ATP-citrate lyase multimerization is required for coenzyme-A substrate binding and catalysis

ATP-citrate lyase (ACLY) is a major source of nucleocytosolic acetyl-CoA, a fundamental building block of carbon metabolism in eukaryotes. ACLY is aberrantly regulated in many cancers, cardiovascular disease, and metabolic disorders. However, the molecular mechanisms determining ACLY activity and function are unclear. To this end, we investigated the role of the uncharacterized ACLY C-terminal citrate synthase homology domain in the mechanism of acetyl-CoA formation. Using recombinant, purified ACLY and a suite of biochemical and biophysical approaches, including analytical ultracentrifugation, dynamic light scattering, and thermal stability assays, we demonstrated that the C terminus maintains ACLY tetramerization, a conserved and essential quaternary structure in vitro and likely also in vivo. Furthermore, we show that the C terminus, only in the context of the full-length enzyme, is necessary for full ACLY binding to CoA. Together, we demonstrate that ACLY forms a homotetramer through the C terminus to facilitate CoA binding and acetyl-CoA production. Our findings highlight a novel and unique role of the C-terminal citrate synthase homology domain in ACLY function and catalysis, adding to the understanding of the molecular basis for acetyl-CoA synthesis by ACLY. This newly discovered means of ACLY regulation has implications for the development of novel ACLY modulators to target acetyl-CoA–dependent cellular processes for potential therapeutic use.

eukaryotes, nucleocytosolic acetyl-CoA is primarily derived from glucose (3) by the enzyme ATP-citrate lyase (ACLY; EC 2.3.3.8). 2 The ubiquity of acetyl-CoA utilization makes ACLY a central node in the intersection of carbon metabolism and cell homeostasis and is a potent therapeutic target for the control of cell proliferation and metabolic dysregulation (4).
The ACLY catalytic mechanism proceeds through an elaborate series of events that convert MgATP, citrate, and CoA into ADP, acetyl-CoA, and oxaloacetate (8 -10). Briefly, the enzyme ATP-grasp domains 4 and 5 bind ATP with Mg 2ϩ and rearrange to autophosphorylate the catalytic histidine 760, which then complexes with citrate at the junction of two power helices from domains 2 and 5. The histidine transfers the phosphate to citrate to form a phospho-citryl intermediate. CoA then enters into the active site and carries out a nucleophilic attack to form a citryl-CoA intermediate. This reaction also liberates P i . Finally, a general base within the enzyme cleaves the citryl-CoA to produce acetyl-CoA and oxaloacetate.
Although the chemistry is clear, the regulation of these reactions remains poorly understood. Many studies have focused on the regulation of ACLY activity in vivo, such as during insulin stimulation or the Warburg effect (11,12). This includes regulation by phosphorylation (13)(14)(15)(16), acetylation, and ubiquitination (17). These studies have all focused on the regulation of the SCS superdomain. However, a significant portion of ACLY remains uncharacterized; the role of the C-terminal citrate synthase homology domain (CSHD) is largely unknown. Improved understanding of the CSHD could present a new dimension to the study of ACLY regulation. The CSHD is proposed by BLAST (18) to be structurally similar to the CS superfamily. CS catalyzes the conversion of oxaloacetate and acetyl-CoA to citrate and CoA, in the reverse of the ACLY mechanism, with the same reaction intermediates. Therefore, it was possible that ACLY inherited the chemistry of CS for noncanonical CoA binding through the CSHD (19). However, this parallel was limited by only partial conservation of the aminoacyl sequence. Only some of the CS residues that interacted with CoA and stabilized the CoA intermediates (20) were conserved in ACLY. Thus, ACLY may have evolved to take advantage of another feature of CS, independent of substrate binding.
CS is present across evolutionary space as a dimeric complex (7). The dimerization of CS is an essential feature, because the citrate synthase dimer interface forms the enzyme active site (20,21). CS self-association not only increased structural stability, but also allowed novel chemistry to occur at the association interface (22).
ACLY is also multimeric. Size-exclusion and sedimentation studies of plant and murine homologs suggest that ACLY forms a homotetramer (23,24), although this has not been confirmed for the human homolog. The conserved multimerization of ACLY parallels the conserved multimerization of CS. Thus, multimerization may play a structural and/or a catalytic role in ACLY function. In this study we hypothesized that ACLY multimerization through CSHD is an essential characteristic that is necessary for the full activity and stability of the enzyme.
To address our hypothesis, we prepared and analyzed the activity and multimerization properties of rationally designed, recombinant human ACLY truncations. We demonstrate that human ACLY forms a tetramer in vitro and in vivo. This tetramerization proceeds through the CSHD. Moreover, we show that the CSHD is required for catalysis. We show that the CSHD interacts with the SCS superdomain, facilitating the binding of the CoA substrate. We propose a new paradigm for the molecular mechanism of ACLY, in which the superdomains of ACLY cooperate through multimerization to produce acetyl-CoA.

Multimerization is mediated by the ACLY C-terminal citrate synthase homology domain
To study the functional role of human ACLY multimerization, we first designed and recombinantly prepared ACLY truncations to isolate the SCS and CS superdomains. We produced the protein as full-length (ACLY-FL, residues 1-1101), the SCS-homology superdomain alone (ACLY-Nterm, residues 1-820), and the CHSD alone (ACLY-Cterm, residues 820 -1101). Size-exclusion chromatography (SEC) confirmed that the loss of the CSHD collapsed the enzyme complex to a stable monomeric state, in agreement with previous studies of recombinant human ACLY (19). In contrast, when we produced the CSHD alone we observed the migration of the purified species at a significantly earlier retention volume than expected for the 34-kDa CSHD monomer (Fig. 1C). This result indicated that the CSHD was a multimer in solution. However, migration during size-exclusion depends on species size and shape (25) and therefore is an insufficient predictor of molecular mass and stoichiometry. Also, the SEC profiles of the full-length enzyme and the CSHD did reveal the presence of multiple species. The migration profiles were broad, possibly as the result of some

Role of ATP-citrate lyase multimerization in catalysis
instability of these constructs, especially in our buffers that purposefully excluded stabilizing co-factors or other additives. The ACLY-FL and ACLY-Nterm constructs may have undergone some degradation over the course of the experiment (Fig. 1B). For this reason, we proceeded with analytical ultracentrifugation and light-scattering experiments to better characterize the multimeric state of these constructs.
Sedimentation equilibrium (SE) analysis confirms that the CSHD drives ACLY tetramerization. A simple, ideal-species global analysis of the SE data for ACLY-FL calculated a molecular mass of 452 kDa, close to the expected tetramer molecular mass of 495 kDa. When linearized, these data aligned closest to the expected distribution for a homotetramer ( Fig. 2A). Parallel experiments of ACLY-Nterm provided a global molecular mass of 82 kDa, close to the expected molecular mass of 93 kDa. When linearized, these data aligned well to the expected distribution for a single monomeric state (Fig. 2B). Equally important was the observed molecular mass of the ACLY CSHD, which yielded a single species molecular mass of 138 kDa. Just as with the full-length enzyme, these data aligned closest to the expected distribution for a homotetramer of 138 kDa ( Fig. 2C; summarized in Table S1). These trends held across a range of concentrations and centrifugation speeds (Fig. S1). We took these data to support our observation that the CSHD is sufficient for ACLY multimerization.
However, there were inconsistencies in the linearization of the ACLY-FL and ACLY-Cterm SE data that, just as with the SEC data, suggested the presence of multiple states. Although the SE data described above are indicative of the multimeric state for each of these constructs, SE can only provide a global average molecular mass. This approach could mask the presence of an equilibrium among different subspecies. Therefore, we looked into the potential existence of different protein states within each sample.
Dynamic light scattering is a fast and nondestructive analysis of protein tumbling in solution (26) that we employed to obtain the ACLY average hydrodynamic radius, as well as to detect any sizeable populations of soluble aggregates or breakdown products. We did not observe significant amounts of any such artifacts that may not have separated during SEC (Fig. S2).
We next employed sedimentation velocity (SV) for detection of equilibria among multiple species under cell-free solution conditions. Our SV experiments confirmed that ACLY-FL was predominantly a single species in solution with an average s 20,w of 14.6 S, whereas ACLY-Nterm predominantly sedimented at 4.96 S, and the CSHD predominantly sedimented at 6.98 S ( Fig.  3B; summarized in Table S2). We noticed some broadness and kurtosis of the SV distributions and so went back to our SE data to test different equilibrium models that might explain these trends.
Upon further analysis of our SE data, the ACLY-FL distributions also fit to a dimer-tetramer equilibrium model, with a predicted K d of 4.71 M, although with poorer statistics. The ACLY-Cterm data could also be fit to a dimer-tetramer equilibrium with a predicted K d of 97.7 M, although even less well (Table S1). ACLY-Nterm could not be meaningfully fit to any equilibrium model.
Together, these data demonstrate that ACLY-FL exists as a multimer in solution, whereas ACLY Nterm is monomeric. The ACLY CSHD appears to drive multimerization to a state most consistent with a tetramer. Therefore, we present the CSHD as the means of ACLY multimerization.
The biological relevance of our cell-free studies is supported by the observation of ACLY tetramerization in protein derived directly from plant and murine tissue. However, we endeavored to validate our findings in human cells. To that end, we cotransfected HEK-293T cells with ACLY-FL with an N-terminal FLAG and/or Myc epitope tag. We found that our epitopetagged ACLY-FL constructs could immunoprecipitate one another (Fig. 4). Consistent with our cell-free data, we conclude that ACLY forms a stable complex with itself in the context of the human cellular milieu.

Role of ATP-citrate lyase multimerization in catalysis Multimerization increases the thermal stability of ACLY
The formation of homomultimers is a widespread, evolutionarily conserved feature of eukaryotic proteins with several potential advantages, such as increased stability, faster evolution of novel functionality, and the development of new regulatory sites (22). These same advantages may hold for the ACLY multimer. Therefore, we were curious about the role of the CSHD in ACLY intrinsic stability, substrate binding, and catalytic efficiency. We used differential scanning calorimetry (DSC) to assess the thermal stability of ACLY-FL (Fig. 5). We observed that ACLY-FL harbored two melting transitions, with average melting transitions (T M ) of 53.8°C (referred to as peak 1) and 69.3°C (referred to as peak 2).
Of note, the full-length protein could not be refolded after denaturation (Fig. S3). This result suggested that the intramolecular interactions maintaining the various domains of ACLY are stable ones, i.e. we could detect no equilibrium between different conformations of the enzyme. In addition, the unequal ⌬H vH /⌬H CAL ratio of both peaks 1 and 2 (Table S3) also suggested underlying complexity (27,28). However, the ACLY-FL peak 1 could only be fit by a single-peak in a non-two-state model. We speculate that the hidden, thermodynamically stable states hinted at by the unequal ⌬H vH /⌬H CAL ratio may be the separate unfolding of the ATP-grasp and citrate-binding regions, separated as they are by the bridging, unstructured domain 5. Further DSC with a slower ramp rate did not help to reveal the likely underlying transitions.
In contrast, ACLY-Nterm and ACLY-Cterm had only single transitions. These constructs had average T M values of 46.2 and 73.0°C, respectively ( Fig. 5; summarized in Table S3). As expected, multimerization clearly increased the overall stability of the protein.
We were initially surprised at the consistent bimodal melt profile of the FL enzyme. The similar T M values of the CHSD and the 2nd peak of the full-length enzyme suggested that ACLY-FL underwent two distinct unfolding events, one for each superdomain. The CSHD data best fit to a simultaneous unfolding/dissociation model (with the lowest ⌾ 2 /DF value), as did the ACLY-FL peak 2. Therefore, we propose that the ACLY-FL peak 1 represents the unfolding of the N terminus. The difference between the two peaks of ACLY-FL and the individual T M of the truncations may be due to the multimeric context of the SCS superdomain in the full length holoenzyme.
The direction of stabilization relative to ACLY-FL provided a clue to the relationship between the two superdomains in the context of the enzyme complex. These data suggested that the  ACLY N and C termini interacted with one another. The SCS superdomain is strongly stabilized by the self-association; the SCS superdomain gains ϳ8 degrees of stability when brought into the ACLY tetramer. We would not expect such a change were the SCS superdomain a self-contained molecular unit, isolated from the CSHD. In contrast to the stabilization of the SCS superdomain, ACLY appears to sacrifice some of the stability of the CSHD for the sake of the holoenzyme. That we see this decrease suggests that CSHD may provide more than an inert tether to the other ACLY subunits. However, the communication between the N and C termini is not so strong as to produce a single melting event for the ACLY complex. Therefore, the two superdomains may have only a few points of contact.
This conclusion was supported by SEC of ACLY-Nterm mixed with excess ACLY-Cterm. This assay demonstrated that the two fragments elute closer to each other than when ACLY-Nterm and ACLY-Cterm are run separately (Fig. 6). We conclude that in the absence of an intramolecular tether, the two domains may form a weak interaction. We next asked what the mechanism of this interaction may be and what may be the functional consequence of cross-talk between the N and C termini.

ACLY enzymatic activity requires tethered N-and C-terminal cooperation
Current models of ACLY function are guided by the ACLY homology to acyl-CoA synthetases, with little known about the contribution of the CSHD. The SCS-homology model for ACLY function suggests that the ACLY-Nterm construct should be capable of binding all ACLY co-factors and catalyzing the formation of acetyl-CoA and oxaloacetate. Contrary to this model, we observed that the loss of the CSHD completely nullified acetyl-CoA production (Fig. 7). This observation offered the possibility that the CSHD contacted the N terminus to directly interact with co-factors as a part of the molecular mechanism of turnover. To test this hypothesis, we assayed whether ACLY-Cterm added in trans can restore ACLY-Nterm activity. We observed that incubation of ACLY-Nterm with up to 10-fold molar excess of ACLY-Cterm did not rescue catalysis. Consistent with the DSC data, these titration experiments suggest that the CSHD allosterically regulates the N-terminal domain in cis.

The ACLY N terminus optimally binds CoA in the context of the full enzyme
Sequence alignment of the CSHD to CS suggests that the domain could contain a CoA-binding pocket and so effect catalysis through direct interaction with CoA. The crystal structures of eubacterial CS with citrate and CoA define an essential histidine and aspartate in the CS active site to stabilize the citryl-CoA enolate intermediate (20), as well as several others that contact the adenine and pantetheine moieties (29). Some, but not all, of these same residues are present in the CSHD. We used differential scanning fluorimetry (DSF) to assay the potential interaction of our ACLY N-and C-terminal constructs with CoA, citrate, and MgATP in different combinations at 20-fold molar excess (Fig. 8). For ACLY-FL we observed statistically significant thermal stabilization through Mg 2ϩ -dependent ATP binding (2.9°C), as well as stabilization through the binding of citrate (2.0°C) and the binding of CoA (3.3°C). Moreover, the stabilization was additive when the cofactors were combined (summarized in Table S4). For ACLY-Nterm, we observed significant stabilization by citrate (3.5°C), and to a lesser extent, ATP (2.5°C). Although the calculated means suggested a small, potential stabilization of ACLY-Nterm by CoA, this shift was not consistent. When the N terminus with MgCl 2 /ATP/citrate/CoA was compared against the N terminus with MgCl 2 /ATP/citrate alone, the effect disappeared (p ϭ 0.276). It therefore seemed possible that the synthetase-homology model for CoA binding was inaccurate, with the CSHD instead coordinating with CoA. Instead, we were surprised to find that ACLY-Cterm was not stabilized by any of these cofactors.
To more sensitively probe the effect of CoA binding to ACLY, we carried out a DSC analysis of ACLY-FL, ACLY-Nterm, and ACLY-Cterm in the presence and absence of CoA, again at 20-fold molar excess. We observed a clear shift for ACLY-FL upon incubation with CoA (1.0°C), whereas the addition of CoA did not have a statistically significant effect on the thermal stabilities of ACLY-Nterm or ACLY-Cterm alone (Fig. 9). Interestingly, the addition of CoA only affected the stabilization of ACLY-FL peak 1. CoA also sharpened the peak 1 melt transition, producing a better fit of the data. Together with our DSF analysis, we conclude that this shift is indicative of the ACLY N terminus binding to CoA only when in the context of the intact protein.

Discussion
ACLY is a major node in the transduction of insulin signaling to changes in gene expression. It is a highly prized target for metabolic diseases, including atherosclerosis, hypercholesterolemia, and also oncogenesis (1,4). In the latter case, the dependence of tumors on glucose for energy makes ACLY inhibition especially appealing. ACLY knockdown can drastically decrease cancer cell proliferation without significantly affecting nearby healthy tissue (30,31). ACLY may also be a target for anti-aging studies, because decreased nucleocytosolic acetyl-

Role of ATP-citrate lyase multimerization in catalysis
CoA levels and ACLY inhibition may delay the onset of agerelated decline (32,33).
Although ACLY has been the focus of myriad clinical studies (4, 34, and 35), many of these efforts have stalled. There are few promising small-molecule ACLY inhibitors. Therefore, there exists a clear need for a better understanding of the catalytic and allosteric biochemistry of the enzyme. Although a great deal of effort has gone into understanding the signaling cascades and post-translational modifications that tune ACLY function in the cell (36), limited knowledge about the molecular basis for ACLY function has held back the development of ACLY therapeutic treatments.
In this study, we highlight the importance of the ACLY C-terminal domain. First, we demonstrate that the evolution-arily conserved CSHD is the driver of ACLY multimerization. Next, we show that ACLY cannot function without the CSHD. Last, we define the CSHD as necessary for CoA binding through coordination with the ACLY SCS superdomain. This is a surprising finding for two primary reasons: the citrate synthase homology domain of citrate synthetase is a dimer and not a tetramer, and although the ACLY N-terminal acyl-synthetase homology domain contains a consensus sequence for binding CoA, it is unable to fully bind CoA in the absence of the C-terminal CSHD.
Each of these findings presents new avenues for ACLY regulation. We connect the evolutionary relationship between ACLY and the CS superfamily to a molecular framework for ACLY function. Our light-scattering, analytical ultracentrifugation, and co-IP experiments show that human ACLY is a stable and consistent homotetramer. To date, we have been unable to isolate stable, monomeric ACLY through point mutagenesis of the CSHD. It stands to reason that structural destabilization through the loss of tetramerization would have severe deleterious effects on catalysis. Thus, multimerization may be as important for ACLY as it is for CS. We posit that a genetic rearrangement event, early in the evolution of citrate utilization for acetyl-CoA production, created the link between SCS and CS as a ready means of increasing enzyme stability and

Role of ATP-citrate lyase multimerization in catalysis
acetyl-CoA output that has endured across millions of years of evolution (37). The potential for a dimer-tetramer equilibrium of ACLY at a biologically relevant concentration also speaks to this relationship. ACLY tetramerization may be the product of a dimer of CSHD dimers coming together. This connection between the two enzymes can inform on the unknown structure of the CSHD.
Citrate synthase dissociation is also a powerful regulatory mechanism that responds to protein levels, the local pH, and the local ionic environment to toggle CS between a dimeric active state and a monomeric inactive state, with a K d of 0.2 M (38). The same could also a means of inactivating ACLY. During periods of high glycolytic flux, cellular ACLY levels also increase severalfold (17). Therefore, ACLY may also be controlled by a similar concentration-dependent switch, offering new frameworks for regulating ACLY levels in vivo.
We conclude from our activity assays that the CSHD is necessary for ACLY function. Moreover, our substrate-binding assays contradict our initial hypothesis that the CSHD directly borrows the substrate-binding residues of CS. Instead, ACLY has adapted the CSHD for a more nuanced role as a facilitator for the activity of the N-terminal SCS superdomain. We observed that the two major regions of the enzyme communi-cate with one another. Furthermore, we conclude that the CSHD participates in the chemistry at the active site by facilitating CoA binding.
Ultimately, the ACLY CSHD may be a potent and specific target for rational drug design. Together, these studies highlight a novel and unique role of the C-terminal CSHD in ACLY function and catalysis and a novel avenue for targeting the C-terminal domain of ACLY for regulating acetyl-CoA dependent cellular processes for potential therapeutic use.

Expression and purification of recombinant ACLY
The Homo sapiens ACLY isoform 1 (accession no. NM_ 001096.2) was used in this study. To generate the plasmid DNA constructs for recombinant protein expression, this sequence was PCR-amplified with DNA primers adding a C-terminal 6ϫ histidine tag sequence and EcoRI/SalI cut sites for ligation. The PCR product was ligated into a custom-engineered pET Duet-1 vector for Escherichia coli expression or into the pFastBac HT vector (Novagen) for expression in Spodoptera frugiperda (Sf9). ACLY-FL and ACLY-Cterm expressed equally in both systems, whereas ACLY-Nterm could only be produced in the Sf9 sys-

Analytical ultracentrifugation
Sedimentation equilibrium of the recombinant protein was carried out in size-exclusion buffer at concentrations of 0.2, 0.4, 0.8 mg ml Ϫ1 at 4°C in a Beckman Optima XL-I analytical ultracentrifuge (Beckman Coulter, Palo Alto, CA). Absorbance optics at 280 nm were taken using quartz windows in six-chamber charcoal centerpiece in an An-60 Ti four-hole rotor (Beckman Coulter) with 6-h scans with a radial step size of 0.001 (four scans/speed, with three independent experiments for ACLY-FL and ACLY-Cterm, two experiments for ACLY-Nterm). The speeds used were 6000, 9000, 12,000, and 16,000 rpm as recommended based the expected monomer and tetramer molecular masses of the ACLY constructs (39,40). The molecular masses were calculated by global ideal species analysis by SEDPHAT (41). The error associated with the global analysis was determined by Monte Carlo simulation with 1000 iterations at a confidence level of 0.95. We only analyzed those data between the sample meniscus (variable) and Absorbance 280 nm Յ 1.5 AU, to avoid violating Beer-Lambert linearity between concentration and light attenuation. Equilibrium fitting was done with strict mass conservation and rotor stretch correction, as well as sample bottom, baseline, and radial-invariant noise fitting. Goodness of fit was determined by global reduced 2 . The sample parameters of protein partial specific volume, buffer density, and buffer viscosity calculated by SEDNTERP (42). Protein molar mass and extinction coefficients were calculated by ExPASy Protparam (43).
Sedimentation velocity experiments were carried out in sizeexclusion buffer at a concentration of 0.3 mg ml Ϫ1 for all proteins. The experiments were run at 4°C and 42,000 rpm with a radial step-size of 0.001 and a 5-min delay between scans (200 scans total, two independent experiments). Only the minimum number of scans were analyzed (41), as recommended for the maximum molecular mass by DCDTϩ version 2.4.3 by John Philo (44,45). The data were analyzed by SEDFIT (46) with a continuous c(s 20,w ) distribution model with mass conservation and a confidence level (F ratio) of 0.95, where the frictional ratio (f/f 0 ) was allowed to float. The c(s 20,w ) distribution were plotted using Prism version 7.4 (GraphPad, La Jolla, CA).

Cell culture
HEK-293T cells were maintained in Dulbecco's modified Eagle's medium ϩ 10% fetal bovine serum. The cells were plated at 1 million cells/plate in 10-cm dishes for 24 h prior to transfection. 10 g of pEF6/ACLY-FL/Myc and/or pEF6/ ACLY-FL/FLAG was combined with 24 l of polyethyleneimine (Sigma-Aldrich, catalog no. 408727) in 800 l of Opti-MEM and incubated with cells for 6 h. Following incubation, the cells were washed with PBS, placed in maintenance medium for 48 h, and then harvested.

Co-immunoprecipitation
Harvested cells were lysed in 400 l of NP-40 buffer containing 1 mM DTT and 1ϫ Proteinase Inhibitor. 1 mg of protein was incubated for 1.5 h with anti-MYC beads (Cell Signaling, catalog no. 9B11). Following incubation, the beads were washed three times and then boiled in 2ϫ SDS buffer. Boiled IP samples and 30 g of input for each sample were run on SDS-PAGE gels, transferred to nitrocellulose membranes, and incubated with primary anti-␤-actin (Cell Signaling, catalog no. 8H10D10), anti-FLAG (Sigma-Aldrich, catalog no. F1804), anti-Myc (Cell Signaling Technology, Danvers, MA, catalog no. 9B11), or anti-ACLY (ProteinTech, catalog no. 15421-1-1AP) antibodies overnight. The blots were washed in PBS-T (3.2 mM Na 2 HPO 4 , 0.5 mM KH 2 PO 4 , 1.3 mM KCl, 135 mM NaCl, 0.05% Tween 20, pH 7.4) two times for 10 min each and then incubated for 30 min with secondary antibodies against mouse or rabbit. The blots were again washed two times for 10 min with PBS-T and then imaged on the LI-COR Odyssey 675 fluorescent imager (LI-COR Biotechnology, Lincoln, NE) as previously described (47).

Dynamic light scattering
Light scattering experiments were performed on a Zetasizer V (90°scattering apparatus; Malvern Panalytical, Malvern, UK) in a 2-l SUPRASIL quartz cuvette (Hellma Analytics, Müllheim, Germany) at 3.3°C. The protein was tested at 8 M in size-exclusion buffer in two independent experiments, six replicates (33 scans/replicate). The buffer viscosity and refractive index were calculated with Zetasizer software, version 7.12, and the protein refractive index was set to 1.45. The protein hydrodynamic distribution was calculated by the Malvern Zetasizer DTS software, and correlation functions were calculated by SEDFIT. Prism was used to plot the data.

Kinetic assays
The activity of ACLY was measured according to the method of Linn and Srere (48). Briefly, enzyme activity was determined for protein at 20 nM in 1 ml of 200 mM Tris, pH 8.8, 20 mM MgCl 2 , 20 mM potassium citrate tribasic, pH 7.5, 20 mM ATP, and 10 mM DTT with 5 units of porcine malate dehydrogenase (Millipore) and 200 M NADH (Sigma-Aldrich). The reaction was initiated with 400 M CoA and allowed to proceed for 20 min at room temperature during continuous Absorbance 340 nm

Role of ATP-citrate lyase multimerization in catalysis
detection in a BioMate 3S spectrophotometer (Thermo Fischer Scientific). The activity was determined across three independent experiments, n ϭ 6 each. Excel (Microsoft, Redmond, WA) was used to select the linear range of the data (where second derivative ϭ 0) and to convert ⌬Absorbance into enzyme velocity using the extinction coefficient of NADH. The data were analyzed for statistical significance and plotted with the median and 95% confidence interval using Prism. The significance of the difference in V max was determined by two-way analysis of variance with Sidak's multiple comparisons test.

Differential scanning fluorimetry
DSF was carried out with 0.1 mg ml Ϫ1 protein that had been dialyzed overnight into DSF buffer (25 mM HEPES, pH 7.5, 200 mM NaCl, 1 mM DTT). SYPRO Orange protein dye (Thermo Fischer Scientific) was added to the protein at a final concentration of 5ϫ, along with additional buffer, 10 mM MgCl 2 , 10 mM ATP disodium salt, 10 mM potassium citrate tribasic, pH 7.5, and/or 1 mM CoA (all reagents from Sigma-Aldrich) to bring the total volume to 20 l. The concentrations of these additives were chosen to be at least 20ϫ above the published K d (or K m , where the K d was unavailable) (10,19) and thus considered to be saturating. The reactions were set up in a 384-well optical plate (Applied Biosystems, Foster City, CA). The plate was read in a 7900HT Real-Time PCR (Applied Biosystems) from 20 to 95°C at a 2% ramp rate with a 10-min 20°C preequilibration step. The raw fluorescence was quantified using the SDS 2.4 software package. The data were subsequently normalized and taken to the first derivative using Excel, with the inflection point of the melt curve (the maximum dy/dx value) used as the T M . To account for inflections at the beginning of the PCR stage, data outliers were excluded using ROUT analysis, and the remaining values were tested for significance with a two-way analysis of variance with Dunnett's multiple comparisons test in Prism. Individual tests of significance were determined by t tests with Sidak's multiple hypothesis correction. The data were plotted with the median and 95% confidence interval.

Differential scanning calorimetry
The proteins were dialyzed overnight in DSC buffer (25 mM HEPES, pH 8.0, 200 mM NaCl) to remove any reducing agent, adjusted with the dialysis buffer to 25 M, and spin-filtered (EMD Millipore, Burlington, MA) at 18,000 ϫ g to remove particulates. Prior to loading into the cell, the samples were degassed with stirring in an evacuation chamber for 5 min at 10°C. Each sample was analyzed by MicroCal VP-Capillary DSC (Malvern Panalytical), with buffer in the reference cell. When DSC was used to assess CoA-mediated stabilization, either CoA or water was added to the protein after dialysis to account for the effect of dilution. DSC buffer with either CoA or water was used to provide the buffer-only scan. The samples were scanned from 10.0 to 90.0°C with a 10-min 10.0°C preequilibration and a 30°C h Ϫ1 or 60°C h Ϫ1 ramp rate. Protein refolding (equilibrium) analysis was carried out by cooling the sample from 90 to 10°C at 60°C/h, with a 10-min re-equilibration at 10°C and a second 60°C h Ϫ1 run. The data were analyzed with the ORIGIN 7.0 VP-DSC package. The data were baselined through a linear connect model followed by manual adjustment by cursor. The T M , ⌬H CAL , and ⌬H vH were calculated through a non-two-state Levenberg-Marquardt fitting model and evaluated by reduced 2 analysis. The average T M , ⌬H CAL , and ⌬H vH of the data were calculated using Prism. We report the average weighted estimate ⌬H WA (49) as a measure of data quality, which corrects for experimental baseline fluctuations and the differences between experiments. The significance of the difference in T M was determined by an unpaired t test. All data sets are available at http://dx.doi.org/10.17632/ m6pz3rh5n4.1 under a CC 4.0 license.