Peptide binding induces large scale changes in inter-domain mobility in human Pin1.

Pin1 is a peptidyl-prolyl cis/trans isomerase (PPIase) essential for cell cycle regulation. Pin1-catalyzed peptidyl-prolyl isomerization provides a key conformational switch to activate phosphorylation sites with the common phospho-Ser/Thr-Pro sequence motif. This motif is ubiquitously exploited in cellular response to a variety of signals. Pin1 is able to bind phospho-Ser/Thr-Pro-containing sequences at two different sites that compete for the same substrate. One binding site is located within the N-terminal WW domain, which is essential for protein targeting and localization. The other binding site is located in the C-terminal catalytic domain, which is structural homologous to the FK506-binding protein (FKBP) class of PPIases. A flexible linker of 12 residues connects the WW and catalytic domain. To characterize the structure and dynamics of full-length Pin1 in solution, high resolution NMR methods have been used to map the nature of interactions between the two domains of Pin1. In addition, the influence of target peptides on domain interactions has been investigated. The studies reveal a dynamic picture of the domain interactions. 15N spin relaxation data, differential chemical shift mapping, and residual dipolar coupling data indicate that Pin1 can either behave as two independent domains connected by the flexible linker or as a single intact domain with some amount of hinge bending motion depending on the sequence of the bound peptide. The functional importance of the modulation of relative domain flexibility in light of the multitude of interaction partners of Pin1 is discussed.

trans isomerization of peptidyl-prolyl imide bonds (1)(2)(3)(4). Recently, with the discovery of the parvulin class of PPIases, peptidyl-prolyl isomerization has been identified as a key step in cell cycle regulation, oncogenesis, signal transduction, and a multitude of other cellular processes. Pin1 is the best-characterized PPIase of the parvulins and selectively catalyzes cis/ trans isomerization of peptide sequences containing the phospho-Ser/Pro or phospho-Thr/Pro motif (pS/T-P) (5-7). Lu et al. (5) have proposed that, in the presence of certain kinases, Pin1 participates in a "tag and twist" mechanism to activate its substrate: the kinase phosphorylates the serine or threonine side chain (tagging), and Pin1 subsequently cis/trans isomerizes (twists) the peptide bond preceding proline in a di-peptide pS/T-P, thus introducing a marked kink in the backbone of the peptide chain.
Pin1 is a two-domain protein of 18.4 kDa consisting of an N-terminal WW domain (Pin1 WW ) important for substrate targeting (15), and a C-terminal catalytic domain (Pin1 CAT ), which changes the substrate conformation by catalyzing the cis/trans isomerization of peptidyl-prolyl imide bonds. Previously reported x-ray and NMR structures of full-length Pin1 (Pin1 FL ) reveal a global similarity of Pin1 CAT with the FKBP-like class of PPIases and of Pin1 WW with other WW domains found in single or tandem repeats in over 25 unrelated cell-signaling proteins (16).
Both domains selectively bind pS/T-P-containing substrate peptides, but only the catalytic domain isomerizes this sequence motif (7). Selectivity is achieved by specific phosphatebinding sites, of which one is located next to the active site of Pin1 CAT and the other within the peptide binding epitope of Pin1 WW . Two x-ray studies and one NMR study have characterized the Pin1-peptide interactions previously: Ranganathan et al. (17) have solved the structure Pin1 FL in complex with the Ala-Pro-dipeptide (PDB entry 1PIN), which is presumed to bind to the active site of the catalytic domain. Verdecia et al. (18) have elucidated the structure of Pin1 FL complexed to the doubly phosphorylated peptide YpSPTpSPS of the C-terminal domain of the RNA polymerase II large subunit (PDB entry 1F8A). The NMR study of Wintjens et al. (19) compares the structure of free Pin1 WW with Pin1 WW in complex with peptides from the Cdc25 phosphatase (Cdc25) and the cytoskeletal protein tau (PDB entries 1I6C, 1I8G, and 1I8H). Wintjens et al. suggest a unique binding scheme in which the peptide substrates all bind in the same orientation with the peptidyl-prolyl bond being in the trans conformation.
Pin1 encodes two functional elements, localization and catalysis, each in one separate domain. A flexible linker tethers both domains to one another. In terms of specificity toward proteins, it is functionally important to establish whether these domains are independent of one another or whether they interact in the absence and/or presence of the substrate. To address the impact of peptide binding on the inter-domain motion of the two domains of Pin1, we applied standard NMR methodology (20,21). We have determined the domain-domain interaction surface of Pin1 under solution conditions by differential chemical shift mapping. Estimated overall correlation times derived from both 15 N spin relaxation data and HydroNMR calculations were analyzed to characterize the effects of peptide binding on the domain flexibility of Pin1 FL . Furthermore, residual dipolar couplings provided additional supporting information about domain-domain orientation and motion (22)(23)(24)(25). In work now submitted for publication, 2 Brownian Dynamic simulations were performed to interpret the NMR relaxation properties of Pin1 with respect to global and inter-domain motion.
Full-length Pin1 (Pin1 FL ) and the two domain fragments (Pin1 WWϩL and Pin1 LϩCAT ) have been studied in the presence of the following three different peptides: (i) the peptideWFYp-SPR (Pintide), which was originally designed from a peptide library to be an optimal substrate for Pin1 (38); (ii) the doubly phosphorylated peptide YpSPTpSPS (CTD), which occurs as a heptad repeat in the C-terminal domain of the large subunit of RNA polymerase II; and (iii) the mono-phosphorylated peptide EQPLpTPVTDL (Cdc25), which is a model peptide from the essential mitotic phosphatase Cdc25 (1).
Binding Assays-The Pin1-peptide binding studies were conducted by monitoring changes in 1 H-15 N HSQC spectra. The chemical shift changes, ⌬, were calculated using Equation 1.
Generally, 0.25 mM protein solutions were employed. All three peptides were added in 8-fold excess.
To measure the dissociation binding constants (K D ) of Pintide and the CTD peptide binding to Pin1, HSQC spectra were recorded for a set of successively diluted samples with protein concentrations of 0.5, 0.2, 0.1, 0.05, and 0.02 mM. For these the protein-peptide ratio was fixed at 1:4, and the chemical shift changes due to the dissociation of the protein-peptide complex at successively lower protein (and peptide) concentrations were monitored. The observed chemical shift changes (⌬ 1Hϩ15N ) were then fitted to the following expression by adjusting K D and A, where P 0 is the total protein concentration, and A ϭ ␦ free Ϫ ␦ bound , i.e. the difference in chemical shift between the free (␦ free ) and bound (␦ bound ) states. This equation has been derived from the mass law for the fixed protein-ligand ratio of 1:4 and results after substitution of Equa-  15 N R 2 (Carr-Purcell-Meiboom-Gill (CPMG)) values were measured at 600 MHz with conventional pulse sequences (28). The protein concentration was 0.5 mM. The CTD and Cdc25 peptides were added in 6-fold excess, whereas the concentration of Pintide was 2 mM due to its limited solubility. The 15 N R 1 and the 15 N R 2 relaxation decays were sampled at eight different time points each (T 1 delays ϭ 0.005, 0.079, 0.079, 0.159, 0.239, 0.239, 0.359, 0.519, 0.759, and 1.119 s; T 2 delays ϭ 0.009, 0.017, 0.017, 0.034, 0.060, 0.086, 0.086, 0.112, 0.148, and 0.190 s) with duplicate spectra for each time point. All spectra were acquired with 160 ϫ 1024 complex points. A recycle delay of 4 s and eight transients were applied for the R 1 and R 2 experiments. For the unsaturated { 1 H}-15 N NOE measurement, a recycle delay of 5 s was used. This recycle delay was substituted in the presaturated { 1 H}-15 N NOE experiment with a 0.5-s delay followed by a 4.5-s long series of non-selective 120°1H pulses separated by 5-ms delay. The NOE spectra were recorded in an interleaved manner with 32 transients each.
Backbone amide 15 N R 1 and R 2 relaxation rates were determined by fitting peak heights as functions of relaxation decay times to singleexponential decay functions. Steady-state NOE values were calculated from the ratios of the peak intensities with and without proton saturation (29). The standard errors for R 1 , R 2 , and { 1 H}-15 N NOE were estimated to 5%.
From a total of 156 backbone 15 N-amides, relaxation data were extracted for 118 residues of Pin1 FL , for 111 of Pintide-bound Pin1 WT , for 126 of CTD-bound Pin1 FL , and for 120 of Cdc25-bound Pin1 FL . For Pin WWϩL 46 out of 54 residues and for Pin1 LϩCAT 76 out of 109 residues were used for analysis. The residues Arg-17, Ser-18, Ser-19, Ser-42, Ser-43, Gly-44, Gln-75, Glu-76, and Glu-145 are not assigned. However, Ser-18 appears in the HSQC spectrum of peptide-bound Pin1 FL . 29 residues in Pin1 FL , 37 residues in Pintide-bound Pin1 FL , 22 residues in CTD-bound Pin1 FL , 28 in Cdc25-bound Pin1 FL , and 27 residues in Pin1 CAT exhibit spectral overlap. The residues from which relaxation data were obtained and those used to determine the estimated overall correlation times are listed in the supplementary material. The isotropic correlation times were calculated using the program Tensor 2.0 (30). Two hundred Monte Carlo simulations were carried out to assess the validity of the isotropic diffusion model. Because the averaged correlation times are biased by the choice of residues, a range of correlation times was calculated resulting in deviations of 0.4 ns for Pin1 WW and 0.1 ns for Pin1 CAT .
Because Pintide-bound Pin1 FL is only saturated to ϳ85%, the relaxation rates are extrapolated to 100% saturation as follows, where R 1,2 extr , R 1,2 obs , and R 1,2 PinFL are the R 1 and R 2 relaxation rates of saturated Pin1 FL -Pintide complex, the non-saturated Pin1 FL -Pintide complex, and the Apo-Pin1 FL , respectively, and bound is the fraction bound or degree of saturation. The above extrapolation requires that bound and free peptide be in fast exchange, which can be safely assumed because we observe an averaging of the chemical shifts of the bound and free state in all titration studies used to determine the K D values of the peptides.
Dipolar Couplings-We employed two types of media for the partially aligned media, a nonionic liquid crystalline medium and bacteriophages. The liquid crystalline medium (49) was prepared from 1-hexanol and C12E5 (Fluka). A 10% solution of this medium was prepared in buffer containing 10% D 2 O. This solution was added in equal amounts to a 1 mM protein stock solution, yielding a final 5% medium/0.5 mM protein solution. For samples containing peptide, solid peptide was added until a solid precipitate emerged. Aligning samples containing phages (50) were prepared by adding a concentrated solution of filamentous phage PF11 (resulting concentration 21 mg/ml; Asla Ltd., Riga, Latvia) to the protein solution.
Alignment of the samples was monitored via the splitting of the deuterium signal. N-HN heteronuclear dipolar couplings of the sample containing phages were measured at 800 MHz using the IPAP scheme (31). N-HN heteronuclear dipolar couplings of the samples containing liquid crystalline medium were measured at 600 MHz as S 3 E spectra (32). In each case, 256 T 1 data points were acquired and zero-filled to 1024 points. The splitting was evaluated using NMRView 5.0.4 (B. A. Johnson, Merck Research Laboratories). The RDC values were obtained by subtracting the reference value of the protein in isotropic solution.
Like for evaluation of relaxation data, the dipolar couplings were extrapolated to 100% saturation. The fraction bound was 85% for the phage solution and only 40% for the nonionic liquid crystals. The low saturation degree for the nonionic liquid crystals is due to peptide binding to the liquid crystals. Because the extrapolation of the data from the nonionic liquid crystals is less accurate, the data from phage solution were assumed to be more reliable. For analysis of the data we used Pales (33) and Module (34). In Pales, the tensor values were obtained using the SVD (35) and a maximum likelihood method (36). The Q value (21) was used to assess the agreement between experimental and theoretical values.
The x-ray structures of Apo-Pin1 (PDB entry 1PIN) and of Pin1 bound to the CTD peptide (PDB entry 1F8A) were used for all models; for consistency, we used the residue numbering of 1PIN for both structures. Protons were added to both structures. In all cases, the WW domain loop (residues Ser-16 to Arg-21) and the flexible linker (residues Glu-35 to Ala-53) have been omitted from the analysis. Coordinate files with an aligned arrangement of the two domain were generated semimanually with Module (37).

RESULTS
The Two Domain Nature of Pin1 and the Role of the Flexible Linker-To characterize domain-domain interactions and dynamics NMR experiments were carried out on three different constructs of Pin1: (i) Pin1 FL , the full-length protein (residues 1-163); (ii) Pin1 WWϩL , the isolated WW domain plus linker (residues 1-54); and (iii) Pin1 LϩCAT , the isolated catalytic domain plus linker (residues 44 -163). The residue numbering refers to that of the PDB structure 1PIN. Notably, each of the single-domain constructs includes the linker. Hence, the effect of the absence of a domain is studied rather than the absence of the covalently attached linker. This ensures minimal structural and chemical shift perturbation within the single domain constructs. All three constructs have been studied in the presence of three different peptides, namely WFYpSPR (Pintide), YpSPTpSPS (CTD), and EQPLpTPVTDL (Cdc25).
Differential Chemical Shift Mapping: Effect of the Linker- Fig. 1 shows the differences in chemical shifts between Pin1 FL and either Pin1 LϩCAT or Pin1 WWϩL , mapped onto the x-ray structure of Ranganathan et al. (17). As also seen in Fig. 2A, the strongest chemical shift perturbations due to the presence of the second domain are found in the ␤3 strand as well as the ␤1/␤2 and ␤2/␤3 loops of the WW domain forming the ends of its concave surface. In the catalytic domain, the chemical shift changes are clustered around the strand ␤6 and helix ␣4. Observed perturbations within strand ␤7 are likely due to the influence of the linker communicating the absence of the WW domain. Weaker shift perturbations are found for individual residues distributed throughout the catalytic domain. Interestingly, the most affected regions describe the hydrophobic cavity comprising the concave surface of the WW domain and the back side (strand ␤6 and helix ␣4) of the catalytic domain. Ranganathan et al. (17) have also identified this inter-domain interaction surface in their x-ray structure. However, in this study helix ␣1, which is only weakly perturbed in solution, is also included in the interaction surface. Notably, Pin1At, a homologous cis/trans isomerase from Arabidopsis thaliana lacking the WW domain, consists of negatively charged residues at the comparable surface (Pin1At: Asp-105, Asp-108, Glu-51, and Asp-52; hPin1 FL : Gly-148, Phe-151, Gln-94, and Lys-95) (39). Pin1 WW shields this surface of the catalytic domain. Hence, the two domains are most likely coupled by weak hydrophobic and hydrophilic inter-domain interactions and not directly by the linker, because this linker is too flexible to give rise to specific interactions. Interestingly, attractive forces between the domains are not large enough to persist without the linker. Mixing Pin1 WWϩL and Pin1 LϩCAT , at 0.25 mM concentration, did not show any chemical shift perturbations when compared with the shifts of the individual domains. These studies clearly establish that WW and catalytic domains interact and that one role of the linker is to increase the local concentrations of each of the domains with respect to each other. The recently solved solution structure of Pin1 by Bayer et al. (40) confirms this weak interaction of both domains across the common binding interface, because no NOEs between the domains could be observed.
Effect of the Peptides-To study the interaction of three model peptides with Pin1 FL and its isolated domains, it was necessary to know the saturation of the peptide-binding site located within the WW domain. In our studies the saturation of Pin1 with peptide was estimated to be 85% for Pintide (K D ϭ 200 -400 M), 90% for the CTD peptide (K D ϭ 200 M), and 93% for the Cdc25 peptide (K D ϭ 117 M (19)). The affinities determined in this study for Pintide and the CTD peptide are about one order of magnitude larger than the dissociation constants reported by Verdecia et al. (18), who found values of 17 M for Pintide and 10 M for the CTD peptide. The differences are likely due to the lower pH in our study (6.6 versus 7.5), which alters the protonation state of the phosphate group and thus affects binding (41,42). Besides, increasing salt concentration decreases dramatically the PPIase activity of Pin1 implying that charges within or near the PPIase active site are also critical for a productive interaction between Pin1 with its substrate (42). Nevertheless the relative affinities for both peptides are reproduced as Pintide binds more strongly than the CTD peptide. Verdecia et al. (18) have reported that full-length Pin1 has higher apparent affinities to its substrates than its isolated domains. In Fig. 2 (B-D) chemical shift changes of three different peptides bound to each of the constructs (Pin1 FL , Pin1 LϩCAT , or Pin1 WWϩL ) are shown as a function of residue number.
Peptide Binding to the Isolated Domains-The yellow data points in Fig. 2 show that all three peptides induce strong shift changes in Pin1 WWϩL . Almost all residues within the WW domain are affected indicating that the peptide binding is as- sociated with conformational changes (19,43). The pattern of chemical shift changes along the sequence is similar for all peptides supporting Wintjens' hypothesis (19) of conserved interactions between the WW domain and its various peptide substrates. In contrast, peptide binding to Pin1 LϩCAT is much more variable (red data points in Fig. 2): The Cdc25 peptide does not bind to the Pin1 LϩCAT construct at all, whereas Pintide provokes chemical shift changes spread over the entire catalytic domain. The CTD peptide only produces few shift changes localized mainly in helix ␣3.
Peptide Binding to Pin1 FL -As seen in Fig. 2  Unlike the isolated SH3-SH2 fragment in the Src kinase whose linker residues are involved in nonspecific, hydrophobic domain-linker contacts and thus exhibit low positive { 1 H}-15 N NOEs (20), the linker of Pin1 FL is completely disordered. The flexibility of the linker is not notably reduced upon peptide binding. This suggests that the linker loosely connects the two domains but does not interact with the domains or the bound peptide. This is consistent with the fact that linker residues do not show electron density in the x-ray structures. Backbone 15 N longitudinal and transverse relaxation rates, R 1 and R 2 , are sensitive to rotational diffusive motion. To monitor inter-domain flexibility we estimate overall correlation times ( c ) of the two domains in their free, linked, and complexed states from the R 2 /R 1 ratio (Fig. 3B). In this approach, both internal and anisotropic motions are neglected. Only those residues in secondary structure elements without appreciable internal motion are taken into account. In addition, those residues are excluded that exhibit R 2 /R 1 ratios that considerably deviate from the average value. This procedure still samples all possible orientations of the NH vectors in a reliable manner. The accuracy of the estimated c depends on the number of selected residues, internal motion, and/or anisotropic motion. It is higher for Pin1 CAT than for Pin1 WW , because significantly more residues could be selected in the larger catalytic domain and because Pin1 WW is known to undergo significant conformational motion on the millisecond time scale (43). A summary of the estimated c and a listing of residues they were derived from are given in Table SI (see Supplementary Material).
For further analysis of the estimated c the following rigid sphere model depicted in Fig. 4 is assumed. Pin1 is thought to consist of two rigid spheres of different sizes connected by a flexible linker. The model assumes further that the domains tumble isotropically as rigid bodies. In this case, 15 N relaxation rates R 1 and R 2 are only a function of the overall c , which is correlated with the volume via the Stokes-Einstein-Debye law and hence is proportional to the molecular weight of the domains. The model defines two limiting cases where (i) the two domains tumble independently of one another with correlation times c WWϩL and c LϩCAT and (ii) the domains stick together and tumble as one intact moiety with a c FL-rigid for the fulllength rigid Pin1. According to the model, these limiting c values are proportional to the masses of the constructs (m WW ϭ 4.51 kDa; m WWϩL ϭ 5.55 kDa; m CAT ϭ 12.71 kDa; m LϩCAT ϭ 13.75 kDa; and m FL ϭ 18.26 kDa). Any deviation from these extreme cases is described by an inter-domain interaction parameter x, which quantifies in a simple additive manner how much the c of one domain will depend on the c of the other domain as follows: c WW (x) ϭ c WWϩL ϩ x c CAT and c CAT (x) ϭ c LϩCAT ϩ x c WW (with 0 Ͻ x Ͻ 1 and c WW and c CAT proportional to the molecular weights of the respective domains without the linker). The y-axis intercepts of these equations reflect the fact that a given domain will always interact with the entire linker, whose function is to increase the local concentration of the two domains. Each domain also interacts directly with the other domain via the domain-domain interaction surface as is described by the second term involving the x parameter. Fig. 4 depicts these theoretical curves and shows how well the experimental data fit to a model where the only adjustable parameter, c FL-rigid , is chosen to be 11.6 ns to yield a residual R 2 value of 0.95.
We find that the experimental c values of the isolated do- mains are 0.5-1 ns longer than the expected theoretical values. Aggregation may be one explanation for this discrepancy. However, this systematic error is more likely due to an underestimation of complex effects of the flexible linker, which are not adequately captured in the model. Brownian Dynamic simulations 2 may provide a better model for the flexible linker, and future relaxation studies of constructs without the linker will help validate these models.
Comparing the correlation times of Apo-Pin1 FL with those of the isolated domains, the c values for Pin1 WW and Pin1 CAT in Apo-Pin1 FL are significantly larger than those for the isolated domains resulting in an inter-domain interaction parameter of 0.47. Thus, on the one hand the linker couples both domains and restricts their flexibility. On the other hand, the difference of about 2 ns between the correlation times of the two domains indicates that Apo-Pin1 FL does not tumble as one single rigid body but shows that the two domains tumble independently to a significant extent. Fig. 4 shows that significantly differing inter-domain interaction parameters account for the experimental data of the three peptides bound to Pin1 FL , suggesting that distinguishable inter-domain flexibility is induced upon peptide binding. The inter-domain interaction parameter of 0.43 for the Cdc25 peptide bound to Pin1 FL is slightly lower than the value for Apo-Pin1 FL . Therefore, binding of this peptide clearly does not result in any additional flexibility and may even disrupt the interaction among the domains. In contrast to the Cdc25 peptide, the CTD peptide mediates stronger contacts between the two domains as evidenced by the larger x value of 0.68. The highest inter-domain interaction parameter of 0.79 is obtained for the Pintide peptide. Thus, this peptide significantly restricts the flexibility of the two domains. An even higher value of x ϭ 0.85 is obtained if one extrapolates R 2 /R 1 and c values from the actual 85% protein peptide saturation to full saturation. This confirms that Pintide significantly restricts the flexibility of the two domains. Nevertheless, total rigidity (x ϭ 1) is not attained, indicating that residual flexibility and possibly hinge bending motions are still present.
In agreement with the chemical shift perturbation, the interdomain flexibility is reduced in the order Cdc25 Ͻ CTD Ͻ Pintide. Assuming that peptides bind in a similar orientation as they do in the crystal and NMR structures, the N terminus of the peptides is found to be orientated toward the hydrophobic parts of the inter-domain interaction surface. Hence, hydrophobic residues in the N terminus of the peptide may be decisive for the induction of favorable inter-domain interactions. Indeed, the hydrophobicity of the N terminus increases in the order Cdc25 Ͻ CTD Ͻ Pintide. However, the size of the peptides decreases in the same order as well. Hence, it cannot be excluded that the two domains form a floppy hydrophobic cleft accommodating molecules only of limited size and that the observed trend is predominantly a steric hindrance effect.
Hydrodynamic Calculations-Hydrodynamic calculations establish the connection between an arbitrarily shaped molecule and its hydrodynamic properties in solution. The program HydroNMR computes NMR relaxation rates R 1 , R 2 , and { 1 H}-15 N NOEs from a rigid structure (45). Using the atomic coordinate file, a primary hydrodynamic model is built from beads replacing each nonhydrogen atom and having a so-called atomic element radius a. In this model the only adjustable parameter is the atomic element radius a, with a ϭ 3.30 Å being the natural value (46). Larger a values indicate aggregation or oligomerization, whereas lower a values hint at interdomain mobility. In our case, the R 2 /R 1 ratios of the catalytic domain and the WW domain differ systematically due to the combination of domain size mismatch and the above discussed domain flexibility. Hence, the a values were derived from fitting to the experimental R 2 /R 1 ratios of the catalytic domain only and from using the crystal structure of Ranganathan et al. (17) (PDB entry 1PIN). As a result, the calculated R 2 /R 1 values agree with the experimental values of the catalytic domain but overestimate the ratios of the WW domain (see Fig. S6, Supplementary Material).
For Pintide bound to Pin1 FL , the atomic radius parameter a was found to be 3.3 Å. This suggests that both domains interact tightly and tumble almost as a single intact domain. In comparison, the experimental R 2 /R 1 of Apo-Pin1 FL could only be fitted with an a value of 2.5 Å. For Cdc25 bound to Pin1 FL and CTD bound to Pin1 FL , the a values were calculated to 2.5 and 3.1 Å, respectively. These a values can be interpreted as increasing inter-domain mobility in comparison to Pintide bound to Pin1 FL . They show the same trend as the inter-domain interaction parameters x.
The degree of anisotropy was calculated from the rotational diffusion tensors computed by HydroNMR and is tabulated in Table SI (Supplementary Material). Evidently, Pin1 WWϩL (D ʈ /D Ќ ϭ 1.39) tumbles more anisotropically than Pin1 LϩCAT (D ʈ /D Ќ ϭ 1.24). This anisotropy cannot be neglected demonstrating that the assumption of isotropic tumbling is valid only to a first approximation for the discussion above. In comparison to Pin1 LϩCAT , the anisotropy of Pin1 FL (D ʈ /D Ќ ϭ 1.29) does not increase significantly. This does not necessarily indicate that no additional anisotropy is induced by domain contacts, because changes of both the orientation and the size of the diffusion tensors need to be monitored and interpreted.
Fit of Dipolar Couplings to the Isolated Domains-Relative domain-domain orientations are generally obtained by aligning the tensor orientations of the isolated domains to one orientation for the whole protein. Backbone N-HN heteronuclear RDCs of Apo-Pin1 FL and Pintide-bound Pin1 FL were measured using two different alignment media, a nonionic liquid crystalline medium (49) and bacteriophages (50).
Our fit of the experimental RDCs to the isolated domains agrees well with the crystallographic data. With respect to Pin1 CAT , our data fit better to the crystal structure of PDB entry 1PIN (17) than to the crystal structure of PDB entry 1F8A (18). Both structures differ in the region of the phosphate binding loop due to enhanced flexibility and crystal-packing interactions (18). Concerning the flexible loop (residues Ser-16 to Arg-21) of Pin1 WW , we generally find poor agreement between experimental and predicted data. This loop adopts different conformations in both the x-ray (17,18) and NMR structures (19,43) and has been assumed to display a high degree of conformational flexibility.
Size of the Alignment Tensor-In the case of Apo-Pin1 FL aligned in both liquid crystalline medium and in phage solution, the distribution of experimental RDCs is much broader for Pin1 CAT than for Pin1 WW , indicating a higher degree of alignment for Pin1 CAT than for Pin1 WW . As seen in Fig. 5 the axial components of the tensors calculated from these values yield a better quantification. In both alignment media, they differ significantly between both domains (D axial (Pin1 CAT ) ϭ 9.6 and D axial (Pin1 WW ) ϭ 2.7 for liquid crystalline medium; D axial (Pin1 CAT ) ϭ 12.6 and D axial (Pin1 WW ) ϭ 3.7 for phages). It is noteworthy that the alignment tensors of Pin1 WW are highly imprecise due to the following reasons: (i) Most of the N-H bond vectors in Pin1 WW of the available structures have similar orientations. (ii) The small size of Pin1 WW results in less RDCs than for Pin1 CAT . (iii) The available structures differ even in the structurally conserved parts without the flexible loop (residues Ser-16 to Arg-21), thus adding "structural noise." Nevertheless, values found using three different methods are comparable (single value decomposition method (35), the maximum likelihood method (36), and Metropolis Monte Carlo method (37)). The significant differences of the alignment tensors of the two domains can be interpreted in terms of inter-domain motions (22,23). Pin1 CAT is the larger domain and accounts for 76% of the mass and 71% of the surface area of Pin1 FL . Assuming that the alignment arises solely from steric factors (33), most of the orientation is expected to be due to interactions between Pin1 CAT and the cosolvent. Thus, the greatly diminished alignment tensor of Pin1 WW may reflect uncorrelated rotation of this domain with respect to fixed Pin CAT .
A partially different picture is found in the case of Pintidebound Pin1 FL . For Pin1 CAT the dipolar coupling histogram is not changed considerably by the peptide. Also, the resulting principal values and Euler angles of the alignment tensor are similar. Thus, Pin1 CAT adopts a similar orientation in comparison to Apo-Pin1 FL . In contrast, as seen in Fig. 5, both the histogram and the resulting alignment tensor are much larger for peptide-bound Pin1 WW than for free Pin1 WW (D axial (Pin1 CAT ) ϭ 13.1 and D axial (Pin1 WW ) ϭ 8.3 for liquid crystalline medium; D axial (Pin1 CAT ) ϭ 11.9 and D axial (Pin1 WW ) ϭ 8.9 for phages). As a result, the mobility of Pin1 WW with respect to Pin1 CAT is reduced significantly upon peptide binding.
Aligning the Two Domains: Structural Implications-A prerequisite for determining the relative orientation of the two domains is that the alignment tensors of both domains are of equal size. Because the alignment tensors differ less in peptidebound Pin1 FL than in Apo-Pin1 FL and because the peptide saturation in phages is higher, we concentrated on the alignment of Pintide-bound Pin1 FL obtained from the phage solution. Due to the 180°degeneracy of the residual dipolar couplings, the alignment of the two domains normally results in four possible solutions for the average structure with a common tensor orientation (34). In our case, the number of possible structures fulfilling the dipolar coupling data is even higher due to the high rhombicity (Pin1 WW : 0.57 for PDB entry 1PIN). This leads to a second set of four possible solutions, which are tilted by 90°with respect to the first four possibilities (47), because the zz-and yy-axes cannot be unambiguously assigned any longer. As a result, one of the eight possible structures agrees with the relative orientation of the crystal structure of PDB entry 1PIN. The other possible solutions can be excluded either because of topological reasons, i.e. the distance between the ends of the linker is too large, or because of the disagreement with the observed chemical shift perturbations. This confirms that the relative orientation of the domains in solution is consistent with the x-ray structure. DISCUSSION Pin1 uses two domains to facilitate targeting to and isomerization of the pS/T-P binding motif. However, in this process it is unclear whether inter-domain interactions are functionally relevant. Therefore, we investigated the two-domain nature of Pin1 using various NMR methods to address the questions whether the domains interact and, if so, whether different peptides are able to induce inter-domain interactions to a different extent.
Chemical shift perturbation and dynamic studies, including 15 N relaxation studies, residual dipolar couplings, and Hy-droNMR calculation, reveal that the two domains of Pin1 FL weakly interact and exhibit significant inter-domain flexibility. The domains interact via hydrophobic contacts between the concave peptide binding surface of the WW domain and the back side (helix ␣4 and strand ␤6) of the catalytic domain. Helix ␣1 does not take part in this interaction. Flexibility of the two domains is markedly modulated by the three peptide substrates used in our study. The inter-domain flexibility is restricted in the order of Cdc25 Ͻ CTD Ͻ Pintide. The Pintide peptide induces sufficiently strong interactions such that both domains nearly tumble as a single rigid molecule, whereas the Cdc25 peptide induces no additional inter-domain interactions but seems to weaken them. The linker is flexible in solution and is not involved in domain-linker contacts. The interaction site encompasses a hydrophobic cavity comprising the concave peptide-binding surface of the WW domain and the back side (helix ␣4 and strand ␤6) of the catalytic domain. For Pintide bound to Pin1 FL the crystallographic structure is consistent with residual dipolar coupling data.
Considering that short peptide sequences potentially do not exploit the complete length of the binding site, it remains speculative to what extent our findings are transferable to the full-length substrates. Nevertheless, in the following, two mechanistic models for the interaction of Pin1 with its substrates are discussed in the light of the significant inter-domain flexibility observed.
Wintjens et al. (19) and Zhou et al. (7) have put forward a sequential model in which phosphorylation of the S/T-P motif is a prerequisite before targeting (via the WW domain) and isomerization (via the catalytic domain) of the same pS/T-P peptide. These two processes can occur subsequently and independently. The sequential order of these events directs the isomerization equilibrium to one specific conformer. Independently of the order, this scenario holds true for one and/or several binding motifs in the substrate. For substrates with repetitive targeting motifs, like Cdc25, an initial binding event may trigger a succession of alternating binding and isomerization steps, each one revealing a further site to be isomerized (1). This hypothesis is consistent with the observation that at high cellular concentrations Pin1 CAT is sufficient to carry out the essential function of Pin1 (51). The higher catalytic activity of Pin1 FL in comparison to Pin1 CAT can be explained by the role of Pin1 WW in targeting trans pS/T-P motifs and increasing thereby the local concentration of this peptide to be isomerized. In this scenario, inter-domain contacts may serve as an extended recognition surface to control the accessibility of different substrates. However, two functionally separated domains are not mandatory for this sequential model, because the WW domain needs to dissociate in order for the catalytic domain to bind and isomerize the pS/T-P peptide substrate.
The above model can be distinguished from a model in which the WW domain anchors Pin1 to an already phosphorylated pS/T-P motif on another protein, possibly an activated kinase or phosphatase or a protein, which is part of a multienzyme complex involving a kinase or phosphatase. This scenario would enable an efficient "tag and twist" mechanism proposed by Lu et al. (5) in which the kinase would "tag" the substrate via phosphorylation and Pin1 would subsequently isomerize or "twist" the pS/T-P imide bond. In this case domain flexibility of Pin1 is crucial if it is to function as a generic "twisting" module working together with a multitude of kinases and phosphatases. A key example for this mechanism could be the C-terminal domain of RNA polymerase II, which exists in a dynamic equilibrium between hypophosphorylated and hyperphosphorylated forms resulting from antagonistic actions of specific kinases and phosphatases. In the context of these kinases and phosphatases, Pin1 is believed to contribute to the regulation of transcription and pre-mRNA maturation by inducing conformational changes after phosphorylation and/or before dephosphorylation of the C terminus (52)(53)(54). Similarly, Cdc25C is multiply phosphorylated and activated in part by Cdc2/cyclin B in a positive feedback loop during the G 2 /M transition of the cell cycle (1). In this context, Pin1 WW may bind to phosphorylated Plk I to provide a complex that first phosphorylates Cdc25C (55) and subsequently activates this substrate by cis/ trans isomerization. Moreover, Pin1 is recruited to the cyclin D1 multienzyme complex (14) and appears to be required for the checkpoint delaying the onset of mitosis in response to incomplete replication (51,56). We hypothesize that Pin1 can only be recruited and catalytically active in such a multitude of different target protein complexes due to its long flexible linker between the targeting WW domain and the catalytic domain. We believe that Pin1 acts as versatile prolyl cis/trans isomerization module in conjunction with kinases and phosphatases involved in pS/T-P-mediated signaling, because many of these kinases and phosphatases have pS/T-P motifs themselves.
In conclusion, we have shown that the two domains of Pin1 are loosely connected modules whose mobility is modulated by different peptides. Provided that the substrates can trigger similar modulations in flexibility, variable domain flexibility may enable Pin1 to function in a multitude of different contexts. This near simultaneous execution of phosphorylation and isomerization events offers an intriguing alternative to a sequential and context free targeting and isomerization model.
In the context of protein folding a similar mechanism may also be valid for multidomain PPIases such as FKBP52 (57) or trigger factor (58).