Synchronized motion of interface residues for evaluating protein–RNA complex binding affinity: Application to aptamer‐mediated inhibition of TDP‐43 aggregates Authors: Francesco Paolo Panei, Lorenzo Di Rienzo, Elsa Zacco, Alexandros Armaos, Gian Gaetano Tartaglia, Giancarlo Ruocco, Edoardo Milanetti Published: 2024 ABSTRACT AbstractInvestigating the binding between proteins and aptamers, such as peptides or RNA molecules, is of crucial importance both for understanding the molecular mechanisms that regulate cellular activities and for therapeutic applications in several pathologies. Here, a new computational procedure, employing mainly docking, clustering analysis, and molecular dynamics simulations, was designed to estimate the binding affinities between a protein and some RNA aptamers, through the investigation of the dynamical behavior of the predicted molecular complex. Using the state‐of‐the‐art software catRAPID, we computationally designed a set of RNA aptamers interacting with the TAR DNA‐binding protein 43 (TDP‐43), a protein associated with several neurodegenerative diseases, including amyotrophic lateral sclerosis (ALS). We thus devised a computational protocol to predict the RNA–protein molecular complex, so that the structural and dynamical behavior of such a complex can be investigated through extensive molecular dynamics simulation. We hypothesized that the coordinated and synchronized motion of the protein‐binding residues, when in contact with RNA molecule, is a critical requisite in order to have a stable binding. Indeed, we calculated the motion covariance exhibited by the interface residues during molecular dynamics simulation: we tested the results against experimental measurements of binding affinity (in this case, the dissociation constant) for six RNA molecules, resulting in a linear correlation of about 0.9. Our findings suggest that the synchronized movement of interface residues plays a pivotal role in ensuring the stability within RNA–protein complexes, moreover providing insights into the contribution of each interface residue. This promising pipeline could thus contribute to the design of RNA aptamers interacting with proteins.Keywords: binding affinity prediction, computational method, computational molecular biophysics, molecular dynamics simulations, protein–RNA interactions, synchronized motion INTRODUCTION Assessing the binding between proteins and RNA plays a key role in many molecular aspects, both for understanding the fundamental mechanisms of cellular molecular processes and for more technological purposes related to diagnostics and the treatment of many diseases (Boeynaems et al., 2019; Cirillo et al., 2013; Li et al., 2023). Among the possible actions of RNA molecules in both physiological and pathological contexts, there is a growing focus on their inhibitory action on protein aggregates (Aarum et al., 2020; Louka et al., 2020). Indeed, biomolecular condensates are defined as cellular membraneless assemblies where biomolecules concentrate with no fixed stoichiometry (Banani et al., 2017; Shin & Brangwynne, 2017). They perform a variety of biological functions, such as the regulation of biochemical reactions, modulating the protein concentrations, or the application of mechanical forces (Alberti & Hyman, 2021). In particular, as a response to various forms of cellular stress, the formation of the so‐called stress granules work to segregate proteins and nucleic acids to promote the rate of cellular survival (Guzikowski et al., 2019). However, the aberrant behavior of stress granules has been identified as a primary driver for cellular disfunction in neurodegenerative diseases (Lin et al., 2015; Protter & Parker, 2016; Van Treeck & Parker, 2019). Protein misfolding and aggregation may be prevented by promoting competitive mechanisms of interaction with other molecules that could represent a molecular partner for proteins (Laura et al., 2011; Pastore & Temussi, 2012). Indeed, possible therapies involve biomolecules that might antagonize protein aggregates formation, altering stress granules' dynamical properties and promoting their disassembly. In this context, some recent experiments have demonstrated that RNA concentration in stress granules, and the subsequent RNA–protein interactions, have a key role in determining the solubility of protein aggregates (Mann et al., 2019; Zacco et al., 2019). In particular, short strands of nucleotides, called RNA aptamers, might bind to RNA‐binding proteins recruited in the stress granules and directly enhance or disadvantage their aggregation (Zacco et al., 2018). Within this framework, the design and the validation of RNA aptamers as potential aggregation inhibitors are usually based on time‐consuming and costly experimental attempts (Affinito et al., 2019; Radom et al., 2013). On the other side, a reliable computational evaluation of the protein‐aptamer binding properties, such as the binding affinity as a measure of the strength of the interaction between the two molecules involved in binding (Seo et al., 2021), could drive the design of RNA aptamers and represent powerful support for any experimental investigation. To address this challenge, several computational tools that are able to identify and characterize the molecular recognition between the two molecules, either identifying the set of protein‐binding site residues or predicting the complex 3D structure with molecular docking, have been developed recently (Blanco et al., 2019; Carson et al., 2010; Kumar et al., 2008; Livi et al., 2016; Ma et al., 2011; Milanetti et al., 2021; Murakami et al., 2010; Terribilini et al., 2007; Wang et al., 2010; Wang & Brown, 2006). These tools rely on learning methods or physico‐chemical approaches based on a static representation of the protein–RNA biomolecular complex (Bheemireddy et al., 2022). Nevertheless, it has been well established that proteins are dynamic entities, and their conformational dynamics is often crucial to perform their physiological functions (Henzler‐Wildman & Kern, 2007; Karplus & Kuriyan, 2005; Yang et al., 2014), as well as the description of the binding properties. From this perspective, gaining insight into the interplay between three‐dimensional structure and dynamics would support the development of new RNA aptamers and, more generally, ease the comprehension of its molecular recognition with the target protein. In order to address the complex topic of RNA–protein interaction prediction, we have recently employed (Zacco et al., 2022) an approach based on the CatRAPID algorithm (Bellucci et al., 2011; Cirillo et al., 2017) and investigated, only for a few systems, the dynamic behavior of the predicted RNA–protein complex. The previously obtained results have prompted us to design a new computational procedure, primarily based on molecular dynamics simulations, to provide predictions on protein–RNA interaction in a more general and systematic manner. In light of these considerations and our previous results, we designed a computational procedure to reproduce in silico the molecular complex formed by a protein and an RNA aptamer and to estimate its binding affinity from its dynamical properties. The protocol starts from a resolved 3D structure of the protein and the sequence of the RNA aptamer, involving the following steps. First, we combine molecular dynamics (MD) simulations, principal component analysis (PCA), and clustering methods to obtain three‐dimensional structures of the RNA aptamer: this have to be suitable, as much as possible, for the binding with the receptor protein. Then, we generate a 3D structure of the protein–RNA complex by docking the identified RNA conformations to the target protein. Starting from the top‐scored structure, which should correspond to a minimum of free energy, we finally perform an extensive MD simulation to optimize the binding between the two molecules and to evaluate their binding affinity. This is achieved by defining two dynamic‐structural descriptors, which relate to the behavior of protein‐binding site residues in the presence of the molecular partner (in this case an RNA aptamer). The first descriptor is a measure of the compactness of the binding interfaces, defined as the average distance between protein residues within the binding region and RNA nucleotides, presuming that the proximity of RNA to the protein reflects the stability of the complex. The second descriptor focuses on the synergy of motion among interacting residues through their average covariance, postulating that correlated motions at the interface indicate the system's enhanced capacity to maintain molecular binding amid thermal noise. The definition of these descriptors relies on the idea that the protein–RNA system, after binding, forms a single molecular entity, and the ability to remain a single entity will depend on the compactness between the two objects (molecules) and their coordinated motion. The idea is that, in order to maintain interaction with the ligand, not only is the distance between the two interacting molecules generally smaller in high‐affinity complexes but also the synergy of motion among the binding residues is greater. This suggests that correlated motion ensures a more stable interaction in high‐affinity molecular complexes. Therefore, the measurement of the covariance at the interface as a measure of the synchronous motion between residues of the binding site is to quantify the effect of the ligand on the protein structure. In fact, a protein in its unbound state is free to move at room temperature, showing a certain degree of motion correlation. The binding with the molecular partner, especially with molecules with a significant net charge, such as RNA, acts as a constraint for the residues involved in the binding, which could not only reduce their mobility (Zen et al., 2010) but, as hypothesized in this work, could increase the synergy of the movement (precisely because they are subjected to the same constraint, the ligand). Here, we propose a study aimed at correlating the binding affinity between protein and RNA with the average covariance of the residues at the binding interface. We apply this approach for the first time on a specific case of interest: the TAR DNA‐binding protein 43 (TDP‐43) and a set of six RNA aptamers, specifically designed for binding TDP‐43 RNA recognition domains. TDP‐43 is among the biological components recruited in stress granules, and it has been found to be one of the most implicated proteins in several pathologies, such as amyotrophic lateral sclerosis (ALS), frontotemporal dementia (FTD), and Alzheimer's disease (AD) (Grassmann et al., 2021; Gu et al., 2017; Jo et al., 2020; Josephs et al., 2014). These are relentlessly progressive neurodegenerative conditions without a fully curative treatment (Al‐Chalabi et al., 2016). In these conditions, the incorrect folding of the protein, and its self‐aggregation, highly interferes with the physiological solubility of granules (Gopal et al., 2017; Neumann et al., 2006; Scotter et al., 2015). The misplacement of nuclear TDP‐43 to the cytoplasm and its subsequent aggregation into toxic inclusions is a common pathological characteristic in more than 97% of all ALS cases (Kwong et al., 2007). Previous investigations across various molecular systems have shown that aptamers have the potential to inhibit protein aggregation. However, the exact mechanisms by which aptamers exert their influence on or regulate aggregation are not fully understood (Tran et al., 2022). In addition to the importance of considering the TDP‐43 protein, as it is implicated in certain neurodegenerative diseases, we have also leveraged its ability to recognize RNA molecules through its central domain. In fact, TDP‐43 is a modular RNA‐binding protein, and its structure includes an N‐terminal domain, two RNA recognition motif (RRM) domains that have a preference for binding single‐stranded GU‐rich RNA sequences (Tollervey et al., 2011), and a low‐complexity C‐terminal domain. Here, we select the two RRM1 and RRM2 motifs, which are capable of binding to RNA molecules, with the aim to analyze the binding, characterized through the binding affinity, between the central domain of TDP‐43 and a set of six aptamers proposed as potential interacting molecules. These short RNA aptamers are crafted using catRAPID (Bellucci et al., 2011; Cirillo et al., 2017), an algorithm designed to predict the binding propensity of protein–RNA pairs accurately by integrating secondary structure, hydrogen bonding, and van der Waals contributions (Agostini et al., 2013). As demonstrated in Zacco et al. (Zacco et al., 2022), catRAPID can be customized to design short RNA aptamers for the TDP‐43 protein by integrating it with other computational tools (Bellucci et al., 2011; Cirillo et al., 2017). To confirm the prediction we obtained computationally, we experimentally measured the dissociation constants for the six protein–RNA systems considered in this study. As the main purpose of the work, we compared the dynamical descriptors obtained by our pipeline with the experimental data, showcasing the ability of the descriptor based on the synergistic motion of interface residues to accurately describe experimental molecular affinity data. In fact, we demonstrate that the synchronized motion of residues interacting with the RNA molecule is correlated with the binding affinity, with a Pearson coefficient of approximately 0.90, thus highlighting not only the importance of MD in the analysis of protein–aptamer complexes but also providing the opportunity to design new potential aptamer candidates, useful also as probes for visualizing TDP‐43 condensates with super‐resolution microscopy, as demonstrated in (). Finally, taking advantage of the atomic resolution of molecular complexes treated in MD, the proposed methodology allows for the analysis of the role of each residue within the binding site in terms of its covariance contribution. RESULTS This section is organized as follows. First, we present the computational protocol we developed to obtain an eligible model of the biological complex composed of the TDP‐43 protein and a small RNA aptamer. Since the objective of this procedure is to analyze the dynamic properties of the binding between the two molecules, we need to generate an initial static conformation of the protein–RNA complex and then conduct MD simulations until equilibrium conditions are reached. Thus, we report the study of two dynamical quantities, that is, the distance between the protein‐binding site residues and the RNA aptamer and the covariance of the protein‐binding site residues, as two possible descriptors able to estimate the binding affinity of each protein–RNA complex. To validate our approach, the two descriptors are evaluated as predictors of the experimentally determined dissociation constant for a set of RNA aptamers in complex with the TDP‐43 protein, which was measured specifically for the set of aptamers proposed in this work for TDP‐43. Finally, we perform an analysis on the contribution of each single residue to the synchronized motion among the residues belonging to the binding region with the RNA molecule. The analysis presented in the following section concerns the binding of a selected set of aptamers with the TDP‐43 protein. The results concerning two additional and independent systems are reported in the Data S1. Computational protocol to generate and simulate a protein–RNA aptamer complex Here, we report briefly the outline of the computational protocol that we implemented to generate and simulate each TDP‐43 protein–RNA aptamer complex, starting from the RNA sequence and the 3D structure of TDP‐43 protein RRMs as deposited in the PDB entry 4bs2 (Lukavsky et al., 2013). A detailed description of the procedure can be found in Section 4. Before implementing the protocol designed to characterize protein–RNA binding in terms of binding affinity, a set of RNA molecules was suggested as potential candidates interacting with the TDP‐43 protein. In this context, RNA aptamers were designed using the catRAPID‐based procedure, which outputs the nucleotide sequence for each RNA. Following this preliminary step, elaborated further in Section 4, the initial stage of the procedure involves generating the atomic coordinates of the RNA aptamers based on their sequence (Figure 1‐I). Once an initial 3D structure is obtained, the next step is to explore the conformational landscape of the unbound RNA aptamer to identify configurations most suitable for binding with the TDP‐43 protein. In analogy with the experimental complex composed of TDP‐43 and a RNA molecule (Lukavsky et al., 2013), we assume that the RNA configuration most suitable for the binding has an overall open (unfolded) conformation, characterized by a high distance between 5′ and 3′ terminal nucleotides. Therefore, we perform MD simulation of the RNA aptamer in water at a temperature progressively higher than room temperature to increase the probability to explore open conformations (Figure 1‐II). By studying the distance between the terminal atoms of RNA in the trajectory obtained at the first temperature, we categorized the explored RNA conformational space into open and closed states (Figure 1‐III). If the defined open conformational space is not sufficiently populated, we repeat the MD simulation at a higher temperature. Once this condition is fulfilled, we identify the most representative RNA open conformations along the entire trajectory (Figure 1‐IV), performing a cluster analysis of the conformational states of the RNA which are projected into the essential space of the first two principal components of the PCA (see Section 4). Subsequently, these representative RNA conformations are docked with the experimental structure of the TDP‐43 protein (Figure 1‐V). A single complex structure is then chosen based on both its docking score and the TDP‐43 residues experimentally known to be involved in the binding with a specific RNA molecule, aiming to closely resemble the experimental bound conformation. Finally, the selected structure undergoes MD simulation under physiological conditions until an equilibrium condition is satisfied (Figure 1‐VI). The resulting trajectory constitutes the dynamical representation of the protein–RNA complex obtained from the RNA sequence and forms the basis for the subsequent analysis of binding affinity. Schematical representations of all the steps involved in the procedure to obtain a reliable reproduction of a biological complex composed by the TAR DNA‐binding protein 43 (TDP‐43) protein and RNA aptamers. (I) We generate a 3D structure of the RNA aptamer using SimRNA webserver. (II) We explore the RNA conformational space by molecular dynamics (MD) simulation of the RNA aptamer in water. (III) The conformational space is splitted into two parts, separating the elongated and close configurations. (IV) We carry out a principal components analysis (PCA), projecting the atomic coordinates obtained during the whole trajectory onto the two principal components. (V) The structures of the k cluster centroids of the elongated populations are finally processed through the web server HDOCK. (VI) Simulation of the protein–RNA complex, which are used to define two descriptors based on the dynamic behavior of the studied molecular system. Despite a significant increase in computational cost compared to a high‐throughput screening approach based on sequence information, molecular docking and extensive MD simulations for the refinement of the initial pose represent crucial steps for investigating the behavior of the molecular complex from a dynamic–structural perspective. This approach enhances the information gained compared to sequence‐based approaches but obviously limits the analysis to a smaller number of molecular complexes due to the substantial computational cost. Comparison test between docking pose and Nuclear magnetic resonance (NMR) structure To verify the capability of the docking algorithm for this specific system, we utilized one of the aptamers for which the protein–RNA complex is known via the NMR technique (PDB code: 4bs2) (Lukavsky et al., 2013). The first pose provided by the docking algorithm, which had the best score, was compared to the experimental complex. The root mean square deviation (RMSD) between the model and experimental structure was found to be 5.106 Å. As shown in Figure 2a of Supplementary Information, where a cartoon representation of the two aligned structures is reported, there is a significant structural overlap between the two aptamers (predicted model and experimental). Importantly, the majority of intermolecular chemical interactions between the model and experimental structures are preserved, as demonstrated by the following analysis: for each residue, we calculated the average distance between its C‐alpha atom and each phosphorus atom of the RNA. This residue‐specific value was calculated for both the predicted docking pose and the first NMR model. The relationship between the two variables is shown in Figure 2b of the Supplementary Information, highlighting a strong correlation between the two sets of distances (Pearson coefficient = 0.82). This demonstrates the structural similarity between the docking pose and the experimental NMR structure. Analysis of the molecular dynamics of the protein–RNA molecular system. (a) Cartoon representation of the protein–RNA complex as obtained from the Nuclear magnetic resonance (NMR) experiment (PDB code: 4bs2). In particular, the RNA‐binding region is depicted in green, and the RNA molecule is represented in light red. (b) The gyration radius descriptor as a function of time is reported for both the RNA‐1 system (on the left) and the RNA‐6 system (on the right). (c) The root mean square deviation (RMSD) with respect to the initial configuration as a function of time is reported for both the RNA‐1 system (on the left) and the RNA‐6 system (on the right). The RMSD analysis is reported for the entire system (gray curve), protein‐binding site (green curve), and the RNA aptamer (light red curve). (d) The root mean square fluctuations (RMSF) descriptor for each residue is reported for both the RNA‐1 system (on the left) and the RNA‐6 system (on the right). Dynamical descriptors for binding characterization The protocol was applied to the TDP‐43 protein, using 4bs2 PDB code (Lukavsky et al., 2013), the NMR‐resolved structure of the two RRMs (RRM1 and RRM2), we which interact with an RNA molecule. The experimentally resolved structure provides information about the residues most involved in binding to the RNA molecule, mainly located in the linker connecting the two RRM motifs. These residues can serve as constraints to identify the most likely binding region, including for other proposed aptamers. Indeed, in this work, a set of six RNA aptamers were selected, whose sequences and experimentally determined dissociation constants are reported in Table 1 (see Section 4 for details about the design and the experiments). In this section, we analyzed the six MD simulations we performed involving the molecular complex between TDP‐43 and the RNA aptamers. A detailed overview of the MD simulations of the RNA aptamers can be found in the Data S1 (more specifically in Table S1, the equilibrium range for each simulation is reported). TABLE 1The sequences of nucleotides of the RNA aptamers for the TAR DNA‐binding protein 43 (TDP‐43) protein, starting from 5′ to 3′. A is adenine, C is cytosine, G is guanine, and U is uracil, following the IUPAC nomenclature. We report here also the K d corresponding to each aptamer, as obtained from the experimental procedure (see Section 4).NameSequence K d (nM) σKd (nM)RNA‐1GUGUGAAUGAAU40.9RNA‐2AUUCAUUCACAC1360600RNA‐3GUGGUCCCCG750150RNA‐4AGCAACACCG1320350RNA‐5GCGGGGCCCG1800400RNA‐6CGGUGUUGCU7010 The sequences of nucleotides of the RNA aptamers for the TAR DNA‐binding protein 43 (TDP‐43) protein, starting from 5′ to 3′. A is adenine, C is cytosine, G is guanine, and U is uracil, following the IUPAC nomenclature. We report here also the K d corresponding to each aptamer, as obtained from the experimental procedure (see Section 4). The main purpose of this work is to assess the stability of the protein–aptamer complex, evaluating its dynamical properties. First, we analyzed the stability of such a simulation studying the RMSD, the gyration radius, and the root mean square fluctuations (RMSF), reported in Figure 2. For more details, see Figure S1, where the RMSD, RMSF, and radius of gyration plots are reported for each system. Figure 2 shows the RMSD and RMSF plots for two of the simulated protein–RNA systems. In particular, we present, by way of example, the analyses for the systems related to the RNA‐1 and RNA‐6 aptamers, those with the best binding affinity to the TDP‐43 protein. Plots for all other systems are shown in the Data S1. The RMSD analysis reveals that all systems eventually reach equilibrium, although not all achieve this within the initial microsecond of simulation. In many of the studied cases, the average fluctuation of RNA nucleotides is higher than that of protein residues, especially at the terminal regions of the RNA molecule. This outcome is expected, as the initial and final regions of the aptamer may have fewer interactions with protein residues, making them more susceptible to fluctuations induced by thermal noise. Overall, in accordance with our expectations, the interacting regions (highlighted in color) exhibit lower average fluctuations. Despite the importance of these preliminary analyses in characterizing the dynamic behavior of the considered complexes, the descriptors used are insufficient to describe the protein–RNA binding in terms of interaction energy. Therefore, in order to quantify the potential relationship between binding properties (in this case in terms of binding affinity) and the motion of the residues involved at the binding interface, we need to estimate the structural–dynamical stability of the entire molecular complex. To this end, we rely on the assumption that a high binding affinity between the protein and the RNA aptamer corresponds to the capability of the complex itself to preserve its structure as effectively as possible under typical thermal noise (Gergely et al., 2000). To maintain such stability, we consider two factors: the ability of the two interacting molecules to retain their typical binding distance during the simulation and the coordination of the motion of the interacting residues. In this regard, we introduce two new descriptors for analyzing both the compactness of the two interfaces and the correlated motions between the atoms of the two molecules involved in the interaction. The comparison between our computational estimates and the experimental data of the dissociation constant would represent an important result and confirmation of the efficacy of the procedure. For this reason, the dissociation constants for each protein–RNA complex analyzed have also been experimentally determined. The binding interface compactness descriptor: Average distance between protein and RNA structures The first approach employed in this study to correlate the binding strength of a protein–RNA complex with dynamic–structural properties is the mean distance between the two binding partners: the TDP‐43 protein and the corresponding RNA aptamer. To this end, we calculate for each C‐alpha atom belonging to the binding residues the distance to the nearest phosphorus atom of the RNA aptamer. Therefore, for each frame of the simulation, we can calculate the average value of this distance, indicating the spatial compactness of the two bonded interfaces. Additionally, by considering all frames of the simulation, we calculate both the average value and the standard deviation. We compared the results of the distance‐based descriptor (Equation 1) with the corresponding experimental values of the dissociation constant (Table 1), which is an experimental measurement of the binding affinity between the two molecules. The idea is that in the aqueous cellular environment the RNA is subject to external stresses, and therefore, an unstable pairing would cause the RNA aptamer to separate from the receptor protein and their relative distance to increase. In other words, the greater the distance between the two interacting molecules, the lower the binding affinity should be. Setting a distance threshold between the C‐alpha atom and the phosphorus atom of nucleotides interacting with the protein allows us to define the two binding regions. Following this, for each binding site, we compute the Pearson coefficient to assess the linear relationship between the distance‐based descriptor, which is called D (see Section 4) and the dissociation constant K d. A higher Pearson coefficient indicates a stronger and more reliable linear relationship between the theoretical descriptor derived from simulations and the experimental binding affinity value. Therefore, we examined how the Pearson coefficient between these two variables varies as we adjust the threshold used to define the binding regions. We observed an overall greater distance in correspondence of a weak binding in the range between 10.0 and 18.0 Å (Figure 3a), with the strongest correlation (Pearson coefficient of 0.46) for 17 Å (Figure 3b). Nevertheless, as indicated from the corresponding p‐value (0.3), the significance of these results is not enough to validate our starting hypothesis, and it is necessary to develop a more complex descriptor to serve as a proxy for the dissociation constant. Furthermore, this result indicates that there is not a strong dependency between the compactness of binding regions and the strength of interaction, suggesting that interface residues have the freedom to fluctuate and move away from their partner even in complexes with high binding affinity. Therefore, despite an overall trend of positive correlation for all considered distance thresholds, further analysis is needed to better understand the nature of interface residue motions, exploring the role of potential correlated motions. Analysis of the descriptor based on interface atom distance. (a) The Pearson correlation value was computed with variations in the distance threshold that defined the binding region. (b) A scatterplot shows the relationship between the theoretical descriptor D (distance) and the experimental K d value, which is associated with the distance threshold that yielded the highest Pearson correlation. The line on the plot represents the linear fit between the data points. The descriptor based on the synergy of binding residue motions: The covariance of interface residues Typically, as also in the case studied in this work, aptamer–protein binding is characterized by a remarkable difference between the size scale of the aptamer and the receptor molecule (Dua et al., 2011). Indeed, in the equilibrium time range of our MD simulations, the gyration radius of the protein is stably about twice that of RNA. The protein spatial organization is indeed the result of many interactions between a large number of residues. On the contrary, a RNA aptamer is a short strand of nucleotides, and its 3D structure may be assumed, as a first approximation, as a linear chain (although characterized by non‐trivial minimum conformations, some of which are more compact than others) with respect to the protein. In this framework, we assumed that a key factor, in order to preserve the complex structure in the presence of external stresses, is the synergy of the motion of binding interface residues. A higher positive correlation of the atomic fluctuations between these residues would reveal a greater binding affinity, ensuring that eventual solicitations are absorbed and dumped. We selected the atomic covariance (Equations 2 and 4) as a measure of such property, since it describes at the same time how two residues communicate during the motion of the simulation and the absolute value of their atomic displacements (Section 4). In analogy with what we did for the distance descriptor, the covariance was then associated with the experimental value of the dissociation constant of each complex. For almost the whole range of the threshold of definition of the protein‐binding site (Section 4), we found a correspondence between highly synergistic residues and greater binding affinity (Figure 4a). Starting from 11.0 Å and until 19.0 Å, the validation of our hypothesis is very strong (a Pearson coefficient lower than −0.73 corresponds to a p‐value less than 0.01). The shape of the curve shown in Figure 4a clearly indicates the presence of an optimum, as the anti‐correlation values progressively decrease until they reach a minimum, and then rise for larger distance threshold values. In particular, the optimal correlation is obtained with a threshold of 18.0 Å (Pearson coefficient = −0.89 and p‐value = 0.007): in this case, the computational predictions of the binding affinity are almost proportional to the experimental values (Figure 3b). It is worth noting that the best values of correlation are obtained for the covariance descriptor with a very similar definition of binding site, probably identifying the actual set of residues involved in RNA recognition. Analysis of the descriptor based on correlated motion of residues within the binding region. (a) We calculated the Pearson correlation value as the distance threshold defining the binding site varied. (b) A scatterplot was generated between the descriptor based on the covariance of residues within the binding site and the experimental K d value associated with the distance threshold producing the highest Pearson correlation. The line on the plot represents the linear fit between the data points. Testing the covariance‐based descriptor in two additional complexes with high and low affinity In this work, we specifically studied the affinity between TDP‐43 and a set of RNA aptamers, testing the hypothesis that the distance between the two interfaces and the covariance among the binding site residues may be a good predictor of their binding affinity. The choice of the TDP‐43 protein represents an interesting object of study since its aberrant behavior has been implicated in the development of multiple neurodegenerative diseases. By examining the differences due to different RNA aptamers on the TDP‐43 protein, we were able to effectively measure the differences between the bound and unbound forms of the protein. Indeed, the advantage of selecting a single case of interest, in addition to the considerable biological importance of the system, is that it automatically subtracts the contribution of the covariance between the binding site residues when the protein is in its unbound state, which was not directly investigated because it is common to all the protein–RNA systems in this study. Thus, considering the same protein in interaction with different RNA aptamers allows us to characterize the difference in covariance between one aptamer's interaction and another's. Despite this basic motivation underlying our work, an additional test with two different protein–RNA complexes, both with known experimental K d, was also conducted. In particular, from (Hong et al., 2023), we selected two complexes: pdb 3AEV, with a K d of 9.80E−10 mol, and pdb 2KH9, with a K d of 9.00E−05 mol. These two systems were selected in order to maximize the difference between their K d values, among the protein–RNA complexes of the dataset where the aptamer sizes are comparable to those used in our study (since in many cases, the RNA completely envelops the interacting protein). For each of the two protein–aptamer complexes, a 500 ns MD simulation was conducted, and the RMSD analysis is shown in Figure 3a. After discarding the initial equilibration phase for both systems, we analyzed the average covariance of the binding residues, which are selected based on a distance cutoff between the C‐alpha of the protein and the phosphorus (P) atom of the interacting RNA. For almost all the cutoff values considered, the complex with higher binding affinity (low K d) has a significantly higher average covariance among the binding site residues compared to the complex with low binding affinity (high K d), see Figure 3b of SI. Despite the selection of only two additional complexes, this analysis suggests the possible generalization of the approach adopted in this work, which obviously requires further investigation. Analysis of the average covariance of each residue across all protein–RNA complexes Given the importance of the covariance‐based descriptor in capturing protein–RNA binding information in terms of binding affinity, a more in‐depth investigation into the synergy of interface residue motions is necessary to better understand the role of correlated motions both in general and at the single residue level. To this end, we considered the covariance value of all residue–residue pairs within the binding regions of all six protein–RNA complexes analyzed. We thus calculated the average covariance for each residue within the binding site of the six molecular systems, taking into account covariance values of each residue i with each other residue j belonging to the binding site. This procedure allowed us to analyze the role of each residue. To obtain an initial overall result based on this analysis, the distribution of the average covariance values is shown in Figure 5a. The area under the probability density function curve is colored based on the covariance value. The positive covariance values represent 68% of the total area of the distribution. Interestingly, we show that the majority of residues at the binding interface exhibit a covariate motion, while only a smaller portion follows an anticovariate motion. As expected, the peak of the distribution is located near the zero covariance value, indicating that most residue pairs at the interface have very low covariance. Analysis of the synergistic motion between interface residues mediated for each residue. (a) Distribution of the average covariance values for each residue within the binding site of all protein–RNA systems. The gray area represents the region under the curve corresponding to positive covariance values (AUC = 0.68). (b) Average covariance for each residue, which was obtained by averaging the average covariance for each system across the six protein–RNA systems. Only the residues belonging to the binding site of all protein–RNA systems were selected. (c) Colormap of the average covariances for each residue within each protein–RNA complex. For each residue and each complex, the average covariance is displayed, obtained by averaging the covariance of each residue pair in which the residue is involved. The values have been normalized, ranging from a minimum of −1 to a maximum of +1. To explore the average contribution of each residue in the six protein–RNA systems analyzed in this study, we calculated the mean of the average covariance for each residue across all the pairs it interacts with in each protein–RNA complex across the six molecular systems. Specifically, for this analysis, we considered only the residues involved in the binding sites of all protein–RNA complexes. Following the procedure, we selected 54 residues, whose average covariance values are shown in Figure 5b. To highlight the contribution of each individual residue in each protein–RNA complex, we present the colormap in Figure 5c, where the average covariance value of each residue for each complex is reported. Analyzing the results for each residue across all molecular systems, we identified that residues G245, C244, S258, C198, R197, G196, V195, A228, P187, and F229 predominantly contribute to the correlated motion of binding site residues at the protein–RNA binding interface. DISCUSSION In silico investigation of the molecular mechanisms that regulate interactions between molecules is an important issue in the framework of computational structural biology, since it represents a powerful tool for investigating system properties, typically at the atomic level, which are not easily available through experimental techniques. Here, we present a new computational protocol aiming at estimating the binding affinity between protein and RNA aptamers, based on the analysis of the covariance of the binding residues at the end of the entire computational process. More specifically, we focused on the application of the computational strategy to a specific biologically relevant case. In the biological context of neurodegenerative diseases, it has been demonstrated that interactions between proteins and relatively small RNA strands play a pivotal role not only in terms of diagnostics but also in therapies for such pathologies. In particular, in this work, the protein TDP‐43 was analyzed, since it is one of the most implicated in neurodegeneration. The approach presented here is based on the MD technique and defines dynamic descriptors to investigate the nature of the binding between the two molecules. This binding was quantified in terms of the dissociation constant, experimentally determined in this study to provide information on the specific biological case selected. Starting from the established concept of “dynamic personality” of proteins (Henzler‐Wildman & Kern, 2007), for which the functions of proteins are governed ultimately by their dynamic character, we designed a computational procedure able to predict in silico the structure of a complex composed of a protein and a RNA aptamer, starting only from a 3D structure of the protein and the genetic sequence of the RNA aptamer. Although the method may appear to be contingent on the requirement to know the experimental structure of the protein involved in binding with the aptamer, in the current era of AlphaFold (Jumper et al., 2021), it is possible to extend this methodology to all proteins with unknown structures by using the AlphaFold algorithm for predicting, taking into account its error estimate on each individual residue, the three‐dimensional structure. The main hypothesis of this work is that a high binding affinity of a given protein–aptamer system corresponds to the capability of a complex to preserve its structural and dynamical properties in the presence of its molecular partner. The overall compactness of the predicted complex is assessed through the average distance between the ligand and the receptor. The results concerning the distance descriptor (see Figure 3) present a low degree of statistical significance. This is probably due to the general nature of such a descriptor, determined by a high number of factors that are not only related with binding affinity (such as electrostatic repulsion and solvation in water environment). However, the analysis is of considerable importance because it examines the role of compactness between the two interacting regions. Despite not showing a strong linear relationship between this descriptor and the experimental binding affinity for all analyzed distance threshold values (used for the definition of the binding region), we observed a positive correlation value for almost all threshold values considered. Furthermore, the analysis reveals a potential and unexpected spatial separation between the two interacting molecules, even in systems with a high binding affinity. The considerable distance present in all systems (even those with other binding affinity) could highlight the role of the fluctuations of the interacting residues. To address this, a new descriptor based on the covariance of motions of binding residues belonging to the protein has been developed. On the other hand, the results regarding the covariance descriptor are very promising. From Figure 4, we note that the coordinate motion of interface residues is an effective indicator of the protein–RNA binding affinity. Moreover, looking at panel a of both Figures 3 and 4, we note that it exists in an optimal region of the protein whose motion is correlated when the protein is in interaction to a relatively small ligand such as a RNA aptamer, suggesting that this condition could be a possible definition of biding region for a protein–RNA system. Indeed, the binding site thresholds providing the maximum value of correlation with binding affinity are very similar for the two descriptors (17–18.0 Å). This result provides a biophysical definition of the set of residues actually involved in RNA recognition. Given the capability of the covariance‐based descriptor of interface residues to estimate the binding affinity of each protein–RNA complex, we conducted a more detailed analysis of the covariance between pairs of residues involved in the binding site, also investigating the contribution of each individual residue. We initially demonstrated that the majority of interface residues are characterized by positive covariance, highlighting that, on average, correlated motions are largely predominant in protein interface. Furthermore, in order to link the analysis of the correlated motion of each residue with the binding affinity of the complex, we analyzed only the average covariance of each residue belonging to the RNA‐1 and RNA‐6 systems, as these are the two systems with the highest binding affinity in terms of experimental dissociation constant. The residues P187, G245, A260, C244, S190, N259, S258, G196, I257, H256, and R197 are considered to be key residues in explaining the mechanism related to a high binding affinity, given that these residues in both systems have a very high covariance. A set of residues around residue 195 and residue 260, as computationally predicted with the approach proposed here, have also been found to be critical residues in terms of chemical‐shift perturbations of TDP‐43 RBD upon RNA binding, as these residues undergo major structural rearrangements upon binding with the aptamer (Lukavsky et al., 2013). Interestingly, RNA‐1 and RNA‐6 systems, both characterized by high binding affinity values, display an opposite behavior in terms of the covariance path at the individual residue level. In this case, residues with high covariance belonging to the RNA‐1 system have significantly lower covariance in the RNA‐6 system. These results show that residues with high average covariance are not always the same in the analyzed systems, suggesting further investigation in the future in this regard, since the key residues for binding affinity may depend on the molecular system under consideration. In conclusion, the computational strategy and analyses presented in this work can function as a reliable computational protocol for predicting the molecular complex between proteins and RNA aptamers, as evidenced by the observed stability in their MD simulations. Moreover, the covariance of interface residues has demonstrated its efficacy in estimating the experimental binding affinity value, measured in terms of dissociation constant. This could potentially pave the way for its prediction and the development of similar protocols in aptamer design. MATERIALS AND METHODS Design and quantification of the binding constants of the aptamers The aptamers were designed following the identical procedure detailed in (). Biolayer interferometry experiments were carried out using an Octet Red instrument (ForteBio, Inc., Menlo Park, CA) at 25°C. The binding assays were executed in 10 mM potassium phosphate buffer (pH 7.2) with 150 mM KCl and 0.01% Tween20. Streptavidin‐coated biosensors were loaded with 1 μg/mL RNA aptamer, which had been modified with biotin on the 3′ end and exposed to increasing protein concentrations varying from 20 nM to 20 μM, depending on the strength of the binding. K d values were determined by fitting the response intensity (wavelength shift upon binding) relative to the protein concentration at a steady state. The assay was repeated at least three times, with each repetition carried out in triplicate. Protein purification The purification of RRM1, RRM2, and RRM1‐2 from kanamycin‐resistant pET‐Sumo expression vectors containing a six‐histidine tag (his‐tag) and expressed in Rosetta (DE3) Escherichia coli cells was conducted following previously established procedures (Zacco et al., 2018). In summary, after cell lysis and centrifugation, the protein fractions were resuspended and passed through a nickel‐coated resin. Non‐His‐tagged proteins were washed away, and the constructs were eluted using a high imidazole concentration solution (10 mM potassium phosphate buffer pH 7.2, 150 mM KCl, and 250 mM imidazole). The eluate was dialyzed to remove excess of imidazole, and the His‐tag and the Sumo domain were cleaved using Tobacco Etch Virus protease (at a 1:20 enzyme‐to‐protein molar ratio). Subsequently, an additional affinity purification step with a nickel‐coated resin was performed to remove nucleic acids. The protein was then eluted using 1.5 M KCl and subjected to size‐exclusion chromatography with a HiLoad 16/60 Superdex 75 column. Protein identity and purity were confirmed through polyacrylamide gel electrophoresis (PAGE). Finally, the proteins were rapidly frozen and stored in a low salt buffer (10 mM potassium phosphate pH 7.2, 15 mM KCl) at −80°C. Design of aptamers for TDP‐43 using catRAPID We initiated the catRAPID predictions with a pool of over 3 million RNA fragments, identified through iCLIP, that bind TDP‐43 (Tollervey et al., 2011). From this pool, we selected contiguous sequences enriched over the experimental control, characterized by interacting nucleotides close to each other (with a maximum distance of three nucleotides between two peaks and a minimum region size of 50 nucleotides). For each interaction site, the iCLIP score was computed as the sum of experimental reads divided by the length of the sequence. A positive set was generated, comprising 100 sequences with the highest iCLIP scores. As a negative counterpart, a pool of 100 sequences from the iCLIP data was selected, each having the same length as the positive set but with contiguous non‐interacting nucleotides. To prioritize candidate RNA aptamers, RNA and protein fitness scores were computed. For each aptamer candidate “s,” the RNA fitness score (1), ranging between 0 and 1, was introduced to assess the effect of a random mutation “i” on the catRAPID interaction propensity for TDP‐43. Similarly, the protein fitness score (2) of a given sequence, ranging between 0 and 1, evaluated how strong the catRAPID interaction propensity was for TDP‐43 in comparison with a protein sequence “i” having the same length and amino acid composition. Molecular dynamics simulations The MD simulations described in the next section, as well as their setting up, are performed with the GROMACS software (Abraham et al., 2015) (v. 2019.3) and have the following common settings. The topology of the system is generated with the AMBER force field ff99 (Wang et al., 2000). The system is solvated in a cubic box such that the edge of the box was 1.0 nm away from the closest atom. Water molecules are modeled with the SPCE model (Berendsen et al., 1987). Na+ and Cl− counterions are added to ensure charge neutrality. The system undergoes an equilibration procedure composed of: (i) energy minimization with steepest descent; (ii) a 100 ps‐long equilibration in the NPT ensemble using the Parrinello‐Rahman barostat (Parrinello & Rahman, 1981) at 1 bar; and (iii) a 100 ps‐long equilibration in the NVT ensemble using the Bussi‐Donadio‐Parrinello thermostat (Bussi et al., 2007) at the given temperature. The equations of motion are integrated by a leap‐frog algorithm with timestep equal to 2 fs. The smooth particle mesh Ewald method (Essmann et al., 1995) is used to calculate electrostatic interactions with a cutoff equal to 1.0 nm. Van der Waals interactions were cutoff at 1.0 nm. Each molecular simulation was conducted for a duration of 1 μs. If the system did not show stability in terms of RMSD during the last 0.6 μs of the simulation, an extra microsecond was incrementally added until the desired equilibrium condition was reached. Computational protocol to generate and simulate a protein–RNA aptamer complex The computational protocol we developed requires as inputs a 3D structure of the target protein and the sequence of an RNA aptamer, and it is composed of the following six steps. The settings partially depend on the TDP‐43 target systems, but the steps are generally applicable to a generic protein–RNA aptamer system. Initial RNA 3D structure The first step consists of generating a 3D conformation of the RNA aptamer starting from the nucleotide sequence. This is achieved with the SimRNA webserver (Magnus et al., 2016), relying on a coarse‐grained Monte Carlo simulation with the default number of 500 steps. RNA simulations The RNA conformer output of a is not optimized for the binding with a protein, and we therefore proceed with a deeper exploration of its conformational space via MD simulations. The RNA undergoes an MD simulation with the described settings, at the initial temperature of 350 K and for a total time of 1 μs. RNA open conformational space In analogy with the experimental complex composed of TDP‐43 and RNA1 (Lukavsky et al., 2013), we assume that the RNA configuration most suitable for the binding has an overall open (unfolded) configuration, characterized by a high distance between 5′ and 3′ terminal nucleotides. We therefore look for these RNA conformations among the ones explored during the simulation. For each frame of the trajectory, we calculate the euclidean distance D ter of the P atoms in the 5′ and 3′ terminals of the aptamer. Then, we perform a N‐modal fit of the distribution of D ter, starting with N = 2 Gaussian functions and increasing until convergence of the fit. The open configurations are finally defined as the set of structures whose terminal distance D ter was higher than the mean of the Gaussian function with greater mean among the N ones. If the population of the open states is lower than 10% of the total trajectory, we again perform the RNA simulation in b, increasing the temperature to 400 K. If this condition was not satisfied for 400 K either, we start a dynamic with T = 500 K and we stop it when the aptamer does explore a single open configuration. The latter is then used as the starting one again with T =350 K. Representative RNA conformations Among the explored open conformations, we want to select the representative ones, which will be the ones docked with the target protein (see Section 4.5.5). First, we carry out a PCA, projecting the atomic coordinates obtained during the whole trajectory onto the two principal components. The resulting points in the 2D essential space corresponding to the open conformations are clustered using the k‐means clustering algorithm as implemented in scikit‐learn python library (Pedregosa et al., 2011). The number k of clusters is calculated with the silhouette algorithm (Rousseeuw, 1987), also implemented in scikit‐learn. The centers of the k identified clusters are selected as the representative conformations of the RNA aptamer open states. Molecular docking of the RNA with TDP‐43 protein simulations From the identified RNA conformations and the starting structure of the protein, we need to generate a 3D complex structure. The k identified cluster centers in Section 4.5.4. undergo a molecular docking operation with the TDP‐43 protein resolved in pdb 4bs2 (Lukavsky et al., 2013) using the web server HDOCK (Yan et al., 2020). It is known from in vivo experiment that TDP‐43, being a ribonucleoprotein (RNP) composed by two RRMs, interacts with RNA with both the domains, especially for high‐affinity complex configurations (Bhardwaj et al., 2013). In this work, we exploit this experimental knowledge by assuming that all the aptamers bind in the same protein region, constraining the molecular docking of the RNA aptamer in this region. The top‐scored output conformation of the software that satisfies the double‐domain interaction is selected for the subsequent MD simulation. MD production of the protein–RNA complex In our approach, the protein–RNA complexes are investigated from a dynamical point of view with the aim to take into account the inherent flexibility of biomolecules (Henzler‐Wildman & Kern, 2007) in the study of their binding affinity. The protein–RNA complex obtained in Section 4.5.5. is used as a starting structure for an MD simulation with the described settings, at room temperature and for an initial simulation time of 1 μs. The equilibrium of the produced trajectory is evaluated through an analysis of the RMSD of the whole structure with respect to the initial configuration and the gyration radius of the protein‐binding site. The total time is increased by further steps of 1 μs, until these two quantities do not vary significantly. Binding affinity evaluation Protein‐binding site All the analyses below focus locally on the protein‐binding site, namely the set of the protein aminoacids involved in the binding with the RNA aptamer, determined in the complex structure conformation corresponding to the average value of RMSD with respect to the starting structure within the equilibrium range (Table S1). We identify the alpha carbon Cα atom of the backbone chain to be a representative atom for the whole protein residue. In analogy with this last assumption, we identify the phosphorus atom P of the backbone chain to be a representative atom for the whole RNA residue. Therefore, the protein binding site is defined as the set of Cα atoms within a distance threshold d BS. The analysis of the binding is performed by varying the threshold d BS between 8.0 and 20.0 Å with step of 0.5 Å. Distance between protein and RNA aptamer To study the distance between the protein and the RNA aptamer, we define the quantity D as the average over the trajectory of the euclidean distance between the Cα atoms of the protein binding site with the closest P atom of the RNA aptamer: where riRNAt and rjBSt are the coordinates at the frame t, respectively, of the ith RNA nucleotide and the jth protein‐binding site residue; N RNA is the total number of RNA nucleotides; N BS is the total number of protein‐binding site residues, and the sum runs over all the frames of the corresponding equilibrium range (Table S1). Covariance of the protein functional residues To study the synergy of the motion of the protein‐binding site residues, we introduce the quantity C that is their average covariance along the simulation: where σrjrk is the covariance between the jth and kth protein‐binding site residues. where μj and μk are the average coordinates of the jth and kth protein‐binding site residues, and the sum runs over all the frames of the corresponding equilibrium range (Table S1). Finally, we take the values of covariance between the TDP‐43 protein‐binding site and the RNA1 aptamer (see Table S1) as reference, and we define the following z‐score quantity Z where C 1 is the average covariance of the protein‐binding site residues (Equation 2) for the protein‐RNA1 complex and Σ1 their standard deviation. AUTHOR CONTRIBUTIONS Francesco Paolo Panei: Investigation; validation; formal analysis; data curation; writing – original draft; writing – review and editing. Lorenzo Di Rienzo: Investigation; validation; methodology; formal analysis; data curation; writing – original draft; writing – review and editing; conceptualization. Elsa Zacco: Investigation; data curation; methodology; conceptualization; writing – review and editing. Alexandros Armaos: Formal analysis; validation; methodology; writing – review and editing. Gian Gaetano Tartaglia: Conceptualization; visualization; funding acquisition; writing – original draft; writing – review and editing. Giancarlo Ruocco: Conceptualization; funding acquisition; writing – review and editing; visualization. Edoardo Milanetti: Conceptualization; methodology; data curation; supervision; formal analysis; validation; visualization; writing – review and editing; writing – original draft; investigation. Supporting information Data S1. Supporting Information. ACKNOWLEDGMENTS This research was partially funded by grants from ERC‐2019‐Synergy Grant (ASTRA, n. 855923); EIC‐2022‐PathfinderOpen (ivBM‐4PAP, n. 101098989); and Project “National Center for Gene Therapy and Drugs Based on RNA Technology” (CN00000041) financed by NextGeneration EU PNRR MUR—M4C2—Action 1.4—Call “Potenziamento Strutture di Ricerca e Creazione di Campioni Nazionali Di R&S” (CUP J33C22001130001). Panei FP, Di Rienzo L, Zacco E, Armaos A, Tartaglia GG, Ruocco G, et al. Synchronized motion of interface residues for evaluating protein–RNA complex binding affinity: Application to aptamer‐mediated inhibition of TDP‐43 aggregates. Protein Science. 2024;33(12):e5201. 10.1002/pro.5201 Review Editor: Nir Ben‐Tal REFERENCES Associated Data Supplementary Materials Data S1. Supporting Information.