Dynamics-modulated Biological Activity of Transforming Growth Factor (cid:1) 3* □ S

Transforming growth factor (cid:1) 3 (TGF- (cid:1) 3) is an important mediator of growth, maintenance, and repair processes in human cells. Internal dynamic properties have been derived from 15 N NMR relaxation data and mapped onto the spatial structure of TGF- (cid:1) 3. The pattern of internal dynamics in the structure identifies potential “hot spots” of binding free energy and reveals the importance of conformational entropy in the interaction of TGF- (cid:1) 3 with the receptors. The observed internal dynamics set TGF- (cid:1) 3 apart from other TGF- (cid:1) isoforms, with which it shares the same fold. These findings may explain functional differences among the various TGF- (cid:1) isoforms and thus prove essential in the search for related therapeutic agents.

Transforming growth factors ␤ (TGF-␤) 1 form a group of multifunctional cytokines that account for a substantial portion of the intercellular signals governing cell fate (1). The TGF-␤ superfamily includes bone morphogenetic proteins (BMP), growth and differentiation factors, and activins/inhibins. Three mammalian TGF-␤ isoforms exhibit sequence homologies higher than 70% and are functionally closely related. They, however, show different biological activities in certain cell types or systems. Appropriate levels of TGF-␤ activity are essential to an organism's well being (2). Lack of sufficient TGF-␤ can result in immunological and inflammatory disturbances, developmental abnormalities, deficient wound healing, and increased tumorigenesis. Conversely, excessive TGF-␤ ac-tivity leads to scarring, the development of fibrotic diseases in multiple organ systems, and immune suppression.
TGF-␤ is produced by virtually all cell types as an inactive precursor, which is then cleaved into a latent complex. Activation of latent TGF-␤ in vivo is caused by proteolytic cleavage of the latency-associated peptide (LAP), by enzymatic deglycosylation of LAP, by conformational changes of the latent complex following binding to thrombospondin, and by the acidification of the pericellular space (3,4). To propagate signals across the cell membrane, the members of the TGF-␤ superfamily require two structurally related receptors (type I and type II), both having a short extracellular domain (ectodomain), a single membrane-spanning region, and an intracellular serine/threonine kinase domain (5). The members of the TGF-␤ superfamily are homodimers held together by a disulfide bond, and each monomer has binding sites for type I and type II receptors (6). During the signaling process TGF-␤ binds first to its type II receptor (T␤R2) and then to type I receptor (T␤R1). The ectodomain of T␤R2 binds with higher affinity to TGF-␤3 than to TGF-␤1, whereas the recognition of TGF-␤2 has to be supported by the presence of betaglycan (7). In the heteromeric complex of TGF-␤ with the receptors, the kinase domain of T␤R2 phosphorylates T␤R1, which in turn phosphorylates downstream intracellular signaling components (8).
The molecular basis for the diverse biological activities of TGF-␤ is not well understood. X-ray crystallographic (9,10) and NMR (11) analyses have revealed identical folds of the TGF-␤ isoforms. However, recent studies have shown important structural differences of TGF-␤3 in solution with respect to its crystal structure and the structures of other TGF-␤s (12,13). In view of the therapeutic potential of TGF-␤, we present here a dynamic study of TGF-␤3 aimed at explaining its behavior in aqueous solution and characterizing determinants involved in the interaction with receptors.

15
N Relaxation Measurements-Biologically active, recombinant human TGF-␤3 was prepared at Novartis Pharma AG (Basel, Switzerland) by refolding in vitro the monomeric, denatured protein overexpressed in Escherichia coli (14). NMR measurements were performed with a sample of 1.4 mM 15 N-labeled TGF-␤3 in mixed solvent of 87% H 2 O, 5% D 2 O, 6% dioxane-d 8 and 2% methanol-d 3 at pH 2.9 and 40°C. It was shown that these conditions do not affect the TGF-␤3 structure in solution, but provide optimal resolution in 1 H-15 N correlation spectra of TGF-␤3 (see Supplementary Material) and fully remove aggregation of TGF-␤3, which prevent NMR relaxation analysis (12,13). The backbone 15 N longitudinal R 1 and transverse R 2 relaxation rates and 15 N{ 1 H} NOE were measured on 11.7, 14.1, and 18.8 T Varian Unity and Inova spectrometers using pulse sequences described elsewhere (15). The values of 15 N R 1 and R 2 were obtained by fitting of an exponential function to the decay of signal intensities in spectra recorded with 12 * This work was supported by a grant from the Russian Foundation of Basic Research and a grant from the Royal Swedish Academy of Sciences. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.  1 The abbreviations used are: TGF, tumor necrosis factor; BMP, bone morphogenetic protein; LAP, latency-associated peptide; T␤R1 and T␤R2, type I and type II receptors of TGF-␤; BR1A and BR2, type IA and type II receptors of BMP-2; ActR2, type II receptor of activin; ec or relaxation delays ranging from 10 to 1500 ms in R 1 and from 0 to 200 ms in R 2 experiments. The R 2 experiments were carried out using a CPMG sequence with 1.0-ms delay between consecutive 15 N 180 o pulses. To account for off-resonance effects in CPMG the R 2 values were numerically corrected (16). The 15 N{ 1 H} NOEs were calculated as a ratio of signal intensities in spectra recorded with and without prior saturation of amide protons. The delay between scans in the NOE experiment was 6.0 s. The proton saturation in the NOE experiment was achieved by a sequence of 120 o 1 H pulses applied during 5.0 s. All spectra were processed and quantified using a macro within the VNMR software. Each of the R 1 , R 2 , and NOE experiments were repeated twice, and the data were averaged over the measured data sets. To account for possible small systematic errors in the experimental data, minimal uncertainties of 3% for R 1 and R 2 and Ϯ0.05 for NOE were assumed for the subsequent "model-free" data analysis. The evidence that TGF-␤3 does not aggregate at 1.4 mM concentration was provided by a R 1 experiment repeated on a diluted 0.14 mM protein sample.
Hydrodynamic Calculations-For comparison with the experimental data the rotation diffusion tensor D r of the TGF-␤3 dimer was estimated by hydrodynamic calculations using the beads model approximation (17) based on the crystal structure of the TGF-␤3 dimer (10) (PDB code 1tgj). Each residue of TGF-␤3 was represented by a spherical friction element (bead) of a 3.5-Å radius centered at the position of its C ␣ atom (18). Hydrodynamic interactions between beads were accounted for by a modified Oseen tensor. The calculations were performed with the DIFFC program (19).
Model-free Analysis of the Relaxation Data-Model-free analysis (20,21) of the relaxation data for TGF-␤3 was performed with the DASHA software (19) using a procedure described elsewhere (22,23). Values of 1.02 Å for the NH internuclear distance and Ϫ170 ppm for the 15 N chemical shift anisotropy were used in the analysis. The extensive set of nine independent relaxation measurements for each NH group allowed us to use the most complex anticipated model-free spectral-density function. Namely, intramolecular motion of an NH group was described by four adjustable parameters of the "extended" reorientation correlation function (24): order parameters S f 2 and S s 2 and correlation times f and s of pico-and nanosecond motions, respectively, and as an extra parameter by the adjustable exchange term R ex proportional to the square of the magnetic field strength. Anisotropic rotational diffusion of TGF-␤3 was accounted for by five-exponential overall rotation correlation functions of the NH groups (25). The parameters of intramolecular motions of all nuclei were optimized simultaneously with six parameters of anisotropic overall rotation: rotation correlation time R , ratios of principal components of rotational diffusion tensor D r , D x /D z and D y /D z , and the Euler angles ␣, ␤, and ␥ defining the orientation of the molecular frame where D r has diagonal form with respect to the reference molecular frame. The uncertainties of the optimized parameters were obtained from the covariance matrix of the optimized model (26). The directions of the NH vectors in the reference molecular coordinate frame were taken from the crystal structure of TGF-␤3 (10) (PDB code 1tgj).

RESULTS
Experimental Relaxation Data-15 N relaxation data were obtained for the backbone amides of 86 out of the 103 nonproline residues of TGF-␤3 at magnetic fields of 11.7, 14.1, and 18.8 T. The relaxation data for the remaining residues were not extracted due to spectral overlap and/or resonance broadening (especially for the N-terminal residues 6 -17). The measured 15 N{ 1 H} NOE, 15 N R 1 , and R 2 values ( Fig. 1, a- Overall Rotation of the TGF-␤3 Dimer in Solution-A more rigorous analysis of the relaxation data was performed using the model-free approach (20). The experimental relaxation data were fit by simultaneous optimization of the parameters of anisotropic rotation of TGF-␤3 and the parameters of intramolecular motions of individual NH groups for all 15 N nuclei. This procedure yields an overall rotation correlation time R of 16.0 Ϯ 0.3 ns, which is somewhat higher than the 14.3 Ϯ 0.6 ns calculated from the mean of the 15 N R 2 /R 1 ratios of all NH groups in regular secondary structure elements with 15 N{ 1 H} NOE values higher than 0.6. Such an underestimation is inherent to R values derived from R 2 /R 1 ratios with the assumption of very fast internal dynamics due to the unaccounted contributions of nanosecond time scale motions (22). The ratios of principal components of the rotational diffusion tensor D x /D z ϭ 0.72 Ϯ 0.14 and D y /D z ϭ 0.78 Ϯ 0.16 derived from the relaxation data reveal moderate anisotropy of TGF-␤3 tumbling in solution. Hydrodynamics calculations based on the crystal structure of the TGF-␤3 dimer predicts similar rotation diffusion tensor although with more pronounced anisotropy: D x /D z ϭ 0.42 and D y /D z ϭ 0.45. These results suggest a less extended fold of the TGF-␤3 dimer in solution than in crystal, probably due to less compact packing of regular structure elements. This is in agreement with our previous NMR study (13), which showed that the secondary structure and the global fold of the TGF-␤3 dimer in solution are close to those obtained for the protein crystal, but some regions are in dynamic equilibrium with an unfolded conformation. Moreover the CD studies in aqueous solution show that the ␣-helical content of TGF-␤3 in solution is less then expected from the known crystal and solution structures of TGF-␤s, including free TGF-␤3 in crystal (12). In contrast, a good agreement between the experimental CD spectrum measured in aqueous solution at pH 3.0 and CD spectrum calculated from crystal structure was found for TGF-␤2 (12).
Internal Dynamics of the TGF-␤3 Dimer-Model-free analysis of the relaxation data yields the order parameters S f 2 and S s 2 describing motions in the picosecond and nanosecond time scales, respectively, and the corresponding correlation times f and s (Fig. 1, d-f). The generalized order parameter S 2 ϭ S f 2 S s 2 , plotted in Fig. 1d, characterizes therefore fast motions, in contrast to exchange contribution R ex (Fig. 1g) to R 2 and resonance broadening, which indicate slow motions in the micro-millisecond time range. The generalized order parameter S 2 provides a measure of the amplitude of internal motion, where S 2 ϭ 1 means that the given N-H bond vector is fixed on the pico-to nanosecond time scale, and S 2 ϭ 0 indicates that the motion is unrestricted. These internal dynamic properties derived from the relaxation analysis are summarized in Fig. 2 by color-coded ribbon diagrams of the TGF-␤3 crystal structure (10). Each monomer of the TGF-␤3 dimer adopts an extended fold made up of four helices and four ␤-strands arranged in two irregular antiparallel ␤-sheets, ␤A/␤B and ␤C/␤D (see Fig. 3c) (10,13). The ␤-sheets represent the most rigid part of TGF-␤3 having high order parameters. Slightly enhanced pico-to nanosecond mobility with lowered S 2 values is observed in the vicinity of the short breaks of both ␤-sheets located at residues 19 -20/41-42 and 81-82/107-108, especially for residues 18, 21, 41, 82, and 107. Additionally, the spatially adjacent segments 18 -21/40 -43 and 86 -87/102-104, which are close to the break and twist of the first and the second ␤-sheets, respectively, are involved in micro-millisecond motion as revealed by R ex and resonance broadening observed for the residues from the first ␤-sheet. Residues 77 and 79 in the hydrophobic dimer interface and residues 34 and 90 in the hydrophobic pocket near the first ␤-sheet loop are also affected by slow conformational exchange. The largely remote region 91-97 comprising the hairpin loop formed by ␤-strands ␤C and ␤D undergoes rapid internal motions on the 20 -30-ps time scale with lowered S f 2 values, whereas this region is relatively rigid on the nanosecond time scale. Besides, residues 93 and 95 from this region exhibit slow conformational exchange.
In the crystal structure residues 4 -9 form ␣-helix H1 followed by a disordered, solvent-exposed loop delimited by disulfide bridge 7-16. The segment 6 -17 is involved in extensive millisecond time scale motions limiting our ability to obtain reliable 15 N relaxation data due to the strong resonance broadening. Another region having slow conformational dynamics is the segment between the ␤-strands ␤A and ␤B, including the one-turn ␣-helix H2 (residues 24 -28) and an extended loop 29 -32. Note that the loop 29 -32 is also flexible on the pico-to nanosecond time scale revealed by lowered order parameters for residues 29 and 31.
The most flexible regions of the molecule are the N-terminal residues 2-5 and the region between the ␤-sheet strands ␤B and ␤C with residues 51-75 comprising the central ␣-helix H3 and the one-turn 3 10 -helix H4. In this region, the low and non-uniform order parameters S s 2 and S f 2 and the high f of 40 -70 ps suggest a wide range of processes occurring in the pico-to nanosecond time scale presumably connected with temporary ruptures of the helices H3 and H4. The helix-coil equilibrium in TGF-␤3 was previously suggested based on the chemical shift indices of 1 H␣ and on 1 H-NOE connectivities (13). Additionally, the region 48 -75 is involved in another slow process resulting in disproportional doubling and/or tailing of the amide resonances. This process likely occurs on the second time scale with characteristic times slower than 15 N T 1 revealed by the absence of detectable exchange cross peaks in ZZ-exchange spectra (27). A possible source of this slow exchange is cis-trans isomerization of the Cys 48 -Pro 49 and/or Ser 75 -Pro 76 peptide bonds.

Dynamic Mapping the TGF-␤3 Receptor Recognition Sites-
The members of the TGF-␤ superfamily mediate their functions by binding to cell surface receptors. Several functional regions in TGF-␤ were mapped using site-directed mutagenesis of TGF-␤1 and domain swap approach applied to TGF-␤1/␤2 chimeras (28,29). It was shown that region 52-55 is very important for the interaction of TGF-␤ with the cell receptors, whereas region 92-98 regulates specific binding of isoforms to T␤R2 only. Lately, for BMP-2 (6), activin A (30), and glial cell-line derived neurotrophic factor (31), i.e. other members of the TGF-␤ superfamily, mutant proteins that exhibit altered biological activity and receptor binding affinity were constructed and analyzed. Furthermore, the crystal structure of BMP-2 in complex with two ectodomains of its type I receptor (ecBR1A 2 ⅐BMP-2) and the recently obtained crystal structure of TGF-␤3 in complex with two ectodomains of its type II receptor (ecT␤R2 2 ⅐TGF-␤3) now serve as the prototypes for the intermediate complexes of the TGF-␤ superfamily (32,33). Comparison of the ecT␤R2⅐TGF-␤3 interface with either the FIG. 1. 15 N dynamics  ; e and f, the correlation times f and s of pico-and nanosecond time scale motions, respectively; g, exchange contribution R ex to R 2 quantified at 11.7 T. The R ex contributions are very similar to those (see Supplementary Material) estimated using field dependence of R 2 -R 1 /2 (45). The uncertainties are shown by bars. Broken lines indicate the region 6 -17, which is poorly sampled due to strong broadening of NMR signals.
activin-binding site on ecActR2 or the ecBR2-binding site on BMP-2 suggests that there is significant structural diversity in the manner by which type II receptors of the TGF-␤ superfamily interact with their cognate ligands (33). As a result, two epitopes designated as "wrist" and "knuckle" interacting with BMP type I and type II receptors, respectively, were found and mapped onto the spatial structure of BMP-2. In addition the "fingertips" epitope recognizing the TGF-␤ type II receptor was found in the TGF-␤3 ligand. Overall, these findings provide a framework for the molecular description of receptor recognition and activation in the TGF-␤ superfamily.
The monomer fold of the TGF-␤ dimer was described as an FIG. 2. TGF-␤3 backbone dynamics in solution mapped to its crystal structure (10) (PDB code 1tgj) using color coding. The NMR data (13) are consistent with x-ray structure that justifies the use of the x-ray structure for the pictures. a, fast pico-to nanosecond dynamics revealed by the generalized order parameter S 2 . Secondary structure elements and chain termini are labeled on the side and top views of the TGF-␤3 dimer. The TGF-␤ type II receptor-binding epitope, named (33) as fingertips, the BMP type IA and putative type II receptor-binding epitopes, designated (6) as the wrist and knuckle, and additional thumb epitope (33) are highlighted by broken oval curves. The hydrophobic pocket capable of binding a dioxane molecule or the side chain of Phe 84 of T␤R1 is indicated. Amide groups not characterized are shown in light gray. b, conformational exchange. Residues with micro-millisecond time scale motions identified by having exchange contributions R ex to R 2 exceeding 2 s Ϫ1 at 11.7 T are colored in magenta. Residues 6 -17 and 48 -75 exhibiting strong broadening and pronounced doubling of backbone NH resonances, respectively, indicative of slow conformational exchange, are shown in cyan. c, model for the relative positioning of the type I and type II receptor ectodomains in the TGF-␤-receptor signaling complex. The model is based upon superposition of the ligand components of the two ecBR1A⅐BMP-2 (32) and two ecT␤R2⅐TGF-␤3 (33) asymmetric units with the crystal structure of free TGF-␤3 (10) dimer. The crystal structure of free TGF-␤3 was used, because the regions 1-12 and 55-72 of ecT␤R2-bound TGF-␤3 are not observable in the crystal (33). In the case of ecBR1A⅐BMP-2 (PDB code 1es7), the root meant square deviations in backbone atom positions between residues 16 -23, 31-47, 59 -68, 78 -94, and 100 -114 in BMP-2 and 17-24, 32-48, 57-66, 77-93, and 98 -112 in free TGF-␤3 dimer were minimized (amount to 1.38 Å). In the case of ecT␤R2⅐TGF-␤3 (PDB code 1ktz), the root mean square deviations in backbone atom positions between residues 16 -45 and 83-106 in TGF-␤3 were minimized (amount to 0.85 Å). The ribbon diagrams of the side and top view of ecBR1A 2 ⅐ecT␤R2 2 ⅐TGF-␤3 complex are shown. The two TGF-␤3 monomers are color-coded according to S 2 ; the two ecBR1A and two ecT␤R2 molecules are depicted in green and blue, respectively. The chain termini of ectodomains and the membrane proximal side of the signaling complex are labeled on the side view. The figure was prepared with MolMol (46). outstretched "hand" with the N-terminal ␣-helix H1 as the "thumb" and the extended sheets as slightly curled "fingers" (9, 10) (Fig. 2a). It was determined that the residues of the loops at the tips of both ␤-sheets form the high affinity fingertips epitope (33). The wrist epitope comprises residues of ␣-helix H3 and its pre-helix loop of one subunit and contacts both the inner side of the loop at the tip of the first ␤-sheet and the C-terminal part of ␤-strand ␤C of the other subunit of the dimer (32). In turn the putative knuckle epitope mapped (6) onto the outer convex surfaces of both ␤-sheets is spatially adjacent to the fingertips epitope and opposite the wrist epitope.
The color-coded ribbon diagrams (Fig. 2, a and b) demonstrate that the most dynamic regions of the TGF-␤3 dimer exactly correspond to the receptor-binding sites of the fingertips as well as wrist epitopes in contrast to the knuckle epitope. For clarity we constructed the model of TGF-␤3-receptor signaling complex based on the crystal structures of ecBR1A 2 ⅐BMP2 (32) and ecT␤R2 2 ⅐TGF-␤3 (33) complexes (Fig.  2c). Interestingly the interface between subunits of free TGF-␤3 is flexible, which suggests the possibility of the structural accommodation of the TGF-␤3 dimeric fold upon receptor binding. This is consistent with the fact that ecT␤R2-bound TGF-␤3 shows an altered arrangement of its monomeric subunits (33). Assuming that the backbone mobility of TGF-␤3 active sites is reduced upon receptor binding, one would expect a strongly unfavorable entropic contribution to the free energy of complex formation (34). All sources of changes in free energy are critical in the delicate energy balance that determines the equilibrium populations of protein-ligand or protein-protein interactions. This implies that the loss of conformational entropy due to the ordering of TGF-␤3 binding surface must be efficiently compensated by hydrophobic interactions and enthalpic factors such as H-bond or electrostatic interactions. Furthermore, the enhancement of TGF-␤3 internal mobility tends to cluster at both epitopes that may indicate localized "hot spots" (35) of binding free energy.
The hairpin loop region 92-95, the one-turn ␣-helix H2, and the extended loop 29 -32 are suitable for such hot spots in the fingertips epitope of TGF-␤3. The residues Arg 25 and Arg 94 from these regions play an important role in determining binding specificity and affinity to T␤R2 (29,33). Several potential hot spots of binding free energy are observed in the more spacious wrist epitope putative for recognizing the TGF-␤ type I receptor. The most prominent of these is the central ␣-helix H3 with its pre-helix loop identified as an active site by mutagenesis analysis (28) of TGF-␤1. Another candidate common with the fingertips epitope is region 29 -32 situated in a hydrophobic pocket nearby the outer part of the ␤-sheet loops. This pocket in TGF-␤3 can accommodate a dioxane molecule both in crystal (10) and in solution (13). Furthermore, in the BR1A 2 ⅐BMP-2 complex this hydrophobic pocket interacts with the side chain of Phe 85 of BR1A corresponding to Phe 84 of T␤R1. That is perhaps a key feature for the recognition of type I receptor within the TGF-␤ superfamily (32). The next possible hot spot of binding free energy is the N-terminal part of TGF-␤3, designated as thumb epitope, which is spatially close to wrist epitope and contains many non-conservative residues in the solvent-exposed flexible loop following ␣-helix H1 (Fig. 3c). This segment is found only among the TGF-␤s and the activins but not in other members of the TGF-␤ superfamily and may also participate in the TGF-␤ recognition of type I receptor. Thus, the low affinity wrist epitope of TGF-␤3 is more flexible in comparison with the fingertips epitope that apparently results in the most pronounced conformational entropy loss upon receptor binding. In the vacant knuckle epitope of TGF-␤3 two interesting dynamic clusters adjacent to the fingertips and thumb epitope may be identified as potential hot spots of binding free energy. The first is the region 34 -37 from the Nterminal part of ␤-strand ␤B, which partially participates in the interaction with ecT␤R2. The second is the hydrophobic patch nearby the break of the first ␤-sheet, containing residues Tyr 21 and Tyr 40 . These hot spots may serve to recognizing other TGF-␤-binding proteins, such as betaglycan. Betaglycan has two binding sites for TGF-␤ and plays an important role in assembling a potentially more productive TGF-␤-receptor signaling complex (36). Moreover TGF-␤-binding sites to different proteins probably overlap, and described above hot spots in the TGF-␤ epitopes may also interact with other TGF-␤ activity regulation proteins such as LAP and thrombospondin. LAP binding to TGF-␤ inhibits the ligand binding to receptor sites on the T␤R2 and betaglycan (36). Thrombospondin binds to a site in the active domain of TGF-␤ through three repeats of hydrophobic WXXW motif (3), which are known to become posttranslational C-mannosylated (37). As shown previously the hydrophobic pocket close to the fingertips epitope is potentially capable of binding a carbohydrate ring (10) that suggests the thrombospondin-binding site in the TGF-␤ molecule. The suggestions discussed above are based on the well known fact FIG. 3. Comparison of internal dynamics among TGF-␤ isoforms. a, 15 N{ 1 H} NOEs (black) of TGF-␤3 measured at 11.7 T in the present study are plotted versus the 15 N{ 1 H} NOEs (red) for the backbone 15 N nuclei of TGF-␤1 obtained at 11.7 T, pH 4.2, 45°C (11). The broken line indicates the region 6 -17, which is poorly sampled due to strong broadening of NMR signals. b, average isotropic thermal Bfactors (Å 2 ) for the backbone atoms of free TGF-␤2 (green; PDB code 2tgi), free TGF-␤3 (black; PDB code 1tgj), and ecT␤R2-bound TGF-␤3 (blue; PDB code 1ktz). The absent regions 1-12 and 55-72 of ecT␤R2bound TGF-␤3 are disordered in the crystal (33). c, sequence alignment of TGF-␤ isoforms and positions of secondary structure elements of TGF-␤3 suggested by the x-ray (10) and NMR (13) data. Amino acid variations of TGF-␤1 and TGF-␤2 in comparison with TGF-␤3 sequence are highlighted by red and green, respectively. that intrinsically unstructured proteins offer important advantages in cellular signaling and regulation. Their inherent flexibility allows their local and global structure to be modified in response to different molecular targets, allowing one protein to interact with multiple cellular partners and allowing fine control over binding affinity (38).
Comparison of Internal Dynamics for TGF-␤ Isoforms-The results presented above and literature data reveal that TGF-␤3 dynamics are different from that of other TGF-␤ isoforms. In Fig. 3a the backbone 15 N{ 1 H} NOEs of TGF-␤3 obtained in the present study are plotted versus 15 N{ 1 H} NOEs of TGF-␤1. This plot is chosen for comparison, since the order parameters S 2 were not reported in the NMR study of TGF-␤1 in solution (11). The lowered 15 N{ 1 H} NOEs clearly indicate the regions with enhanced mobility on the subnanosecond time scale. Although direct comparison of NMR and x-ray data on protein dynamics may be considered problematic (39,40), in Fig. 3b we also list, as a reference information, the thermal B-factors taken from crystallographic studies of TGF-␤s (9, 10, 33) reflecting the local mobility of a protein constrained by a crystal lattice. Thermal B-factors of free TGF-␤2 and free TGF-␤3 suggest that the ␤-sheets and the helix H3 represent the most rigid parts of these TGF-␤ isoforms, while internal mobility is increased for loop regions connecting secondary structure elements. The 15 N NMR relaxation study of TGF-␤1 suggests an overall rather restricted pico-to nanosecond mobility for the protein with somewhat more flexible regions 2-4, 10 -13, 50 -56, 71-75, and 91-98 (11). The same regions of TGF-␤3 in solution also exhibit internal motions over a wide range of time scales. However, there are several differences in internal dynamics of TGF-␤3 in solution from that anticipated from previous studies of TGF-␤ isoforms. The major difference is that in TGF-␤3 the central ␣-helix H3 and the C-terminal part of its pre-helix loop are extremely flexible in solution. Note that this region is the least conserved among the TGF-␤s (Fig. 3c). Transient disruptions of helical structure in TGF-␤3 are presumably caused by the presence of four helix-destabilizing residues (Thr 57 , Thr 60 , Gly 63 , and Thr 67 ) on the outer face of the ␣-helix H3 alongside its pre-helix loop. Similarly, the regions 1-12 and 55-72 of ecT␤R2-bound TGF-␤3 are also disordered in the crystal (33). In contrast, the crystal structure of free TGF-␤3 displays low thermal B-factors and well defined electron density throughout ␣-helix H3. Crystal packing forces are responsible for such ordering, as several residues of this segment participate in intermolecular interactions in the crystal lattice (10). These observations emphasize the importance of the NMR information on protein dynamics in solution for this family of proteins.
Relationship between the Unique Dynamics and Specific Macroscopic Behavior of TGF-␤3-Our results indicate that TGF-␤3 is significantly more flexible than other proteins of this family. Thus, certain protein properties should be very sensitive to mild variations of local environment. The high backbone flexibility is related to the local unfolding of the protein. This, in turn, results in exposure of the hydrophobic regions and may cause aggregation. As shown previously, the association of oligomeric proteins is generally an entropy-driven process (41). Although TGF-␤3 does not form aggregates under the conditions of the experiments (pH 2.9), the flexible nature of the molecule is correlated with the tendency to aggregate and bind to surfaces at higher pH (12). In contrast, TGF-␤1 and TGF-␤2 easily solubilize in aqueous solutions at physiological pH. All that allows us to propose that the unusual dynamics of TGF-␤3 is related to its specific macroscopic behavior and biological activity. The good solubility of TGF-␤1 and TGF-␤2 at physiological conditions may be associated with the significantly elevated plasma levels of TGF-␤1 and TGF-␤2, but not of TGF-␤3 in patients with disseminated malignant melanoma (42). TGF-␤3 is possibly deposited in its aggregated form and stays at its place of secretion. The poor solubility of TGF-␤3 between pH 5 and pH 8 and its good solubility below pH 4 could explain localized biological activity of the molecule in processes like bone repair. Local bone remodeling involves low pH digestion of the bone by osteoclasts at the ruffled border. Bone proteins, including TGF-␤, are released from the bone by osteoclasts in the ruffled border compartment (pH Ͻ 4) and presumably transcytosed to the osteoclasts apical surface (4,43,44). The growth factors activated by low pH exposure can locally stimulate osteoblasts and contribute to the formation of new bone matrix. The fact that flexibility and aggregation of TGF-␤3 are likely correlated with its function (the growth factor should act close to the place of secretion) has the implication that this protein is beneficial in local therapies by remaining confined at the application site.
Concluding Remarks-The study presented here shows that highly dynamic regions of TGF-␤3 cluster at the receptor-binding epitopes. This manifests an important role of internal dynamics in providing the specificity and affinity of TGF-␤3 to receptors. Moreover, the differences in internal dynamics of TGF-␤ isoforms provide the basis for an explanation of the variations in their macroscopic behavior and interactions with receptors. Those are related to the control of TGF-␤ release and activity, which are critical in many diseased states and injury repair processes. This once again confirms that function (and thus potential clinical application) cannot be based on the knowledge of spatial structures alone, but that dynamic aspects are of utmost importance.