand-free struc-ture has, as noted previously, a separate groove-distal pocket site(identified as Site 1) formed between the two TE subdomains(26). Site 1 is still present in the TE–MGLP structure, togetherwith the groove-tunnel site (named Site 2) (Fig. 1C). The exis-tence of Site 1, but not Site 2, in the ligand-free structure has beencorroborated by computational analysis (28) (see also Fig. S1A,demonstrating a very similar result). A similar analysis of theTE–MGLP structure indicates Site 2 as the major site, exhibitinga cluster volume and energy significantly more favorable thanSite 1 (Fig. S1B). Although a site-directed mutagenesis experi-ment suggested possible involvement of Site 1 in the activity ofthe TE domain (26) and computational studies indicated dockingof fatty acids to the site (28), the functional significance of the siteremains to be further clarified or explored.

PUFA Inhibitions of the Thioesterase Activity of the TE Domain.Although dietary PUFAs have long been known to provide a vari-ety of beneficial effects on human health, the precise molecularmechanism(s) by which they exert these effects, and their specific

enzyme/protein target(s), remain unclear. Our structural work,demonstrating the precise binding of the γ-linolenyl tail of MGLPto the TE domain, may have uncovered a potential moleculartarget, and thus, a unique mechanism of action of PUFAs. Tofurther test these ideas, we examined the effects of related ω or(n-6) linolenic acids (GLA, DGLA, and ALA) (see Scheme S1)on the TE domain esterase activity with palmitoyl-CoA as thesubstrate (see SI Methods). Like MGLFP and MGLP, DGLA,ALA, and GLA have an identical set of three cis double bonds,although its location differs relative to the fatty acid chain length(Scheme S1). At a concentration of 16 μM in the assay mixture,DGLA caused about 54% reduction in activity, but ALA andGLA produced only 16% inhibition (Table 1). DGLA acts as acompetitive inhibitor, with a Ki value of 6.36 μM (Fig. S2). [Highsubstrate concentrations cause substrate or product inhibition(Fig. S2), which may explain the observation that palmitate inhi-bits TE activity comparable to that by DGLA.] Esterification ofDGLA (as in the compound dihomo-γ-linolenic acid ethyl ester)abolished inhibition of the TE domain activity. DGLA is lesseffective in inhibiting intact hFAS, causing only 31% inhibition.Phenylmethanesulfonyl fluoride (PMSF), a general reagent forrapid covalent modification of active site OH or SH groups, alsocaused almost complete loss of TE domain activity (Table 1). Inthe study of Menendez et al. (18), PMSF was used in the extrac-tion of FAS from human breast SKBR3 cancer cells treated withhigh concentrations of methyl ester of GLA and ALA (18) [ashigh as approximately 11 times the DGLA concentration usedin our studies (Table 1)], and may therefore have prevented amore accurate determination of the reduced activity of FAS fromextracts of cells treated with PUFAs.

The TE–DGLA complex has resisted crystallization, despiteexhaustive attempts. To get an idea of the possible mode ofbinding of DGLA, we carried out a straightforward modelingexperiment, because DGLA more closely aligns with MGLP(Scheme S1). The C3–C20 acyl segment of DGLA can be matchedexactly with the entire 18-carbon γ-linolenyl tail of MGLP(Fig. S3A). The C2C1OO− head group can be accommodatedwith no bad contacts in the area occupied by the methylphospho-nate, which deploys hydrophobic residues (Fig. 1B) and, withslight torsional rotation, its two carboxylate oxygens positionedwithin hydrogen-bonding distances to the His 2481 and Ser 2308residues of the catalytic triad (Fig S3 A and B).

Effects of DGLA on Palmitate Biosynthesis in 3T3-L1 Preadipocytes andon Breast Cancer Cell Lines.To test whether DGLA has an effect onde novo palmitate biosynthesis, 3T3-L1 preadipocyte cells were

Table 1. Effect of PUFAs on the thiosesterase activity of FAS and its TE domain

ProteinCompound

Specific activity‡ Normalized activity (%)Identification* Concentration (μM)†

TE domain (0.10 μM)† None (control) — 1,055±47 100Methyl γ-linolenylfluorophosphonate (MGLFP) 0.54 35±12 3Dihomo-γ-linolenic acid (DGLA) 16 481±21 46Dihomo-γ-linolenic acid ethyl ester 16 967±43 92γ-Linolenic acid (GLA) 16 884±40 84α-Linolenic acid (ALA) 16 885±40 84Palmitic acid (PA) 16 491±46 47PMSF 0.54 38±22 3

Intact human FAS (0.10 μM)† None — 92±46 100MGLFP 0.30 2±0.2 2DGLA 16 66±33 69

*Several of the abbreviations in parentheses correspond to those shown in Scheme S1. PMSF, phenylmethanesulfonylfluoride. All fatty acids were obtainedfrom Cayman Chemicals, stored at −80 °C and used from freshly made solutions in ethanol or DMSO. After obtaining the requisite aliquot, each vial ofPUFA was purged with nitrogen before storage.

†Concentrations refer to those in the assay mixture.‡Nanomoles product formed/min/mg of enzyme (see SI Methods). Average value from three assays. All assays contained about 0.8% ethanol, which onlycaused about <10% reduction in activity of TE in the absence of solvent.

Zhang et al. PNAS ∣ September 20, 2011 ∣ vol. 108 ∣ no. 38 ∣ 15759

BIOCH

EMISTR

Y

Dow

nloa

ded

by g

uest

on

June

7, 2

020

incubated for 24 h with 20 μM DGLA. Eight hours into the in-cubation, [13C]-sodium acetate was added. At the end of the 24-hincubation, the lipid was extracted and [13C]-palmitate was ana-lyzed by gas chromatography/mass spectrometry (GC/MS). Asshown in Fig. 2A, DGLA treatment of 3T3–L1 preadipocytesreduced the [13C]-acetate incorporation into [13C]-palmitate by51% (P ¼ 0.079).

We also tested the effect of the ethyl ester of DGLA on cellviability and fatty acid synthesis on some selected breast cancercell lines. As shown in Fig. 2B, the DGLA ethyl ester impactsthe viability of SKBR3 and MDAMB231 cell lines compared tononcancerous cells (HMEC) when exposed for 48 h at 20 μMconcentrations. Moreover, when the breast cancer cell linesSKBR3 and MDAMB23 were exposed to the DGLA ethyl ester(20 μM) for 12 h, then treated with [14C]-acetate for 6 h, theradioactivity incorporated into the fatty acid fractions (fattyacids) was 40 to 50% lower than that of the control cell linesthat were not treated with the DGLA ethyl ester. The overall FASactivities in the extracts of the DGLA ethyl ester-treated cellsvaried, but generally were 20–40% lower than the FAS activitiesin the extracts of the nontreated cells.

Comparison of the Structures of TE Complexes with MGLP and Orlistat.X-ray structure analysis has revealed two modes of binding of theinhibitor Orlistat to two independent TE domain molecules Aand B (see Fig 3) in the crystal structure (27). One mode, seenin molecule A, has a covalently bound inhibitor due to the acyla-tion of the active site serine residue by the β-lactone ring (Fig. 3B),and the other mode, observed in molecule B, has a noncovalentlybound product owing to the hydrolysis of the β-lactone ring(Fig. 3C). Consequently, the geometries of the three groups(palmitic core, peptidyl moiety, and hexanoyl tail) of the Orlistatdiffer between the two modes (Fig. 3 A–C).

The two TE–Orlistat and TE–MGLP complex structures arevery similar; the rmsds for α-carbon atoms between differentpairs of superimposed structures are 0.52 and 0.63 Å. The slightdifferences in the rmsds could be related to the significantlyhigher resolution of the TE–MGLP structure than that of theOrlistat-bound structure. Comparisons of the active site regionsof the TE–MGLP and the two TE–Orlistat structures (Fig. 3) ledus to the following major observations. First, the emergence ofthe helix flap in the active site region is, as already indicatedabove, unique to the TE–MGLP structure (Fig. 3 A–C). As in theligand-free TE domain structure (26), the segment correspondingto the helix is disordered in the two independent structures of

the TE–Orlistat complexes. Second, although the TE–Orlistatstructures lack the well-defined groove-tunnel site (Site 2), theyshow, as in the ligand-free and TE–MGLP structures (Fig. 1C andFig. S1A), the presence of groove-pocket Site 1. Third, given theirpositions in the respective TE structures (Fig. 3 D and E), thegatekeeper helix in the TE–MGLP complex would clash withthe hexanoyl tails of the two bound Orlistats (Fig. 3 D and E).Due to the difference in the relative orientations of the tailsbetween the two Orlistats, the extent of the clashes would differ,with the one in the noncovalently bound inhibitor being muchmore severe (Fig. 3 D and E). Fourth, major differences betweenthe modes of binding of MGLP and the two Orlistat forms areconfined to the region close to the catalytic triad (Fig. 3 D and E).This is attributed to the differences in the chemical and confor-mational features between MGLP and the two forms of Orlistat.Consequently, approximately the first seven carbons of the pal-mitic core of Orlistat in both forms are close to the proteinsurface, whereas those of the γ-linolenyl acyl group of MGLPlie in the groove. Interestingly, despite the differences, the cata-lytic triad is nearly superimposable in the overlapped structures,attesting to the stability of the active site residues. Fifth, due todifferences in binding geometries close to the catalytic triad, onlythe segments of approximately C7–C16 of the γ-linolenyl chain ofMGLP and C8–C16 of the palmitic core of the covalently or non-covalently bound Orlistat are in close proximity in the superim-posed structures (Fig. 3 D and E). However, none of the carbonatoms of the γ-linolenyl chain and palmitic core superimpose, andthe terminal C17–C18 carbons of MGLP extend further.

DiscussionOur collective studies show two remarkable and unprecedentedresults: the formation of a site in a completely unlikely domainof FAS, the TE domain, with a geometry for binding a long poly-unsaturated fatty acyl chain, and the inhibition of the intacthomodimeric enzyme and its TE domain by a PUFA, DGLA.These results are remarkable because the TE domain has beenknown to act solely on long fatty acyl chains in the key terminalstep of fatty acid biosynthesis. These results are also consistentwith our preliminary investigation indicating that DGLA is cap-able of inhibiting the biosynthesis of palmitate in 3T3–L1 prea-dipocyte cells. Moreover, the ethyl ester of DGLA inhibits thesynthesis of fatty acids in breast cancer cell lines by inhibitingFAS. It also affects the viability of these cells, as shown in Fig. 2B.The exact mechanism involved in the impact on the cells’ viabilityis not known at this time and will be worth investigating.

The reactivity of MGLFP to the TE domain is attributedalmost entirely to the phosphorylation by the phosphonate headgroup of the catalytic serine residue. Nevertheless, surprisingly,its γ-linolenyl chain is bound and makes excellent complementaryinteractions with amino acid residues in the groove-tunnel site(Site 2), which itself is formed and shaped mainly by the emer-gence of the helix flap (Fig. 1). With an invisible (disordered)helix, the groove-tunnel is completely absent in the ligand-freeTE structure (Fig. S1A).

DGLA imparts greater inhibition of the thioesterase activityof the TE domain than GLA and ALA. If the DGLA modeledin the groove-tunnel site portrays its mode of binding (Fig. S3),its specificity would be mainly governed by three factors com-bined: the electrostatic interactions of the carboxylate head groupwith the catalytic triad Ser 2308 and His 2481 residues, the place-ment of the three double bonds, together with the rest of thedihomo-γ-linolenyl chain to fully occupy the groove-tunnel, andthe appearance of the helix flap. We infer that all three factorsmay be more difficult to attain by ALA or GLA. For example,if the carboxylate of ALA or GLA is made to engage in interac-tions identical to those of the modeled DGLA, the set of threedouble bonds of ALA and GLA will not superimpose with that of

Fig. 2. Inhibition by DGLA. (A) Inhibition of palmitate biosynthesis in 3T3-L1preadipocyte cells. DMSO, which was used to dissolve the DGLA, served ascontrol. Incorporation of C13-acetate into [13C]-palmitate was measured bygas chromatography/mass spectrometry (GC/MS) (see SI Methods). Valuesare means� SEM, n ¼ 5 replicates. (B) Effects of ethyl ester of DGLA (0, 10,and 20 μM) on the cell viability of cancer cell lines and noncancerous cells(HMEC). Cell viability was determined as described in SI Methods. Valuesare the means� SD (bars) of at least four experiments made in triplicate(*P < 0.005).

15760 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1112334108 Zhang et al.

Dow

nloa

ded

by g

uest

on

June

7, 2

020

MGLP or the modeled DGLA, and their polyunsaturated acylchain will fall short of extending to the end of the tunnel.

The prospect of the TE domain binding a long-chain fattyacyl group and a PUFA is very intriguing. However, whereasour studies have revealed a molecular mechanism of binding ofγ-linolenyl acyl chain and by extension, DGLA, they and otherprevious structural analyses of the TE domain have yet to providea more complete atomic-level understanding of how the TE do-main optimally binds the palmitoyl acyl chain of the substrateor the palmitate product. Although the binding of the two palmi-tic cores of the Orlistats in the two TE structures differs signifi-cantly from that of the γ-linolenyl tail of MGLP (describedabove), owing almost solely to the formation of the groove-tunneland its extensive involvement in interacting with the γ-linolenyltail, Site 2 could be the site for binding the palmitoyl acyl chainor palmitate. In this scenario, binding of the acyl chain of sub-strates would require concomitant ordering and involvement ofthe helix flap in acyl chain recognition and then its subsequentrelease for the dissociation of the palmitate after substrate hydro-lysis. How Site 2 confers fatty acyl chain specificity in this case isunclear. The proposal for palmitoyl binding in Site 2 is com-pounded, on the one hand, by the data showing a sharp declinein activity for fatty acyl substrates with chain lengths longer than18 carbons (29) and, on the other hand, by the structure showing agroove-tunnel (Site 2) capable of accommodating 20-carbon-longfatty acyl chains (Figs. 1 A and C and S3A). Interestingly, thegroove-pocket Site 1 (Fig. 1C) may be more restrictive to C16 toC18 fatty acyl chains (26).

Our results make the case for DGLA as the PUFA of choicefor further or more thorough examination of its beneficial effectson a variety of chronic diseases, suppression of the transcriptional

activity of the Her-2 oncogene, and synergistic amplificationof anticancer drugs. Moreover, they pave the way for developinga new class of compounds to combat cancer, inflammation, orother health conditions and diseases that are linked to FAS, spe-cifically targeting the critical TE domain. Key to drug discoveryis the unique finding of the polyunsaturated fatty acyl chain-induced formation of a well-defined groove-tunnel site (Site 2)in the active site region of TE. Finally, our findings, combinedwith the knowledge of the presence of putative Site 1, provide atantalizing unique avenue of approach in developing FAS TEinhibitors with greater specificity and potency.

MethodsFull methods and associated references are described in SI Methods. Proteinexpression and purification of the human TE domain were performed follow-ing the procedures described previously (26). The thioesterase activity assaysof FAS and its TE domain were carried out using a similar procedure, as de-scribed previously (26, 29). The TE–MGLFP complex was prepared by mixingTE protein in a buffer solution, pH 8.0, with twofold molar excess of MGLFP,and incubating overnight in the cold room. Cocrystals were grown from8–10 mg∕mL TE–MGLP, 18–23% PEG 3350, 0.1 M Bis-Tris buffer, pH 5.5, atroom temperature by using the hanging drop method. A native datasetwas collected at the SBC-19ID beamline at the Advanced Photon Sourceand processed on site with DENZO and SCALEPACK (30). The structure wasdetermined by molecular replacement using the unliganded structure asthe search model (26), and refined using the CCP4 program suite (31). Themethods for the investigations of the the effects of DGLA on palmitate bio-synthesis in 3T3-L1 preadipocytes and on the viability of breast cancer celllines are described in SI Text.

ACKNOWLEDGMENTS. We thank the staff at the Advanced Photon SourceStructural Biology Center-Collaborative Access Team (APS SBC-CAT) 19IDbeamline for their assistance with the X-ray data collection. We also thankV. Nannegari and Dr. William C. Heird of the Department of Pediatrics, Baylor

Fig. 3. Comparisons of the active site region of the structures of complexes of TE with MGLP and Orlistat. Crystallographically unobserved or disorderedsegments are represented by dashed lines. (A) Ribbon representation of the active site with MGLP covalently attached to Ser 2308. MGLP and Ser 2308are represented by ball-and-stick model with carbons in green and the phosphonate group in orange. Even-numbered carbon atoms of the γ-linolenyl acylchain are labeled as in Fig 1C. (B) Similar representation as in (A) with covalently bound Orlistat (orange ball-stick model) [Protein Data Bank (PDB) code 2PX6-A]in one independent conformation (27). (C) Similar to (B) with the hydrolyzed Orlistat (blue model) in the other independent conformation (PDB code 2PX6-B)(27). In both panels B and C, the three groups of Orlistat are labeled as palmitic core with even-numbered carbons, peptidyl moiety (formylamino-4-methyl-pentanoic acid) linked to the C5 of the core and hexanoyl tail linked to the C2 carbon of the core. Moreover, the ordered α-helix (helix flap) in (A) is representedby a dashed line to signify a disordered segment. (D and E) Superpositions of structures depicted in A and B and A and C, respectively.

Zhang et al. PNAS ∣ September 20, 2011 ∣ vol. 108 ∣ no. 38 ∣ 15761

BIOCH

EMISTR

Y

Dow

nloa

ded

by g

uest

on

June

7, 2

020

College of Medicine, for the GC/MS analysis; and Dr. Trey Westbrook of theDepartment of Biochemistry, Baylor College of Medicine, for providing thevarious breast cancer cell lines. This research was supported by grants

from the National Institutes of Health (NIH) (R01GM088803) and The WelchFoundation Grant (Q-581) to F.A.Q., and from the NIH (R01GM0063115), theMedallion Foundation, and the Hefni Tech Training Foundation to S.J.W.

1. Wakil SJ (1989) Fatty acid synthase, a proficient multifunctional enzyme. Biochemistry28:4523–4530.

2. Kuhajida FP (2006) Fatty acid synthase and cancer: New application of an old pathway.Cancer Res 66:5977–5980.

3. Menendez JA, and Lupu R (2007) Fatty acid synthase and the lipogenic phenotype incancer pathogenesis. Nat Rev Cancer 7:763–777.

4. Kuhajda FP, Piantadosi S, Pasternack GR (1989) Haptoglobin-related protein (Hpr)epitopes in breast cancer as a predictor of recurrence of the disease. N Engl J Med321:636–641.

5. Kuhajda FP, Katumuluwa AI, Pasternack GR (1989) Expression of haptoglobin-relatedprotein and its potential role as a tumor antigen. Proc Natl Acad Sci USA 86:1188–1192.

6. Lupu R, Menendez JA (2006) Pharmacological inhibitors of fatty acid synthase (FASN)-catalyzed endogenous fatty acid biogenesis: A new family of anti-cancer agents? CurrPharm Biotechnol 7:483–494.

7. Omura S (1976) The antibiotic cerulenin, a novel tool for biochemistry as an inhibitorof fatty acid synthesis. Bacteriol Rev 40:681–697.

8. Funabashi H, et al. (1989) Binding site of cerulenin in fatty acid synthetase. J Biochem105:751–755.

9. Kuhajda FP, et al. (2000) Synthesis and antitumor activity of an inhibitor of fatty acidsynthase. Proc Natl Acad Sci USA 97:3450–3454.

10. Wang X, Tian W (2001) Green tea epigallocatechin gallate: A natural inhbitor offatty-acid synthase. Biochem Biophys Res Commun 288:1200–1206.

11. Liu B, Wang Y, Fillgrove KL, Anderson VE (2002) Triclosan inhibits enoyl-reductase oftype I fatty acid synthase in vitro and is cytotoxic to MCF-7 and SKBr-3 breast cancercells. Cancer Chemother Pharmacol 49:187–193.

12. Kridel SJ, Axelrod F, Rozenkrantz N, Smith JW (2004) Orlistat is a novel inhibitor offatty acid synthase with antitumor activity. Cancer Res 64:2070–2075.

13. Jiang WG, Bryce RP, Horrobin DF (1998) Essential fatty acids: Molecular and cellularbasis of their anti-cancer action and clinical implications. Crit Rev Oncol Hematol27:179–209.

14. Clarke SD, Romsos DR, Leveille GA (1977) Differential effects of dietary methyl estersof long-chain saturated and polyunsaturated fatty acids on rat liver and adipose tissuelipogenesis. J Nutr 107:1170–1181.

15. Clarke BA, Armstrong MK, Jump DB (1990) Nutritional control of rat liver fatty acidsynthase and S14 mRNA abundance. J Nutr 120:218–224.

16. Katsurada A, et al. (1990) Effects of nutrients and hormones on transcriptional andpost-transcriptional regulation of fatty acid synthase in rat liver. Eur J Biochem190:427–433.

17. Begin ME, Ells G, Das AK, Horrobin DF (1986) Differential killing of human carcinomacells supplemented with n-3 and n-6 polyunsaturated fatty acids. J Natl Cancer Inst77:1053–1062.

18. Menendez JA, et al. (2004) Overexpression and hyperactivity of breast cancer-asso-ciated fatty acid synthase (oncogenic antigen-519) is insensitive to normal arachidonicfatty acid-induced suppression in lipogenic tissues but it is selectively inhibited bytumoricidal alpha-linolenic and gamma-linolenic fatty acids: A novel mechanism bywhich dietary fat can alter mammary tumorigenesis. Int J Oncology 24:1369–1383.

19. Menendez JA, Vellon L, Colomer R, Lupu R (2005) Effect of gamma-linolenic acid onthe transcriptional activity of the Her-2/new (erbB-2) oncogene. J Natl Cancer Inst97:1611–1615.

20. Johnson MM, et al. (1997) Dietary supplementation with gamma-linolenic acidalters fatty acid content and eicosanoid production in healthy humans. J Nutr127:1435–1444.

21. Umeda-Sawada R, et al. (2006) Distribution and metabolism of dihomo-gamma-lino-lenic aid (DGLA, 20∶3n-6) by oral supplementation in rats. Biosci Biotechnol Biochem70:2121–2130.

22. Nwankwo JO (2001) Repression of cellular anaplerosis as the hypothesizedmechanismof gamma-linolenic acid-induced toxicity to tumor cells. Med Hypotheses 56:582–588.

23. Kim H, Choi S, Lee H-J, Lee J-H, Choi HC (2003) Suppression of fatty acid synthaseby dietary polyunsaturated fatty acids is mediated by fat itself, not by peroxidativemechanism. J Biochem Mol Biol 36:258–264.

24. Lio YC, Reynolds LJ, Balsinde J, Dennis EA (1996) Irreversible inhibition of Ca(2+)-independent phospholipase A2 by methyl arachidonyl fluorophosphonate. BiochimBiophys Acta 1302:55–60.

25. Ghomashchi F, et al. (1999) Trifluoromethyl ketones and methyl fluorophosphonatesas inhibitors of group IV and VI phospholipases A2: Structure-function studies withvesicle, micelle, and membrane assays. Biochim Biophys Acta 1420:45–56.

26. Chakravarty B, Gu Z, Chirala SS, Wakil SJ, Quiocho FA (2004) Human fatty acidsynthase: Structure and substrate selectivity of the thioesterase domain. Proc NatlAcad Sci USA 101:15567–15572.

27. Pemble CW, Johnson LC, Kridel SJ, Lowther WT (2007) Crystal structure of the thioes-terase domain of human fatty acid synthase inhibited by Orlistat. Nat Struct Mol Biol14:704–709.

28. Cheng F, Wang QH, Chen MZ, Quiocho FA, Ma J (2008) Molecular docking study of theinteractions between the thioesterase domain of human fatty acid synthase and itsligands. Proteins 70:1228–1234.

29. Pazirandeh M, Chirala S, Huang W, Wakil S (1989) Characterization of recombinantthioesterase and acyl carrier protein domains of chicken fatty acid synthase expressedin Escherichia coli. J Biol Chem 264:18195–18201.

30. Otwinowski Z, Minor W (1997) Processing of X-ray diffraction data collected in oscilla-tion mode. Methods Enzymol 276:307–326.

31. Collaborative Computational Project Number 4 (1994) The CCP4 suite: Programs forprotein crystallography. Acta Crystallogr D Biol Crystallogr 50:760–763.

15762 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1112334108 Zhang et al.

Dow

nloa

ded

by g

uest

on

June

7, 2

020