Competitive Binding of Bismuth to Transferrin and Albumin in Aqueous Solution and in Blood Plasma*

Several bismuth compounds are currently used as antiulcer drugs, but their mechanism of action is not well established. Proteins are thought to be target sites. In this work we establish that the competitive binding of Bi3+ to the blood serum proteins albumin and transferrin, as isolated proteins and in blood plasma, can be monitored via observation of 1H and13C NMR resonances of isotopically labeled [ε-13C]Met transferrin. We show that Met132 in the I132M recombinant N-lobe transferrin mutant is a sensitive indicator of N-lobe metal binding. Bi3+binds to the specific Fe3+ sites of transferrin and the observed shifts of Met resonances suggest that Bi3+ induces similar conformational changes in the N-lobe of transferrin in aqueous solution and plasma. Bi3+ binding to albumin is nonspecific and Cys34 is not a major binding site, which is surprising because Bi3+ has a high affinity for thiolate sulfur. This illustrates that the potential target sites for metals (in this case Bi3+) in proteins depend not only on their presence but also on their accessibility. Bi3+ binds to transferrin in preference to albumin both in aqueous solution and in blood plasma.

Bismuth compounds have long been associated with medicine for the treatment of a variety of gastrointestinal disorders including diarrhea, constipation, gastritis, and ulcers (1)(2)(3)(4). The effectiveness of bismuth has been attributed to its bactericidal action against the Gram-negative bacterium, Helicobacter pylori. There is also a growing interest in using compounds containing radioactive bismuth isotopes as targeted radiotherapeutic agents (5). However, the molecular basis for the mechanism of action of bismuth drugs is not well understood, including bismuth-induced toxicity, especially encephalopathy, which led to the withdrawal of bismuth drugs in France and Australia in the 1970s (1). The diagnosis of encephalopathy is generally defined by the detection of bismuth in blood, plasma or serum, the so-called "Hillemand safety level" (6,7). Bismuth is primarily present in red blood cells, possibly binding to glutathione, with the remainder in serum or plasma (8 -10). The speciation of bismuth in blood plasma, and in particular the nature of interactions of Bi 3ϩ with plasma proteins, are in need of investigation.
Recently we have found that the binding of Bi 3ϩ to human serum transferrin (hTF) 1 and recombinant N-lobe of transferrin is unexpectedly strong (11,12). Transferrin is a singlechain glycoprotein (80 kDa) present in blood at a concentration of about 35 M, and consists of two similar lobes, each of 40 kDa, connected by a short peptide. Its normal function in blood is to carry iron between sites of uptake, utilization, and storage (13)(14)(15)(16). It contains two specific iron-binding sites per molecule, one in the N-terminal lobe and one in the C-terminal lobe. Iron binds as Fe 3ϩ in a cleft formed by two domains in each lobe. Iron cannot bind strongly without concomitant binding of a synergistic anion. Since transferrin is only about 30% saturated with iron in normal serum (13,17,18), there is potential binding capacity for other metal ions that enter the blood. This has led to the idea that transferrin acts as a "delivery system" for therapeutic, diagnostic or toxic ions, including Ga 3ϩ , Ru 3ϩ , and Al 3ϩ (19 -21). Recently we have shown that Bi 2 -hTF can block both membrane binding and cellular uptake of 59 Fe-hTF into BeWo placental cancer cells (22). It is therefore now important to establish whether Bi 3ϩ binding to transferrin can occur under physiologically relevant conditions, especially in the presence of excess albumin and in blood plasma itself. We have shown previously that the order of lobe loading of hTF with metal ions can readily be determined via two-dimensional 1 H, 13 C NMR studies of recombinant [⑀-13 C]Met-hTF (23). It is known that the strength of binding to the two lobes is slightly different, and that Fe 3ϩ is primarily situated in N-lobe in serum (14,18).
Previous investigations of the interaction of Bi 3ϩ with serum albumin has led to the suggestion that albumin may be the major target for Bi 3ϩ in plasma (8), especially since albumin has a free thiolate group at Cys 34 . Human serum albumin, the most abundant protein in blood at a concentration of about 40 mg ml Ϫ1 (about 0.63 mM, Ͼ 10 times that of transferrin), is a single-chain 66.5-kDa protein, which is largely ␣-helical, and consists of three structurally homologous domains (24). It is the major transport protein for unesterified fatty acids, but is also capable of binding an extraordinarily diverse range of metabolites, drugs, organic compounds, and metal ions, e.g. Ca 2ϩ , Zn 2ϩ , Cu 2ϩ , and Ni 2ϩ (25,26).
In the present work, the binding of a bismuth antiulcer drug to human serum transferrin in aqueous solution in the presence of a large excess of albumin and to recombinant transferrin (N-lobe of transferrin and the mutant I132M of N-lobe of transferrin labeled with [⑀-13 C]Met) in intact blood plasma has been monitored directly under physiologically relevant conditions using 1 H, 13 C NMR spectroscopy. The introduction of Met 132 into the N-lobe provides a convenient monitor for metal binding since this residue occupies a similar site within helix 5 of the N-lobe and forms part of the hydrophobic patch around Trp 128 (Leu 122 -Trp 128 -Ile 132 ) as Met 464 in the C-lobe (Val 454 -Trp 460 -Met 464 ). The interaction of bismuth with human albumin was also studied. Surprisingly, we found that Bi 3ϩ still binds to the iron-binding sites of transferrin even in the presence of a large excess of albumin.

EXPERIMENTAL PROCEDURES
Materials-Recombinant N-lobe hTF/2N (residues 1-337) was expressed in baby hamster kidney cells using a pNUT plasmid with L-[⑀- 13 C]methionine in the growth medium, and purified as described previously (27,28). A gene for the mutant I132M protein was created by site-directed mutagenesis using previously published methods (27,29). Iron was removed from proteins by treatment with a metal-removal buffer containing 1 mM NTA, 1 mM EDTA, and 0.5 M sodium acetate, pH 4.9, using ultrafiltration Centricon 10 ultrafilters (Amicon). Human serum albumin (HSA) was purchased from Sigma as essentially globulin-free and fatty acid-free and was purified via ultrafiltration (Centricon 10) using 0.1 M KCl and washing 3 times (each 1 h). It was then lyophilized. Recombinant human albumin (rHA) was supplied by Delta Biotechnology Ltd. (batches GA 950202 and R970103). Samples of rHA were dialyzed against 100 mM ammonium bicarbonate, pH 7.9, and freeze-dried. Ranitidine bismuth citrate (RBC) and bismuth citrate [Bi(Hcit)] were provided by GlaxoWellcome plc. NaHCO 3 , KCl, 5,5Јdithiobis(2-nitrobenzoic acid), and other chemicals were purchased either from Aldrich or Sigma with the highest quality and used as received. Crystalline [Bi(NTA)] was synthesized according to a literature procedure (30), and had a satisfactory elemental analysis.
A 50 mM stock solution of [Bi(cit)] Ϫ was prepared by addition the minimum amount of ammonia solution to a suspension of [Bi(Hcit)] until the solution became clear. The final pH of this solution was about 7, and it was then diluted before use. A solution of Fe(NTA) 2 was prepared from an iron atomic absorption standard (1000 ppm in 1% HNO 3 , Aldrich) and 2 mol eq of H 3 NTA (Aldrich), followed by pH adjustment to between 6.0 and 7.0. This was lyophilized and redissolved in D 2 O before use. A [Bi(NTA)] solution was prepared by dissolving a known amount of [Bi(NTA)] in D 2 O.
Preparation of NMR Samples-I132M [⑀-13 C]Met hTF/2N (0.15 mM) and mixtures with either rHA (1.8 mM) or HSA (1.8 mM) were prepared in D 2 O containing 0.1 M KCl. Prior to Bi 3ϩ or Fe 3ϩ titrations, an aliquot of concentrated NaHCO 3 (0.25 M) was added to the samples to give a final concentration of 10 mM, and pH* values were adjusted using NaOD or DCl. For the intact blood plasma experiments, blood from a male healthy volunteer was collected by venipuncture into lithiumheparinized vacutainers, and the plasma was separated by centrifugation at 6000 rmp for 20 min at 277 K, and stored frozen until used for NMR measurements. I132M hTF/2N was added to 1.2 ml of intact blood plasma to give a hTF/2N concentration of 50 M and the concentration of this sample was doubled by freeze-drying and reconstitution in 0.6 ml of D 2 O, followed by addition of concentrated NaHCO 3 to a final concentration of 20 mM. The pH* was then measured. After addition of Bi 3ϩ , samples were left to equilibrate for at least 30 min at 310 K. All experiments on reconstituted blood plasma were carried out at pH* 7.8 since it was possible to maintain this as a stable pH* value during the course of the long NMR data accumulations. If the initial pH* value was lower, it tended to drift upwards during the experiment. The 1 H, 13 C chemical shifts of the [⑀- 13 C]Met resonances of transferrin were insensitive to pH over the range of 7-8.8, and NMR experiments on Bi 3ϩ loading of transferrin with [Bi(NTA)] and ranitidine bismuth citrate gave the same results at pH* 7.4 and pH* 7.8.
NMR Spectroscopy-NMR spectra were recorded on a Bruker DMX500 spectrometer at 310 K. For one-dimensional 1 H NMR spectra, 400 to 1200 transients were acquired with 6-s (50 o ) pulses and 16,384 data points during the 2-s pulse delay and the water resonance was suppressed via presaturation. For two-dimensional 1 H, 13 C heteronuclear single quantum coherence spectra (HSQC) experiments, the sequence was optimized for 1 J ( 1 H-13 C) ϭ 140 Hz, and 16 to 32 tran-sients were acquired using 2,048 data points in the f 2 dimension ( 1 H), 32 to 64 increments in the f 1 dimension, 13 C frequency width of 3 kHz, and relaxation delay of 1.6 s. The 13 C spins were decoupled using the GARP sequence (31). After zero-filling to 4,096 ϫ 1,024 data points, unshifted Gaussian functions were used for processing in both dimensions. Water suppression was achieved by a combination of presaturation and pulsed-field gradients. One-dimensional HSQC NMR spectra (or 13 Cedited 1 H NMR spectra) were recorded using the first increment of the two-dimensional HSQC sequence. Resolution enhancement was achieved using a combination of exponential (1.5 to 10 Hz line-broadening) and unshifted sine-bell functions (32). Peaks were referenced to sodium 3-(trimethylsilyl)propionate-2,2,3,3-d 4 via the external ⑀-CH 3 peak of L-methionine (15.14 ppm) for 13 C and via formate (8.465 ppm, a minor impurity always present in the protein) for 1 H.
Determination of Bi-HSA (concentration about 0.1 mM) was incubated with different mole ratios of RBC in 0.1 M Tris-HCl buffer, pH 7.4, overnight at 310 K. Unbound RBC was separated via gel filtration using a Superose 12 column and a FPLC system (Amersham Pharmacia Biotech). The concentration of albumin samples was about 6 mg/ml and 500 l of the protein solution was loaded onto the column. Elution conditions were 0.1 M Tris-HCl, pH 7.4, and flow rate 0.5 ml/min. The fractions eluting from 10 to 14 ml were collected and the bismuth content was determined using a CETAC Microneb 2000 direct injection nebulizer (CETAC Technologies, Omaha, NE) coupled with a VG Plas-maQuad PQ2 ICP-MS instrument (VG Elemental, Winsford, Cheshire, UK). Details of the optimization procedure for the DIN-ICP-MS system and measurement conditions for bismuth have been described previously (33).

NMR Studies of Bi 3ϩ Binding to I132M hTF/2N and Comparison with Fe 3ϩ
One-dimensional 1 H NMR-It has been shown previously that 1 H NMR spectra of human serum transferrin are complicated by the overlap of the very large number of resonances present and by their broadening due to the slow tumbling of this 80-kDa protein (36). However, the high-field region of the spectrum of the N-lobe is relatively well resolved. High fieldshifted resonances have been assigned to protons from residues around Trp 128 , i.e. Leu 122 , and Ile 132 (37,38). The mutation of Ile 132 to Met should lead to the disappearance of the resonance at Ϫ0.603 ppm, which has been previously assigned as ␥CH 2 of Ile 132 . Indeed, this was found to be the case (Fig. 1). Other changes were also observed in the spectrum of the mutant in comparison to that of wild-type hTF/2N. For example, the resonance at Ϫ0.170 ppm for hTF/2N disappeared, and the peak for ␦CH 3 of Leu 122 (Ϫ0.339 ppm in hTF/2N) shifted to Ϫ0.324 ppm. Addition of 0.5 mol eq of Bi 3ϩ (as [Bi(NTA)]) caused new peaks to appear at Ϫ0.160 ppm (CЈ), and Ϫ0.255 ppm (BЈ), and further addition of Bi 3ϩ (total 1.0 mol eq) increased the intensity of both CЈ and BЈ but decreased that of peak B significantly (Fig. 1A).
For comparison, titrations of Fe 3ϩ (added as Fe(NTA) 2 ) with the mutant protein were also performed under similar conditions (pH* 7.8, 310 K and 10 mM bicarbonate) and the results are shown in Fig. 1B. Upon addition of 0.5 mol eq of Fe 3ϩ , the resonance at Ϫ0.324 ppm decreased in intensity, and almost completely disappeared with 0.9 mol eq Fe 3ϩ present. Broad new peaks at Ϫ0.004 and Ϫ0.392 ppm appeared and increased in intensity. It is reasonable to assume from these titration studies that the resonance at Ϫ0.324 ppm consists of two overlapped peaks, one of which (peak B) can be assigned to the ␦CH 3 of Leu 122 as judged from the change in pattern of this peak on titration of both wild-type and mutant proteins with metal ions. The other peak (C) cannot be assigned. Two-dimensional total correlation spectroscopy and NOESY experiments support these assignments (data not shown). Peak B has identical associated NOESY cross-peaks for both wild-type and mutant hTF/2N, which suggests that peak B belongs to the ␦CH 3 of Leu 122 . Similar cross-peak patterns were observed for B and BЈ in the NOESY spectrum, which indicates that peak BЈ for the metal-bound protein is the analogue of peak B. A comparison of the changes in the chemical shifts of the ␦CH 3 peak of Leu 122 after binding of hTF/2N and I132M-hTF/2N to Bi 3ϩ and Fe 3ϩ is shown in Fig. 1C.
Two-dimensional 1 H, 13 C HSQC NMR-The two-dimensional 1 H, 13 C HSQC NMR spectrum of I132M hTF/2N and after addition of 0.5 mol eq of Bi 3ϩ or Fe 3ϩ , in the presence of 10 mM bicarbonate, are compared in Fig. 2. As expected, six cross-peaks were observed for the apo-protein, five of which have similar chemical shifts as those observed for the wild-type protein. These have been assigned previously on the basis of single-site mutations combined with other considerations (39,40). Therefore the sixth peak, at 1.51/17.98 ppm, can be assigned to Met 132 . On addition of 0.5 mol eq of Bi 3ϩ (added as either RBC or as [Bi(NTA)]), a notable decrease in intensity of the latter peak occurred and a new peak appeared at 1.45/17.93 ppm, which can be associated with the bound form of the protein. Other new peaks, which can be assigned to Met 109 and Met 309 , also appeared and shifted slightly in both 1 H and 13 C dimensions (Table I). Further addition of Bi 3ϩ (1.0 mol eq) caused the disappearance of the peaks at 1.51/17.98, 1.94/ 16.15, and 2.15/16.16 ppm and led to a further increase in the intensity of the new peaks. These changes were observed more clearly in high-resolution one-dimensional 13 C-edited 1 H spectra (see below, Fig. 3).
Addition of 0.5 mol eq of Fe 3ϩ (added as [Fe(NTA)] 2 ) to I132M hTF/2N caused similar changes in the two-dimensional HSQC spectrum as for 0.5 mol eq Bi 3ϩ , except that the peak for Met 132 for the bound form was significantly broadened. The resonance for Met 132 in the apo-protein disappeared after addition of 1.0 mol eq of Fe 3ϩ . The changes to the shifts of the other Met resonances were identical to those observed on addition of 1.0 mol eq of Bi 3ϩ .
Binding of Bi 3ϩ to I132M hTF/2N in the Presence of Excess of Albumin-The 1 H and 13 C NMR chemical shift changes induced by metal ions (e.g. Fe 3ϩ and Bi 3ϩ ) provide convenient probes for investigation of Bi 3ϩ translocation between transferrin and proteins such as albumin. These experiments were performed using low concentrations of I132M hTF/2N (150 M) in the presence of 12 mol eq of HSA or rHA (1.8 mM), pH* 7.8, 10 mM bicarbonate, 310 K. We choose an [albumin]/[transferrin] ratio of 12:1 to mimic biological conditions. The concentration of albumin in blood plasma (about 0.63 mM) is about 18 times higher than that of transferrin (about 35 M), but hTF is only about 30% saturated with Fe 3ϩ . The 1 H, 13 C two-dimensional HSQC NMR spectrum of this protein mixture shows sharp resonances from the six labeled Met residues of transferrin and broadened (natural abundance) resonances from albumin. The Met 132 peak was overlapped with peaks from albumin (data not shown). Since I132M hTF/2N is present at low concentration, one-dimensional 13 C-edited 1 H NMR spectra were recorded over a period of 30 min each. Fig. 3 shows the  10 ppm from albumin were filtered out with resolution enhancement using a combination of unshifted sine-bell and exponential functions prior to Fourier transformation. The antiulcer compound, ranitidine bismuth citrate, was added in 0.25 mol eq steps to I132M hTF/2N in the presence of 12 mol eq of rHA, 10 mM bicarbonate. With increase in Bi 3ϩ concentration, a new peak at 1.45 ppm appeared. This can be assigned to Met 132 in Bi-I132M hTF/2N, and it gradually increased in intensity, reaching a maximum after 1 mol eq of Bi 3ϩ had been added. The peak for Met 132 in the apo-protein simultaneously decreased in intensity and finally disappeared (Fig. 3). The change in the peak for Met 309 (2.15 ppm) was observable more clearly in one-dimensional 1 H{ 13 C} spectra: it gradually decreased in intensity with the addition of Bi 3ϩ and finally disappeared. The resonance for Met 309 is probably that at 2.14 ppm, which was overlapped with Met 256 , as can be judged from the increased intensity in the two-dimensional HSQC spectrum (data not shown). The peak for Met 109 also appears to shift to low field by 0.03 ppm. The same chemical shift changes were observed after addition of Bi 3ϩ to I132M hTF/2N in the presence of 12 mol eq of HSA. The lack of effect of albumin on Bi 3ϩ binding to this transferrin N-lobe is clearly illustrated in Fig. 3C, which shows the integrated intensity of the peak for Met 132 after addition of various amounts of Bi 3ϩ in the absence and presence of albumin.
Since the concentration of albumin present in the sample is much higher than that of transferrin, the 1 H NMR spectrum is dominated by peaks from albumin. The aromatic region of spectra of these mixed protein solutions (hTF/2N ϩ 12rHA or ϩ 12HSA) was almost identical with and without addition of Bi 3ϩ , particularly the resonance at 7.632 ppm, which has been previously assigned (41,42) to His 3 of albumin (data not shown).

Uptake of Bi 3ϩ by Transferrin in Plasma
The concentration of transferrin in human plasma is about 35 M. It is only about 30% saturated with iron (18) and therefore has about 50 M capacity for binding to other metal ions. To determine if transferrin is a target for bismuth, isotopically labeled [⑀-13 C]Met I132M hTF/2N (50 M) was directly added to human plasma. The whole plasma concentration (including the added transferrin) was lyophilized, and the sample was redissolved in half-volume of the original plasma solution. This gave an I132M hTF/2N concentration of 100 M.
Even with resolution enhancement, the peak for Met 132 was still overlapped with other peaks in the 13 C-edited 1 H NMR spectrum (data not shown). Therefore only the two-dimensional HSQC method was used. The two-dimensional HSQC spectrum of this solution containing 100 mM KCl and 20 mM bicarbonate is shown in Fig. 4. Surprisingly, the peak for Met 109 became severely broadened but the rest of the Met cross-peaks from transferrin were clearly observed. Many other cross-peaks are present but are difficult to assign, partly due to the limited frequency width used (12 ppm in 13 C dimension). The peaks at about 1.46/17.2 ppm, and 1.67/15.8 ppm (folded in 13 C dimension) can be assigned to Ala and Lys residues, respectively, of albumin, and the peaks at about 1.24/19.2 ppm to lipids in plasma. After addition of 0.5 mol eq of RBC (relative to the available transferrin-binding sites), the peak for Met 132 (1.51/ 17.98 ppm) in apo-hTF/2N decreased in intensity and the peak at 1.45/17.93 ppm for Bi 3ϩ -I132M hTF/2N increased in intensity. Similarly, the peak for Met 109 (1.94/16.15 ppm) disappeared and a new peak (bound form) appeared at slightly lower field, and that for Met 309 (2.15/16.16 ppm) shifted to high field. The cross-peak for Met 132 in the apo-protein almost disappeared and the analogous peak for the bound-form further increased in intensity (Fig. 4). After addition of 1.0 mol eq of RBC, this behavior was similar to that observed for I132M hTF/2N with and without 12 mol eq of serum albumin or recombinant albumin under same conditions. Interestingly, with Bi 3ϩ bound to the protein, the peak for Met 26 became sharper and observable.
A second experiment was carried out with blood plasma containing twice the concentration of I132M (200 M), but the behavior of the Met two-dimensional cross-peaks on titration with RBC was similar. The normal one-dimensional 1 2N. A, before; B, after addition of Bi 3ϩ (ranitidine bismuth citrate); and C, slices through two-dimensional 1 H, 13 C HSQC spectra in the 1 H dimension (corresponding to the 13 C signal of ϪSCH 3 of Met 132 ). The shift of the peak for Met 132 , which is in the hydrophobic patch of helix 5, is similar to that observed after direct addition of Bi 3ϩ to I132M apo-hTF/2N (Fig.  2). spectrum of plasma in the His region was identical in the absence and presence of Bi 3ϩ , especially the peak for His 3 of albumin (7.632 ppm), which has previously been used as an indicator for drug binding at Cys 34 (41).
The intact protein as [⑀-13 C]Met-transferrin was also added directly to 1.2 ml of human plasma to give a concentration of 35 M, and the concentration of the sample was doubled by freezedrying and dissolving in 0.6 ml of D 2 O in the presence of 20 mM sodium bicarbonate. Most of the Met resonances were observable in the two-dimensional HSQC spectrum except those for Met 26 , Met 309 , and Met 389 (Fig. 5). Other peaks from plasma were also observed from groups present at relatively high concentrations such as Lys (from albumin) and lipids. The notable change was for Met 464 in the C-lobe from its apo-position (1.38/ 16.30 ppm for 1 H/ 13 C) to Bi 3ϩ bound form (1.18/18.2 ppm for 1 H/ 13 C) after addition of 1 mol eq of ranitidine bismuth citrate. A similar change also occurred for Met 499 in the C-lobe.

Interactions of Bismuth Complexes with Albumin
Effect of Bismuth on the Free Thiol Content of Albumin-The free thiol of albumin at Cys 34 is a potentially strong binding site for Bi 3ϩ . The thiol contents of human serum albumin and recombinant human albumin were determined before and after reaction with bismuth citrate (either RBC or [Bi(cit)] Ϫ ) by the 5,5Ј-dithiobis(2-nitrobenzoic acid) method. The rHA (recombinant) sample contained 0.77 Ϯ 0.01 mol of thiol/mol of protein, while thiol content of (isolated) HSA was significantly lower, only 0.29 Ϯ 0.01 mol/mol HSA. After reaction with various amounts of bismuth citrate in 0.1 M Tris-HCl buffer at pH 7.4 for 12 h, the SH contents decreased by less than 12%, from 0.77 to 0.68 for rHA and from 0.29 to 0.26 for HSA, respectively. This suggests that little Bi 3ϩ binds to Cys 34 of albumin.
Determination of Amount of Bismuth Bound to Human Serum Albumin-Various mole ratios of ranitidine bismuth citrate were reacted with albumin in 0.1 M Tris-HCl buffer at pH 7.4 and equilibrated overnight at 310 K. Albumin-bound bismuth was then separated from free bismuth by gel filtration chromatography. The Bi 3ϩ content of the albumin fractions was measured by DIN-ICP-MS (data not shown). The amount of Bi 3ϩ bound to albumin increased almost linearly with increase in added RBC and did not reach saturation even with 25 mol eq of RBC present. The gel filtration chromatograms of control albumin and its complex with bismuth were very similar both in terms of peak intensity and retention time (data not shown) suggesting that bismuth does not cause aggregation of the protein. When 40 mol eq of glutathione (relative to the measured Bi 3ϩ ) was added to the albumin fraction, a new broad band centred at about 350 nm gradually increased in intensity in a multiphase process, and reached a maximum intensity over a period of 3 h (Fig. 6). This is in contrast to the reaction of bismuth citrate alone with 40 mol eq of glutathione under similar conditions which was complete within minutes. The band at 350 nm is a typical Bi-S absorbance indicating formation of [Bi(SG) 3 ] (9).

DISCUSSION
Bismuth compounds are widely used as antiulcer drugs and recently we have shown (22) that bismuth transferrin, Bi 2 -hTF, exhibits marked dose-dependent effects on membrane binding and cell uptake of 59 Fe-hTF by placental BeWo cells. This suggested that bismuth transferrin is recognized by the transferrin receptor. The present study was undertaken to determine whether Bi 3ϩ can bind to transferrin under physiological conditions, especially in the presence of excess albumin, and in blood plasma itself. Previously we have shown that NMR can be used to monitor the uptake of metals into the individual lobes of transferrin (23). The 1 H, 13 C NMR cross-peak for Met 464 of human transferrin is a sensitive indicator of metal binding to the C-lobe since significant chemical shift changes are induced in both 1 H and 13 C dimensions. In the N-lobe of intact hTF, however, there is lack of this kind of sensitive indicator. Met 464 is situated in the hydrophobic patch (Val 454 -Trp 460 -Met 464 ) of helix 5 in the C-lobe (Fig. 7), which backs onto the metalbinding site and H-bonds to the synergistic anion (43,44). In the N-lobe there is a similar hydrophobic patch in helix 5 near the metal-binding site, consisting of Leu 122 , Trp 128 , and Ile 132 (Fig. 7) (45). The analogue of Met 464 is Ile 132 in human serum transferrin, but is Met 132 in cow and pig transferrin (46), which suggests that I132M is a structurally conservative substitution. To provide a possible sensitive indicator for metal ion binding in the N-lobe of human serum transferrin and to investigate the similarity between the two lobes of transferrin, Ile 132 was mutated to Met using site-directed mutagenesis. It is easy to produce N-lobe protein in this way in the quantities required for NMR. In contrast, recombinant C-lobe is difficult to prepare, but the N-lobe and C-lobe metal binding constants are usually close (47).
The 1 H NMR spectrum of I132M hTF/2N was similar to that of wild-type hTF/2N in the both high-field and His C2H regions, except for the disappearance of the peak for the ␥CH 2 of Ile 132 at Ϫ0.603 ppm. Both Bi 3ϩ and Fe 3ϩ induce similar chemical shift changes for the high field-shifted peak for ␦CH 3 of Leu 122 in the mutant and wild-type hTF/2N (Fig. 1). This suggests that the overall structure of the mutant is similar to that of wild-type hTF/2N. This was also confirmed by molecular modeling, which showed that the protein backbone fold of the mutant is almost identical to that of the wild-type protein. Six of the 10 lowest energy structures placed the side chain of Met 132 above Trp 128 (data not shown), a situation which would give rise to a ring current shift for the ⑀-CH 3 of Met 132 .
In the two-dimensional 1 H, 13 C NMR spectrum of apo-I132M hTF/2N, the ⑀-13 CH 3 resonance of Met 132 exhibits a significant 1 H NMR high-field shift compared with the rest of the Met peaks, as does the analogous cross-peak for Met 464 in the Clobe. Only small changes in shifts of the Met 132 resonance (⌬␦ Ϫ0.06/Ϫ0.05 ppm for 1 H and 13 C, respectively, Table I) occurred when Bi 3ϩ or Fe 3ϩ binds to the mutant I132M hTF/2N, in contrast to the large shifts for Met 464 (⌬␦ Ϫ0.20/1.90 ppm for 1 H and 13 C, respectively) suggesting that the structural changes in helix 5 on loading the protein with metal ions are slightly different for the N-and C-lobes. X-ray crystallographic studies have shown that when metals bind and domain closure occurs, helix 5 pivots on helix 11 and that a domain movement of about 54°occurs in the N-lobe but only about 15 o rotation in the C-lobe (48,49).
Our studies suggest, for the first time, that transferrin should be considered as a potential mediator for bismuth transport in blood plasma. Previously, it has been assumed (8) that albumin, the most abundant protein in blood serum with a free thiol group at Cys 34 , is a target site for bismuth drugs, since Bi 3ϩ is known to have a high affinity for thiolate sulfur. Glutathione, a thiolate sulfur-containing peptide (GSH), for example, can readily displace Bi 3ϩ from its complexes with citrate and EDTA at biological pH values (9). Recent reports (8) have shown that only 2% of albumin molecules bind to Bi 3ϩ if binding is assumed to occur at the free thiol group of Cys 34 (pK a about 5 (50)). In this work we have demonstrated that binding of Bi 3ϩ to albumin is nonspecific; even 25 mol eq of Bi 3ϩ did not saturate albumin, and Cys 34 is not blocked by Bi 3ϩ binding. Previous 1 H NMR studies of albumin have shown that the imidazole CH resonances of His 3 are sensitive to the oxidation of Cys 34 and to the formation of adducts with gold antiarthritic drugs (41) probably because such reactions lead to movement of the side chain of Cys 34 which is communicated to His 3 via intervening helices. The His regions of 1 H NMR spectra of albumin in the presence of I132M hTF/2N or of blood plasma in the presence of intact hTF were found to be almost identical after addition of bismuth compounds (data not shown) which provide further evidence that Cys 34 is not a major binding site for Bi 3ϩ .
We have successfully used two-dimensional HSQC NMR spectroscopy to probe changes of Met resonances of transferrin in solution in which the concentration of albumin is 10 times higher. The observation of similar changes in the chemical shifts of the Met resides of I132M hTF/2N on binding Bi 3ϩ in the presence or absence of a large excess of albumin, and even in blood plasma, suggests that similar conformational changes are induced by Bi 3ϩ under these conditions. Such structural changes could be important for recognition by the transferrin receptor. Bi 3ϩ was also observed to bind to intact transferrin in the presence of a large excess (12 mol eq) of serum albumin or recombinant albumin and a similar behavior was observed in blood plasma. Our findings may have implications for the mechanism of neurotoxicity of bismuth drugs (encephalopathy). For a long time it has not been clear how bismuth is transferred to the brain. It is generally accepted that the diagnosis of bismuth encephalopathy can be confirmed by the detection of high Bi 3ϩ levels in whole blood, serum, or plasma, the so-called Hillemand safety level (6). It is likely that once bismuth has entered into blood it is transported by transferrin, in a similar manner to Al 3ϩ . Al 3ϩ deposition in the brain is known to cause dialysis encephalopathy and this neurotoxicity is related to transferrin transportation and transferrin receptor recognition in the brain (51). Selective labeling of the protein in combination with inverse NMR detection is a powerful method for probing the structure and dynamics of high molecular mass proteins, and provides an approach for investigating the translocation of metallo-drugs (and other drugs) between proteins and enzymes at concentrations of biologically relevance without separation, and can also be applied to protein-ligand (in this case for drug screening) (52) and protein-protein interactions.