|Home | About | Journals | Submit | Contact Us | Français|
Plasminogen activator inhibitor 1 (PAI-1) is a serpin inhibitor of the plasminogen activators urokinase-type plasminogen activator (uPA) and tissue plasminogen activator, which binds tightly to the clearance and signaling receptor low density lipoprotein receptor-related protein 1 (LRP1) in both proteinase-complexed and uncomplexed forms. Binding sites for PAI-1 within LRP1 have been localized to CR clusters II and IV. Within cluster II, there is a strong preference for the triple CR domain fragment CR456. Previous mutagenesis studies to identify the binding site on PAI-1 for LRP1 have given conflicting results or implied small binding contributions incompatible with the high affinity PAI-1/LRP1 interaction. Using a highly sensitive solution fluorescence assay, we have examined binding of CR456 to arginine and lysine variants of PAI-1 and definitively identified the binding site as composed of four basic residues, Lys-69, Arg-76, Lys-80, and Lys-88. These are highly conserved among mammalian PAI-1s. Individual mutations result in a 13–800-fold increase in Kd values. We present evidence that binding involves engagement of CR4 by Lys-88, CR5 by Arg-76 and Lys-80, and CR6 by Lys-69, with the strongest interactions to CR5 and CR6. Collectively, the individual binding contributions account quantitatively for the overall PAI-1/LRP1 affinity. We propose that the greater efficiency of PAI-1·uPA complex binding and clearance by LRP1, compared with PAI-1 alone, is due solely to simultaneous binding of the uPA moiety in the complex to its receptor, thereby making binding of the PAI-1 moiety to LRP1 a two-dimensional surface-localized association.
Urokinase-type plasminogen activator (uPA)2 is one of two serine proteinases involved in cell surface generation of plasmin from plasminogen. Although plasmin's primary role is to degrade fibrin-containing thrombi, it is also involved in a wide range of other normal and pathological processes that result from its ability to cleave matrix proteins such as laminin and fibronectin. In this way regulation of uPA activity plays an important role in control of processes such as wound healing and angiogenesis, as well as the growth and dissemination of tumors. For example, inhibition of uPA activity has been shown to block tumor invasion in the nude mouse (1). uPA is localized to the cell surface by tight binding to its uPA receptor (uPAR) (2) through the N-terminal region of uPA (3, 4).
The principal inhibitor of uPA activity is plasminogen activator inhibitor 1 (PAI-1), a serpin. Inhibition by serpins involves a unique mechanism of kinetic trapping of the acyl-enzyme intermediate formed during cleavage of the serpin's reactive center loop by the proteinase (5). The mechanism involves a massive conformational change in the serpin, with insertion of the reactive center loop into sheet A of the serpin and translocation and consequent distortion of the proteinase active site as the means of kinetic trapping (6,–9). Because uPA is predominantly bound to its receptor, the result of inhibition by PAI-1 is the formation of a ternary membrane-associated uPAR·uPA·PAI-1 complex. This complex is rapidly internalized, as a result of binding to the low density lipoprotein receptor-related protein (LRP1), and the uPA·PAI-1 component degraded (10, 11). In contrast, uPA alone bound to uPAR is poorly internalized (12). Cross-competition for binding to LRP1 between uPA·PAI-1 complexes and PAI-1 complexes with other proteinases suggests that much or all of the binding interaction between the uPAR·uPA·PAI-1 complex and LRP1 occurs through PAI-1 (13). Additional support for this comes from soluble uPAR blocking the binding of uPA to LRP1, although it has minimal effect on the strength of the uPA·PAI-1 interaction with LRP1 (12).
Knowledge of the binding site on PAI-1 for LRP1 is thus very important for an understanding of the structural basis for internalization and degradation of uPA in its PAI-1-complexed form and hence for regulation of surface-bound uPA activity. However, no structure of PAI-1 in complex with a fragment of LRP1 has yet been determined. Indeed, the only structures of complex between a protein ligand and a receptor fragment from an LDL receptor family member are of human rhinovirus HRV2 bound to a portion of the VLDL receptor (14), of the third domain of RAP bound to a fragment from LDL receptor (15, 16), and of reelin R5 and R6 domains bound to a fragment of apoER2 (17). These structures suggest a common theme of engagement of protein ligands by LDL receptor family members that involves binding of pairs of basic residues (lysine in each case) to individual CR domains of the respective receptors. These CR domains are small independent modules that occur in clusters within the receptors. This mode of binding is qualitatively consistent with mutagenesis studies carried out on PAI-1, in which mutation of certain basic residues reduced binding affinity for LRP1, either alone or in complex with proteinase (13, 18,–20). However, there are two major problems with these mutagenesis studies. The most disconcerting is that basic residues identified as being involved in binding result in no more that an 11-fold increase in Kd when mutated and mostly much less, whereas the affinity of PAI-1 for LRP1 is in the low nanomolar region. Such very small effects are not expected if basic residues are the dominant contributors to overall affinity, which has been quantitatively shown to be the case for the RAP D3/CR56 interaction (21) and for model compounds binding to individual CR domains (22). The second problem is that there is no consensus on which basic residues constitute the binding site, with some residues identified as being important in one study (13, 19, 20) and having little or no effect when mutated in another study (18).
To more definitively establish the binding site on PAI-1 for LRP1, we have developed a sensitive fluorescence assay and examined the solution binding of wild type and variant PAI-1s, all in their native form, to the fragment CR456 from LRP1, which we and others have shown to be a high affinity binding site for PAI-1 and its proteinase complexes within cluster II of LRP1 (23,–25). Comparisons are also made for binding of CR56, CR345, and CR567.
All PAI-1 species used were on the stable 14-1B (the very stable variant of PAI-1 containing four stabilizing mutations (N150H, K154T, Q319L, and M354I) (26)) background that has been shown not to alter the functional properties of PAI-1, but it does greatly slow the conversion to the latent state (26). Subsequent use here of the term WT PAI-1 implies the 14-1B background. An estimate of the t½ for conversion to the latent state at 4 °C, based on t½ at 25 °C of 22 h (27) and an activation energy for the latent transition of 96 kJ mol−1 (28), is well over 200 days for WT PAI-1 and would be very much longer for the 14-1B variant. For the 14-1B variant used here, the protein thus remains completely in the native state, even when kept at 4 °C for several days. The cDNA encoding the 14-1B variant was cloned into pQE30 (29) giving a final product containing an N-terminal His6 tag. ΔW-PAI-1 (PAI-1 with all four tryptophans mutated to phenylalanine) was generated by sequentially mutating the four tryptophans (residues 86, 139, 175, and 262) to phenylalanine. For FRET measurements, single cysteine variants were generated on the ΔW-PAI-1 background (H2C, G70C, G84C, S127C, G194C, G230C, G264C, and Q312C). For monitoring binding of LRP1 fragments from fluorescence perturbation, the H2C mutation was introduced into PAI-1 14-1B and used as the background to generate the following variants: K65A, K69A, K65A/K69A, R76A, K80A, R76A/K80A, K88A, K176A, K88A/K176A, and R115A/R118A. All mutagenesis was performed using the QuikChange protocol (Agilent Technologies), and all variants were verified by DNA sequencing before protein expression.
PAI-1 variants were expressed in SG13009 cells (Qiagen) in 2YT medium. Cultures were grown at 37 °C to an A600 = 0.6–1.0 and induced with 1 mm isopropyl β-d-thiogalactoside, and the temperature was then lowered to 27 °C. Cells were harvested by centrifugation after 4–5 h and stored at −80 °C until needed. PAI-1 variants were purified as described (30). Cells were sonicated in 50 mm sodium phosphate, 300 mm NaCl, and 10 mm imidazole, pH 7.4, containing 5 mm β-ME. After centrifugation, NaCl and glycerol were added to 2 m and 5% v/v, respectively. The cleared lysate was loaded on a Ni2+-chelate column. PAI-1 was eluted with 250 mm imidazole and dialyzed overnight at 4 °C against 20 mm Tris-HCl, 1 m NaCl, 0.5 mm EDTA, 5% glycerol, pH 8.0, containing 14 mm β-ME. DTT was added to 5 mm before final purification on a Superdex 75 column (860 × 16 mm), with elution in dialysis buffer.
The effect of the lysine or arginine to alanine mutations on PAI-1 function was determined by assessing the ability of the PAI-1 variants to inhibit uPA. Both stoichiometries of inhibition and apparent second order rate constants were determined by kinetic assay. For all of the variants, the stoichiometries of inhibition and corrected second order rate constants were very similar to those of WT PAI-1, consistent with the alanine substitutions not affecting the structure of the molecule.
The peak containing monomeric PAI-1 was made 10 mm in DTT, concentrated to 1.5–2.5 ml using Amicon ultracentrifugal filters, and buffer-exchanged into degassed 20 mm Tris-HCl, 1 m NaCl, 0.5 mm EDTA, 5% glycerol, pH 8.0, using a Sephadex G25F column (160 × 16 mm). The protein concentration was determined using UV absorption at 280 nm and calculated extinction coefficients of 35,410 m−1 cm−1 for WT PAI-1, and 13,410 m−1 cm−1 for ΔW-PAI-1. Two molar eq of DTT were added to the PAI-1 sample, followed by the addition of 12 eq of IAEDANS (Molecular Probes). The reaction was allowed to proceed on ice overnight. The next day, the reaction was quenched by adding 14 mm β-ME, and the sample was concentrated to 1.5–2.5 ml before excess IAEDANS was removed by G25F gel filtration. After concentrating the samples, the concentrations of PAI-1 and dansyl were determined spectrophotometrically, using an extinction coefficient of 4700 m−1 cm−1 at 340 nm for dansyl and a correction factor of 0.38 for the absorption at 280 nm (A280, corr = A280 − 0.38 × A340). Samples typically contained 50–100 μm PAI-1 and 1.3–1.5 dansyl per PAI-1. A control PAI-1 sample, without cysteine, was reacted similarly and gave 0.3 dansyl per PAI-1. This nonspecific dansyl label was insensitive to binding of LRP1 fragments.
Expression, purification, and refolding of the various CR fragments was according to procedures we have successfully used for many years (25, 31,–33). All CR constructs were expressed in 2YT medium. CR56 and CR567 were cloned in pGEX-2T, modified to contain a TEV proteinase cleavage site, and expressed as GST fusion proteins in BL21 cells. Cells were grown to A600 = 0.6–1.0 before induction with 1 mm isopropyl β-d-thiogalactoside and harvested after 5–6 h at 37 °C. The GST fusion proteins were purified from cleared cell lysate by GSH-Sepharose chromatography, and the GST tag was removed by TEV proteinase cleavage during overnight dialysis against 4 liters of 20 mm Tris-HCl, pH 8.0, 50 mm NaCl, 4 mm EDTA, containing 14 mm β-ME. GST and uncleaved GST-CR fusion proteins were removed by passage through the GSH column.
CR345 and CR456 were cloned into pQE-30, modified to contain a GB1 fusion partner (for increased solubility), and a TEV proteinase cleavage site. These His6-tagged fusion proteins were expressed in SG13009 cells containing the plasmid pRARE. The CR fusion proteins were purified from cell lysate by Ni2+-chelate chromatography, and the fusion partners were removed by TEV proteinase cleavage during overnight dialysis against PBS containing 14 mm β-ME.
Before refolding, all CRs were further purified by Q-Sepharose HP chromatography, using a gradient of 0–1000 mm NaCl in 20 mm Tris-HCl, pH 8.0, and 6 m urea. The denatured CRs were diluted with 50 mm Tris-HCl, 50 mm NaCl, 10 mm CaCl2, pH 8.5, containing 14 mm β-ME and 8 mm 2-hydroxyethyl disulfide (refolding buffer). For increased folding efficiency, all CR constructs were mixed with the chaperone RAP as a GST fusion protein (GST-RAP) at a 1:1 ratio. Final CR concentration in the refolding mixture was 0.1 mg/ml. The CR species were refolded by dialysis against refolding buffer for 24 h at room temperature with N2 bubbling, followed by 24 h at 4 °C without N2. Finally, the mixture was dialyzed twice against 4 liters of 20 mm Tris-HCl, pH 7.8, 50 mm NaCl, 1 mm CaCl2 at 4 °C. The refolding mixture was loaded on GSH-Sepharose, and folded CR constructs, i.e. those capable of binding to RAP and therefore retained on the column, were eluted with 40 mm Tris-HCl, pH 8.0, 100 mm NaCl, 8 mm EDTA. Folded CR constructs were further purified by Q-Sepharose HP chromatography by salt gradient elution with 50–1000 mm NaCl in 20 mm Tris-HCl, pH 8.0, and 0.1 mm CaCl2. If traces of GST-RAP were still present, calcium was removed from the mixture by adding EDTA, and the mixture was passed through a Superdex 75 size exclusion column equilibrated in 20 mm Tris-HCl, pH 7.8, 150 mm NaCl, and 4 mm EDTA, after which calcium was reintroduced to the purified CR. As a quality control between each step of purification, samples were analyzed by SDS-PAGE (with and without reducing agent) and native-PAGE. The presence of a single band on SDS-PAGE, which shifts upon reduction, indicates homogeneous samples with formed disulfide bridges. A single band on native-PAGE suggests the presence of a single folding product. The final product was judged by SDS-PAGE to be more than 95% pure. The effective concentration of functional CR fragments was determined by stoichiometric titration against RAP fragment D3. Binding was monitored from the shift in tryptophan fluorescence maximum of the CR domains upon binding and the high affinity of D3 for all of the CR fragments used here (31).
RAP cDNA, cloned in pGEX-2T, was a kind gift from Dr. Dudley Strickland. GST-RAP was expressed in BL21 cells, induced with 1 mm isopropyl β-d-thiogalactoside at A600 = 0.6–1.0, and harvested after 5 h at 25 °C. GST-RAP was purified by GSH-Sepharose chromatography. GST-RAP was then dialyzed overnight against 4 liters of 20 mm Tris-HCl, pH 8.0, 50 mm NaCl, 4 mm EDTA, containing 14 mm β-ME. Free cysteines in the GST moiety were blocked by treatment with 20 mm iodoacetamide for 1 h at room temperature, and excess reagent was removed by dialyzing twice against 4 liters of 20 mm Tris-HCl, pH 8.0, 50 mm NaCl, 4 mm EDTA, followed by dialysis against 4 liters of 50 mm Tris-HCl buffer, pH 8.5, containing 50 mm NaCl and 10 mm CaCl2.
All fluorescence measurements were made on a PTI QuantaMaster spectrofluorimeter, equipped with double monochromators on both excitation and emission sides. For FRET measurements, excitation was at 280 nm and emission was monitored from 310 to 650 nm. For direct contact perturbation spectra excitation was at 342 nm and spectra were recorded from 400 to 650 nm. In each case slits of 1 nm for excitation and 4 nm for emission were used, with 2-nm steps, 2 s dwell time, and spectra recorded as the average of four scans. For single wavelength measurements, used to follow binding of CR fragments to H2C-dansyl-labeled PAI-1 variants, excitation was at 342 nm, and emission was monitored at the maximum for dansyl emission (470 nm). Slits of 1 nm for excitation and 8–10 nm for emission were used, with signal monitored for 120 s and averaged.
Samples were 1.5 ml in 1-cm cuvettes and typically contained 0.06–1 μm PAI-1, depending on the anticipated Kd value. The pH 7.4 buffer contained 20 mm Tris-HCl, 1 mm CaCl2, and 0.1% PEG 8000. NaCl was added from a stock 5 m solution to achieve the desired final ionic strength. The effect of added NaCl on pH was checked for each sample and found to result in no more than 0.05 pH unit change. Samples were mixed with a magnetic stir bar in the cuvette. Data were corrected for contributions from a buffer blank and for dilution. In all cases dilution was less than 2%. Temperature was maintained by a circulating water bath. All titrations were carried out at 298 K. Kd values were determined by non-linear least squares fitting of the binding data to a single site binding isotherm using the quadratic equation for equilibrium binding (34). Fitting was done in KaleidaGraph (Synergy Software).
To determine percentage quench of tryptophan fluorescence resulting from FRET to dansyl acceptor, dansyl-labeled ΔW-PAI-1 was titrated into a sample of CR56 until a saturable end point was reached. The intensity of the CR56 tryptophan fluorescence in complex was then compared with that of unbound CR56 using the intensity at the wavelength maximum. For measurement of the enhancement of dansyl fluorescence from FRET, CR56 was titrated into a sample of dansyl-labeled ΔW-PAI-1 to a saturable end point. Enhancements were determined as the relative increase in intensity at the wavelength maximum, compared with unbound dansyl-labeled ΔW-PAI-1, with excitation at 280 nm. To extract the effect due to FRET, correction was made for direct binding effects when excited at the same wavelength.
Although no quantitative use was made of the magnitude of tryptophan quench in this study, for reasons given under “Results,” there is a direct relationship between the quench and the efficiency of FRET given by Equation 1,
where QDA is the donor (tryptophan) quantum yield in the presence of acceptor, and QD is the donor quantum yield in the absence of acceptor. E is related to the separation between donor and acceptor by Equation 2,
where R0 is the separation for that donor-acceptor pair that results in 50% efficiency of FRET, and R is the separation of the pair in the system under study. R0 is a function of the spectral properties of the donor and acceptor, given by Equation 3,
where J is the overlap integral, κ2 is the orientation factor, n is the refractive index, and ΦD is the donor quantum yield. Thus, if R0 can be confidently calculated for the system under study, E can be used to determine the donor-acceptor separation R.
Our initial plan for delineating the binding site for LRP1 on PAI-1 was to use tryptophan-to-dansyl FRET between the single tryptophans within each CR domain and a dansyl fluorophore attached covalently to single cysteines introduced within PAI-1. To this end, we created a number of single cysteine-containing variants of a tryptophan-less PAI-1 (ΔW-PAI-1) to be able to monitor both enhancement of acceptor (dansyl) fluorescence and quench of donor (CR domain tryptophan) fluorescence. The sites for introduction of the single cysteines are shown in Fig. 1. CR56 was chosen as being a high affinity two-domain fragment from cluster II (25, 35), with Kd values only ~20-fold higher than for the tightest binding CR456 fragment (25) and with only two tryptophans, one in each domain, to simplify analysis.
We first examined whether any of the dansyl fluorophores showed perturbation upon CR56 binding that was due directly to binding rather than from FRET. Dansyl was excited directly at 342 nm (well removed from the tryptophan absorption) in free PAI-1 and upon complex formation with CR56. Whereas dansyl at most positions showed small changes in either wavelength maximum or intensity, dansyl attached to His-2–Cys showed a blue shift of 14 nm and ~65% increase in intensity (Table 1 and Fig. 2). Smaller but significant blue shifts were seen with Gly-84–Cys and Gly-70–Cys, although other positions showed minimal spectral shifts (Table 1).
The dramatic effect of CR56 binding on dansyl at position 2 could be due either to close proximity of the fluorophore to the bound CR56 or CR56-induced conformational change within the PAI-1. However, we have previously found that binding of CR fragments to PAI-1 results in only a small perturbation of tryptophan fluorescence (25). From subsequent results presented here (see Fig. 4A), we know that this arises only from tryptophans in the CR domains. This suggests that binding of CR56 does not cause significant conformational change within the PAI-1 and thus that the large perturbation of the dansyl fluorophore upon CR56 binding is due to proximity to the fluorophore. To examine whether the fluorophore had itself altered the binding of CR56 to PAI-1, we used the direct perturbation of dansyl fluorophore to follow binding by titrating CR56 into labeled PAI-1. The data were well fitted to a single binding site (Fig. 3A) and gave a Kd of 52 ± 2 nm (Table 2) that was indistinguishable from the value we found previously for binding of CR56 to unlabeled cleaved PAI-1, monitored by the less sensitive perturbation of tryptophan fluorescence (25). This finding suggested that the Cys-2–dansyl is an extremely sensitive yet non-perturbing reporter of CR fragment binding that might be invaluable for examining binding of both other CR fragments and of CR56 or CR456 to variants of PAI-1 that also contained the Cys-2 mutation.
Pursuing our initial goal of using FRET to delineate the CR56-binding site on PAI-1, we determined both enhancement of dansyl fluorescence, from titration of CR56 into dansyl-labeled ΔW-PAI-1 species and of quench of tryptophan fluorescence from titration of dansyl-labeled ΔW-PAI-1 into CR56. For dansyl enhancements, correction was made for any direct perturbation effects caused by binding itself (see above). In each case a saturable effect was seen, consistent with the dansyl enhancement or quench of CR56 tryptophan arising from FRET between CR domain tryptophan excited at 280 nm and PAI-1-linked dansyl emitting around 470 nm. By far, the largest effect was seen for label at Cys-2, with close to 80% quench of CR56 tryptophan fluorescence and 440% enhancement (227% enhancement after correction for proximity effects) of dansyl fluorescence (Fig. 4). Lesser, but still large, quenches of tryptophan fluorescence were also seen with label at positions 84 (55%) and 230 (37%). Other positions showed smaller quenches of tryptophan fluorescence. Significant enhancements of dansyl fluorescence were seen at all positions except 127 (Table 3).
Although we had hoped to use these FRET measurements to define the location of the CR56-binding site, we were not able to do so, because of the uncertainties in the analysis, which greatly reduced the precision possible. The biggest problem was the large value for R0 (the Förster distance for 50% efficiency of FRET) for the tryptophan-dansyl pair of ~23 Å (7, 36). This made it difficult to find a position on PAI-1 that would give no measurable FRET and conversely resulted in many positions that were close enough to give quite significant FRET (Table 3). Unfortunately, no other donor-acceptor pair could be found that was better (37). A second problem was the length of the linker between the cysteine sulfur and the dansyl fluorophore (intervening CH2CO-NH-CH2-NH). This gave great uncertainty in the location of the fluorophore relative to the cysteine sulfur and consequently might result in a fluorophore at a proximal label site pointing away from the CR56-binding site and another fluorophore at a more distant site pointing toward the CR56 giving similar FRET. Although we examined other labeling species with shorter linkers, including BADAN, NBD-chloride, and dansyl-chloride, the fluorescence of the derivatives suffered from very low quantum yield (data not shown). Finally, uncertainty in the value of the orientation factor, κ2, for each label position introduced another uncertainty in determining R0 for each donor-acceptor pair. As a result, we were unable to use FRET efficiencies to accurately map the location of CR domains relative to the mutated cysteine residues. The data were, however, of more qualitative use in suggesting whether the CR domains were near or far from the site of label attachment. In this way, we confirmed the observations above from direct perturbation of dansyl fluorophore that the CR56-binding site must be close to His-2 and also be in the vicinity of Gly-70, Gly-84, Gly-230, and Gly-264 to account for the large tryptophan quenches due to FRET. This is consistent with other mutagenesis studies that have implicated basic residues in this general region in the binding of PAI-1 to LRP1 (13, 18,–20). Although the Gln-312–Cys variant did give a large dansyl enhancement, this was from a very low intensity emission. It is therefore thought that the small tryptophan quench of CR56 fluorescence is a more reliable indicator of the distance of this residue from the binding site.
Guided by the findings above that the binding site for CR56 must be close to residues 2, 70, 84, and 230 (see Fig. 1), and the expectation that individual CR domains engage proximal pairs of basic residues, we identified four possible pairs of basic residues as potential CR domain binding sites. These were Lys-65 and Lys-69, Arg-76 and Lys-80, Arg-115 and Arg-118, and Lys-88 and Lys-176 (Fig. 1). With the exception of Lys-65 and Lys-176, all of these residues have been implicated in LRP1 binding by one or other of the four previous mutagenesis studies (13, 18,–20). PAI-1 variants containing each of these pairs mutated to alanine were created on a background of His-2–Cys and 14-1B, to permit introduction of the sensitive non-perturbing dansyl reporter at this location.
Binding titrations were carried out for each double basic mutant using direct perturbation of the dansyl fluorophore, as above. Only three of the pairs of mutations affected the affinity for CR56 (Table 2). No effect on affinity was seen for the Lys-88/Lys-176 pair. The affinity for the Lys-65/Lys-69 variant was greatly reduced, with about a 50-fold increase in Kd to ~2.5 μm, whereas the affinity of the Arg-76/Lys-80 variant was so weakened that no reliable Kd value could be obtained, because only a small fraction of the PAI-1 could be bound at the highest concentrations of CR56 used (8 μm). However, assuming similar maximum fluorescence perturbation obtained for tighter binders, we crudely estimated the Kd value for this variant as ~70 μm. Surprisingly, the Arg-115/Arg-118 variant was found to bind CR56 more tightly than WT (Kd of 15 nm versus 52 nm), despite a previous report that each of these residues, when individually mutated, reduced binding of uPA·PAI-1 complexes to VLDL receptor, LRP1, and sorlA, with the Arg-118 mutation having about as large an effect on solid-state LRP1 binding as mutating either Arg-76 or Lys-80 (20).
To further define the role of individual basic residues within the two pairs whose mutation reduced affinity for CR56, we also examined binding of single Lys-65–Ala, Lys-69–Ala, Arg-76–Ala, and Lys-80–Ala variants. The K65A mutation had no effect on binding, whereas the Lys-69–Ala variant had affinity reduced similarly to that of the double K65A/K69A variant (Table 2), suggesting that only Lys-69 of this pair engages one of the CR domains of CR56. Mutation of either Arg-76 or Lys-80 resulted in large reduction in affinity, with Kd values greatly increased to 2.5 μm and an estimated ~30 μm, respectively (Table 2), indicating that both of these residues are involved in binding of PAI-1 to CR56.
Given the proximity of Arg-76 and Lys-80 to one another and their distance from Lys-69 (Fig. 1), it seems likely that the former pair engages one CR domain of CR56 and Lys-69 engages the other. To determine which of the two domains, CR5 or CR6, engages which site, we employed tryptophan-dansyl FRET in a qualitative way, using both WT CR56, in which each CR domain contains a single tryptophan, and two variants in which one or the other of the two tryptophans had been mutated to phenylalanine (variants CR5F6W and CR5W6F).3 When FRET experiments were carried out using dansyl reporter at Cys-2 (on a ΔW-PAI-1 background) and each of these single tryptophan-containing CR56s, approximately the same reduction (~50%) in tryptophan fluorescence was seen resulting from FRET to dansyl, suggesting equidistant positioning of each CR domain with respect to the Cys-2 dansyl group. However, when the dansyl-labeled Cys-230 variant was used, much greater FRET was seen to CR5W6F (58%) than to CR5F6W (17%), compared with 37% when both CR56 domains contained tryptophan. This suggests that CR5 is much closer to dansyl attached to residue 230 than is CR6 (see Fig. 1), and thus that the CR5 domain engages the pair Arg-76/Lys-80 while CR6 engages Lys-69.
Although the above experiments identified a binding site for CR56, involving residues Lys-69, Arg-76, and Lys-80, we wanted to extend our studies to CR456, because this likely represents the full binding site for PAI-1, having been found to bind with Kd ~20-fold smaller than CR56 and to be the tightest binding three-domain fragment from within cluster II (25).
Initially we examined the effect of CR456 binding on direct perturbation of dansyl attached to Cys-2 and found similar large enhancement of the dansyl fluorescence to that seen for binding of CR56 (Fig. 2). This is consistent with the CR56 moiety in both species binding in the same location. A titration of CR456 into dansyl-labeled Cys-2 PAI-1, with binding monitored using this direct perturbation of dansyl fluorescence, confirmed the significantly tighter binding of CR456 at I 0.17 (Fig. 3B and Table 2).
Given the much higher affinity for PAI-1 of CR456 compared with CR56, we used the ionic strength dependence of CR456 binding at higher NaCl concentrations to obtain an accurate Kd at I 0.17 by extrapolation. A Debye-Hückel plot of log10Kd against I0.5 for Kd values obtained between 0.3 and 0.7 m NaCl gave a straight line with a slope of 4.3 (Fig. 5A). The Kd at I 0.17, obtained by extrapolation, was 1.5 nm and is actually in good agreement with the Kd value obtained directly of 1.3 nm (Fig. 2). This compares with the Kd of 54 nm for CR56 obtained directly.
Similar measurements of the ionic strength dependence of Kd were also carried out for CR56 but over the [NaCl] range of 0.15 to 0.4 m. The Debye-Hückel plot for these data also gave a straight line but with slope of 2.9 (Fig. 5B). The slope of 2.9 is close to what would be expected from involvement of only the three basic residues Lys-69, Arg-76, and Lys-80 in binding CR56 identified above. Given this, the finding of a higher slope for binding of CR456 (4.3) suggests that the greater affinity of CR456 arises from engagement of one or more basic residues in addition to the three already identified, with the additional binding probably involving CR4.
To better determine whether the CR56-binding site represents a part of the full CR456-binding site and to determine which, if any, additional basic residues are involved in binding CR456, we examined the binding of CR456 to the same set of four double-basic PAI-1 variants, each containing a dansyl reporter at the Cys-2 residue, as for binding of CR56. Where CR56 showed reduced binding only to the variants containing mutations at the Lys-65/Lys-69 and Arg-76/Lys-80 pairs, CR456 showed reductions for both of these pairs, as well as for the variant with mutations at Lys-88/Lys-176 (Table 2). However, the reduction in affinity for the latter was less than for the other two (about 11-fold versus 50–2700-fold increase in Kd at I 0.17). Interestingly, the variant with mutations of Arg-115 and Arg-118 showed increased affinity compared with WT (Kd values of 4.5 nm versus 39 nm at I 0.52), as was found for CR56 binding. Also similar to binding of CR56, mutation of the Arg-76/Lys-80 pair had the greatest effect (~2700-fold increase in Kd).
We then examined binding of CR456 to variant PAI-1s with single basic residue mutations. Anticipating greatly weakened binding, we carried out titrations at I 0.02. Mutation of each of Lys-69, Arg-76, or Lys-80 had very large effects on affinity (Kd values of 103, 130, and 185 nm, respectively, at I 0.02) (Table 2). Assuming a slope of three for ionic strength dependence in Debye-Hückel plots of each of these variants (only three of the four basic residues remained), values of 670, 845, and 1200 nm, respectively, were calculated at I 0.17, implying 447–800-fold increases in Kd resulting from mutating these residues. Mutation of Lys-88 increased Kd at I 0.17 to 20 nm (~13-fold increase compared with WT). Neither the single K176A nor the K65A mutations perturbed binding compared with WT. Thus CR456 appears to use the same basic residues to bind to PAI-1 as does CR56, but with the addition that CR456 also binds to Lys-88, probably through CR4.
Our previous studies have shown that CR456 binds with about a 10-fold decrease in Kd to WT-cleaved PAI-1 than does CR345 or CR567 (25). This is consistent with earlier qualitative studies that showed that PAI-1 does not bind tightly to CR567 but does to a longer fragment that includes domains CR3 and CR4 (further refinement of site-specificity was not carried out) (23, 24). Because we have identified three well separated binding sites for CR domains (see Fig. 1), the lower affinity of CR345 and CR567 compared with CR456 implies that there is a preference at a given PAI-1 site for particular CR domain(s).
To examine this, we first confirmed that the lower affinities for CR345 and CR567 hold true here for our slightly different system (native PAI-1 Cys-2-dansylated variant versus cleaved WT). Titrations carried out at I 0.17 gave Kd values of 16.5 ± 2 and 28 ± 1.8 nm for binding of CR345 and CR567, respectively, confirming the much weaker binding of each fragment compared with CR456 (Fig. 6).
Next, the effect of binding each fragment on the emission spectrum of the Cys-2-dansyl fluorophore was examined. CR567 gave a large enhancement and a 13-nm blue shift (Fig. 7) similar to what was seen for both CR456 and CR56 (Fig. 2). In contrast, CR345 gave a much smaller enhancement and only a 7-nm blue shift (Fig. 7B). This suggests that the positional relationship between the dansyl fluorophore and the domain(s) in CR56 that gives rise to the large enhancement and blue shift remains the same when CR456 and CR567 bind, whereas it is altered when CR345 binds.
To better determine whether, when CR567 binds, the CR5 domain binds to the Arg-76/Lys-80 pair and the CR6 domain binds to Lys-69 (as in binding of CR56 and CR456), we examined binding of CR567 to the variants containing mutations at the Lys-88/Lys-176 pair, the Arg-76/Lys-80 pair, and the Lys-65/Lys-69 pair. The affinity of the first was unaffected (Kd of 24 ± 1.7 nm) compared with WT, whereas the Arg-76/Lys-80-mutated variant had affinity greatly reduced, with Kd of 8.9 ± 0.4 μm (Table 2). This implies that CR567 does not engage the Lys-88 site utilized by CR4 in CR456, although it does engage the site used by CR5, with similar loss of affinity when mutated. Surprisingly, the variant with Lys-65/Lys-69 mutated showed only a 3-fold reduction in affinity, rather than the expected 50-fold. Unfortunately, there was insufficient material to carry out equivalent titrations on CR345. Together, these results are most simply interpreted as CR5 and CR6 having strong enough preferences for the two sites Arg-76/Lys-80 and Lys-69, respectively, that they direct binding there in fragments containing these domains (CR56, CR456, and CR567). When only CR5 is present (as in CR345), a higher affinity is presumably obtained by a shift in register of one compared with binding of CR456, such that CR3, CR4, and CR5 bind where CR4, CR5, and CR6 are bound. Although the resulting Kd is ~10-fold higher than for CR456, binding is presumably tighter than would be obtained if CR4 and CR5 bound in the same way as they do in CR456, and CR3 remained unengaged. This is consistent with the greater apparent binding contribution from Lys-69 in the “CR6” site (~450-fold increase in Kd when mutated) than from Lys-88 in the “CR4” site (14-fold increase in Kd when mutated). The only seemingly contradictory result is the smaller than expected reduction in affinity of CR567 when the Lys-65/69 pair was mutated. This may, however, result from partial compensation for the loss of the CR6 interaction with Lys-69 by interaction of CR7 with another lysine or arginine that is otherwise inaccessible in WT when the CR6 domain is anchored by Lys-69. Indeed, such an alternative mode of binding, with consequent partial offset of binding losses, might be one explanation for the anomalous results of others using the whole of cluster II in which mutation of residues identified here as contributing a large fraction of the binding energy seemingly had minimal effects (18).
The presence of a highly conserved single tryptophan in CR domains from ligand binding regions of LDL family receptors has suggested that the tryptophan makes an important binding contribution, probably to the hydrophobic side chain of ligand lysines. Consistent with this, mutation of tryptophan to serine in either the CR5 or CR6 domain of CR56 has been shown to greatly reduce binding to a RAP D3 affinity column (35), although there is the possibility that the drastic mutation may have adversely affected folding. Because both single tryptophan-containing CR56 species had been prepared for FRET measurements, with the more conservative mutation to phenylalanine, as well as a double variant in which both tryptophans had been replaced by phenylalanine as a control, we determined Kd values for binding of each of these species to dansyl-labeled Cys-2 PAI-1 to determine the importance of tryptophan in contributing to affinity.
Each species produced the expected saturable large enhancement in dansyl fluorescence upon binding to PAI-1 (data not shown). The titrations were each well fitted to a single binding isotherm and gave Kd values at I 0.17 of 130 nm for CR5F6W, 120 nm for CR5W6F, and 560 nm for CR5F6F, compared with 52 nm for CR5W6W. The mutation of each tryptophan to phenylalanine thus reduces overall affinity in an approximately additive manner.
Using a sensitive fluorescent dansyl probe attached to an engineered cysteine at position 2 in PAI-1, we have determined the affinities of basic residue PAI-1 variants for the high affinity LRP1 fragment CR456 and shown that the binding site on native PAI-1 involves the residues Lys-69, Arg-76, Lys-80, and Lys-88 (Fig. 1C). The proximity of this site to residue 2 explains the extreme sensitivity of the dansyl fluorophore at this position to binding of CR456. Independent support for the involvement of four charged residues comes from the slope of the ionic strength dependence of the affinity (4.3) (Fig. 5A). It should also be noted that this is the first time arginine has been identified as being involved in binding to an LDL receptor family member, although we have previously shown with model compounds binding to CR domains that the strength of the interaction in that case is nearly the same as for lysine (22).
Based on the effects of mutation on affinity and the nearly identical direct perturbation of Cys-2-dansyl fluorescence upon binding, we have also shown here that CR56 binds to Arg-76, Lys-80, and Lys-69 and seems to occupy the same sub-sites as do the equivalent CR domains in CR456. The binding site location is also qualitatively consistent with the very large FRET-derived quench of tryptophan fluorescence when dansyl acceptor is at positions 2, 84, or 230 (see Fig. 1). FRET measurements on WT CR56 and CR56 variants with each tryptophan mutated separately to phenylalanine indicate that CR5 is much closer to residue 230 than is CR6 and thus that the orientation of the CR56 is such that CR5 engages the basic pair Arg-76/Lys-80 while CR6 engages Lys-69. If the CR5 and CR6 domains in CR456 bind in the same sub-sites as when CR56 binds, this implies that CR4 engages Lys-88 as the third binding site. Examination of the location of these residues in PAI-1 (Fig. 1) shows that this is also consistent with the relative positions of each of these basic residues. Thus, binding of the highest affinity cluster II fragment CR456 to PAI-1 involves engagement of CR4 by Lys-88, CR5 by the pair Arg-76/Lys-80, and CR6 by Lys-69. Single mutation of each of these residues to alanine results in increases in Kd of 13, 563, 800, and 447 times, respectively.
Given the apparent independence of adjacent CR domains (38), resulting from their flexible linkers, one might expect binding contributions to a ligand to be additive. Assuming such additivity, the summed binding contributions of the four basic residues, Lys-69, Arg-76, Lys-80, and Lys-88, amount to 13.65 kcal mol−1 at 298 K (calculated from the losses in affinity for each mutation), compared with the measured affinity of CR456 of 12.10 kcal mol−1. This close agreement suggests that we have not only identified basic residues in PAI-1 that are important for binding to LRP1 but have shown that their binding contributions quantitatively account for all of the binding energy. In a similar way, we previously showed that engagement of four lysines in RAP D3 accounted fully for the high affinity interaction with CR56 (21). It is striking however, that, although the mutations reported here resulted in extremely large increases in Kd values, the same mutations reported elsewhere had mostly negligible effects (18). We have no good explanation for this discrepancy, other than to point out that, whereas the present studies were carried out in solution using a sensitive fluorescent reporter, other studies relied on solid-phase binding assays, which might be subject to complicating nonspecific effects. Another possibility is the existence of alternative binding modes when critical sites are deleted by mutagenesis, as was discussed above for CR567 binding to the Lys-65/Lys-69 variant. Also, in contrast to a previous study (20), we find no evidence for involvement of either Arg-115 or Arg-118 in promoting binding of CR456. Indeed, removal of these residues, which lie between the Arg-76/Lys-80 pair and Lys-69, actually increases affinity for both CR56 and CR456. Finally, using CR56 species containing only a single tryptophan, we showed that tryptophan within each CR domain results in ~3-fold lower Kd than when phenylalanine is present, perhaps explaining why tryptophan is the highly preferred residue at this location in CR domains within ligand binding regions of LRP1 and other LDL receptor family members.
From the respective contributions of each of the four basic residues to the three CR domain binding sites, it is clear that the strongest interaction by far is between CR5 and Arg-76 and Lys-80, followed by CR6 binding Lys-69, and the weakest interaction is between CR4 and Lys-88. This preference holds when CR56 binds resulting in CR5 and CR6 occupying the same sites. Even for CR567, the benefit of CR5 occupying the “middle” site (see Fig. 1C) appears to outweigh the loss from not occupying the much weaker Lys-88 site. Only with CR345, when occupation of the middle site by CR5 would result in the tight “bottom” site (Lys-69) being empty (see Fig. 1C), does there appear to be a shift in register, such that CR5 probably binds to Lys-69, while CR4 binds to the Arg-76/Lys-80 pair and CR3 binds to Lys-88. We have previously shown that the single CR domains CR3, CR5, and CR8 show little variation in affinity for the model compounds arginine methyl ester or lysine methyl ester (22). It is thus initially puzzling that the middle site, containing Arg-76 and Lys-80, seems to strongly prefer CR5. One possibility is that CR5 has a more negatively charged surface at the engagement site, with the motif Glu-Xaa-Glu-Glu-Glu present in CR5 compared with Glu-Xaa-Glu-Xaa-Glu for each of the domains CR3, CR4, and CR6 (CR7, CR8, and CR9 also have the latter motif). This may result in stronger interaction with the two basic residues in the middle site.
Whereas the residues involved in CR456 binding were identified solely by mutagenesis, it turns out that they are also well positioned relative to one another to be able to engage the three separate CR domains. Guidance is provided by the x-ray structure of the two CR domain fragment from LDL receptor complexed to the RAP D3 domain (15) and of the structure of LDL receptor extracellular domain at low pH in which the YWTD domain interacts with two CR domains of the ligand binding region (39). In the former structure the separation between equivalent positions in the two CR domains (Cα-Cα between Phe-105 and Trp-144) is ~22 Å, while in the latter it is ~19 Å between equivalent tryptophans (Trp-144 and Trp-193). For the ligand D3 in the complex, the separation between the ϵ-amino groups of the principal lysines involved in binding (Lys-256 and Lys-270) is 21 Å. In a study in which Lys-256 was removed from D3, leaving Lys-253 and Lys-270 as the principal lysines, with separation between ϵ-amino groups of 25 Å, CR56 was still able to bind, although with greatly reduced affinity (21). In the highest resolution PAI-1 structure available, that of latent PAI-1 at 1.8 Å (1LJ5), the separation between the ϵ-amino groups of Lys-88 and Lys-80 is 25 Å and between Lys-80 and Lys-69 is 19.3 Å. Although only one of these is similar to what is seen in actual protein complexes, it should be noted that the Cα-Cα separations for the same residue pairs are 13.9 and 19.1 Å (Table 4), so that side chain positioning in solution, which might be different from in the crystal, could easily give ~19 Å separation even for the pair Lys-88 and Lys-80. Independent support for the proposed binding site comes from a study on a peptide inhibitor of PAI-1 binding to LRP1 (40). Modeling of the binding site on PAI-1 suggested that the peptide binds close to the residues Arg-115, Lys-69, Arg-76, and Lys-80, the last three of which we have identified here as part of the CR456-binding site. Significantly, each of the four basic residues constituting the CR456-binding site is highly conserved among mammalian species, with variation only within the other basic residues (Table 5).
An important question about our identification of the CR456-binding site is whether it is the same in all conformational forms of PAI-1, i.e. native, latent, cleaved, and proteinase-complexed. Comparison of the relative positions of the four basic residues identified in the three states of PAI-1 for which there are x-ray structures (native, latent, and cleaved) (Table 4) shows that there is only a small variation for any of the Cα-Cα separations, which implies that CR456 should be able to engage the same four residues in the same way in each of these conformational states, especially given the length and flexibility of the basic residue side chains. This is in keeping with reports from this laboratory and others that native, latent, and cleaved PAI-1 bind with effectively identical affinity to LRP1 or LRP1 fragments (12, 18, 25). It therefore seems likely that the high affinity binding site we have identified on native PAI-1 is present with the same affinity in both latent and cleaved PAI-1, even though latent and cleaved have undergone major conformational changes relative to native. Importantly, these structural changes do not affect the region of PAI-1 where the LRP1-binding site is located (41).
An even more important question than whether different conformations of PAI-1 bind similarly to LRP or its fragments is whether the same binding site is present in the PAI-1 moiety in covalent complex with uPA or other proteinases. Although there is no x-ray structure of PAI-1·proteinase complex to directly answer the question, there are x-ray structures of other serpin·proteinase complexes, viz. α1-proteinase inhibitor in complex with either trypsin (9) or porcine pancreatic elastase (8), as well as NMR structural data on the trypsin complex (42). All of these show that the secondary structure of the serpin moiety is unchanged from that of the cleaved serpin. Given this, the presence of a “cryptic” LRP1-binding site within PAI-1, which is only uncovered in complex (13, 18), seems unlikely. The suggestion of a cryptic site was based on the greater efficiency of cellular clearance of uPA·PAI-1 complex compared with any of the uncomplexed forms of PAI-1, even the structurally equivalent cleaved form. However, it seems more likely that the explanation for this lies with the involvement of uPAR as a co-receptor to LRP1 in clearance of uPA·PAI-1 complexes but not for cleaved, latent, or native PAI-1, which rely only on LRP1 for clearance. By localizing the PAI-1 to the cell surface, through its covalent attachment to uPA, which is in turn non-covalently associated with the membrane-anchored receptor uPAR, the interaction of the PAI-1 moiety with LRP1 would be greatly facilitated, even without any alteration in its binding site for PAI-1 or any need to invoke a binding contribution to LRP1 from the proteinase moiety. Although it is reported that uPA·PAI-1 binds to LRP1 with Kd about 10-fold smaller than PAI-1, the experimental setup does not mimic the situation in vivo (12, 18). Thus, the uPA·PAI-1 complex has the N-terminal region of the proteinase exposed, whereas in vivo, this would be obscured through binding to uPAR (3, 4, 43). Because this N-terminal region has an exposed single lysine that becomes deeply buried upon receptor binding (Lys-23) and a pair of lysines that rest at the edge of the binding pocket (Lys-34 and Lys-35) and would likely be inaccessible to LRP1, it is quite possible that high affinity binding site(s) on uPA for LRP1 are only available when uPA is not in complex with its receptor. This is supported by a study that showed that the N-terminal region of free uPA is involved in binding to LRP1 and that uPAR protects against such binding (12). Furthermore, the separate affinities of PAI-1 and uPA for LRP are so high that if both binding sites were engaged by LRP when uPA·PAI-1 complex binds, the Kd would be in the attomolar (10−18) region, whereas measured values are only ~2 × 10−10 m (20). Taken together, these findings suggest a model for PAI-1·proteinase complex binding to LRP1 shown in Fig. 8. Here, the orientation of the PAI-1 moiety in complex with uPA, derived from the two x-ray structures of serpin·proteinase covalent complexes (8, 9), aligns the PAI-1-binding site correctly with the membrane-anchored LRP1. The only contact between the uPA·PAI-1 complex is with the PAI-1-binding site identified here, but the intrinsic high affinity (1.5 nm) together with the localization of the complex to the cell surface, through being bound to uPAR, would ensure much more efficient clearance than for PAI-1 alone.
Although the intrinsic high affinity of PAI-1 alone reported here for an LRP1 fragment may result in a large fraction of total PAI-1 being complexed to LRP1 at any given time, it nevertheless must result in the observed free concentration reported for circulating PAI-1 of ~2 nm (44). Complex formation with uPA would still result in effective clearance via the proposed co-receptor mechanism, without any further need for a decrease in Kd for the uPA·PAI-1 species for LRP1.
Finally, it is worth considering this model of uPA·PAI-1 binding and clearance with respect to a closely related serpin proteinase nexin 1 (PN1). This is the second of the two human clade E serpins (45) and is also capable of inhibiting uPA by the serpin mechanism. It binds about as tightly to LRP1 as does PAI-1 (46) and with similar affinity to LRP1 fragments in both native and cleaved states (25). X-ray structures of PN1 in complex with heparin or S195A thrombin show that there are basic residues, Arg-63, Lys-71, Lys-74, and Lys-83, at structurally equivalent positions to the four identified here in PAI-1 as composing the LRP1-binding site (Table 4). They are also all highly conserved among mammalian PN1 species. PN1 may thus use the same binding site for binding to LRP1 as does PAI-1, and the same mechanism of co-receptor clearance, at least for complexes with uPA.
P. G. W. G. designed the study, collected, and analyzed the data, and wrote the manuscript. K. D. engineered, expressed, and labeled the proteins. Both authors have read and approved the final version of the manuscript.
We thank Andres Campos for preparing some of the PAI-1 species and Steven Olson for helpful comments on the manuscript.
*This work was supported by National Institutes of Health Grant R01 GM54414. The authors declare no conflict of interest with the contents of this article. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
3In the CR5F6W, CR5W6F, and CR5F6F species, the tryptophans in CR5, CR6, or both, were changed to phenylalanine.
2The abbreviations used are: