Unexpectedly strong binding of a large metal ion (Bi3+) to human serum transferrin.

Large metal ions (>0.9 A ionic radius) have previously been found to bind only weakly to human serum transferrin (hTF, 80 kDa), presumably because the interdomain cleft cannot close around the metal and synergistic anion. Surprisingly, therefore, we report that Bi3+ (ionic radius 1.03 A), a metal ion widely used in anti-ulcer drugs, binds strongly to both the N- and C-lobes with log K1* = 19.42 and log K2* = 18.58 (10 mM Hepes, 5 mM bicarbonate, 310 K). The uptake of Bi3+ by apo-hTF from bismuth citrate complexes is very slow (hours), whereas that from bismuth nitrilotriacetate is rapid (minutes). Evidence from absorption and NMR spectroscopy is presented to show that Bi3+ binds to the specific Fe3+ binding sites along with carbonate as the synergistic anion. Under the conditions used, preferential binding of Bi3+ to the C-lobe of hTF is observed. Linear free energy relationships show that there is a strong correlation between the strength of binding of Bi3+ and Fe3+ to a wide variety of ligands which include transferrin. Therefore we conclude that the strength of metal ion binding to transferrin is determined more by the ligand donor set than by the size of the ion.

Transferrin (80 kDa) is a serum iron transport glycoprotein with a concentration in blood of ϳ2-4 mg/ml. Its normal function is to carry iron as Fe 3ϩ between sites of uptake, utilization, and storage (1)(2)(3)(4). It contains two specific Fe 3ϩ binding sites per molecule, each approximately octahedral and consisting of two tyrosinates, one histidine, one aspartate, and a bidentate carbonate anion (the synergistic anion) derived from the buffer (5)(6)(7). Iron is situated in N-lobe and C-lobe interdomain clefts, which close around the metal and synergistic anion (4). Apo-hTF 1 is known to bind to a wide variety of metal ions. The strength of binding appears to be dependent on the size of the metal ion, being optimum for Fe 3ϩ (ionic radius 0.65 Å), weaker for either slightly smaller, e.g. Ga 3ϩ (0.62 Å), or slightly larger, e.g. In 3ϩ (0.80 Å), ions and much weaker for very small, e.g. Al 3ϩ (0.54 Å), or very large, e.g. lanthanide (0.86 -1.03 Å) ions (Table I). These data appear to indicate that strong binding arises from the matching of the ionic radius to the size of the binding cleft (4). In the case of the larger ions, the interdomain cleft may not close at all.
Since transferrin is only about 30% saturated with iron in normal serum (8), there is substantial binding capacity for other metal ions that enter the blood. Thus transferrin is an important serum transport agent for metal ions of therapeutic, diagnostic or toxic importance including Al 3ϩ (9) Ga 3ϩ , In 3ϩ (10), and Ru 3ϩ (11).
There are no previous reports of the binding of Bi 3ϩ to transferrin, although such binding might be expected to be weak on account of the large size of Bi 3ϩ (ionic radius 1.03 Å, log K ϳ 5; predicted from Table I). Bismuth compounds have been used in medicine for more than 200 years for a variety of gastrointestinal disorders (12,13). There is particular interest in bismuth(III) citrate solubilized by ammonium and potassium hydroxide, which forms the basis of the colloidal bismuth subcitrate present, for example, in the drugs Telen ® (Byk Gulden) and De-Nol ® (Gist Brocade) (14,15). Recently a new adduct of ranitidine with bismuth citrate (Glaxo Wellcome plc), which combines the antisecretory action of ranitidine with mucosal protectant and bactericidal properties of bismuth(III), has been granted a product license in the UK (16,17). There is also interest in using compounds containing radioactive bismuth isotopes as targeted radiotherapeutic agents (18). Despite this medicinal interest, the speciation of bismuth in blood plasma, in particular the binding of Bi 3ϩ to blood plasma proteins, is poorly understood.
In this paper we report the first detection of the binding of Bi 3ϩ to serum transferrin (19). We determine the binding constants via electronic absorption spectroscopy and show, by absorption and NMR spectroscopy, that Bi 3ϩ is taken up into the specific iron sites of transferrin and is accompanied by concomitant binding of the synergistic anion carbonate. We also show that there is preferential uptake into the C-lobe site. Linear free energy plots correlating the strength of binding of Bi 3ϩ and Fe 3ϩ to a wide variety of ligands provide an insight into the reasons why Bi 3ϩ binds tightly to transferrin.

EXPERIMENTAL PROCEDURES
Materials-Apo-hTF was purchased from Sigma (catalog no. T0519) and was washed three times with 0.1 M KCl to remove low molecular mass impurities using Centricon 30 ultrafilters (Amicon). [Bi(Hcit)] was provided by Glaxo Wellcome plc. Crystalline [Bi(NTA)] was synthesized according to a literature procedure (20) and had a satisfactory elemental analysis. NaHCO 3 (Aldrich), NaH 13 CO 3 (99% enriched 13  [Bi(NTA)] in water (pH ϳ4.7) by adding the appropriate amount of NTA. The bismuth concentration was determined by inductively coupled-plasma atomic emission spectroscopy (Perkin Elmer).
A solution of [Fe(NTA) 2 ] was prepared from an iron atomic absorption standard solution (1000 ppm in 1% HNO 3 , Aldrich) and 2 mol eq of NTA. The pH was gently raised to 5-6 with microliter amounts of NaOH (1 M). This solution was then further diluted to give the required [Fe(NTA) 2 ] concentration in 10 mM Hepes buffer.
Preparation of NMR Samples-To allow H-D exchange, apo-hTF was dissolved in 0.1 M KCl in D 2 O, and after adjustment of the pH* to 7.25, was left for 4 h at ambient temperature. After lyophilization, the sample was redissolved in D 2 O containing 10 mM NaHCO 3 . The pH* was then readjusted to 7.25 Ϯ 0.05, when necessary, using NaOD and DCl (0.2 M). The pH* values of NMR solutions were recorded before and after NMR measurements.
Samples for UV Spectroscopy-These were prepared by diluting aliquots of a stock apo-hTF solution to ϳ10 Ϫ5 M with 10 mM Hepes buffer, pH 7.4. The concentration of apo-hTF was determined spectrophotometrically on the basis of ⑀ 280 93,000 M Ϫ1 cm Ϫ1 (21). Immediately before adding Bi 3ϩ , an aliquot of a concentrated solution of NaHCO 3 (0.25 M) was added to give a 5 mM concentration of HCO 3 Ϫ . For kinetic experiments, a solution containing 10 mM Hepes buffer was used for background correction. The spectrum of the apo-hTF solution (containing 5 mM bicarbonate) was recorded after equilibration at 310 K for 10 min. Aliquots of the stock solution of bismuth (usually 5-20 l of [Bi(Hcit)] or [Bi(NTA)] solutions) were then added to the apo-hTF cuvette, and the spectrum was recorded at various time intervals. For titration experiments (at equilibrium), a similar procedure was used. Each solution was allowed to equilibrate at 310 K for at least 30 min before the spectrum was recorded. The dependence of the stability of bismuth hTF on pH was determined by varying the pH of a solution containing 3 mol eq of [Bi(NTA)] and 10 Ϫ5 M hTF in 5 mM NaHCO 3 and recording the UV spectrum after equilibration for 1 h.
NMR Spectroscopy-1 H NMR spectra were recorded on JEOL GSX500 or Varian Unity 600 spectrometers at 500 and 600 MHz, respectively. Spectra were acquired using 0.7 ml of solution in 5 mm tubes at 310 K, using 400-1200 transients, 6-s (50°) pulses, recycle time 2 s, 16,384 data points, and water suppression via presaturation. The chemical shift reference was TSP via endogenous formate (8.465 ppm, pH* Ͼ 7), which was always present in our samples as a minor impurity. Resolution enhancement of the spectra was achieved by processing the FIDs with a combination of unshifted sine-bell and exponential functions (line broadening of 1.5-5 Hz) on a Silicon Graphics computer using Varian VNMR software. A two-dimensional TOCSY spectrum (mixing time 60 ms) of hTF at pH* 7.25 was acquired using 2,048 data points in the f 2 dimension, acquisition time 0.12 s, 48 scans, and 256 increments in the f 1 dimension, in a total time of 14 h.
Proton-coupled 13 C NMR spectra were recorded on a JEOL GSX 500 instrument operating at 125 MHz. Typically 10,000 -30,000 scans were collected using a pulsewidth of 6 s (50°), relaxation delay 2 s, 16,384 data points. The spectra were processed using exponential functions (line-broadening of 5-8 Hz). The chemical shift reference was TSP via added internal dioxan (67.4 ppm).
Electronic Absorption Spectroscopy-This was performed on a computer-controlled Perkin-Elmer Lambda 16 spectrometer. The temperature of the cells was maintained at 310 Ϯ 0.1 K using a PTP-1 Peltier temperature programmer.
pH Measurements-These were made using a micro-combination electrode (Aldrich) and a Corning 145 pH meter calibrated with standard buffers at pH 4, 7, and 10.
Calculations-Equilibrium titration curves were initially fitted using the modified Hill plot (22), followed by an iterative procedure similar to that used by Harris and Pecoraro (23) to recalculate K 2 . Briefly this involves evaluating the concentrations of free Bi 3ϩ , free NTA, and apo-hTF by iteration to minimize differences between calculated and experimentally fixed values of total bismuth, total NTA, and total apo-hTF, assuming a set of values of K 1 (from modified Hill plot) and K 2 for bismuth binding. Values of K 2 were varied to minimize the difference between observed and calculated ⌬⑀ values. The calculations were carried out using the program Kaleidagraph (Synergy Software).

Electronic Absorption Spectroscopy
The complexation of metal ions to the phenolic groups of the tyrosine residues in the specific metal-binding sites of apo-hTF leads to the production of two new absorption bands at 241 and 295 nm (2). These new bands are readily apparent in difference UV spectra of metal-transferrins and apo-hTF. Fig. 1 shows time-dependent difference UV spectra of apo-hTF in the presence of 2 mol eq of [Bi(Hcit)] at 310 K in 10 mM Hepes buffer, pH 7.4, 5 mM bicarbonate. The two new absorbance bands that appear at 241 and 295 nm are similar to those reported previously for binding of other trivalent metal ions to apotransferrin (2). The variation in absorbance at 241 nm with time, together with data for reactions involving 1 and 4 mol eq of [Bi(Hcit)], are shown in Fig. 2 .The time dependence of the absorbance at 295 nm is similar. It can be clearly seen that the absorbance increases rapidly within the first 10 min, followed by a slower second phase. With 2 mol eq of [Bi(Hcit)] present, about 50% of total bismuth binding occurs within the first fast phase, and within the next 10 h the remaining 50% binds. With increasing bismuth concentration, the rate of the first phase increases, whereas that of the slower step remains almost the same. Detailed experiments that might allow a full kinetic analysis were not attempted. Under similar conditions, the reaction of apo-hTF with [Bi(NTA)] also gave rise to the same absorbance bands at 241 and 295 nm, but at a much faster rate than that observed for [Bi(Hcit)]. Complete reaction of hTF with 1 or 2 mol eq of [Bi(NTA)] occurred within 30 min (data not shown).

Equilibrium Studies
Apo-hTF was titrated with [Bi(NTA) x ] (x ϭ 1, 8, or 20), and the absorbance at 241 nm was monitored. A set of titration curves for NTA:Bi ratios of 1:1 to 20:1 is shown in Fig. 3. For NTA:Bi ϭ 1:1, ⌬⑀ increases linearly with increase in Bi:hTF ratio (r) until a value of r ϭ ϳ1 is reached, showing that all the added bismuth is bound to apo-hTF at low values of r. Beyond r ϭ 1, the absorbance increases further but the plot curves downward indicative of the occupation of a second site with a lower binding constant and of competition between free NTA and hTF. With increasing NTA:Bi ratios (8:1 and 20:1), the plots (Fig. 3) show more pronounced downward curvature at high r values due to the competition between hTF and NTA.
The slope of the initial linear portion of these curves is 21,900 (Ϯ 650) M Ϫ1 cm Ϫ1 and can be equated to the molar absorptivity of transferrin with one site saturated with Bi 3ϩ (⌬⑀ 1 ). Hence if the two sites are equivalent, a ⌬⑀ value of ϳ43,800 M Ϫ1 cm Ϫ1 would be expected when both sites are filled. These values are compared to those of other metal-transferrin complexes in Table I.
The absorbance data obtained at different molar ratios of NTA:Bi were used to calculate the binding constants for bismuth-transferrin complexes, using Equations 1 and 2 below.
The relationships between the equilibrium constants K a1 and K a2 and the stability constants for bismuth transferrin (K 1 and K 2 ) and bismuth NTA are given by the following equations.
Bi ϩ hTF ϭ Bi Ϫ hTF If it is assumed that the two binding sites for bismuth on hTF are independent and equivalent, then where the fractional saturation Y ϭ [Bi-bound]/[hTF] total , n ϭ average number of bismuth ions bound per transferrin molecule, K a is the intrinsic binding constant, and K a1 ϭ 2 K a and K a2 ϭ K a /2. For 20:1 NTA:Bi titration curve, the maximum ⌬⑀ observed (ϳ18,000) never exceeded the calculated molar absorptivity (⌬⑀ 1 ϭ 21,900 M Ϫ1 cm Ϫ1 ) (Fig. 3), even at r ϭ 2.5. This suggested that only one Bi 3ϩ ion binds to hTF under these conditions, and so these data were used to calculate K a1 . The slope of the plot of 1/Y versus [NTA]/[Bi(NTA)] gave log K a1 ϭ 1.88 Ϯ 0.02 (correlation coefficient r ϭ 0.992, n ϭ 1.01).
From the 8:1 NTA:Bi titration curve, it was possible to use a similar plot to calculate log K a1 ϭ 1.87 Ϯ 0.02 (in agreement with that obtained from the 1:20 titration curve) and log K a2 ϭ 1.28 Ϯ 0.02 (r ϭ 0.99, n ϭ 1.99).
Using the known value (24) for the stability constant of [Bi(NTA)] of log K Bi(NTA) ϭ17.75 and Equations 3 and 4, pHindependent binding constants of log K 1 ϭ 19.63 and log K 2 ϭ 19.03 were calculated for bismuth hTF. However, there is good evidence that the two binding sites are not in fact equivalent (see below), and therefore the second binding constant was recalculated using the mass balance equation, followed by fitting the apparent absorptivity at each point on the titration curve (the method of Harris and Pecoraro; see "Experimental Procedures"). This was simplified by the initial calculation of K 1 and gave log K 2 ϭ 18.80 Ϯ 0.05, a value 0.2 log units smaller than that obtained from the Hill plot (Equation 5). The fitting procedure is very sensitive to the value of K 2 , and hence the difference in binding constants between the two sites does not arise from experimental errors. This confirms that the two Bi 3ϩ sites are non-equivalent. The difference in binding constants of log K 1 Ϫ log K 2 ϭ 0.84 is similar to that reported for Al 3ϩ , Ga 3ϩ , and Fe 3ϩ ( Table I).

Competition Reactions
Effect of pH on Bi 3ϩ Binding-The variation of the absorbance at 241 nm with the pH of a solution containing apo-hTF (10 M), 3 mol eq [Bi(NTA)] (i.e. both sites occupied with Bi 3ϩ ) and 5 mM bicarbonate in 10 mM Hepes buffer, pH 7.4, was investigated. As shown in Fig. 4, the optimum pH for Bi 3ϩ binding is ϳ7.5, and an increase or decrease in pH decreases the absorbance at 241 nm.
Ligand Competition-The competition between hTF and NTA for Bi 3ϩ binding was compared to that of citrate. Large excesses of NTA remove Bi 3ϩ from both binding sites, whereas even a 140-fold excess of citrate removes only one of the two bound Bi 3ϩ ions from transferrin (Fig. 5). NTA would be expected to be more effective in removing Bi 3ϩ from transferrin compared to citrate on the basis of the reported stability constants for Bi 3ϩ citrate and NTA complexes (log K 13.48 and 17.75 (24), respectively).
Competition with Fe 3ϩ -When [Fe(NTA) 2 ] was added to a solution of apo-hTF containing a 20-fold molar excess of [Bi(NTA)], a new broad band in the visible region centered at ϳ470 nm gradually increased in intensity (Fig. 6). This band has previously been assigned to a phenolate ()-metal (d*) transition of Fe 3ϩ -hTF (26). Three mol eq of [Fe(NTA) 2 ] were sufficient to completely displace bismuth from hTF showing that Fe 3ϩ binds more tightly to hTF than Bi 3ϩ . In accordance with this, when we added excess (up to 20 mol eq) of [Bi(NTA)] to Fe 2 -hTF, there was no evidence for iron displacement from the protein.

C NMR Studies
These were carried out to investigate whether the binding of Bi 3ϩ to hTF also involves concomitant binding of carbonate as synergistic anion. Fig. 7 shows the carbonyl region of the 13 C NMR spectrum of hTF in the presence of H 13 CO 3 Ϫ (10 mM, enriched to Ͼ99% in 13 C), and after addition of 0.89 and 2.0 mol eq of Bi 3ϩ (as [Bi(NTA)]). In the absence of Bi 3ϩ , a sharp signal at 161.1 ppm due to H 13 CO 3 Ϫ is observed, together with a broad envelope at 170 -183 ppm corresponding to the backbone and side-chain carbonyls of hTF (natural abundance 13 C). When Bi 3ϩ is present, a new peak assignable to bound 13 CO 3 2Ϫ is observed at 165.8 ppm. The intensity of this peak increases on addition of the second equivalent of Bi 3ϩ , suggesting that both sites contain bound carbonate. Addition of the anti-ulcer drug ranitidine bismuth citrate to a similar solution of hTF (allowed to equilibrate overnight) produced the same spectral changes. Although the 13 C NMR spectrum showed that most of the free bicarbonate was readily removed from bismuth hTF solutions by ultrafiltration, this procedure had little effect on the peak for bound carbonate (data not shown).

H NMR Studies
These experiments were carried out in order to determine the order of lobe loading and to investigate bismuth-induced structural changes in hTF. The 500-MHz 1 H NMR spectrum of hTF in the presence of 10 mM bicarbonate was recorded after addition of 0 -2 mol eq of [Bi(NTA)] in steps of 0.5 mol eq. In the high field region, between 0.4 and Ϫ1.2 ppm, some peaks, for example peaks a, e, g, and h, are unchanged throughout the titration (Fig. 8A), whereas others either decrease (peak c) or increase (peak d) in intensity on addition of the first equivalent of Bi 3ϩ and then show no further change on addition of the second equivalent. The most noticeable change in the aromatic region of the spectrum is the decrease in intensity of peak q (6.34 ppm) and increase in intensity of peak qЈ during addition of the first equivalent of Bi 3ϩ (Fig. 8B).
A two-dimensional TOCSY 1 H NMR spectrum of apo-hTF in presence of 10 mM bicarbonate, pH* 7.25 showed a cross-peak at 6.34(peak q)/7.72 ppm, suggesting that this may arise from a C⑀H/C␦H connectivity of the imidazole ring of a histidine residue. Furthermore, in the His C⑀H region (7.5-8.8 ppm, Fig.  8B), peaks sЈ, tЈ, and jЈ appear on addition of the first equivalent of Bi 3ϩ but remain almost unchanged with the second equivalent, whereas peak y decreases in intensity gradually throughout the titration.
The most intense peaks in the region of the spectrum from ϳ2.0 to 2.2 ppm are likely to arise from the N-acetyl groups of the glycan chains in the C-lobe of hTF. On addition of the first equivalent of Bi 3ϩ to apo-hTF, a new apparent singlet peak appears at 2.097 ppm, the peak at 2.080 ppm splits into two, and the shoulder at ␦ 2.038 ppm decreases in intensity (see Fig.  8C). These changes occur progressively from 0 to 1 mol eq of Bi 3ϩ , and no further changes occur in this region on addition of the second equivalent of Bi 3ϩ .

DISCUSSION
There appear to be no previous reports of the binding of transferrin to bismuth, and it might be expected that Bi 3ϩ would bind only weakly since it is a large metal ion (sixcoordinate ionic radius 1.03 Å), and ions of this size, such as the lighter lanthanide ions, have previously been found to bind ϳ10 12 times more weakly than Fe 3ϩ (Table I). The concept has arisen that transferrin cannot exhibit the same closed structure with large metal ions bound because the interdomain clefts cannot close (4). Surprisingly, therefore, our data show that there is strong binding of Bi 3ϩ to human serum transferrin.
The changes in the UV spectra of hTF on binding of Bi 3ϩ are similar to those observed previously for the binding of other metal ions to the specific Fe 3ϩ binding sites. The new bands at 241 and 295 nm are attributable to binding to tyrosine ligands (-* transitions). From the magnitude of the change in extinction coefficient (Table I and Fig. 3), it can deduced that two tyrosines are involved in binding Bi 3ϩ in both the N-and C-lobes (Tyr-95/Tyr-188, and Tyr-426/Tyr-517) as is the case for Fe 3ϩ . The displacement of Bi 3ϩ from transferrin by Fe 3ϩ , and lack of binding to Fe 2 -hTF provided further evidence for specific binding of Bi 3ϩ to the protein. Moreover the 13 C NMR data show that carbonate is directly bound to Bi 3ϩ in each lobe. Therefore Bi 3ϩ can now be added to Cr 3ϩ , VO 2ϩ , Mn 3ϩ , Co 3ϩ , Cu 2ϩ , Ga 3ϩ , and In 3ϩ as metal ions that satisfy the criteria (4,25) for specific metal binding to transferrin.
The two binding constants that we have determined for Bi 3ϩ binding to transferrin are only slightly lower than those for Ga 3ϩ binding (Table I). Several other previous studies have demonstrated the non-equivalence of the two transferrin binding sites, e.g. by NMR (27)(28)(29)(30)(31)(32) and EPR (21,34). The strength of Bi 3ϩ binding to transferrin is not so surprising when a linear free energy relationship (LFER) is used to compare the binding constants for Bi 3ϩ with those of Fe 3ϩ and a wide range of O,N-donor ligands, Fig. 9. There is a good correlation, which is described by Equation 8, with correlation coefficient r ϭ 0.979.
The slope of the LFER is greater than 1.0, and many of the stability constants for Bi 3ϩ complexes are greater than those of Fe 3ϩ . However, the oxygen ligands in the plot are carboxylates and not phenolates, and the former might favor Bi 3ϩ since they are softer donors. The strength of binding of metal ions to hTF therefore appears to be related more to the type of donor set provided by the protein than to the size of the metal ion. A critical maximum metal ion radius of 0.8 -0.9 Å has been suggested for domain closure to occur (36). Thus it has been suggested that the interdomain cleft is open when Ce 3ϩ (r ϭ 1.01 Å) binds to lactoferrin. In the case of Bi 3ϩ , it seems clear from the values of ⌬⑀ 241 (Table I and Fig. 3) that two Tyr ligands are bound to Bi 3ϩ in each lobe, as is the case with Fe 3ϩ . Since the Tyr residues are on domain 2 and an interdomain polypeptide strand, such binding could occur without domain closure (37).
It would therefore be interesting to study bismuth hTF by x-ray scattering to determine the radius of gyration for comparison with other metallotransferrins (38,39).
The NMR studies show that (bi)carbonate binds to both the N-and C-lobes of hTF concomitantly with Bi 3ϩ . The 13 C shift of the bound synergistic anion (165.8 ppm) is close to that observed previously for other metallotransferrins, e.g. 166.0/ 166.2 ppm for Tl 3ϩ (40), 165.4 ppm for Al 3ϩ , and 166.5 ppm for Ga 3ϩ (41), suggesting a similar mode of binding, probably as a bidentate carbonate ion. The anion is strongly bound to bismuth hTF and is not readily removed by dialysis. Studies of Bi 3ϩ binding to transferrin can also be readily monitored by 1 H NMR since apo-hTF and Bi-hTF are in slow exchange on the NMR time scale (indicative of strong Bi 3ϩ binding). Peaks in both the aromatic and the methyl regions of the spectrum suggest preferential binding of Bi 3ϩ to the C-lobe. The relatively sharp peak q at 6.34 ppm is affected by the first mol eq of Bi 3ϩ and has been previously assigned to a C-lobe residue (42). It has an associated low pK a (5.87) and an unusual pH titration shift range (0.75 ppm), and arises from a His C␦ proton (based on two-dimensional TOCSY data) with an unusual high field shift. From examination of the x-ray structure of hTF with Fe 3ϩ in the C-lobe (7), possible candidates are His-473 or His-535. The sharp resonances at 2.0 -2.1 ppm are attributable to the N-acetyls of the NAcGlc and NAcNeu residues in each of the two biantennary glycan chains attached to Asn-413 and Asn-611 in the C-lobe of hTF. Since these resonances were perturbed only on addition of the first mol eq of Bi 3ϩ , we can conclude that preferential binding occurs to the C-lobe of hTF. A similar NMR behavior has been observed for Ga 3ϩ (with oxalate as synergistic anion) and In 3ϩ binding to hTF (28,43), and several other metal ions are known to bind more strongly to the C-lobe of hTF than to the N-lobe (40).
No attempt was made to investigate the kinetics of reactions of hTF with Bi 3ϩ citrate or NTA complexes at this stage, but it is apparent that there at least two distinct kinetic steps. The citrate complex reacts much more slowly with hTF than the NTA complex, and dissociation of the low molecular mass ligand bound to Bi 3ϩ appears to be partly rate-determining. Similar observations have been reported for Fe 3ϩ binding to hTF (33,44).
Conclusions-We expected the binding of Bi 3ϩ to transferrin to be weak because the size of the binding cleft is thought to be matched to that of the metal ion, being optimum for Fe 3ϩ (ionic radius 0.65 Å). Unexpectedly, therefore, we have found that Bi 3ϩ (radius 1.03 Å) binds very tightly to both the N-and C-lobes of transferrin with affinities close to those of the smaller metal ions Ga 3ϩ (0.62 Å) and In 3ϩ (0.80 Å). Hence, ionic size can now be said not to be the major factor determining the strength of metal binding to transferrin. Linear free energy relationships for the binding of Fe 3ϩ and Bi 3ϩ to a range of different ligands demonstrate that the ligand donor set plays a dominant role.
We have shown that Bi 3ϩ binds specifically to the Fe 3ϩ binding sites in transferrin and that (bi)carbonate is also bound as the synergistic anion. Although Bi 3ϩ binds relatively strongly, it can be displaced by Fe 3ϩ . The NMR data show that Bi 3ϩ binds preferentially to the C-lobe, as has been found for several other metal ions. The ligands already bound to Bi 3ϩ play an important role in determining the rate of transfer of Bi 3ϩ from low molecular mass complexes onto transferrin.
It will be of interest for future work to investigate whether the domains of the N-and C-lobes close around the large Bi 3ϩ ions when the metal is bound, and to determine whether Bi 3ϩ binding occurs in intact blood plasma or other fluids and whether transferrin plays any role in determining the biodistribution of bismuth in the body.