INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

The double-stranded RNA (dsRNA)-binding domain is among the most common RNA-binding motifs (Varani, 1997; Varani and Nagai 1998) and is found in many proteins from all kingdoms of life involved in RNA processing, maturation, and localization (Green and Matthews, 1992; St Johnston et al., 1992). In vitro studies have shown that the double-stranded domain (dsRBD) proteins bind to dsRNA but not to dsDNA and only weakly to DNA-RNA hybrids (St Johnston et al., 1992; Bass et al., 1994; Clarke and Matthews, 1995; Bevilacqua and Cech, 1996). dsRBDs bind dsRNA of sufficient length (more than 12 base pairs) regardless of its base composition, and therefore represent general dsRNA binding units. Therefore, binding of the dsRBD to dsRNA represents an example of structure-specific but sequence-independent protein-RNA recognition that is distinct from that of other common RNA-binding motifs. Consistent with its biochemical activity, many dsRBD-containing proteins bind a wide variety of RNAs in the cell, although some dsRBDs exercise their function on a specific set of RNAs (Macdonald and Struhl, 1988; St Johnston et al., 1991; Kim-Ha et al., 1991, 1993; Ferrandon et al., 1994; St Johnston, 1995; Broadus et al., 1998). How this domain can discriminate so effectively between double-stranded nucleic acids of similar structures, and how a domain with a general ability to bind dsRNA regardless of its sequence can function in the metabolism of specific mRNAs in the cell are outstanding questions that must be addressed if we are to understand the function of this ubiquitous RNA-binding module. These questions have begun to be addressed by the three-dimensional structure of dsRBDs (Bycroft et al., 1995; Kharrat et al., 1995) and especially by two complex structures (Ryter and Schultz, 1998; Ramos et al., 2000).

The x-ray structure of the complex of Xenopus Xlrbpa with an RNA duplex (Ryter and Schultz, 1998) and the nuclear magnetic resonance (NMR) structure of the third dsRBD from Drosophila Staufen protein in complex with a stem-loop (Ramos et al., 2000) revealed how the dsRBD binds dsRNA. Together, the two structures provide a satisfactory explanation for the ability of the domain to bind dsRNA and discriminate against dsDNA or hybrids. Discrimination in favor of dsRNA and against dsDNA is attributable to interactions mediated by loop 2 (which binds the RNA minor groove, making RNA-specific contacts with 2'-OH) and by loop 4 (where conserved basic side chains make structure-specific interactions with the phosphodiester backbone across the major groove from the site of loop 2-minor grove contacts). Interactions provided by helix 1 necessary for binding reinforce these interactions. Mutations within each of these three regions of the protein or in a conserved residue that positions loop 2 and loop 4 with respect to each other severely diminish or abolish the RNA-binding activity. Mutations of these same residues abolish the biological activity of at least one dsRBD-containing protein, Drosophila Staufen, establishing the functional significance of these structural observations.

Two significant differences between the NMR and crystal structure revealed the highly dynamic nature of the protein-RNA interface. The degree of order at the interface and the observation of well-ordered interfacial water molecules in the crystal structure reflect (at least in part) the lower quality of NMR structures when compared to high-resolution crystal structures. However, NMR studies of backbone dynamics also revealed unambiguously how regions of the protein critical for RNA recognition (loop 2 and loop 4) retain significant conformational flexibility in the RNA-bound protein (Ramos et al., 2000).

Residual flexibility is increasingly being appreciated as an important element in protein-nucleic acid recognition. However, quantitative properties of such motions have only recently begun to be addressed experimentally through novel NMR methods and computationally through molecular dynamics simulations. Although NMR has a unique ability to investigate experimentally motional properties of biomolecules, it is difficult to provide detailed insight into motional properties because the spectral density function is only sampled by the NMR experiment at a few frequencies (Peng and Wagner, 1992, 1994). Molecular dynamics (MD) techniques are simulations, but are capable of providing detailed insight into the motions that occur during molecular recognition and how motional properties of biomolecules change upon binding (Beveridge and Ravinshanker 1994; Cheatham et al., 1995, 1997; Young and Beveridge, 1998; Hermann and Westhof 1999; Reyes and Kollmann, 1999; Sen and Nilsson, 1999; Tang and Nilsson, 1999; Castrignanò et al., 2000; Tsui et al., 2000). The two methods synergistically reinforce each other when applied to the same system.

Here we report the analysis of 2-ns dynamics of the same complex derived from Staufen protein that was studied by NMR, and a comparison with the molecular dynamics simulation conducted on the separated protein and RNA components. A number of MD simulations have been dedicated to study human U1A protein and its RNA complex, which represent paradigms to understand how the RRM, the largest RNA-binding protein family, recognizes RNA (Reyes and Kollmann 1999, Hermann and Westhof 1999). However, the dsRBD represents a completely different paradigm in RNA recognition that has not yet been analyzed using this computational approach. The results add a new quantitative description of the residual motion present at the RNA interface of this important class of RNA-binding proteins and provides the observation that short-lived interactions to contribute energetically important interactions during molecular recognition.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

The simulations of the free RNA, the free protein and the protein-RNA complex were carried out using the NMR structure closest to the average structures to provide the starting coordinates. The experimental structures consisted of 16, 14, and 16, respectively, structural bundles consistent with the experimental NMR-derived structural constraints. The macromolecules were immersed in rectangular boxes of the following dimensions: 76 × 70 × 50 Å³ (complex) 46 × 49 × 62 Å³ (RNA), and 63 × 43 × 47 Å³ (protein), filled with TIP3P water molecules (Jorgensen et al., 1983). A minimum solute-wall box distance of 10 Å was imposed. The systems were neutralized by addition of Na⁺ cations or Cl anions using the AMBER leap module. The three simulated systems are composed of a total of 14,196 (RNA), 13,090 (protein) and 25,038 (complex) atoms.

Molecular dynamic simulations were conducting on a cluster of Compaq ES40 workstations by modeling each system with the AMBER95 all atom force field (Cornell et al., 1995) and using periodic boundary conditions. A cutoff radius of 9 Å was introduced for nonbonded interactions, updating the neighbor pair list every 10 steps. The electrostatic interactions were calculated with the Particle Mesh Ewald method (Darden et al., 1993). The SHAKE algorithm (Ryckaert et al., 1977) was used to constrain all bond lengths involving hydrogens. Optimization and relaxation of solvent and ions were performed at first by keeping the solute atoms constrained to their initial position with progressively decreasing force constants of 500, 25, 15, and 5 Kcal/mol Å², respectively. The systems were minimized thereafter without any additional constraints and progressively heated up to the temperature at which the simulation was conducted. Density, volume, and overall potential energy of the system were monitored and observed to reach convergence within the first 100 ps. The three 2-ns simulations were carried out at a constant temperature (Berendsen et al., 1984) of 300 K and at a constant pressure of one atmosphere with a 2-fs time step. Pressure and temperature coupling constants were 0.5 ps.

Atomic coordinates were saved every 0.1 ps for analysis during the production run (from 0.5 to 2 ns). The AMBER carnal and ptraj modules were used to analyze structural properties of the complexes and individual molecules (root mean square deviation (RMSD), hydrogen bonds, etc.). The time dependence of the RMSD from the starting structure, as indication of phase-space accessibility, has been calculated by

(1)

where the coordinates {r(t)} are obtained optimally superimposing the instantaneous configurations at time t with the starting structure (t = t₀). The atomic root mean square fluctuations (RMSF) have been computed by using the definition,

(2)

where the averages have been computed over the equilibrated MD trajectory.

The criterion for the occurrence of a hydrogen bond was a maximum donor-acceptor distance of 3.5 Å and a minimum donor-proton-acceptor angle of 120°. The solvent accessible surface area of RNA and protein sites was evaluated with the program NACCESS (Hubbard and Thornton, 1993), using the default probe size of 1.4 Å.

	RESULTS AND DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

Stability of the simulation

A 2-ns MD simulation of the complex between Staufen dsRBD3 and an RNA stem loop containing a 13 base-pairs double-helical region and capped by a tetraloop (Fig. 1) was conducted as described in the methodological section. An analogous set of simulations was also conducted for the separate protein and RNA components. In each case, NMR-derived coordinates were used as starting structures for the simulation. The RMSD of the dsRBD-dsRNA complex relative to the starting structure is reported in Fig. 2 A. The results show that the simulation is stable over the entire 2 ns over which it was conducted, with the RMSD reaching a plateau after ~300 ps and oscillating around 3 Å for the remainder of the trajectory. The analysis of the results reported throughout the manuscript refers to the last 1.5 ns of the simulations, because inspection of all RMSD plots indicate that all three systems are all well equilibrated after the first 500 ps of the simulation.

View larger version (145K): [in this window] [in a new window]   FIGURE 1   View of the structure of the dsRBD-dsRNA complex used as the starting structure for the simulation. The lateral chain of the lysines, which have been found to contact RNA during the simulation, have been explicitly represented.

View larger version (26K): [in this window] [in a new window]   FIGURE 2   Time evolution of the root mean square deviations relative to the starting coordinates. (A) dsRBD-dsRNa complex. (B) Free (black) and bound (red) dsRNA. (C) Free (black) and bound (red) dsRBD protein domain.

The RMSDs with respect to their respective starting structures of the RNA free and in complex with dsrbd3 protein are shown in Fig. 2 B as a function of the simulation time. The RMSD of the free RNA reaches the equilibrium more slowly than it does in the complex and higher fluctuations are also observed. A similar plot for the protein (Fig. 2 C) indicates that the conformational space sampled by the protein is lower in the complex than in the free protein, the plateau being ~3.5 and 2.5 Å for the bound and the free protein, respectively. The molecular origin of these differences in RNA and protein dynamic behavior is discussed below.

The relatively small RMSDs and overall stability of the simulation indicate that all structures simulated here are well maintained over the course of the simulation. Because NMR structures are generated as structural bundles consistent with the experimental data (14 for the complex, 16 for the RNA, 14 for the protein), we are able to compare the conformational distribution represented by the experimental structures and by the MD simulation. Table 1 compares the average, minimum, and maximum RMSDs within the NMR structures (relative to the starting structure); these values are comparable to the RMSDs of the MD trajectories. The minimum RMSD of the MD trajectory (relative to each experimental NMR structure) is reported in Table 2. These results indicate that both the complex and the free protein and RNA sample conformational spaces during the MD simulation are close to those observed by NMR.

Analysis of the conformational fluctuations and their changes upon complex formation

A detailed picture of the local motions that occur on a residue-by-residue basis within the protein and the RNA in their free and bound forms was obtained by analyzing RMSFs averaged over each protein and RNA residue. Figure 3 A reports the per residue protein RMSFs calculated over the C- atoms. The RMSFs calculated from blocks of 300 or 500 ps give comparable values. From this analysis, we estimate the error on the reported values to be lower than 10%. The plot reveals several regions of the protein that are considerably more mobile than the well-ordered secondary structured regions of the dsRBD. In the free protein, these are the C-terminal amino acids, which are disordered as expected, and, significantly, loops 2, 3, and 4 (residues 25-30, 38-40, and 47-51, respectively). Surprisingly, flexibility of loop 3 is quenched in the complex, although this loop is not in direct contact with the RNA. The excess mobility (compared to the well-ordered hydrophobic core of the protein) of residues within loop 2 is reduced for some residues (i.e., 25, 30) but increased for others (i.e., 26, 27), while residues within loop 4 retain similar mobility in the presence or absence of RNA. These results indicate that loops 2 and 4, two of the three regions of the protein that participate to the RNA interface, are more highly mobile than the remainder of the protein both in the free and in the RNA-bound form of the protein. In fact, loop 2 appears to be the region of highest mobility, as evidenced by the value of the RMSF for residues within this critical loop. These results are consistent with the conclusions of NMR relaxation measurements (Ramos et al., 2000) that reported enhanced levels of flexibility within loop 2 and loop 4. Moreover, several amino acids could not be quantitatively analyzed in the complex because of line broadening indicative of a motion on a slower time scale than sampled by the heteronuclear NOE measurements. Finally, NOE interactions involving many side chain protons were quenched due to motion on the nanosecond time scale (A. Ramos and G. Varani, et al., unpublished results).

View larger version (25K): [in this window] [in a new window]   FIGURE 3   Root mean square fluctuations observed in the course of the simulation: (A) calculated for each protein residue for the free (black) and bound (red) dsRBD over the C- atoms; (B) averaged for each protein residue for the free (black) and bound (red) dsRBD over the side chain; (C) averaged over each RNA base for the free (black) and bound (red) dsRNA. The secondary structure of the protein is represented in the upper part of the figure.

The quenching of the loop 3 fluctuations must be due to long-range effects, because this region is not directly involved in interactions with RNA. An intriguing possibility is suggested by the observation that interactions involving loop 2 and loop 4 must be precisely spaced to allow for recognition of dsRNA. Loop 4 binds across the major groove from the site of interaction with loop 2 and the minor groove (Ryter and Schultz, 1998; Ramos et al., 2000), and it is very likely that this spacing (which differs between A-form RNA and B-form DNA) is important for discrimination between DNA and RNA double helices (Ramos et al., 2000). In support of this suggestion, mutation of Phe-32, a universally conserved residue that fixes the relative position of the two loops, abolishes RNA binding, though this residue does not contact RNA at all. Perhaps the increased rigidity of loop 3 reflects the long-range consequences of the establishment of interactions involving both loop 2 and loop 4, which leads to increased structural rigidity in the protein.

To further characterize conformational flexibility, we calculated side chain RMSFs for free and bound protein as well (Fig. 3 B). The plot indicates that the positively charged side chains, i.e., lysines and arginines, have the largest fluctuations, as expected from their long length. Those that localize within loops 2 and 4 (e.g., Lys-30, Lys-50, and Lys-51) and that contact the RNA retain comparable high levels of mobility in the free and bound forms. However, positive side chains that do not reside within these loops have large fluctuations. Overall, it appears that surface accessibility and chain length dictates side-chain mobility, but it is relevant and consistent with other results discussed here that RNA binding does not affect significantly the mobility of side chains at the RNA-protein interface.

The RNA per-base RMSFs weighted per mass unit over all the atoms is reported in Fig. 3 C. We attribute the higher fluctuations observed in the RNA to relatively long-range motions (bending and twisting of the RNA helix) and therefore to the absence of long-range interactions (other than base pairing) between RNA residues equivalent to the packing interactions observed in the hydrophobic core of the protein. The plot of Fig. 3 indicates that conformational fluctuations are higher in the free than in the bound RNA. In both cases, the highest fluctuations are localized to the helix termini (as expected) and to the tetraloop This result was unexpected, because UUCG tetraloops are very stable structural units (Cheong et al., 1990), although it is perhaps to be expected that it is more rigid than an uninterrupted stretch of perfect A-form RNA. As a matter of fact, bases 15, 16, and 17, the most mobile ones, have a smaller number of structural restraints when compared to the other ones. Protein binding causes a strong quenching of the fluctuations for almost all the RNA bases, including the tetraloop, which remains nonetheless the most mobile region of the structure. Interestingly, base 15, among the most mobile residues in the free RNA, shows the largest quenching in fluctuations, probably because it becomes involved in direct and water-mediated protein interaction (see below).

Direct hydrogen bonds at the proteinRNA interface form a network of short-lived, continuously exchanging interactions

The intermolecular dsRBD-RNA interface is small compared to other RNA-protein complexes and involves relatively few protein residues (Ryter and Schultz, 1998; Ramos et al., 2000). The buried area of 1450 Å² obtained from the experimental results is confirmed from the average value observed in our simulation (1550 Å²). This value does not change significantly over the entire trajectory (data not shown). Clearly, this protein-RNA complex displays a relatively small number of long-lasting contacts when compared to other protein-RNA or protein-DNA complexes (Eriksson et al., 1995; Nilsson, 1998; Tang and Nilsson 1998; Reyes and Kollman, 1999; Chillemi et al., 2001). A schematic representation of the direct protein-RNA hydrogen bonds present for more than 30% of the simulation time, a cutoff used in MD simulation studies of similar systems (Tang and Nilsson 1998), is shown in Fig. 4. The figure indicates that dsRBD and dsRNA do not form a tight complex with an extensive and intricate intermolecular interface, but rather conserve a relatively high degree of freedom with respect to each other. Among the residues involved in the RNA interaction (primarily loops 2 and 4 and helix , plus a few interactions involving residues from the 1 strand), only loop 2 and loop 4 residues maintain direct hydrogen bonds with RNA atoms for a long percentage of time. In helix 1, only His-6 forms a hydrogen bond for a relatively long percentage (38%) of the simulation time. This residue is one of only three amino acids involved in base-specific interactions, the other being Pro-26 and Lys-30 within loop 2. All other contacts involve nonsequence-specific interactions with the sugar-phosphate backbone of the double-stranded RNA.

View larger version (32K): [in this window] [in a new window]   FIGURE 4   Schematic representation of direct protein-RNA direct contact observed for more than 50% (thick line) and between 30% and 50% (thin line) of the total simulation time. Amino acids are color coded according to different regions of the protein, as shown at the bottom.

The simulation suggests that interactions involving Lys-30 and Lys-54 have a major role in the recognition process, because these amino-acid side chains form numerous direct hydrogen bonds with RNA atoms for a high percentage of the simulation time. Both Lys-30 and Lys-54 are very highly conserved in dsRBD proteins, and mutations of either residue strongly affect RNA binding (Ramos et al., 2000). However, a detailed analysis of the interactions formed by these two residues in the course of the simulation reveals that individual amino-acid side chains undergo large amplitude motions that lead to contacts with the RNA remaining present for only a few tens of picoseconds before a new set of contacts is established. However, these protein side chains remain associated with the RNA for a high percentage of the simulation by continuously switching among different interactions even if individual contacts are short lived. An illustration of the time-dependent network of interactions observed at the interface during the simulation is reported in Fig. 5, where we show how the side chain of Lys-54 rotates quickly, continuously changing its hydrogen atoms involved in the interaction with the O3' atom of G19. Much more remarkable is the observation that each of the two Lys amino groups contacts a different oxygen atom of the RNA backbone in the course of the simulation. This is shown in Fig. 6, where five different snapshots of the Lys-54-RNA contacts are shown. In detail, three of them (A, B, and C) show the occurrence of different direct Lys-54-RNA contacts, whereas the remaining two (D and E) show the occurrence of different Lys-54-RNA water-mediated contacts. Similar observations could be made for the other lysines that contact the RNA, namely K30, K50, and K51. We conclude that, rather than forming single well-defined interactions with the RNA, these side chains continuously switch from one polar interaction to the other on a very fast time scale, yet retain a nearly continuous association with RNA polar groups.

View larger version (41K): [in this window] [in a new window]   FIGURE 5   Time evolution of the distance between each of the three hydrogen atoms of the amino group of Lys-54 and the O3' atom of G19.

View larger version (46K): [in this window] [in a new window]   FIGURE 6   Five snapshots of the MD simulation showing the variety of RNA contacts explored by the lateral chain of Lys-54. (A, B, C) Direct protein-RNA contacts. (D, E) Water-mediated protein-RNA contacts.

Residual interfacial flexibility may be responsible for the relatively high per-residue RMSF (Fig. 3, A and B) and provides a molecular explanation for it. This result also supports the conclusion that the apparent disorder highlighted by the bundle of NMR structures reflects a true physical property of the interface. It is interesting to compare the behavior of K50, K51, and K54 observed in the MD trajectory with that described by NMR (Ramos et al., 2000). Figure 3 A indicates that the backbone of Lys-54 is rigid, and indeed this residue belongs to an -helix, whereas Lys-50 and Lys-51 are highly mobile even at the backbone level and are found on the loop connecting the third strand of the -sheet with the terminal -helix. However, all these side chains have high levels of flexibility (Fig. 3 B), although they all participate in RNA recognition during the simulation and in the NMR structure. Consistent with this observation, superposition of the NMR structures shows a spread of structures for these side chains (Fig. 6 of Ramos et al., 2000). Unfortunately this conclusion, based on superposition of structures, cannot be corroborated with direct observations of local mobility, because there are no effective methods available to monitor side-chain mobility (other than for methyl groups; see e.g., Mittermaier et al., 1999). The behavior described herein may be responsible for the high levels of fast local dynamics observed in ¹⁵N NMR relaxation experiments (Ramos et al., 2000). Although the two methods are distinct, NMR and MD simulations may both have monitored a fundamental property of the dsRBD-dsRNA interface. An attractive explanation for these phenomena is that retaining residual flexibility upon binding reduces entropic costs associated with the rigidification of long basic side chains. Because favorable electrostatic interactions are retained in the different hydrogen bonding conformations observed in the simulation, enthalpic gains can be achieved at relatively small entropic costs through this mechanism.

Water-mediated hydrogen bonds at the proteinRNA interface are a short-lived yet important component of the interface

Analysis of the simulation trajectories allowed us to analyze protein-RNA interactions mediated by water molecules as well. As for the direct interactions discussed in the previous section, we only considered water-mediated protein-RNA hydrogen bonds that are present for more than 30% of the simulation time. Water molecules mediate, at least transiently, interactions that are key to recognition (Fig. 6). As reported in Fig. 7, water-mediated interactions are numerous, comparable to the number of direct hydrogen bonds reported in Fig. 3, yet lower than reported for protein-DNA complexes (Chillemi et al., 2001). Because of their number, these interactions are likely to play a thermodynamic and structural role comparable to that of direct interactions discussed in the previous section.

View larger version (36K): [in this window] [in a new window]   FIGURE 7   Schematic representation of water-mediated contacts observed for more than 50% (thick line) and between 30% and 50% (thin line) of the total simulation time. Amino acids are color coded according to different regions of the protein, as shown at the bottom.

Water-mediated hydrogen bonds involve His-17, Phe-18, Lys-19, and Glu-23 within strand 1, Lys-30 within loop 2, Lys-50 and Lys-51 within loop 4, and Lys-54 within helix 2. Phe-18, Lys-19, Lys-50, Lys-51, and Lys-54 have water-mediated RNA interactions for more than 50% of simulation time, suggesting that they are preferred hydration sites. These residues are important for binding, because mutation of Phe-18, Lys-30, Lys-50, and Lys-51 to Ala all reduce binding affinity (Ramos et al., 2000). None of the water molecules mediating the interactions are single long-lasting water molecules, but rather they all represent transient water molecules that are in continuous fast (typically 1-10-ps) exchange with the bulk water. For example, the water-mediated interaction between the NH group of Phe-18 and the O5' atom of G19, which is present for 87% of the simulation time, is mediated by 147 different individual water molecules throughout the simulation, and the longest-lasting water molecule bridges the two atoms for 67 ps. The observation that all water-mediated contacts involve water molecules continuously exchanging with solvent is consistent with the high flexibility observed at the interface.

It is very unlikely that any of the water-mediated contacts observed in the simulation could have been detected by NMR using current methods, due to their very short residence time and to difficulties in distinguishing signals from RNA 2'-OH from the water signals (Otting et al., 1991; Kubinec and Wemmer, 1992; Liepinsh et al., 1992). In contrast, the crystallographic structure of the dsRBD from Xenopus laevis Xlrbpa protein bound to a dsRNA duplex (Ryter and Schultz; 1998) reports 18 water-mediated contacts. This number is close to what is observed in our simulation (15), but only three of these contacts are common between experimental work and simulation. This difference can, in large part, be explained based on sequence divergence between the two dsRBDs: the only common residues involved in water-mediated contacts are Arg-143, Lys-163, and Lys-164 of Xlrbpa (these correspond to Lys-30, Lys-50, and Lys-51 of Staufen). Both Lys-50 and Lys-51 are involved in water-mediated contacts analogous to those observed in the crystal structure. A second significant difference concerns the residence time of these water molecules. In the simulation, the water molecules bound to these residues are not fixed but rather in fast exchange with the solvent. For example, the contact between Lys-50 and the phosphate oxygens of A13 is mediated by 396 different water molecules during the simulation. Although the longest-lasting molecule bridges the two chemical groups for 29 ps, this position is occupied by a water molecule for 79% of the simulation time. Perhaps this large fractional occupation allows its identification through x-ray diffraction. A similar behavior is found for the water molecules close to residue Lys-30 and Lys-51. These results indicate that the water molecules seen through x-ray diffraction are found in a specific position with high probability, but are nonetheless in fast exchange with the solvent. This behavior may again confer flexibility to the protein-RNA complex and stabilize the complex by providing favorable hydration while reducing entropic costs.

	CONCLUSIONS

TOP ABSTRACT INTRODUCTION METHODS RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

The MD simulations of the complex between Staufen dsRBd3 and RNA and of the separate protein and RNA components have allowed us to examine the role of conformational flexibility and of water-mediated interactions in protein-RNA recognition. The simulations indicate that the protein retains a substantial flexibility in the complex. In particular, loops 2 and 4, the regions of the protein that mediate recognition, maintain a high degree of flexibility upon binding, with critical basic protein side chains retaining favorable interactions with the RNA but migrating from one contact to another with typical residence times of a few tens of picoseconds. The simulation also highlights the crucial role of water molecules in mediating recognition, because water-mediated contacts are comparable in number to direct hydrogen-bonding interactions. The presence of many water-mediated contacts was observed in the x-ray structure of a related dsRBD-dsRNA complex. However, the MD simulation indicates that contacts mediated by molecules are in fast exchange with the solvent (every few picoseconds typically), thus contributing to the substantial flexibility of the intermolecular interface. This is in contrast to the static picture provided by x-ray diffraction.

A high degree of flexibility at the protein-RNA interface is likely to provide entropic advantages, because it reduces the entropic loss due to rigidification of protein side chains, as observed instead in other examples of RNA-protein recognition. A strong correlation between interfacial order and specificity was observed in a study of sequence-specific RNA-protein recognition in the RRM family of RNA-binding proteins (Mittermaier et al., 1999). From the few examples where dynamic processes have begun to be studied at RNA-protein surfaces, sequence-specific recognition requires a complex distribution of dynamics, with some side chains being highly ordered and others retaining high degree of flexibility. In contrast, high levels of interfacial flexibility may be prevalent in cases, such as the present, of nonsequence-specific recognition. The present work highlights the important role of flexibility in molecular recognition and the ability of MD simulations to reveal quantitative features of the underlying dynamic processes.