Crystal Structure of Human ZAG, a Fat-Depleting Factor Related to MHC Molecules

See allHide authors and affiliations

Science  19 Mar 1999:
Vol. 283, Issue 5409, pp. 1914-1919
DOI: 10.1126/science.283.5409.1914


Zn-α2-glycoprotein (ZAG) is a soluble protein that is present in serum and other body fluids. ZAG stimulates lipid degradation in adipocytes and causes the extensive fat losses associated with some advanced cancers. The 2.8 angstrom crystal structure of ZAG resembles a class I major histocompatibility complex (MHC) heavy chain, but ZAG does not bind the class I light chain β2-microglobulin. The ZAG structure includes a large groove analogous to class I MHC peptide binding grooves. Instead of a peptide, the ZAG groove contains a nonpeptidic compound that may be implicated in lipid catabolism under normal or pathological conditions.

ZAG is a soluble protein whose name derives from its tendency to precipitate with zinc salts and its electrophoretic mobility in the region of the α2globulins (1). ZAG is normally present in most body fluids including serum, sweat, saliva, cerebrospinal fluid, seminal plasma, milk, amniotic fluid, and urine (1). In addition, ZAG accumulates in breast cysts as well as in 40% of breast carcinomas, and is induced by glucocorticoids and androgens in breast cancer cell lines. Hence, ZAG may participate in breast diseases, including cancer (2).

The function of ZAG was elucidated when a lipid-catabolizing factor with the same amino acid sequence as ZAG was isolated from the urine of cancer patients with cachexia (3). Cachexia is a wasting syndrome caused by depletion of muscle and adipose tissue that is present in the majority of patients with cancer, AIDS, and other life-threatening diseases (3). ZAG appears to be responsible for the fat-depletion component of cachexia, since it stimulates lipid breakdown in adipocytes and reduces fat stores in laboratory animals (3). ZAG is overexpressed in carcinomas that induce fat loss but not in other tumors. Application of ZAG to adipocyte membranes activates a guanosine triphosphate–dependent adenylate cyclase activity, perhaps through direct or indirect interactions with a G protein–coupled receptor (3). Thus, its mode of action could be similar to that of lipolytic hormones. These results suggest that ZAG normally functions to regulate lipid degradation, which increases to a pathological extent in cachexia.

ZAG shares 30 to 40% amino acid sequence identity with the extracellular portions of class I major histocompatibility complex (MHC) heavy chains (4). Class I MHC molecules present peptide antigens to cytotoxic T cells (5). Other proteins related to class I MHC molecules include CD1, which presents hydrophobic antigens to T cells (6), the neonatal Fc receptor (FcRn), which transports immunoglobulin G across epithelia (7), and HFE, which binds transferrin receptor and regulates iron homeostasis (8). These MHC homologs are membrane-bound heterodimers that use the soluble protein β2-microglobulin (β2M) as a light chain. ZAG, however, is a secreted protein, and it does not associate with β2M (9). The latter property is shared by MIC-A, a divergent membrane-bound member of the class I family (10).

Like FcRn (11), HFE (11), and MIC-A (10), ZAG does not bind endogenous peptides (9), but it appears to carry a small proteinase-resistant compound whose injection induces glomerulonephritis in experimental animals (12). In peptide-binding class I MHC molecules, a large groove located between two α helices in the α1-α2 superdomain of the heavy chain serves as the binding site (5). An analogous groove acts as the antigen binding site in CD1, but is narrowed in HFE and closed in FcRn, the class I–related proteins that do not bind small molecular weight ligands (11).

Here, we present the 2.8 Å crystal structure of human ZAG, which reveals an MHC-like fold without a β2M light chain and a groove that closely resembles the peptide-binding sites of classical class I MHC molecules. Rather than containing a peptide, the ZAG groove includes an unidentified nonpeptidic ligand that may be relevant for ZAG's function in lipid catabolism.

ZAG was purified from human serum and crystallized as described (9). The crystal structure was determined by multiple isomorphous replacement with anomalous scattering (MIRAS) aided by fourfold noncrystallographic symmetry (NCS) averaging (Table 1) (13). The overall structure of ZAG is similar to those of class I MHC heavy chains (Fig. 1A). The α1-α2 superdomains of ZAG, class I, and class I–related proteins form a single eight-stranded antiparallel β sheet topped by two α helices, and the α3 domain adopts a fold resembling immunoglobulin constant domains (5,11). Electron density corresponding to carbohydrate is present at three of the four potential N-linked glycosylation sites in ZAG, with an unusually large number of ordered carbohydrate residues visible at Asn239 (Fig. 1B) (14). The α3 domain of human, but not mouse or rat, ZAG contains an RGD sequence (Arg231, Gly232, Asp233) suggested to be involved in cell adhesion (15). However, unlike the RGD sequences in characterized adhesion molecules such as fibronectin III domains (15), the ZAG RGD is located in a β strand rather than a loop.

Figure 1

(A) Ribbon diagram of the structure of human ZAG. Density corresponding to the nonpeptidic ligand is shown in green as derived from a 2.9 Å MIRAS, NCS averaged, figure-of-merit weighted electron density map contoured at 1σ. Ordered N-linked carbohydrates are shown in ball-and-stick representation. (B) The ZAG model (molecule 2) in the region of the N-linked carbohydrate attached to Asn239 superimposed on a 2.8 Å SIGMAA-weighted 2F oF cannealed omit electron density map (13). The carbohydrate electron density at Asn239 shows nine carbohydrate residues arranged in a biantennary structure (14) (two of the carbohydrate residues are omitted from the figure for clarity). The large number of ordered carbohydrate residues at Asn239 is likely to result from stabilization of the flexible carbohydrate by crystal contacts with protein residues in symmetry-related ZAG molecules (shown in red sticks; bottom). Figures were made as described (23).

Table 1

Data collection, heavy-atom phasing, and refinement statistics for ZAG. Statistics in parentheses refer to the highest resolution bin. ZAG was crystallized in space group P21212 with four molecules per asymmetric unit and cryoprotected as described (9). Native and heavy-atom derivative data sets were collected at –165°C from multiple crystals using an R-AXIS IIc imaging plate system mounted on a Rigaku R200 rotating anode generator. Data were processed and scaled with the HKL package (13). Heavy-atom refinement was done with SHARP (13), which treated different data sets of derivatives with the same sites as separate “crystals” of the same “compound” and refined separate heavy-atom occupancies and temperature fac- tors for each “crystal” data set. Electron density maps calculated from MIRAS phases derived from all eight heavy-atom data sets were superior to maps derived from various combinations of mercury and platinum data sets. Eight NCS operators were determined from least squares superposition of HLA-A2 α1-α2 and α3 domains manually placed in the electron density, four relating the α1-α2 regions and four relating the α3 domains. After solvent flipping using Solomon (13), subdomain NCS averaging of the four molecules and phase extension from 5.0 to 2.9 Å were carried out with DM in the CCP4 suite (13) (final average correlation >0.8). When the same averaging and phase extension procedure was done without prior solvent flipping, the resulting electron density maps were inferior. The model was built using O (13) and refined as described (13).

View this table:

The quaternary arrangement of the ZAG α3 domain with respect to the α1-α2 platform differs from that found in β2M-binding class I and class I–related proteins. The overall shape of ZAG is similar to an inverted “L,” in which the long axis of the α3 domain is roughly perpendicular to the flat side of the α1-α2 platform, whereas the comparable angle is acute in β2M-binding proteins (Fig. 2A). In those proteins, β2M interacts with the α3 domain and the underside of the α1-α2 platform and is typically required for stability (16). The displacement of the ZAG α3 domain compared to its class I counterpart results in the inability of β2M to optimally contact the α3 and α1-α2 domains of ZAG (17), contributing to ZAG's lack of affinity for β2M. The high thermal stability of ZAG in the absence of β2M (9) can be explained by a network of hydrogen bonds between α3 and α1-α2 that are not present in β2M-binding class I proteins (Fig. 2B). In addition to the extra hydrogen bonds, the loop connecting β strand 4 to the helical region of the ZAG α1 domain platform (residues 51 to 54) and the loop connecting strands D to E in the α3 domain (residues 236 to 241) are closer together than their class I counterparts, contributing to the burial of a larger interdomain surface in ZAG (970 Å2 total) than in classical class I molecules (660 Å2 in HLA-A2) (11). There is some flexibility in the position of the ZAG α3 domain relative to α1-α2, as demonstrated by different interdomain relationships of the four ZAG molecules in the crystallographic asymmetric unit (Fig. 2A), yet ZAG is not particularly protease-sensitive at the platform-α3 hinge or any other region (18). Although the quaternary structure of ZAG differs significantly from the heavy chains of class I and class I–like structures, the differences are less than anticipated from the absence of the β2M light chain. The overall similarity between ZAG and class I MHC heavy chains contrasts with the large interdomain rearrangements observed in the crystal structure of MIC-A (10).

Figure 2

Structural comparisons of ZAG, class I, and class I–related proteins. (A) (Left) Comparison of the structures of ZAG and HLA-A2 (11). (Right) Comparison of the four ZAG molecules in the crystallographic asymmetric unit (magenta) with the heavy chains of human class I MHC (yellow; PDB codes 1hhh, 1vac, 2clr) and β2M-binding class I MHC homologs (green; FcRn, CD1, HFE) (11). Superpositions were based on the Cα atoms in the platform domains, using 2clr as a reference molecule. The position of the ZAG α3 domain with respect to its α1-α2 platform falls out of the range of the positions of the α3 domains of the β2M-binding class I proteins. Differences in the platform-α3 interdomain relationships in the four ZAG molecules demonstrate that there is flexibility in the position of the ZAG α3 domain relative to α1-α2. However, the overall similarity of the four molecules, which are subjected to different crystal packing forces, rule out that ZAG's shape is an artifact of crystallization. (B) Close-up comparison of the interface between α1-α2 (blue) and α3 (green) in ZAG and HLA-A2 (11). Additional H-bonds and a larger interdomain surface area stabilize ZAG compared with class I molecules, whose heavy chains are stabilized by interactions with β2M. Figures were made as described (23).

Despite the similarity between ZAG and class I molecules, structural features of the ZAG α3 domain make it unlikely to associate with the T cell coreceptor CD8. Of 15 class I heavy-chain residues identified at the CD8 binding site in the HLA-A2/CD8 cocrystal structure (19), only one is conserved between class I and ZAG sequences (class I Asp122, ZAG Asp123) (4). It is not possible to rule out an interaction between ZAG and T cell receptors, because the class I MHC residues that contact these receptors are not particularly conserved (20).

Although ZAG does not associate with peptides (9), the helices in both the α1 and α2 domains are almost identically positioned to their counterparts in peptide-binding class I MHC molecules (Fig. 3A) (21). Thus, the ZAG platform includes an open groove, by contrast to the narrowed or closed grooves observed in other class I homologs that do not bind peptides [FcRn (11), HFE (11), and MIC-A (10)]. Despite the fact that most of the residues within the COOH-terminal portion of the ZAG α2 domain helix are chemically identical to their class I counterparts (Table 2), the ZAG groove has a different shape than a typical class I MHC peptide-binding groove (Fig. 3B). Differences in side chain conformations (Fig. 3A) and side chain substitutions create the altered shape of the ZAG groove, which can be described as containing three pockets: a large, predominantly hydrophobic central pocket and two smaller, more acidic flanking pockets (Fig. 3B). The boundary on the left of the central pocket (Fig. 3B) is created by the side chain of Arg73, which points into the groove to separate the central and left pockets. Class I MHC grooves contain a smaller side chain at this position (His70 in HLA-A2) (4). Because of Arg73 and other side chain substitutions or conformational differences relative to class I molecules, the ZAG groove cannot accommodate even a polyalanine version of an eight- or nine-residue peptide in a class I–binding conformation (Table 2).

Figure 3

Comparisons of the grooves of ZAG, class I, and class I–related proteins. (A) Stereoview comparisons of the α1-α2 platforms of ZAG and class I proteins that contain open antigen-binding grooves. (Top) ZAG is compared with HLA-A2, a classical class I MHC molecule. “N” and “C” indicate the orientation with respect to the NH2- and COOH-termini of a peptide bound in the HLA-A2 groove. Side chains in common between the ZAG and HLA-A2 grooves (Table 2) are highlighted on the Cα backbones. Although many of the ZAG groove residues are chemically identical to their counterparts in HLA-A2, the conformations of the side chain, the backbone, or both, of these residues are generally different; thus, the grooves have different shapes [see (B)]. (Bottom) Comparison of ZAG and CD1. Although the ZAG and CD1 grooves both bind nonpeptidic ligands, the ZAG groove is smaller and shallower. Prolines within the α2 domain helices of ZAG (Pro167) and CD1 (Pro169) are highlighted. The proline in the ZAG helix is accommodated without significant distortion (24), as previously seen for the analogous proline within the HFE helix (Pro166) (11). (B) Extruded groove pockets of ZAG and HLA-A2. Cut-away groove molecular surfaces (11) are shown from above (top) and the side (bottom) with electrostatic potentials (23), in which positive potential is blue, neutral is white, and negative potential is red. The approximate locations of pockets A through F (Table 2) are indicated on the HLA-A2 surfaces. The molecular surface of the central portion of the ZAG groove is nearly neutral except for the contribution of Arg73 and, like CD1 (11), is hydrophobic compared to the grooves of the class I molecules. Calculated groove surface areas (11) are indicated for ZAG and HLA-A2. For comparison, the groove surface areas for other MHC homologs are ∼1440 Å2 (CD1), ∼235 Å2 (FcRn), and ∼415 Å2 (HFE) (11). (C) Stereoview of electron density (from a 2.9 Å MIRAS, NCS averaged, figure-of-merit weighted electron density map contoured at 1σ) corresponding to the ZAG ligand superimposed upon a ribbon diagram of the ZAG α1-α2 platform. Residues within 4.5 Å of the density (Table 2) are highlighted in ball-and-stick representation. Figures were made as described (23).

Table 2

Comparison of residues in α1-α2 grooves of ZAG and a class I MHC molecule. Uppercase letters in any of the HLA-A2 and ZAG columns indicate residues that are conserved in human class I MHC sequences (4) or residues that are conserved in human, mouse, and rat ZAG (4). Pocket residues in the peptide-binding groove of HLA-A2 are defined as having ≥5.0 Å2 accessible surface area to a 1.4 Å probe, but <5.0 Å2 accessible surface area to a 5 Å probe (11). HLA-A2 residues marked with an asterisk were originally defined as pocket residues by M. A. Saper et al. (11) but are accessible to a 5.0 Å probe. The HLA-A2 residues designated as “NONE” (conserved) or “none” (not conserved) do not meet the criteria for being in a pocket, but are listed for comparison with ZAG. Pocket letter names (uppercase letters, conserved residues) refer to well-characterized pockets in the peptide-binding grooves of class I MHC molecules (5, 11). ZAG residues analogous to class I pocket residues are designated as “POCKET” (conserved) or “pocket” (not conserved) if they meet the criteria for pocket residues in HLA-A2, “BURIED” or “buried” if they have ≤5.0 Å2 accessible surface area to a 1.4 Å probe and ≤5.0 Å2 accessible surface area to a 5 Å probe, and “EXPOSED” or “exposed” if they have ≥5.0 Å2 accessible surface area to a 1.4 Å probe and ≥5.0 Å2 accessible surface area to a 5 Å probe. Surface areas were calculated (11) using the coordinates of HLA-A2 (excluding water molecules and bound peptide) and ZAG. Steric clashes with polyalanine peptides bound to ZAG in their class I–binding configuration were defined as described in the structural analysis of HFE (11).

View this table:

Instead of a peptide, the ZAG groove contains an as yet unidentified ligand that cocrystallizes with the protein. Electron density that cannot be accounted for by the amino acid sequence or N-linked carbohydrates is found in the central pocket of the ZAG groove (Figs. 1A and 3C). The density is a curved nonbranched tube, lacking the characteristic protrusions of peptides, that is situated near a cluster of hydrophobic amino acids (three tryptophans, four tyrosines, an isoleucine, and two phenylalanines) and the positively charged side chain of Arg73 (Table 2). Although it is not possible to unambiguously identify the compound or compounds in the ZAG groove at the current resolution of the electron density maps, previous biochemical studies rule out that the density corresponds to peptide or mixture of peptides (9). The composition of the ZAG groove residues near the density (Table 2) suggests that it represents one or more small hydrophobic molecules, perhaps negatively charged. Chloroform-methanol extractions and acid eluates of ZAG as well as the intact protein have been analyzed by gas chromatography, electrospray, or matrix-assisted laser desorption/ionization mass spectrometry, or a combination of these methods (22). The ligand has not yet been detected, but in the absence of information about the chemical nature, size, and ionic state of the ligand, these results cannot be considered conclusive. Electrospray analyses of proteolytic fragments of ZAG reveal that the ligand is not covalently bound (22).

Classical class I MHC molecules are extremely polymorphic, whereas ZAG exhibits species-specific variations but little or no genetic polymorphism (4, 5). Most of the allele-specific variations of class I MHC molecules map to residues within the peptide-binding groove that interact with peptides (Table 2) (5). This property of class I molecules results in allele-specific peptide-binding motifs, such that individual class I molecules show distinct preferences for binding peptides (5). In contrast, residues within the ZAG groove are mostly conserved, and those residues closest to the ZAG ligand are completely conserved (Table 2), even though human and rodent ZAG share only ∼56% sequence identity (4). Taken together with ZAG's lack of polymorphism, these observations imply that human and rodent ZAG carry a single compound, or single class of compounds, related to their function in lipid catabolism.

The crystallographic analysis of ZAG reveals a structure with surprising similarity to classical class I molecules despite ZAG's inability to bind peptides or β2M (9). ZAG and MHC-related proteins such as FcRn (11) and HFE (11) have adapted the same basic fold to perform widely different roles within and outside of the immune system. These molecules illustrate the versatility of the MHC fold and raise intriguing questions about the ancestral function and evolutionary relationships of MHC and MHC-related proteins.

  • * These authors contributed equally to this work.

  • To whom correspondence should be addressed. E-mail: bjorkman{at}


Stay Connected to Science

Navigate This Article