Role of Disulfide Bonds in Acrp30/Adiponectin Structure and Signaling Specificity

Acrp30/adiponectin is an adipocyte-derived serum protein with important roles in regulation of lipid and glucose metabolism, but which of its isoforms are biologically active remains controversial. We addressed this issue by first characterizing the structure of each individual Acrp30 oligomer and the determinants responsible for multimer formation. Freeze etch electron microscopy showed the trimer to exhibit a ball-and- stick-like structure containing a large globular sphere, an extended collagen stalk, and a smaller sphere on the opposite end of the stalk. The hexamer consists of two adjacent trimeric globular domains and a single stalk composed of collagen domains from two trimers. Although not necessary for trimer formation or stability, two of the three monomers in an Acrp30 trimer are covalently linked by a disulfide bond between cysteine residues at position 22. In contrast, assembly of hexameric and higher molecular weight (HMW) forms of Acrp30 depends upon formation of Cys22-mediated disulfide bonds because their reduction with dithiothreitol or substitution of Cys22 with alanine led exclusively to trimers. HMW and hexamer isoforms of Acrp30 activated NF-κB in C2C12 cells, but trimers, either natural, formed by reduction of Acrp30 hexamer, or formed by the C22A mutant, did not. In contrast, incubation of isolated rat extensor digitorum longus with naturally formed Acrp30 trimers or trimeric C22A Acrp30 led to increased phosphorylation of AMP-activated protein kinase-α at Thr172 and its activation. Hexameric and HMW Acrp30 could not activate AMP-activated protein kinase. Thus, trimeric and HMW/hexameric Acrp30 activate different signal transduction pathways, and Acrp30 represents a novel example of the control of ligand signaling via changes in its oligomerization state.

decreased in obese or diabetic humans and animals (2,3). In Pima Indians, occurrence of diabetes later in life is accompanied by decreased Acrp30 levels before onset of diabetes (4). Acrp30 exerts multiple metabolic actions at multiple tissue sites. The isolated globular domain of Acrp30 (gAcrp30) simulates fatty acid oxidation in skeletal muscle, whereas fulllength Acrp30 synergizes with insulin to inhibit hepatic glucose production (5)(6)(7). In mice, disruption of Acrp30 locus leading to its ablation resulted in impaired fatty acid clearance, increased tumor necrosis factor ␣ levels, and aggravated insulin resistance in animals fed a high fat diet (8,9).
How Acrp30 acts as a hormone to regulate these physiological processes remains unknown. We addressed this issue by analyzing the structure of Acrp30 secreted from cells and in serum. Acrp30 purified from transfected human embryonic kidney (HEK) 1 293T cells or Escherichia coli exists as trimers and hexamers (10). Transfected HEK cells also secrete an even higher molecular weight (HMW) isoform of Acrp30 (10). All three isoforms of Acrp30 are present in mouse serum and the conditioned medium of differentiated 3T3-L1 adipocytes, albeit with different relative abundances (10). Purified isoforms are stable in PBS and do not interconvert (10). The HMW and hexameric Acrp30 can activate transcription factor NF-B in undifferentiated or differentiated C2C12 cells, but trimeric Acrp30 or gAcrp30 cannot. Rather, gAcrp30, but not full-length Acrp30 hexamer, enhances muscle fatty acid oxidation by inactivating acetyl-CoA carboxylase following stimulation of AMP-activated protein kinase (AMPK) (11,12).
Because different isoforms of Acrp30 have different activities, it is important to understand how they are formed. Acrp30 contains three easily recognizable domains: an N-terminal signal sequence, a collagen domain, and a C-terminal globular domain. Because the crystal structure of the Acrp30 globular domain shows a trimeric fold similar to tumor necrosis factor ␣, the collagen domain of Acrp30 may mediate multimerization into hexamer and HMW isoforms. Alternatively, disulfide bonds may hold Acrp30 oligomers together because they are required for multimerization of Acrp30 homologs C1q and emilin (13,14). Acrp30 contains two cysteine residues that are conserved among all species with available sequence information. One is located in the globular domain at position 138 of mouse Acrp30 (numbering begins after the signal peptide). The other is at position 22 in a conserved 17-amino acid segment with the sequence APALVPPPKGTCAGWMA.
To understand mechanisms used by Acrp30 to multimerize, we employed deep etch cryo-electron microscopy to determine the structures of each oligomer. We then tested whether disulfide bonds are required for Acrp30 oligomerization and which cysteine residues are involved in their formation. Acrp30 trimer is best described as a "ball on a stick" model, whereas the hexamer consists of two adjacent parallel trimers that resemble a "Y." The HMW structure is currently not interpretable. Reduction of HMW or hexameric Acrp30 with dithiothreitol (DTT) collapsed them into trimers. In parallel, site-directed mutagenesis of each of the two cysteines in Acrp30 showed cysteine at position 22 to be crucial for formation of the HMW and hexamer species. In contrast, cysteine at position 138 in the globular domain had no role in directing the formation of Acrp30 oligomers. Consistent with our prior results that naturally produced Acrp30 trimers cannot activate NF-B, we show that neither trimeric C22A Acrp30 nor trimers formed by the reduction of Acrp30 hexamers can activate NF-B. Previously we showed that trimeric gAcrp30 could activate AMPK, and here we show that full-length C22A trimers as well as C138A trimers activated AMPK in rat extensor digitorum longus (EDL) muscle, as measured by phosphorylation at Thr 172 of AMPK␣. In contrast, hexameric and HMW C138A Acrp30 could not. Thus, trimeric Acrp30 activates AMPK and hexameric/HMW Acrp30 activates NF-B, and we propose that the oligomerization state of Acrp30 determines the signaling pathway it can activate.

EXPERIMENTAL PROCEDURES
Cysteine Mutagenesis-Site-directed mutagenesis of Acrp30 in pcDNA3.1 vector was performed using a QuikChange Mutagenesis kit (Stratagene) according to the manufacturer's instructions. Primers used to construct C22A mutant have the sequences 5Ј-CCC AAG GGA ACT GCT GCA GGT TGG ATG GC-3Ј and its reverse complement. Primers used to construct C138A mutant are 5Ј-C ACT GGC AAG TTC TAC GCC AAC ATT CCG GGA CTC-3Ј and its reverse complement. The mutations were confirmed by DNA sequencing.
Production and Gel Filtration Analysis of Wild Type and Mutant Acrp30 -As previously described, Acrp30 (residues 1-230, without the 17-residue signal sequence) was purified from E. coli strain BL21 (DE3) as a His 6 -tagged precursor that was subsequently treated with PreScission protease (Amersham Biosciences) to remove the tag (10). Wild type (GenBank TM accession number U37222), C22A, and C138A Acrp30 (nucleotides 46 -789 of U37222) were purified from conditioned serumfree medium of transfected HEK 293T cells as described previously (5,10). Purified protein samples were loaded into an Akta fast protein liquid chromatography system and fractionated either through a 16/60 or a 10/30 Superdex 200 column (Amersham Biosciences) and eluted with PBS.
Electron Microscopic Imaging of Acrp30 Oligomers-HEK 293T Acrp30 oligomers purified by gel filtration chromatography were suspended with finely ground mica flakes and quickly frozen using liquid helium in a cryopress as described previously (15,16). The samples were then fractured using a cryomicrotome and partially freeze-dried for deep etching before rotary replication with a 2-nm film of platinum (16,17). Transmission electron micrographs were taken in stereo with a 10°tilt at 70,000ϫ magnification (16,17).
Equilibrium Sedimentation-The samples were centrifuged at 12,000 rpm for 18 h at 10°C in a Beckman XL-A analytical ultracentrifuge before absorbance was recorded at 230 nm. The data were fit globally using MacNonlin PPC (18) to the following equation that describes sedimentation of a homogeneous species: where A is the absorbance at radius x, AЈ is the absorbance at reference radius x o , H ϭ (1 Ϫ ) 2 /2RT (where R is the gas constant, T is the temperature in degrees Kelvin, is the partial specific volume ϭ 0.71896131 ml/g, is the density of solvent ϭ 1.0061 g/ml, is the angular velocity in radians/s) M is the apparent molecular weight, and B is the solvent absorbance (blank).
Luciferase Reporter Assay-C2C12 cells (ATCC) were co-transfected with plasmids encoding firefly luciferase under the control of E-selectin promoter (19) and ␤-galactosidase driven by the cytomegalovirus promoter using FuGene 6 (Roche Applied Science) according to the manufacturer's instructions. Following overnight or 8 h of incubation with various forms of Acrp30 or lipopolysaccharide, the cells were washed, and the luciferase and ␤-galactosidase activities were tested using kits from Promega and Clontech.
Muscle Incubations-The rats were anesthetized with intraperitoneal pentobarbital sodium injection (60 mg/kg), and EDL muscles strips were prepared and tied to stainless steel clips as described previously (20). The muscles were preincubated for 20 min at 37°C in oxygenated (95% O 2 , 5% CO 2 ) Krebs-Henseleit solution containing 6 mM glucose and then incubated for 30 min in the same medium in the absence or presence of 2 g/ml C22A Acrp30 or 2 g/ml HMW, hexamer, or trimer isoforms of C138A Acrp30. At the end of this incubation the muscles were frozen in liquid N 2 until Western blot analysis.
Western Blot Analysis-Twenty-five g of crude muscle homogenate were fractionated by SDS-PAGE and transferred to polyvinylidene difluoride membrane (Bio-Rad). Following transfer, the membranes were blocked with 5% bovine serum albumin in 25 mM Tris, pH 7.5, 135 mM NaCl, 2.5 mM KCl, and 0.05% Tween 20 for 1 h at room temperature. The membranes were incubated with Phospho-AMPK␣ (Thr 172 ) or total AMPK␣ antibodies from Cell Signaling Technology (Beverly, MA) and then with secondary antibody conjugated to horseradish peroxidase from Amersham Biosciences. The bands were visualized by enhanced chemiluminescence and quantified by laser scanning densitometry. The immunoblots were performed under conditions in which autographic detection was in the linear response range.

RESULTS
We performed freeze etch electron microscopy to visualize the architecture of HEK 293T cell-produced Acrp30 trimer, hexamer, and HMW isoforms following their separation by gel filtration chromatography (Fig. 1). The trimer exhibits a balland-stick structure with the globular domain forming the ball and the collagen triple helix forming the rigid stick (Fig. 1A). In the majority of the pictures a ball-like structure smaller than the globular head domain was also observed. This probably represents the N-terminal region of Acrp30 upstream of the collagen domain. The length of the collagen domain is ϳ15 nm. Because the collagen triple helix has 10 Gly-X-Y repeats per turn and a pitch of 86 Å, the expected length of the Acrp30 collagen domain with 21 repeats is ϳ18 nm, similar to what was observed. The architecture of hexameric Acrp30 shows two trimers lying adjacent to each other in parallel head-to-head fashion and is reminiscent of the letter Y (Fig. 1B). Possibly because of strong interactions between the globular domain with the mica flakes, we only obtained what we believe to be tail end views of the HMW isoform (Fig. 1C). More apparent in the stereo views in supplemental Fig. S1, these HMW molecules resemble cones standing upside down, with the globular heads adhering to the mica and the tails sticking up into the air. However, we cannot infer the precise structure of HMW Acrp30 from these images.
To investigate the possibility that disulfide bonds are responsible for formation or maintenance of hexamer or HMW Acrp30, we analyzed elution profiles of isolated hexamers and trimers purified from E. coli in gel filtration columns following incubation with 100 mM DTT. In the absence of DTT, the apparent molecular masses of E. coli-produced Acrp30 hexamer ( Fig (10), it is an artifact of the bacterial production process and was not studied further. In the present study we refer to the trimer A species simply as trimer. As shown in Fig. 2, 100 mM DTT treatment reduced the apparent molecular masses of E. coli-derived hexamer ( Fig. 2A) and HEK cell-derived hexamer (Fig. 2C) and HMW Acrp30 (Fig. 2D) to that of a trimer (Fig. 2B). Reduction of E. coli hexamer into trimer could also be achieved with 10 mM of DTT or 1 mM ␤ME (data not shown).
As shown in Fig. 3, Acrp30 contains two conserved cysteines, one in the conserved region between the signal sequence and the collagen domain (Cys 22 ) and the other in the globular domain (Cys 138 ). The conserved region around Cys 22 of mouse Acrp30 has the sequence APALVPPPKGTCAGWMA. To investigate whether one or both of these cysteines are responsible for oligomerization of Acrp30, each was substituted with alanine. Size exclusion chromatography of the C138A Acrp30 secreted from transfected HEK cells revealed a migration profile similar to that of wild type protein (Fig. 4A). Three peaks corresponding to HMW, hexamer, and trimer in proportions analogous to those of wild type Acrp30 were observed (Fig. 4A). Gel filtration analysis of purified C22A Acrp30 demonstrates that this mutant migrates only as a single, apparently trimeric species (Fig.  4B). Equilibrium sedimentation in the analytical ultracentrifuge confirmed this species to be a trimer (Fig. 5A); the molecular mass determined by this analysis is 78 kDa, identical to that of the wild type Acrp30 trimer A (10). C22A Acrp30 was also visualized by freeze etch electron microscopy (Fig. 5B). Similar to wild type Acrp30 trimer (Fig. 1A), a large sphere consisting the globular domain, a collagen stalk, and a smaller sphere most likely representing the N-terminal region could be seen in images of C22A Acrp30. These results indicate that the C138A mutation has no observable effect on oligomer formation and distribution, whereas the presence of Cys 22 is critical for formation of hexamer and HMW complexes.
To assess the degree to which Acrp30 monomers are disulfide-linked in each of the three isoforms, we analyzed the proportion of Acrp30 disulfide-linked dimers and monomers in nonreducing SDS-PAGE (Fig. 5C). As control, after reduction all of the Acrp30 species migrate as monomers on SDS-PAGE (data not shown). Trimers from both wild type and C138A Acrp30 contain dimers and monomers in an approximately 2:1 proportion, indicating one Cys 22 -Cys 22 disulfide bond between two monomers and a free Cys 22 sulfhydryl group in the third monomer. Both wild type and C138A HMW and hexamer isoforms contain virtually no monomers, indicating that all Acrp30 polypeptides are linked together by disulfide bonds. Because C138A HMW and hexamer have the same dimer to monomer ratios as the corresponding wild type Acrp30 species, the disulfide bonds linking monomers together must be formed by Cys 22 . Indeed, the C22A mutant trimer contains no dimers in nonreducing SDS-PAGE and consists only of monomers that are not disulfide linked.
Previously we showed that HMW and hexameric Acrp30 activated NF-B, but trimer could not (10). To confirm the causal relationship between the oligomerization state and NF-B activation, we examined the ability of ␤ME-reduced E. coli-produced hexamer as well as C22A and C138A oligomers produced in transfected mammalian cells to activate NF-B, as measured by an E-selectin promoter reporter gene assay. Following treatment with 0.5 mM ␤ME at 2 h at 37°C and then centrifugation under vacuum for 15 min to reduce the concentration of ␤ME, E. coli-derived Acrp30 hexamer was collapsed to trimer (data not shown) and lost its ability to activate NF-B (Fig. 6A). Reduction of ␤ME concentration by centrifugation under vacuum did not reoxidize Acrp30 (data not shown). Consistent with results obtained using wild type Acrp30, the C138A HMW and hexamer species potently acti-vated NF-B (Fig. 6B). Neither the C138A trimer nor the mutant C22A trimer activated NF-B (Fig. 6B), corroborating the observation that only hexamer and higher order oligomers of Acrp30 can activate NF-B.
We previously reported that gAcrp30 activated AMPK in isolated rat skeletal muscle but that full-length Acrp30 hexamer produced in E. coli did not (11). We thus tested whether activation of AMPK also depends on the oligomerization state of Acrp30. Incubation of isolated rat EDL muscles with trimeric C22A Acrp30 or with purified C138A trimer led to 2.5-and 1.5-fold increases, respectively, in the proportion of AMPK␣ phosphorylated at Thr 172 (Fig. 7); phosphorylation at Thr 172 is a metric of AMPK activation. These changes in phosphorylation of AMPK␣ were not due to changes in AMPK␣ abundance (Fig. 7). Neither HMW (Fig. 7) nor hexamer (data not shown) isoform of C138A Acrp30 was able to increase AMPK␣ phosphorylation in isolated rat EDL muscles.
Thus, gAcrp30 and trimeric Acrp30, but not hexameric or HMW Acrp30 isoforms, activate AMPK in skeletal muscle, whereas hexameric or HMW Acrp30 isoforms, but not gAcrp30 or trimeric Acrp30, activate NF-B (Table I). Clearly the state of oligomerization of Acrp30 determines the signal transduction pathway it activates. DISCUSSION Here we defined the oligomeric structures of two Acrp30 isoforms and demonstrated the crucial role of Cys 22 disulfide bonds in formation of the hexamer and HMW species. Importantly, we showed that only hexameric and HMW Acrp30, but not trimeric Acrp30 species, can activate NF-B, and that only gAcrp30 and trimeric Acrp30, but not hexameric or HMW Acrp30 isoforms, activate AMPK in skeletal muscle. Our results with the C22A mutant are broadly consistent with the recent findings of Pajvani et al. (21). Although Pajvani et al. (21) reported that substitution of Cys 22 with serine resulted in N-terminal truncation of Acrp30, we did not observe any proteolysis in the C22A mutant. Together, these studies indicate that individual Acrp30 oligomers have different biological activities, because Pajvani et al. also showed that trimeric Acrp30 was the most potent isoform in suppressing hepatocyte glucose production (21).
Multiple Oligomeric Forms of Acrp30 - Table I summarizes the activities of different Acrp30 oligomers purified from different sources. Because trimeric full-length Acrp30 and gAcrp30 activate AMPK but not NF-B and because hexameric/HMW Acrp30 have the opposite actions in that they activate NF-B but not AMPK, the signaling specificity of Acrp30 depends critically upon its oligomerization state. Acrp30 thus represents a novel example where the signaling specificity of a hormone is regulated by its oligomerization state. FIG. 3. Domain structure of Acrp30 and alignment of N-terminal region among human, monkey, canine, bovine, mouse, and rat orthologs. At the very N terminus of Acrp30 is a signal sequence; between the signal sequence and collagen domain is a region whose first half bears little similarity among the species represented. The second half of this region is remarkably conserved and contains a conserved cysteine. Amino acid residue numbers on the bottom refer to the mouse sequence only. Residue 1 refers to the first amino acid found in mouse serum Acrp30, glutamic acid (34). The 17 residues N-terminal to residue 1 presumably form the signal sequence. The two cysteines, one in the conserved domain just outside the collagenous region and the other in the globular domain, are shaded. The alignment was produced by the ClustalV method.
All three oligomeric isoforms of Acrp30 are present in serum (10), but at least in mice the relative proportion of the different oligomers depends on gender; serum from females contains a higher proportion of HMW isoforms than that from males (21). Injection of mice with glucose or insulin causes a transient decrease of the HMW isoform (21). Little else is known to affect the distribution of Acrp30 oligomers. Given our observation that different oligomers activate different signaling pathways, current clinical data correlating the levels of Acrp30 with the degree of insulin resistance and body mass index (2, 3) should be reconsidered. In particular, we do not know whether the current enzyme-linked immunosorbent assay or radioimmunoassay procedures used to measure total Acrp30 levels (22,23) exhibit any bias toward one or more isoforms. Our data suggest that only some of these oligomers affect insulin sensitivity, presumably via activation of AMPK. A comparison of oligomer distribution between normal and insulin resistant individuals might elucidate which signaling pathway, NF-B, AMPK, or perhaps others, is more important for maintenance of insulin sensitivity.
Disulfide Bonds and the Structure of Acrp30 Oligomers-In deep etched electron micrographs the collagen stalks of Acrp30 trimers and hexamers appeared to be extended and straight with little or no flexibility. The addition of this collagen "rod" to a globular domain would extend the spherical radius of the protein, resulting in an abnormally large molecular weight in gel filtration chromatography. Acrp30 hexamers are composed of two parallel trimers connected in a head to head fashion. We cannot draw firm conclusions about HMW structure. The finding that the trimeric globular heads are adjacent to each other in the hexamer is significant because it may provide additional structural complexity that could allow the hexamer and HMW species to interact with cell surface receptors differently than the trimer.
Disulfide bonds are present in all three isoforms of Acrp30. Nonreducing SDS-PAGE data indicate the hexamer and the HMW isoforms are almost totally composed of disulfide-linked dimers. Nonreducing SDS-PAGE data also indicate that each trimer consists of one disulfide-linked dimer and one monomer. The crystal structure of the trimeric Acrp30 globular domain shows that none of the three Cys 138 residues participates in disulfide linkages (24). Because Cys 22 is the only other cysteine in Acrp30, it must be the residue responsible for disulfide formation. We confirmed this experimentally by demonstrating the lack of disulfide-linked dimers in the trimeric C22A Acrp30. In contrast, C138A Acrp30 had the same 2:1 dimer to monomer ratio as wild type Acrp30 in nonreducing SDS-PAGE. The Cys 22 -Cys 22 disulfide bond in Acrp30 trimers may help to stabilize the collagen triple helix. Interestingly, the asymmetry introduced by the disulfide bond at the N terminus is mirrored by an asymmetry of the three monomers making up the trimeric globular domain in the crystal structure (24).
Disulfide bonds are required for assembly of hexameric and HMW Acrp30 because reduction of either isoform with DTT collapsed them into trimers. Although residue Cys 138 does not form disulfide bonds in the crystal structure of the trimeric gAcrp30 (24), it may in hexamer or HMW forms of Acrp30. However, our observation that the C138A mutant can assemble normally into hexamer and HMW isoforms makes this possibility unlikely. Instead, disulfide bonds linking Cys 22 residues are required for the formation of Acrp30 hexamer and HMW oligomers because Acrp30 with the Cys 22 A substitution forms only trimers. The trimeric organization of C22A Acrp30 was confirmed by gel filtration chromatography and equilibrium sedimentation in the analytical ultracentrifuge. Close inspection of the primary sequences of Acrp30 orthologs (Fig. 3) shows a remarkable degree of amino acid sequence conservation around Cys 22 ; amino acids other than Cys 22 in this conserved region may be important in mediating the formation of disulfides in hexameric and HMW Acrp30.
Either reduction of high order Acrp30 isoforms with ␤ME or substitution of Cys 22 with alanine abolished the ability of Acrp30 to activate NF-B. This confirms our previous observation that only hexameric and HMW Acrp30 isoforms can effectively activate NF-B. Perhaps one Acrp30 monomer binds one cell surface receptor, and receptor clustering is required for activation of NF-B by Acrp30. The notion that a hexameric but not trimeric complex is required for receptor activation has recently been extended to Fas ligand, a member of the tumor necrosis factor ␣-C1q superfamily. Whereas trimeric soluble Fas ligand is unable to induce cell death, a soluble hexameric Fas ligand generated by fusion with the Acrp30 collagen domain could signal apoptosis (25).
Activation of the AMPK and NF-B Pathways by Acrp30 -Presently, the significance of NF-B activation by hexameric and HMW Acrp30 is unclear. NF-B does not seem to mediate Acrp30-stimulated fatty acid oxidation or suppression of he- patic glucose output because gAcrp30 potently enhances fatty oxidation but does not activate NF-B (10,11). Recently Pajvani et al. (21) reported that the trimer form of (full-length) Acrp30 is more active toward suppression of hepatocyte glucose output than the higher molecular weight isoforms. Similar to gAcrp30, trimeric Acrp30 is not effective in NF-B activation, making it unlikely that NF-B activation is involved in affecting liver glucose production. NF-B activation results in increased interleukin-6 gene expression and release from C2C12 myotubes (26,27). Interleukin-6 is produced at high levels by skeletal muscle during exercise (28) and is thought to trigger increased fatty acid and glucose production from adipose tissue and liver, respectively, thus providing an increase in circulating fuel for use by muscles (28). Perhaps hexameric and HMW Acrp30 play a role in the release of IL-6 from skeletal muscle through NF-B activation and thereby indirectly stimulate lipolysis from fat tissue and glucose production from liver.
As described above, unlike the hexamer and HMW isoforms, gAcrp30 does not activate NF-B (10). Instead it signals through AMPK (11,12). Whether full-length Acrp30 can also stimulate AMPK is controversial. We found that Acrp30 hexamer could not activate AMPK (11), whereas Yamauchi et al. (12) reported that full-length Acrp30 could. One possible explanation for the difference is that "full-length Acrp30" used by Yamauchi et al. contained mixtures of different Acrp30 oligomers. In the present study we showed that trimeric fulllength Acrp30, but not the higher molecular weight isoforms, could activate AMPK. The presence of some trimer in the preparation used by Yamauchi et al. could account for the apparent differences between these studies (11,12). A recent report indicates that two heptahelical membrane-spanning proteins act as receptors, albeit with different binding affinities, for gAcrp30 and full-length Acrp30 to mediate activation of AMPK in muscle cell lines (29). It is unclear which fulllength Acrp30 isoforms, trimer, hexamer, or HMW species, bind to each of these receptors.
The Role of the Collagen Domain in Acrp30 -The use of disulfide bonds to stabilize oligomeric structures appears to be a common theme in the family of proteins bearing the C1q-like globular domains. Disulfide bonds near the N terminus help the six heterotrimers of C1q to assemble properly (14,30). Similarly, emilin and multimerin both have a cysteine-rich segment at the N terminus that participates in multimerization of these molecules (13). Because neither emilin nor multimerin contains collagen repeats, it brings into question whether the collagen domains are needed for multimerization of this family of molecules. We demonstrated the necessity of disulfide bonds formed by Cys 22 in the assembly of hexameric and HMW Acrp30 but have not shown that these disulfide bonds are sufficient as well. If disulfide bonds alone are sufficient, one might expect the trimer subunits of the hexamer to be more loosely associated rather than tightly linked in the FIG. 6. Differential activation of NF-B by wild type and cysteine-mutated Acrp30 isoforms in C2C12 cells. A, loss of NF-B activation by Acrp30 hexamer following reduction with ␤ME. Luciferase activity was measured by E-selectin promoter-luciferase reporter gene in transfected C2C12 cells following overnight incubation with 4 g/ml Acrp30 hexamer purified from E. coli, before or after a 2-h incubation at 37°C with 0.5 mM ␤ME. As controls, lipopolysaccharide (LPS, 200 ng/ml, E. coli serotype 055:B5, Sigma) and PBS were similarly treated with ␤ME. The samples containing ␤ME were centrifuged in a Speed-vac for 15 min to reduce the concentration of ␤ME before addition to the cells. B, activation of E-selectin promoter-luciferase in undifferentiated C2C12 cells following an 8-h incubation with 2 g/ml of C22A Acrp30 or HMW, hexamer, and trimer isoforms of C138A Acrp30.

FIG. 7. Differential activation of AMPK by cysteine mutated
Acrp30 isoforms in isolated rat EDL muscles. EDL muscles from 60-gram rats were incubated for 30 min in the absence or the presence of C138A trimer, C138A HMW, and C22A trimer at 2 g/ml each (n ϭ 3/group) and then frozen in liquid N 2 until analyzed. Activation of AMPK was measured by Western blot analysis of AMPK␣ phosphorylated at Thr 172 (p-AMPK) using a phosphorylation state-dependent antibody. Total AMPK␣ levels were also assessed using a phosphorylation state independent anti-AMPK␣ antibody. The details are described under "Experimental Procedures." observed parallel, head-to-head, formation. This suggests that interactions other than disulfide bonds contribute to the high order structures of Acrp30. Collagens form long fibrils that lend structural integrity to the extracellular matrix. Self-assembly of the fibrils is driven by hydration forces as well as entropy associated with hydrophobic forces (31)(32)(33). Because Acrp30 hexamer is composed of only two relatively short collagen triple helices, these forces may not be strong enough by themselves to hold the hexamer together. Nevertheless these weak interactions may help to stabilize the tightness of the hexameric structure. These molecular forces that allow fibril formation may contribute significantly to the maintenance of the HMW form of Acrp30.
In conclusion, interchain disulfide bonds formed by residue Cys 22 are necessary for the oligomerization of Acrp30 beyond the basic trimer and formation of the hexamer and HMW structures seen in freeze etch electron microscopy. Further studies are required to resolve the structure of the HMW Acrp30 and molecular forces other than the disulfide bonds that contribute to the particular structures of each Acrp30 isoform. Most importantly, in light of our finding that different Acrp30 oligomers activate different signal transduction pathways, the physiological relevance of the different oligomerization states in living animals must be determined. In particular, measurements of the total level of Acrp30 in human serum are used in studies correlating Acrp30 levels with diabetes and obesity (2,3), whereas the level of only one or more of these isoforms might actually be relevant.