Structure, Properties, and Engineering of the Major Zinc Binding Site on Human Albumin*

Most blood plasma zinc is bound to albumin, but the structure of the binding site has not been determined. Zn K-edge extended x-ray absorption fine structure spectroscopy and modeling studies show that the major Zn2+ site on albumin is a 5-coordinate site with average Zn-O/N distances of 1.98 Å and a weak sixth O/N bond of 2.48 Å, consistent with coordination to His67 and Asn99 from domain I, His247 and Asp249 from domain II (residues conserved in all sequenced mammalian albumins), plus a water ligand. The dynamics of the domain I/II interface, thought to be important to biological function, are affected by Zn2+ binding, which induces cooperative allosteric effects related to those of the pH-dependent neutral-to-base transition. N99D and N99H mutations enhance Zn2+ binding but alter protein stability, whereas mutation of His67 to alanine removes an interdomain H-bond and weakens Zn2+ binding. Both wild-type and mutant albumins promote the safe management of high micromolar zinc concentrations for cells in cultures.

Zinc is not only required for hundreds of essential extra-and intracellular proteins and enzymes but is also recruited by toxins such as anthrax lethal factor (1) and staphylococcal enterotoxin (2). There is a need to understand how zinc transport and distribution is controlled (3).
Although considerable progress has been made in the identification and study of membrane-bound zinc transporters, the molecular mechanism of extracellular zinc transport is still obscure. The total concentration of zinc in blood is high, ϳ15-20 M (4), and plasma zinc concentrations are maintained at a relatively constant level, except during periods of dietary zinc depletion and acute responses to stress or inflammation, when they are depressed (5). In humans, ϳ98% of so-called "exchangeable" zinc in blood plasma (9 -14 M) is bound to serum albumin (6). Studies on perfused rat intestine have implicated albumin in the transport of newly absorbed zinc in portal blood, from the intestine to the liver (5). Albumin has also been shown to promote zinc uptake by endothelial cells, with receptor-mediated endocytosis as the most likely mechanism (7).
Albumin, the most abundant protein in blood plasma (ϳ40 mg ml Ϫ1 and 0.6 mM), is synthesized in the liver and secreted into the blood stream as a 585-residue, single-chain protein after loss of a 24-residue propeptide (8). The protein is largely ␣-helical and folds into three structurally homologous domains (I, II, and III), each of which contains two subdomains (A and B) (see Fig. 1A) (9). There are 35 Cys residues that form six disulfide bridges in each domain, except for domain I, which contains only five bridges and a free thiol at Cys 34 .
Zinc binding to albumin has also been demonstrated in vitro, and the so-called high-affinity site for zinc on albumin displays a binding constant of K Ϸ 10 7 M Ϫ1 (10 -13). Although over 50 x-ray structures of albumin have been reported to date (9, 14 -16), no experimental structural data are available for the zinc site on albumin. On the basis of 111 Cd NMR studies, sitedirected mutagenesis, and molecular modeling, we have proposed that the major zinc site on albumin, termed site A, is located at the interface of domains I and II and is formed by the side chains of His 67 and Asn 99 from domain I and by His 247 and Asp 249 from domain II (Fig. 1B) (17,18).
Here, we report the first direct structural characterization of a zinc site on albumin, using Zn K-edge x-ray absorption fine structure (EXAFS) 2 spectroscopy. We also explored the possibility of engineering albumins with increased or decreased zinc-binding affinity by mutating the postulated zinc-binding ligands. NMR methods were used to identify mutated and zincbinding histidines and to probe the effects of zinc binding on the conformational dynamics of the protein. Finally, we show that albumin at physiological concentrations promotes the culture of hepatocytes at otherwise toxic zinc concentrations.

EXPERIMENTAL PROCEDURES
All albumin samples were prepared from purified, recombinant proteins produced as described previously (17,27).
Mutant albumin sequences were prepared from a native human albumin cDNA by site-directed mutagenesis. 1 H NMR Spectroscopy-Amide proton resonances were eliminated by repeated incubation of samples with D 2 O (99.9%; Aldrich). One-dimensional and two-dimensional 1 H NMR experiments were routinely carried out at 310 K on a Bruker Avance 600 spectrometer operating at 599.82 MHz using a Z-gradient triple-resonance ( 1 H, 13 C, and 15 N) probe head using 1 mM albumin solutions in D 2 O at pH * ϭ 7.3-7.4 (pH * corresponds to the pH meter reading for a solution in D 2 O using H 2 O-based buffers for calibration, without further correction. These can be converted to pD by adding 0.4.), containing 50 mM NaCl, 50 mM Tris-Cl and sodium formate (1 mM, as internal calibration standard; 8.48 ppm relative to sodium 3-(trimethylsilyl)propionate). Typically, 512 transients were acquired for the one-dimensional spectra (90°excitation pulse, 9 kHz sweep width, and 8,192 time domain data points) using a presaturation pulse sequence for residual water suppression. Two-dimensional data are shown and described in supplemental Figs. S4 -S8.
EXAFS Data Acquisition-Albumin samples were 1 mM in 50 mM HEPES, which had been demetallated using Chelex resin (Bio-Rad) prior to albumin addition. Aliquots of metal chlorides (Zn 2ϩ , Cd 2ϩ , and Cu 2ϩ ) were added from 0.2 M stock solutions. Zn K-edge EXAFS data were recorded on EXAFS stations 9.3 and 16.5 at Daresbury Laboratory Synchrotron Radiation Source at 15 K and in fluorescence mode. Total data collection times were between 11 and 19 h per sample.
Model Building and Refinement-The EXAFS data were processed and analyzed as described previously (19) using full (three-dimensional) multiple scattering. We used our model of the zinc site (17), based on a published crystal structure at 2.5 Å of ligand-free (apo)albumin (Protein Data Bank code 1AO6 (9)), as a starting point for EXAFS data refinement. The model included the side chains of His 67 , Asn 99 , and Asp 249 , as well as the side chain and backbone of His 247 and all atoms of Gly 248 , including the peptide bond to Asp 249 (30 atoms). Analysis required application of the full curved wave theory. In addition, it was necessary to iterate between EXAFS data fitting and energy minimization within the molecular modeling package MOE (version 2004.03), using an in-house customized version of the AMBER94 force field. The final model of zinc-albumin satisfies protein-specific constraints as well, validated by PRO-CHECK (20) and WHAT_CHECK (21).
The starting models of the mutants were generated by analogy to the wild-type starting model (17), using Sybyl (version 6.8), employing a customized TRIPOS force field incorporating zinc-specific parameters. Molecular graphics representations were generated using MOLMOL (version 2000 K) (22) and Weblab Viewer (Accelrys).
Equilibrium Dialysis and Stability Comparisons-The affinity of albumin for zinc was determined using an adapted published procedure (23). A 1-ml aliquot of 100 M recombinant human albumin (rHA) was dialyzed for 72 h at 310 K against 250 ml of zinc-containing buffer (25 mM HEPES and 100 mM NaCl). The free zinc concentration was adjusted by using varying concentrations of dipicolinic acid (Sigma). The exact free zinc concentrations were calculated using the "Species" module of the IUPAC Stability Constants Database (29), and the published stability constants for the Zn 2ϩ -dipicolinic acid system (24). After dialysis, the rHA samples were subjected to rapid gel filtration (Ͻ3 min; 25 mM HEPES, pH 7.4; Sephadex G25M, pD 10, GE Healthcare) to remove any unbound or loosely associated zinc; therefore, the measured zinc contents represent zinc that remained bound to albumin under these conditions. Subsequently, the albumin:zinc ratio was determined spectrophotometrically. The albumin concentration was determined by measuring the absorbance at 279 nm, using a molar extinction coefficient of 0.558 mg Ϫ1 cm Ϫ1 (25). The zinc concentration was determined using the zinc indicator 4-(2-pyridylazo)resorcinol (Sigma). Calibration standards (1-25 M Zn 2ϩ in 25 mM HEPES) were always run in parallel to albumin samples. The data presented are neither background-corrected nor normalized; the error bars are based on estimated errors in albumin absorbance measurements.
Cell Culture, Death, and Number Assessment-WRL-68 cells were obtained from the American Type Culture Collection (Rockville, MD) and maintained in 80 cm 2 tissue culture grade flasks (CellStar) in Dulbecco's modified Eagle's medium (containing L-glutamine; Invitrogen) supplemented with 10% newborn calf serum (Invitrogen), penicillin and streptomycin, and non-essential amino acids (Invitrogen), which include the zincbinding histidine and cystine at a concentration of 0.2 mM each.
Cells were grown at 310 K, 5% CO 2 in an incubator and supplemented with fresh medium every 2-3 days or as required. Once cells became confluent, they were harvested using trypsin and EDTA.
Cells were seeded onto 12-well cell culture plates (tissueculture-treated polystyrene, Nunclon) at 0.099 -0.1 ϫ 10 6 FIGURE 1. A, domain structure of human serum albumin (9). Domain I (red), domain II (blue), and domain III (green) are held together solely by linker helices and weak interactions. B, the proposed interdomain zinc site on albumin is formed by two residues from domain I (red) and two residues from domain II (blue). cells/ml using 0.5 ml (for assessment of albumin/zinc effects) (see Fig. 6) or 1 ml (for assessment of zinc toxicity, see supplemental Fig. S10) per well and left overnight to equilibrate. The medium was then aspirated off, and the cell layer was washed once with phosphate-buffered saline. Cells were treated with 0.5 ml of medium supplemented with 0, 60, 300, or 600 M ZnCl 2 , in the absence or presence of wild-type rHA or H67A mutant albumin with medium alone as a control. Following a 48-h incubation, the plates were analyzed using flow cytometry. Propidium iodide (1 g/ml) was added to samples immediately prior to counting to detect cell death. Samples were vortexed and run using a Beckman Coulter EPICS cell counter. The total number of events after 60 s was recorded to determine cell numbers for comparison between groups. All treatments were carried out in triplicate on plates. Each plate was set up in triplicate as well.

RESULTS
To elucidate the residues involved in high-affinity zinc binding on albumin, we studied the metal-binding properties of wild-type rHA and the three mutants H67A, N99D, and N99H by Zn K-edge EXAFS spectroscopy, equilibrium dialysis, and 111 Cd and 1 H NMR spectroscopy. The H67A mutant corresponds to a deletion of one metal ligand and was expected to bind zinc less strongly and in a different mode than wild-type rHA, and the two Asn 99 mutants were designed to replace the weak ligand asparagine (amide oxygen) with the stronger ligands aspartate (carboxylate oxygen) or histidine (imidazole nitrogen).
EXAFS Spectroscopy-For the wild-type rHA, we analyzed Zn K-edge data for three different preparations, namely Zn-rHA, Zn,Cd-rHA, and Zn,Cd,Cu-rHA (see "Experimental Procedures" and Fig. 2A). Because it is known that albumin contains several metal-binding sites (10,13), it was important to establish that the various methods monitor zinc binding to its high-affinity site, rather than mixtures of different species. Based on previously published data (10,11,13,26,27), the latter mixed-metal preparation is expected to yield albumin with Cu 2ϩ bound to the N-terminal ATCUN motif (28), Cd 2ϩ bound to the so-called site B (27), and Zn 2ϩ to site A (17,27). The EXAFS spectra for these differently metallated albumins are essentially identical, indicating that under the conditions applied, the first equivalent of Zn 2ϩ binds predominantly to site A. Evaluation and subsequent experiments for the mutant proteins were carried out on 1:1 mixtures of Zn 2ϩ and albumin.
The EXAFS data for the Zn-H67A, Zn-N99D, and Zn-N99H mutants are compared with that of wild-type Zn-rHA in Fig. 2 (B-D). The data show clearly that each mutation affects the zinc environment. Analysis of the EXAFS data provided a structural model of the zinc site in wild-type albumin. Single scattering analysis of the first shell (ligand or nearest neighbor) atoms for wild-type rHA containing 1 mol eq of bound Zn 2ϩ (Zn-rHA) categorizes the zinc environment as a 5-coordinate O/N site with an average Zn-O/N distance of 1.98 (Ϯ0.02) Å, plus another more distant O/N at 2.48 Å. A theoretical molecular model based on Protein Data Bank code 1AO6 was built, in which zinc is bound by the four postulated ligands and a water molecule. The theoretical EXAFS data for this model display an R-factor of 28% with respect to the experimental data, even prior to any refinement. Using this model as the starting point and applying full multiple scattering EXAFS data fitting (Fig. 3A), combined with force-field-based energy minimization of the surrounding portions of the protein, a refined model of the site was obtained (Fig. 3B).
The imidazole N⑀ of His 67 (1.99 Å), the side chain amide oxygen of Asn 99 (2.03 Å), imidazole N-␦ of His 247 (1.98 Å), and carboxylate oxygen of Asp 249 (1.99 Å), together with a water oxygen (1.91 Å) provide the first ligand shell, whereas the backbone carbonyl oxygen of His 247 provides a sixth, more weakly bound ligand (2.48 Å). Attempts to fit the data by alternative models, e.g. by defining the ␦-oxygen or nitrogen of Asp 249 or Asn 99 as the sixth ligand, did not yield satisfactory fits.
Intriguingly, although ligand backbone atoms and nonbonding residues are usually omitted in this method, it was necessary to include all atoms of residues His 247 -Gly 248 -Asn 249 within 5 Å of the Zn 2ϩ ion in this case, to account for all contributions to the data. However, even though highly constrained, unique positions could not be determined from the EXAFS data alone for a number of the atoms, and a small number of closely related models with similar structures were obtained. Therefore, the positions of the backbone atoms of residues 247-249 shown in Fig. 3B are representative of a family of conformers which provides a reasonable fit to the data (R ϭ 23%).
The EXAFS spectrum of the H67A mutant (Fig. 2B, inset) differs considerably from the spectrum of wild-type rHA and is consistent with a 4(ϩ1)-coordinate site. The loss of one ligand is also reflected in the lower intensity of the first shell peak in the Fourier-transformed spectrum. In contrast, spectra for both the Asn 99 mutants exhibit comparatively small, yet significant differences form the spectrum ( Fig. 2A) of wild-type rHA. All data for the mutants were well fitted using optimized models based on the in silico mutation of the respective residues (supplemental Fig. S1).
Equilibrium Dialysis-The affinity of the mutated albumins for zinc was determined by equilibrium dialysis. The moderate, i.e. high nanomolar, zinc affinity of albumins reported in the literature (log K ϭ 6.1-7.5) (10 -11,12,13) might be attributable to both the exposed nature of the postulated site and the presence of the amide side chain of asparagine as one of the four ligands, a binding group that is relatively weak and uncommon in proteins. Mutation of Asn 99 to His or Asp, both of which have high affinities for zinc, was therefore predicted to increase zinc affinity, whereas the removal of the His 67 ligand was expected to decrease affinity. Fig. 4A shows a comparison of zinc affinities obtained from the equilibrium dialysis experiments. A value of log K ϭ 7-8 can be estimated for the conditional binding constant of the first molar equivalent of Zn 2ϩ to wild-type rHA at pH 7.4, a value in the same range as those reported previously for human serum albumin (10 -11, 12, 13). Importantly, all three mutations have a pronounced effect on the affinity of albumin for zinc. The relative affinities follow the order: N99D, N99H Ͼ wild type Ͼ H67A. The binding constants of the Asn 99 mutants are ϳ5 times greater than that of wild-type rHA, whereas that of the H67A mutant is ϳ10 times less. These data are in agreement with the changes expected on introducing side chains with stronger affinity than Asn (Asp and His) on the one hand and lower affinity (Ala), on the other, into the postulated zinc binding site (29).  AUGUST 21, 2009 • VOLUME 284 • NUMBER 34 NMR Spectroscopy-111 Cd and 1 H NMR spectroscopy were used to probe the binding of Cd 2ϩ and Zn 2ϩ to the mutant albumins in comparison to wild-type rHA. 111 Cd and 113 Cd (both spin one-half nuclei) NMR experiments have been used previously to probe zinc sites in proteins including albumin (11,26,27).

Zinc Binding to Albumin
Consistent with previously reported data (11,17,26,27), wild-type albumin incubated with 2 mol eq 111 Cd gave rise to two peaks in 111 Cd NMR spectra, at 131 (site A) and 28 ppm (site B) (Fig. 4B). The 111 Cd NMR spectrum of Cd 2 -N99H, recorded under identical conditions as that for wild-type rHA, contained peaks at 122 and 28 ppm (Fig. 4B), clearly demonstrating that site A is affected by the Asn 99 mutation, whereas site B remains unchanged. The relative affinities of Zn 2ϩ and Cd 2ϩ are also clearly affected. Even 3 mol eq of Zn 2ϩ were not sufficient to displace 111 Cd 2ϩ completely from site A (supplemental Fig. 2), in contrast to wild-type rHA, for which complete disappearance of peak A occurs after addition of 1 mol eq of Zn 2ϩ (17,27). This enhanced relative affinity for Cd 2ϩ compared with Zn 2ϩ on replacement of an oxygen donor by the softer nitrogen donor ligand in the N99H mutant is in line with the hard-soft acid-base principle (30).
The 111 Cd NMR spectrum of Cd 2 -N99D rHA recorded under standard conditions at 295 K showed only peak B, which is unaffected by the mutation. This was surprising in view of the higher affinity of this mutant for Zn 2ϩ and similarly expected higher affinity for Cd 2ϩ . After addition of 2 more mol eq of Cd 2ϩ and raising the temperature to 310 K, two additional very broad peaks appeared at 67 and 106 ppm (⌬1 ⁄ 2 ϳ2 kHz) (Fig.  4B). The peak at 106 ppm can be assigned to a Cd 2ϩ -Tris complex by comparison with a reference solution (supplemental Fig. S3), whereas the peak at 67 ppm is in the chemical shift range expected for a site with one nitrogen and three to five oxygen donors (31). The temperature dependence of the line widths suggests that exchange processes occur in the N99D mutant, which are at an intermediate rate on the NMR time scale at 295 K.
We used 1 H NMR spectroscopy to identify His residues of interest and to probe the effect of Zn 2ϩ on His residues of wild-type and mutant rHAs. High resolution 1 H NMR studies of albumin are complicated by peak overlap and broadening. Sequential assignments are difficult to make, but with the aid of resolution enhancement techniques, two-dimensional methods, and sitedirected mutagenesis, some assignments of His H-⑀1 peaks in the low field region are possible.
The overall appearance of the wild-type and mutant spectra is very similar (Fig. 5A), but there are substantial changes to two low fieldshifted peaks, labeled 1 and 4, and their corresponding cross peaks in the two-dimensional spectra (supplemental Figs. S4 -S7 and Table S1), for all three mutant proteins, suggesting that these peaks correspond to the two histidines in the mutated zinc site, His 67 and His 247 . As seen in Fig. 5A (bottom spectrum; see also supplemental Fig. S8 for a two-dimensional total correlation spectrum), the peaks for at least four His residues are severely broadened (or shifted) by Zn 2ϩ binding to rHA, most notable being the disappearance of peaks 1, 4 (His 67 and His 247 ), 6 (not assigned), and 11 (His 3 ). In contrast, the peak for H-⑀1 of His 39 (labeled 7) appears to sharpen on Zn 2ϩ binding to rHA. Such effects could arise from chemical exchange, changes in mobility (including the direct effects of Zn 2ϩ binding), or changes in solvent accessibility. It is noteworthy that these spectral perturbations are also observed on addition of substoichiometric amounts of zinc (supplemental Fig. S9).
The effects of Zn 2ϩ on the His H-⑀1 peaks of the H67A mutant (supplemental Fig. S9) were similar to those for wildtype rHA, consistent with zinc binding to the same site, as suggested by the EXAFS data. However, studies of both Asn 99 mutants were hampered by reversible, temperature-dependent Zn 2ϩ -induced protein precipitation at 310 K. This result suggests that there is a complicated relationship between Zn 2ϩinduced structural changes at the interface between domains I and II and the thermostability of the protein fold.
Zinc Binding to Albumin in Mammalian Cell Culture-To investigate the Zn 2ϩ -binding ability of albumin in the presence of mammalian cells, we studied the growth of human WRL-68 cells on standard cell culture plates. This human epithelial heteroploid liver cell line exhibits a morphology similar to primary hepatocyte cultures. Zn 2ϩ concentrations above 72 M were toxic to cells cultured in Dulbecco's modified Eagle's medium supplemented with 10% newborn calf serum (supplemental Fig.  S10, A and B), as seen by cell rounding, loss of adherence, and necrotic cell death, effects similar to those of Zn 2ϩ on other cell types (32,33). Physiological concentrations of rHA (600 M) were sufficient to completely abolish the adverse effects of Zn 2ϩ on hepatocytes, even at Zn 2ϩ concentrations as high as 600 M (Fig. 6, A and B), suggesting that the zinc-albumin complex is stable under in vitro conditions and capable of controlling Zn 2ϩ distribution in cell cultures. It should be noted that although the cell culture medium used in these experiments contains the amino acids cystine and histidine (both 0.2 mM), which can chelate Zn 2ϩ with micromolar affinity (log K Ϸ 6.5 (histidine) or Ϸ 6.7 (cystine) (29)), the differences observed in Fig. 6 (A and B) are due to the presence of albumin only. All albumin mutants studied were also well tolerated by cultured cells, both in the presence and absence of added Zn 2ϩ .

DISCUSSION
The combination of site-directed mutagenesis, molecular modeling, and EXAFS has provided new insight into the structure of the important high-affinity zinc site on human albumin. The residues forming this site are conserved in all known mammalian albumin sequences (supplemental Table S2 and Fig.  S12) except for that of the guinea pig, which conspicuously lacks both domain I residues. No human albumin variants with mutations of any of these four residues are known. These findings further emphasize the importance of this site with its peculiar combination of moderate affinity but high specificity for zinc, properties likely to relate to its biological transport function.
The coordination sphere of albumin-bound zinc can be described as distorted octahedral, if the backbone carbonyl oxygen of His 247 is included as the sixth ligand. Hence, the zinc coordination sphere can be considered as saturated. The zincamino acid distances are within the ranges observed for small molecule zinc complexes and protein structures (34,35). Interestingly, the zinc-oxygen (water) distance is relatively short (1.91 Å) compared with those for small molecule complexes (2.09 Ϯ 0.05) but has numerous precedents in zinc sites in proteins, as an analysis of Zn-water distances in the MESPEUS Database shows (35). This might suggest that this bound aqua ligand might be highly polarized and have a low pK a , a feature of some catalytic zinc sites (e.g. carbonic anhydrase). Albumin is known to exhibit esterase-like activity, but this has previously The dramatic effect of the H67A and Asn 99 mutations on the two low-field shifted resonances labeled 1 and 4 allows these two peaks to be assigned to His 67 and His 247 . His 3 (55) and His 39 (56) have been assigned previously and are labeled 11 and 7, respectively. The bottom spectrum demonstrates the effect of zinc addition on the His H-⑀1 NMR resonances of wild-type rHA. At least six histidines experience major changes upon zinc binding; all low-field shifted resonances (1, 2, and 4; i.e. including His 67 and His 247 ), resonance 6 (not assigned), His 39 (resonance 7), and His 3 (resonance 11) are affected by zinc binding. B, the five interdomain His residues in the x-ray structure of fatty acid-free albumin (Protein Data Bank code 1AO6) (9), including their hydrogen bonding partners. Remarkably, all interdomain histidines of human serum albumin are located between domains I and II. WT, wild type. been assigned to Lys 199 , Trp 214 , and Tyr 411 (36). The catalytic activity of zinc albumin has not yet been studied; perhaps this possibility should be further investigated.
A limited number of examples of metal sites with the same set of ligands as the zinc site in albumin are found in enzymes, in particular some phosphatases; in the latter examples, the pentacoordinate zinc site forms part of a binuclear site with bridging carboxylate and hydroxyl groups and displays a geometry different to that represented in Fig. 3B. A mononuclear site with the same set of four protein ligands is found in the zinc-dependent endopeptidase LytM in its inactive form (37). In contrast to albumin, in LytM, there is no space for coordination of a fifth ligand for zinc. The active form of the enzyme is formed only upon dissociation of the Asn residue, which essentially provides a switch for enzyme activation.
In general, the amide side chain of asparagine (or glutamine) is only rarely observed as a protein ligand for zinc, or indeed for other metal ions, but zinc-carboxamide coordination is common during catalysis, for example in the enzyme-substrate complexes of Zn-containing peptidases and proteases, such as carboxypeptidase A (38). In the interdomain zinc site in albumin, Asn 99 has a dual role; in the absence of zinc, together with the backbone carbonyl oxygen of His 247 , Asn 99 provides one of the few, and therefore likely crucial, interdomain hydrogen bonds in albumin. The finding that the yield from albumin expression of the N99D mutant was typically three times lower than for the wild type and other mutants, together with the observation of 1 H NMR peaks for a second conformer, might suggest that this interdomain H-bond is indeed important for stability of the protein fold. The N99D mutation (replacement of a neutral-NH 2 group by a negatively charged oxygen atom) abolishes the interdomain H-bond, and instead introduces a charge repulsion between Asp 99 O-␦ and the carbonyl oxygen of His 247 (supplemental Fig. 13). In contrast, the N99H mutation allows an H-bond to be preserved in this location (supplemental Fig. S13).
It is clear that albumin provides a large capacity for zinc binding in blood plasma, but it is interesting to note that zinc binding in turn has a substantial effect on the dynamic behavior of albumin, as exemplified by the 1 H NMR data ( Fig. 5 and supplemental Fig. S8). The effects of zinc on the 1 H NMR spectra of human serum albumin bear a remarkable resemblance to the effects induced by raising the pH* (see "Experimental Procedures") above 7.5 (39); both H ϩ release and Zn 2ϩ addition lead to the disappearance of the same His H-⑀1 resonances. It has been suggested that these His residues are implicated in the pH-induced N-to-B conformational transition (40). Conformational changes and allosteric effects are important for the ability of albumin to bind and transport hormones and therapeutic drugs (41,42) and therefore can have direct influence on the pharmacokinetics of these compounds (43).
For example, the N-to-B transition has been shown to influence directly the affinity of albumin toward steroids (44) and warfarin (45,46), which both bind to Sudlow site I. This site contains a hydrophobic binding pocket located in subdomain IIa, which is also the main binding site for other drugs and hormones including aspirin, phenytoin, pregnenolone, testosterone, and thyroxin (41,43). Similarly, the environment of Cys 34 , an important antioxidant and drug binding site, is also affected by this conformational change (47).
Although numerous x-ray structures of albumin in the nitrogen form have been determined in the past decade, no three-dimensional structural data are available for the B conformation. A reexamination of currently available experimental evidence may provide a step toward a better understanding of the molecular mechanisms of the N-to-B transition as well as the effects of zinc binding. It should be noted that chemicophysical studies have confirmed that the solution structure of albumin is well represented by the x-ray crystal structures (48).
It is thought that the N-to-B transition is associated with a "loosening up" of the structure localized in domain I with a contribution from domain II (40,48). A small loss of ␣-helical content in the B form has been localized to the interdomain linker helices, and it has been suggested that in this form mutual domain movements can take place (49), which may be more restricted in the nitrogen form.
The underlying molecular mechanism of the N-to-B transition is thought to involve the breaking of hydrogen bonds from protonated histidine residues (40). Some of these residues have unusually high pK a values (25) (supplemental Table S2); these are expected to be protonated, and therefore strong H-bond An inspection of published x-ray structures of albumin reveals that five histidine residues are situated at domain interfaces and have the capacity to form H-bonds involving their side chains: His 67 , His 146 , His 242 , His 247 , and His 288 (Fig. 5B and supplemental Table S2). Intriguingly, the number of His residues involved in the N-to-B transition has previously been reported as five (39). All five interdomain histidines are located between domains I and II (supplemental Table S2), an observation consistent with 1 H NMR spectroscopic studies of peptic and tryptic fragments of albumin (40).
Direct experimental evidence exists not only for the involvement of one of these five residues, His 242 , in the N-to-B transition but also for its impact on drug binding to Sudlow site I (46). Mutation of His 242 to a valine shifts not only the midpoint of the N-to-B transition from the wild-type value of 7.8 to a value of 7.1-7.2 but also affects the affinity toward warfarin. His 242 is located in domain II and forms a hydrogen bond with Gln 196 (Fig. 5B), which is located in the domain linker helix.
Albumin contains very few interdomain H-bonds that do not involve histidines, and hydrophobic interactions between domains I and II are also scarce. Thus, it is highly likely that the domain I/II interface is indeed subtly controlled by the protonation state of the interdomain His residues, and any manipulation, including zinc binding, of interdomain histidines may have effects on the entire interface.
It is striking that two of the five interdomain histidines are part of the zinc-binding site. The zinc ligand His 247 does not form an interdomain H-bond in fatty acid-free albumin but is H-bonded to Glu 100 in fatty acid-loaded albumin (9,14). In either case, His 247 is usually H-bonded via its N␦ to Asp 249 , which also forms an interdomain H-bond to N⑀ of His 67 in fatty acid-free albumin. The amide-NH 2 group of Asn 99 is an H-bond donor to the backbone carbonyl oxygen of His 247 as discussed above. Considering the scarcity of interactions between domains I and II, it is not surprising that the binding of a zinc ion at this interface should have far reaching consequences, as is evident from the 1 H NMR data. We speculate that zinc can reinforce the interdomain contact and stabilize a specific mutual orientation of the two domains. This would be expected to have an impact on the dynamic behavior of residues at this interface, including the interdomain histidines. Thus, both in the presence and absence of zinc, this particular site is of pivotal importance for stabilizing the contact and relative orientation of domains I and II. Remarkably, a conformational change thought to be similar to the N-to-B transition also occurs when albumin interacts with hepatocyte cell membranes (51,52), and it has been hypothesized that this may assist in the delivery of ligands, including fatty acids, to liver cells.
Our findings suggest that further work to investigate the possible interactive binding of zinc and drugs, metabolites, and xenobiotics is warranted. Fluctuations in zinc together with fatty acid levels in blood may have important effects on the transport of some classes of these molecules.
Our studies show that physiological concentrations of albumin can protect WRL-68 cells from zinc-induced toxicity (Fig.   6), even at a 1:1 ratio of zinc:albumin. A similar protective action of bovine serum albumin against the toxic effects of zinc has been shown with cerebellar granule neurons (53) and epithelial cells (54). These findings emphasize the requirement for a suitable zinc-binding agent to be present in physiological fluids such as blood plasma and cerebrospinal fluid.
Because the affinity of albumin for zinc is only moderate, albumin-bound zinc has been considered to be part of the socalled exchangeable zinc pool. This moderate affinity is likely to assist the transfer of zinc to target sites. The molecular mechanisms by which this transfer is achieved are not known in detail yet, and it is remarkable that albumin, in addition to many other functions, has evolved as a highly specific zinc "buffer" in plasma. Under physiological conditions, the unique ligand composition of the site is tailored and specific for zinc and appears to bind zinc with the appropriate affinity (and probably dynamics) to allow albumin to safely release and deliver zinc when necessary.