Enhancing aptamer selection in alzheimer’s disease: integrating structure prediction and molecular dynamics simulations

Abstract

Alzheimer’s disease is the most frequent neurodegenerative disease and the leading cause of dementia worldwide. With disease-modifying treatments highly requested, numerous aptamers have been experimentally selected, showing high affinity and specificity binding to the main drivers in the pathology. Still, more studies are needed to compare the biochemical properties and target interactions to streamline the generation of high-efficacy therapeutics. With recent improvements in bioinformatics, we predicted the 2D and 3D structures of known aptamers based on literature-derived sequences, followed by molecular dynamics, molecular docking, and MM/PBSA binding affinity simulations of the aptamer-target complexes. We observed a strong correlation between experimental affinity values and predicted binding free energies, demonstrating the value of implementing computational strategies to streamline the selection process. We identified DNA aptamers as most promising due to their high predictability compared to RNA aptamers and the low docking scores of peptide aptamers. Furthermore, we identified hydrophobic and basic amino acids most frequently contributing to the interaction, with the basic amino acids, arginine, histidine, and lysine accounting for most interactions in all groups. This suggests that forming hydrophobic pockets and ionic interactions mediates aptamer binding, allowing a more directed targeting of Alzheimer’s disease and providing the basis for future modifications.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-12186-1.

Introduction

As the most common type of dementia, Alzheimer’s disease causes 70% of the cases¹affecting more than 40 million people worldwide². Symptoms of this progressive neurodegenerative disease range from mild memory loss in the early stages³ to severe symptoms, including disorientation, development of depression^4,5a loss of impulse control, difficulties in reading, writing, and speaking³ and life-threatening complications, such as difficulties in swallowing⁶. The pathology of Alzheimer’s disease is characterized by amyloid-β (Aβ) plaques and tau neurofibrillary tangles (NFT), leading to neuronal cell death and cognitive dysfunction⁷.

Moreover, it is hypothesized that Aβ plaques and tau NFTs play a crucial role in the generation and progression of Alzheimer’s disease. By enzymatic, proteolytic cleavage of the amyloid precursor protein (APP), β-secretase and γ-secretase, beta-amyloid protein (Aβ) fragments are generated⁸. The resulting fragments, Aβ40 and Aβ42, can self-aggregate due to their hydrophobicity, forming soluble oligomers, fibrils, and amyloid plaques, which cause damage to axons, dendrites, and result in the loss of synapses^9,10. While accumulations of Aβ fragments can also be observed in healthy brains, it is suggested that decreased degradation of Aβ drives neurotoxicity and tau pathology induction in Alzheimer’s disease, ultimately leading to neuronal cell death and neurodegeneration¹¹. In addition to Aβ aggregates, hyperphosphorylation of the microtubule-associated tau protein leads to dissociation and instability of the microtubules and the formation of NFTs¹². It is proposed that hyperphosphorylation of the tau protein is highly influenced by Aβ deposits and hyperactivation of mTOR, promoting the formation of paired helical filaments (PHFs) and NFTs¹³ causing instability-related disorders referred to as tauopathies¹⁴ (Fig. 1a).

Fig. 1.

Aptamers targeting key components in the pathology of Alzheimer’s disease and bioinformatic pipeline predicting and analyzing aptamer-protein binding modes. (a) Aptamers targeting key components in the pathology of Alzheimer’s disease were obtained from the literature targeting molecules in the Aβ- or Tau-mediated Alzheimer’s development. For comparison, aptamers were grouped into RNA (red), DNA (yellow), or peptide (blue) structures. (b) The schematic overview of the bioinformatic pipeline for predicting aptamer structures, performing molecular dynamics, and analyzing molecular docking.

Current therapies for Alzheimer’s disease aim to improve symptoms, but disease-modifying therapies (DMTs) are highly needed. Aptamers are a promising approach for treating neurodegenerative diseases as clinical diagnostic tools, either in the form of biomarkers¹⁵ or as therapeutic agents in the form of aptamer drug conjugates (ApDCs)^16,17 or inhibitors^18,19. Targeting the pathophysiological mechanisms, immunotherapies, and small-molecule inhibitors are promising approaches to prevent the development and limit the progression of Alzheimer’s disease³. Previously, it was shown that inhibition of the secretases or aggregation of Aβ and tau has promising effects in different stages of Alzheimer’s disease¹³.

The single-stranded oligomers, composed of DNA, RNA, or peptides, can bind to their targets with high affinity and specificity by folding into tertiary structures^20,21. The interaction is based on non-covalent interactions, hydrogen bonding, electrostatic interactions, and van der Waals forces²²enabling aptamers to recognize and bind targets like monoclonal antibodies²³showing superior thermal stability, modifiability, and causing lower immunogenicity²⁴. Due to their smaller size compared to antibodies, aptamers can bind to targets that are inaccessible to antibodies²⁵. Using systematic evolution of ligands by exponential enrichment (SELEX)-based approaches, aptamers are positively selected against the target molecules, including β-secretase, Aβ fragments, and tau protein, identifying aptamers with high affinity from random libraries²¹ (Fig. 1a).

While many aptamers have been selected and tested experimentally, little research has been conducted to compare different aptamers and classes based on their biochemical properties, molecular structures, and binding characteristics. In contrast to conventional methods, in silico approaches, including molecular dynamics simulation and molecular docking, can be used to predict aptamer conformers and their binding to target proteins. In recent years, it has been shown that molecular dynamics simulations can be used to predict protein-protein and protein-nucleic acid structures and interactions with increasing accuracy and correlation to experimentally determined values, especially for small nucleotide-based molecules²⁶.

Based on the primary sequence and experimentally determined dissociation constant, computational prediction of structures, docking, and molecular dynamics simulations allowed a comparison of structural features, changes in free energy, and the interacting amino acids as crucial factors for improving the binding affinities of aptamers²⁷ and enable a more targeted selection of aptamers with high efficiency in the future.

Results

Secondary structure prediction revealed an increased loop formation and lower structure ΔG values, indicating a higher stability of RNA-based aptamers

Aptamer primary sequences were obtained from the literature, targeting key players in AD, including Aβ40 and Aβ42, tau, and β-secretase (Fig. 1a), along with dissociation constant (K_d) values (Supplementary Table S1, S3, S5) when available. To compare the different classes of RNA, DNA, and peptide aptamers, we predicted their secondary structures using Mfold, selecting the most stable structure for further processing based on the predicted changes in free energy (ΔG) values related to the formation of secondary structure. Additionally, we converted the nucleic acid sequences by thymine or uracil replacement to analyze changes in the free energy of secondary structure. For better visualization of the bioinformatic pipeline, a schematic overview was generated to guide the processes of aptamer preparation, molecular dynamics simulation, molecular docking, and binding analysis (Fig. 1b).

When comparing the different classes of nucleic acid aptamers, experimental K_d values of RNA aptamers showed a high variance and significantly lower binding affinity towards their target proteins than DNA aptamers (Supplementary Table S1 and S3, Fig. 3a). However, predicted secondary structures revealed a higher formation of loop and stem structures, resulting in significantly lowered change in free energy (ΔG) of RNA aptamer structures (Supplementary Table S1 to S4, Fig. 3b), which was also observed for converted sequences. While the change in free energy decreases after thymine to uracil conversion of DNA aptamers, the stability of RNA aptamers was reduced in the corresponding DNA form (Supplementary Table S1 and S3, Fig. 3b), making the RNA aptamers more than twice as energetically favorable. These data suggest that RNA aptamers are more thermally stable due to the increased formation of secondary structures, but exhibit a reduced binding affinity compared to DNA aptamers.

3D structure prediction, molecular dynamics simulations, and molecular Docking reveal the best binding modes for aptamer-target complexes and binding free energies

To compare RNA, DNA, and peptide aptamer classes and characterize the aptamer-target interactions, we predicted aptamer 3d structures using 3dRNA/3dDNA for nucleic acid or PEP-Fold4 for peptide aptamers (Supplementary Table S2, S4, S5), with the most stable conformer selected for further processing, based on the webserver’s change in free energy scoring system. Molecular dynamics (MD) simulations were performed to account for the conformational flexibility of the biological molecules. The simulation trajectories were clustered using a root mean square deviation (RMSD)-based approach to identify structurally stable conformations. The average structures of the three most populated clusters were selected as representative conformations.

RMSD analysis of the simulation trajectories revealed high inter-aptamer differences in structural flexibility among nucleic acid aptamers, with DNA and RNA aptamers exhibiting higher overall flexibility (Supplementary Fig. S1a, S1b) than peptide aptamers and target proteins (Supplementary Fig. S1c, S1d). The variability is likely influenced by differences in aptamer length, sequence composition, and presence of secondary structure elements such as loops or stems. Despite these differences, the overall RMSD values ranged between 0.5 and 1.5 nm, indicating that even the more flexible aptamers reached structural stability throughout the simulation. However, peptide aptamers showed reduced RMSD values below 0.4 nm, with minimal fluctuations over time (Supplementary Fig. S1c), which most likely results from the short length of the molecules. Similarly, target proteins exhibited lower RMSD values than nucleic acid aptamers, indicating reduced conformational flexibility consistent with their more compact folding and extensive intramolecular interactions such as hydrogen bonding and hydrophobic packing. Notably, Tau₄₄₁ displayed a particularly low RMSD, potentially due to its oligomeric nature and intramolecular stabilization, in contrast to the monomeric forms of Aβ₄₀, Aβ₄₂, and β-secretase (Supplementary Fig. S1d). Regarding the pronounced RMSD spike observed in one of the DNA aptamer trajectories during the final nanosecond (Supplementary Fig. S1b), the behavior is likely an artifact caused by periodic boundary conditions (PBC) during the simulation and does not reflect a structural instability of the aptamer itself.

To further evaluate the structural flexibility of the aptamers and targets during molecular dynamics simulations, we compared the initial models to the representative average structures derived from RMSD-based clustering. While most aptamers retained their overall fold, with only minor deviations observed primarily in single-stranded regions, aptamers β19, β55, BI2, and RNV95 displayed localized flexibility in unpaired regions, while their core secondary structures remained intact. For Tau₄₄₁, one cluster conformation exhibited a noticeable shift, likely due to an altered global orientation rather than a true structural rearrangement, and BACE1 showed no significant deviations between the initial and clustered conformations. In contrast, the inherently more flexible targets Aβ₄₀ and Aβ₄₂ exhibited a higher degree of conformational variation (Supplementary Fig. 2), suggesting a more dynamic nature.

Next, the identified average structures were used in an all-to-all molecular docking approach via the HDOCK webserver. Based on docking and confidence scores, the ten highest-ranked binding modes were evaluated, with the top three binding modes for each aptamer–target pair selected for further analysis (Table 1). Given the absence of well-defined or experimentally validated binding sites for Aβ₄₀ and Aβ₄₂, blind docking was employed for all aptamer–target pairs to ensure an unbiased evaluation of potential interaction sites and consistency across docking simulations. The docking results revealed a predominantly targeted binding of BACE1 near its known active site across most aptamer types, although aptamer orientation and contact interface varied among aptamers. In contrast, a less consistent binding was identified for Aβ₄₀ and Aβ₄₂ when docked with nucleic acid aptamers, likely reflecting these peptides’ structural heterogeneity and flexibility. Interestingly, peptide aptamers showed more localized binding preferences. Notably, the preferred binding site appeared to be more strongly influenced by the conformational state of the target protein than by the aptamer structure. This was particularly evident for Aβ species, where variations in protein folding and surface accessibility led to different binding interfaces, highlighting the importance of target flexibility in docking-based binding predictions (Fig. 2).

Table 1.

Molecular Docking of aptamers and target proteins.

Class	Aptamer	Target	Binding mode 1			Binding mode 2			Binding mode 2
			Docking score	Conf. score	MM/PBSA ΔG	Docking score	Conf. score	MM/PBSA ΔG	Docking score	Conf. score	MM/PBSA ΔG
RNA	TH14	BACE1	-451.05	1.00	18.86	-423.95	1.00	15.64	-406.19	0.99	68.69
	S10	BACE1	-417.92	1.00	-31.52	-414.23	1.00	19.42	-404.13	0.99	-4.59
	N2	Aβ40	-391.44	0.99	-5.87	-361.36	0.99	7.95	-359.39	0.99	24.64
	Tau-1	Tau-441	-368.79	0.99	-187.17	-336.60	0.98	-195.52	-334.69	0.98	-200.58
	E2	Aβ40	-338.14	0.98	13.25	-330.62	0.97	14.30	-313.06	0.96	11.42
	β55	Aβ40	-337.38	0.98	-14.64	-316.86	0.97	5.22	-316.86	0.97	-4.20
	β19	Aβ40	-335.94	0.98	9.56	-324.71	0.97	14.33	-323.21	0.97	7.18
	E22-P-AbD43	Aβ42	-326.35	0.97	35.78	-318.04	0.97	2.75	-318.00	0.97	10.82
DNA	IT1	Tau-441	-309.51	0.96	-228.890	-309.37	0.96	-237.5	-305.17	0.96	-214.66
	IT2	Tau-441	-362.35	0.99	-257.32	-334.27	0.98	-305.26	-324.00	0.97	-333.59
	IT3	Tau-441	-367.67	0.99	-253.79	-349.10	0.98	-285.72	-347.22	0.98	-258.88
	IT4	Tau-441	-332.64	0.97	-234.20	-324.71	0.97	-239.66	-323.42	0.97	-299.95
	IT9	Tau-441	-372.53	0.99	-257.83	-323.93	0.97	-235.41	-318.18	0.97	-300.75
	BI1	BACE1	-341.76	0.98	28.31	-337.04	0.98	7.59	-330.05	0.97	-31.38
	BI2	BACE1	-356.65	0.98	65.73	-346.01	0.98	51.70	-335.14	0.98	-16.35
	RNV95	Aβ40	-318.79	0.97	0.70	-313.43	0.96	-28.47	-312.31	0.96	-9.39
	T-SO508	Aβ40	-318.95	0.97	-13.85	-316.58	0.97	-3.69	-316.04	0.97	-10.59
	Aβ-Apt (Aβ7-92-1H1)	Aβ42	-302.18	0.95	-17.93	-282.70	0.93	-9.60	-279.43	0.93	-14.03
Peptide	H102	Aβ42	-207.66	0.76	-23.27	-203.41	0.74	-21.08	-201.86	0.74	-26.99
	iAβ5	Aβ42	-187.97	0.68	-22.24	-185.82	0.67	-36.5	-182.14	0.66	-17.30

Based on the predicted 3d structures (Supplementary Tables 2 and 4), molecular docking was performed using the HDOCK webserver. To characterize the interaction, docking scores and confidence scores (conf. score) for the prediction quality are shown, as well as predicted binding free energies obtained after molecular dynamics simulation and computation of MM/PBSA (MM/PBSA ΔG). Previously selected trajectory clusters were docked in an all-to-all approach, selecting the best three binding modes for each aptamer-target pairing. The best binding mode is highlighted in bold.

Fig. 2.

Top-Ranked Binding Modes of Aptamer–Target Complexes. Molecular docking-derived binding poses of the three best-ranked docking modes for each aptamer–target pair were selected based on docking and confidence scores. Structures represent DNA (yellow), RNA (red), or peptide (blue) bound to the target proteins Aβ₄₀ (PDB-ID 6TI5), Aβ₄₂ (PDB-ID 6SZF), BACE1 (PDB-ID 6EJ3) and Tau₄₄₁ (PDB-ID 5O3L) (grey), illustrating the spatial diversity of predicted binding orientations derived from blind docking using representative conformations obtained from RMSD-based clustering of MD simulations.

Moreover, molecular docking revealed significant differences in the binding between the classes of aptamers. While confidence scores of the best binding modes for nucleic acid aptamers were above 0.96, peptide aptamers showed low confidence scores ranging from 0.66 to 0.76, indicating a less favorable binding to the target proteins. Moreover, peptide aptamers presented a lower docking score of about − 180 to -200 (Table 1). In contrast, docking scores for nucleic acid aptamers ranged between − 300 and − 400, making the binding of nucleic acid aptamers more likely and with increased affinity. When comparing RNA and DNA aptamers, RNA aptamers exhibited a lower docking score, suggesting an even higher binding affinity (Table 1; Fig. 3c).

Fig. 3.

Analysis of aptamer biochemical properties and aptamer-target interaction sites. To compare the different classes of RNA (red), DNA (yellow), and peptide aptamers (blue), box plots of obtained data were generated for (a) literature-derived experimentally determined K_d values, (b) MFold secondary change in free energy (ΔG) of secondary structures, (c) HDOCK docking scores and (d) MM/PBSA binding free energies. (e) To quantify the relationship, Pearson r values of the obtained data are shown as a correlation matrix RNA and DNA aptamers. Significance levels are indicated as ns (p > 0.12), * (p < 0.033), ** (p < 0.002) and *** (p < 0.001). Molecular docking-derived aptamer-target complexes f) were visualized using USCF ChimeraX, showing the contacts with a threshold distance of 3 Å with the complex of BI1 bound to BACE1 (PDB-ID 6EJ3) as representative. Amino acid frequencies analyzed target protein amino acids involved in the interaction for (h) RNA, (j) DNA, or l) peptide aptamers and g, i, k) grouped by their biochemical properties with error bars indicating the standard deviation among aptamers in the group.

Next, we performed a molecular dynamics simulation of the aptamer-protein complexes (Supplementary Fig. S3, S4, S5) and calculated the Molecular Mechanics/Poisson − Boltzmann Surface Area (MM/PBSA) ΔG prediction to quantify the aptamer-target binding further. After the molecular dynamics simulation of the aptamer-target complexes, ΔG values were obtained to compare the binding of the different aptamer classes. RNA aptamers showed the highest ΔG values, while no significant difference was observed between the classes of DNA and peptide aptamers (Fig. 3d). It is to be noted that for nucleic acid aptamers, a high variance between single aptamers within one class was noticed in the RMSD plots as well as binding free energies, suggesting highly different interactions between the aptamers.

Correlation analysis of RNA and DNA aptamers identifies predictive values for simulation-aided aptamer selection

To explore potential predictors of aptamer affinity, we performed a Pearson correlation analysis comparing experimental dissociation constants (K_d) with computational parameters, including changes in free energy of secondary structure, docking scores, and MM/PBSA-derived binding free energies, to assess whether bioinformatically derived metrics can aid in selecting high-affinity aptamers and compare the classes of nucleic acid aptamers, with experimental K_d values obtained from the literature used as the reference for the correlation.

For RNA aptamers, no significant correlation with changes in free energy of secondary structures was observed (r = − 0.119, p = 0.638), while DNA aptamers exhibited a statistically significant moderate negative correlation (r = − 0.657, p < 0.01), suggesting a relation of stabilizing secondary structures to higher binding affinities. Notably, docking scores demonstrated statistically significant moderate positive correlations with experimental K_d values for both RNA (r = 0.501, p = 0.031) and DNA (r = 0.607, p = 0.002) aptamers (Fig. 3e). While a moderate positive correlation was observed, their significance and consistency suggest that docking scores may serve as a useful initial filter in aptamer screening workflows. In addition, binding free energies derived from MM/PBSA calculations showed a moderate positive correlation for DNA aptamers (r = 0.616, p = 0.001), while RNA aptamers showed a moderate negative correlation (r = − 0.514, p = 0.028) (Fig. 3e), indicating class-specific trends. Importantly, parameters derived from three-dimensional structural modeling, including molecular docking and MM/PBSA binding free energies, are more likely to reflect the physical and chemical nature of aptamer–target interactions and thus hold greater potential as predictive indicators. Overall, DNA aptamers showed more consistent and stronger correlations, suggesting a more reliable prediction using bioinformatics-based methods and selection of high-affinity aptamers compared to RNA aptamers.

High frequencies of basic and hydrophobic amino acids modulate the aptamer-protein interaction site

To further analyze the interaction of the simulated aptamer-target complexes identified by molecular dynamics and molecular docking simulations, we visualized and quantified the amino acids of the target protein at the interaction interface (Fig. 3f). Using a cutoff of 3 Å, relative amino acid frequencies identified residues important for the aptamer-target binding. Moreover, amino acids were grouped according to their biochemical properties into hydrophobic, polar, acidic, and basic amino acids. While interactions with RNA aptamers most frequently involved hydrophobic amino acids, including tyrosine, valine, and alanine, the basic amino acids histidine and lysine showed the highest interaction frequencies (Fig. 3g and h). This preference for basic amino acids was even more pronounced for DNA aptamers, where they constituted the majority of contacts (Fig. 3i). Given the larger number of hydrophobic residues compared to basic amino acids, the basic amino acids arginine, histidine, and lysine were most frequently interacting with the nucleic acid aptamers (Fig. 3j). Similarly, peptide aptamers followed this trend, with basic amino acids dominating the interactions, though hydrophobic amino acids also made notable contributions (Fig. 3k and l).

Discussion

Alzheimer’s disease is a progressive neurodegenerative disease causing severe complications, including disorientation, the development of depression⁴ a loss of impulse control, difficulty in reading, writing, and speaking³ and difficulties in swallowing⁶, as hallmarks of the pathology in Alzheimer’s disease are based on the misfolding and accumulation of amyloid-β (Aβ) plaques and tau neurofibrillary tangles (NFT), causing cognitive dysfunction⁷ with the amyloid protein (Aβ) and tau protein being the leading players¹³. Using aptamers, these processes can be inhibited with high affinity and specificity^20,21. Given the growing number of reported aptamers, this study aimed to investigate the basis of a high-affinity interaction with the target proteins to enable a more efficient selection in the future, with recent advancements in molecular docking and dynamics simulations offering powerful computational tools to complement and refine traditional experimental aptamer screening methods. We identified distinct differences in binding characteristics by comparing RNA, DNA, and peptide aptamers based on their biochemical and structural properties, facilitating future planning for high-affinity aptamer selection.

By molecular dynamics and molecular docking simulation combined with MM/PBSA-based ΔG predictions, we identified DNA aptamers as the most promising candidates, showing a strong correlation to experimentally determined binding affinities. Moreover, we identified hydrophobic and basic amino acids accounting for the majority of binding contacts, enabling a directed implementation in future aptamer selection and binding site identification of target proteins.

The biochemical properties of RNA and DNA aptamers were analyzed to identify predictive indicators of stability and binding affinity, allowing for a more directed selection of aptamers in the future. RNA aptamers exhibited significantly higher dissociation constants (Fig. 3a), directly impacting binding affinities, as it was previously shown that decreased dissociation rates (K_d) are associated with an increased aptamer affinity²⁸. Moreover, RNA aptamers exhibited the lowest predicted change in free energy (ΔG) values of secondary structures, suggesting greater intrinsic structural stability than native and DNA-converted aptamers (Fig. 3b), with the increased formation of secondary structures observed for RNA aptamers (Supplementary Table S2) likely influencing the molecules’ stability²⁹. However, correlation analysis revealed a negative correlation between secondary structure ΔG values and experimental K_ds for DNA aptamers and no significant correlation for RNA (Fig. 3e), indicating that the stability of secondary structures has little or a negative influence on the aptamers’ binding and no predictive value for selecting high-affinity aptamers.

To further evaluate aptamer-target interactions and account for molecular flexibility, molecular dynamics simulations were performed to identify representative conformations of both aptamers and target proteins. We employed an all-to-all molecular docking approach, allowing the prediction of optimal binding modes across all aptamer-target pairings (Table 1), with scoring of binding modes used to compare the binding of the different classes regarding the docking interaction (docking score) and its likelihood (confidence score). Although we observed a high variance in docking scores, an overall high confidence scores above 0.95 for nucleic acid aptamers suggests a robust prediction and accuracy of identified binding modes, with DNA aptamers showing significantly less favorable docking scores compared to RNA aptamers (Fig. 3c). Moreover, significantly positive correlations to experimentally K_d values were observed for RNA and DNA (Fig. 3e), suggesting that docking scores are a suited predictive value for identification of both RNA and DNA aptamer candidates. However, given previous reports on the limited accuracy of docking scores, we also applied MM/PBSA calculations to estimate binding free energies more reliably³⁰. Therefore, the positive correlation of binding free energies of DNA aptamers makes them more suitable for prediction-aided selection approaches than RNA aptamers. Unlike RNA and DNA aptamers, docking and confidence scores for the peptide aptamers iAβ5 and H102 were significantly lower (Table 1; Fig. 3c), making them less attractive as a high-efficiency treatment. Since the confidence scores were below 0.75, an alternative docking method might be used to validate the findings for peptide-protein docking. Additionally, the limited sample number of peptide aptamers and the variability in nucleic acid aptamer experimental and predicted values highlight the need for additional data to conclude on the relations.

To further characterize the aptamer-target interactions, the amino acid distribution participating in the interaction was investigated by molecular docking predictions. The interactions most frequently involved hydrophobic and basic residues of the target protein. RNA aptamers exhibited a high frequency of hydrophobic amino acids, including valine, phenylalanine, and tyrosine. In addition, basic residues such as arginine, histidine, and lysine were frequently observed within a 3 Å cutoff (Fig. 3g and h), indicating a significant role in the binding process. This trend was also observed in peptide aptamers, showing high frequencies of both phenylalanine and basic amino acids (Fig. 3i and j). The basic amino acids arginine, histidine, and lysine were observed most frequently for DNA aptamers, suggesting a strong electrostatic component compared to RNA aptamers (Fig. 3g and h), and glutamine was also commonly found across all aptamer classes. While minor differences between the classes were observed, in conclusion, hydrophobic and basic amino acids are most likely to mediate aptamer-target binding.

As described before, hydrophobic contacts significantly contribute to the aptamer–protein interaction by forming favorable hydrophobic pockets²². Moreover, the frequent involvement of basic amino acids, including Lys, Arg, and His indicates a stabilization of electrostatic interactions between the positively charged residues and the negatively charged phosphate backbone of nucleic acids³¹. In addition to charge-based attraction, chemical differences between DNA and RNA can be suggested to further modulate the interaction due to facilitated hydrogen bonding and alterations in local geometry by the RNA-specific 2’-hydroxyl group³²potentially influencing the structure’s conformational stability and flexibility. While our analysis primarily reflects backbone-dominated interactions, such features may contribute to context-dependent differences in binding affinity or specificity between RNA and DNA aptamers.

Our findings suggest that RNA and DNA aptamers outperform peptide aptamers regarding stability and binding potential and show an increased potential for treating Alzheimer’s disease. Although RNA aptamers showed enhanced structural stability by an increased formation of secondary structures and lower secondary structure free energies, experimentally determined K_d values were significantly higher than those of DNA aptamers. While significantly lower docking scores were observed for RNA aptamers, predicted binding free energies demonstrate that DNA aptamers are preferred when selecting high-affinity aptamers. Moreover, correlation analysis revealed significant positive relationships between predicted docking scores and MM/PBSA-derived binding free energies to experimentally determine K_d values for DNA aptamers, allowing a more reliable screening by bioinformatical methods and selection of potential aptamer candidates. Analysis of binding amino acids revealed that aptamer-target interactions are based on hydrophobic and basic amino acids. The affinity and effectiveness might be increased by specific and direct targeting of the most frequent residues or by identifying the optimal binding sites of target proteins.

To further validate our findings and assess if the simulation timescale is sufficient, we extended the molecular dynamics simulations to 100 ns for two representative aptamer-target complexes. A comprehensive analysis of system convergence was performed by comparing the results from the initial 15 ns simulations with the extended 100 ns trajectories. Structural metrics, including RMSD, solvent accessible surface area (SASA), radius of gyration, and the number of hydrogen bonds revealed no significant differences between the two simulation lengths, indicating that our systems reached a stable equilibrium early in the simulation (Supplementary Fig. S6). Importantly, MM/PBSA ΔG calculations also remained consistent, suggesting that the shorter simulation time was sufficient for a robust and reliable quantification of the binding affinities. This validation provides strong evidence that our reported results are methodologically valid and the conclusions drawn are reliable.

In addition, the limited number of experimentally validated aptamers constrains the statistical strength of cross-comparisons, and while computational screening can guide aptamer design, experimental validation remains essential due to the inherent limitations in computational approximations. Therefore, future studies should aim to expand the dataset of aptamers, explore more diverse target proteins, and incorporate extended simulations and ensemble-based docking to further refine predictive models.

Materials and methods

Structure Preparation

Primary nucleic acid or peptide sequences were obtained from the literature. The secondary structure of nucleic acid aptamers was predicted by Mfold³³ (http://www.unafold.org/mfold) with conditions of 1 M NaCl at 25 °C. In addition to the visual output of the secondary structures, the predicted ΔG values were considered for further characterization, with the energetically most favorable structure selected for further procedure. In addition, the sequences of RNA aptamers were converted into DNA by substituting uracil to thymine bases and vice versa to allow a broader comparison of DNA and RNA aptamers. To compute the tertiary structures of the RNA and DNA aptamers, the 3dRNA³⁴ and 3dDNA³⁵ webservers (http://biophy.hust.edu.cn/3dRNA) were used, predicting the secondary structure in the dot-brackets format using the implemented RNAfold algorithm³⁶. Based on the secondary structure, the smallest secondary elements (SSEs) approach was used to build the tertiary structure. For the peptide aptamers iAβ5 (LPFFD)³⁷ and H102 (HKQLPFFEED)³⁸the PEP-FOLD4 webserver (https://bioserv.rpbs.univ-paris-diderot.fr/services/PEP-FOLD4) was used³⁹selecting the most stable structure based on the web server’s ranking. The target protein structures for Aβ₄₀ (PDB-ID 6TI5), Aβ₄₂ (PDB-ID 6SZF), β-Secretase-1 (PDB-ID 6EJ3) and tau filaments (PDB-ID 5O3L) were obtained from the RCSB Protein Data Bank (https://www.rcsb.org/) database.

Molecular dynamics and molecular Docking simulations

Since biomolecules are flexible structures molecular dynamics simulation was performed to obtain relaxed and stable conformations of the aptamers and target proteins. The structure files were prepared using CHARMM-GUI^40–43 for solvation and ionization to 0.15 M NaCl using the TIP3P water model⁴⁴. During the preparation step, missing residues were modeled, and the bound inhibitor was removed for BACE1 (PDB ID 6EJ3). All molecular dynamics simulations were performed with GROMACS (Version 2024.2)^45–47 and Amber ff19SB⁴⁸ OL15⁴⁹ and OL3⁵⁰ force fields. After energy minimization of 5000 steps steepest descent, 125 ps of equilibration at 298.15 K with Nose-Hoover thermostat were performed, followed by production with the addition of Parrinello-Rahman barostat at 1 bar for a 10 ns simulation time. All bonded hydrogens were constrained using the LINCS algorithm⁵¹.

After RMSD-based clustering of the MD-derived trajectories using a cutoff of 0.1 to 0.35 nm, the average structures of the three most populated clusters were selected as representative conformations. These structures were used in an all-to-all molecular docking using the HDOCK webserver (http://hdock.phys.hust.edu.cn)⁵². With limited information on binding sites for some target proteins, molecular docking was performed as blind docking for all targets to ensure a consistent and unbiased computational pipeline across all aptamer–target pairs, allowing a comprehensive exploration of potential binding modes. Up to 4601 binding modes were generated for each docking run, with the 10 top-ranked binding modes further examined. Based on docking and confidence scores, the three best binding poses for each aptamer–protein model were selected for subsequent analysis of interacting amino acids and MM/PBSA ΔG calculations.

Moreover, a threshold distance of 3 Å was chosen to identify interacting amino acids, with frequencies of single amino acids calculated in reference to the total number of contacts. For further analysis, amino acids were grouped based on their biochemical properties to hydrophobic (Ala, Ile, Leu, Met, Val, Phe, Trp, Tyr, Pro, Gly), polar (Asn, Cys, Gln, Ser, Thr), acidic (Asp, Glu) or basic (Arg, His, Lys) amino acids. For visual inspection, UCSF ChimeraX⁵³ was used to visualize aptamer and complex structures or interactions with a center-to-center distance of 3 Å.

Calculation of change in free energy by MM/PBSA

For the calculation of changes in free energies upon aptamer-target binding, predicted binding modes were prepared using CHARMM-GUI. After minimization and equilibration, a production run of 10 to 15 ns was performed for each complex to reach an RMSD equilibrium, indicating structural stabilization. While simulations of 100 ns or longer are standard for calculating absolute changes in free energies⁵⁴, in the context of comparative screening of an extensive candidate library, we prioritized shorter, independent simulations to achieve robust statistical sampling. Using gmx_MMPBSA (Version 1.6.3)⁵⁵ changes in free energy were calculated using the last 25% o each trajectory, with simulations and calculations performed in technical replicates (n = 2).

Data analysis

Statistical analysis was performed using GraphPad Prism (Version 9.5.1). Statistical differences between groups were calculated using a two-sided Mann-Whitney test with error bars indicating the standard deviation. Differences were defined as significant with p < 0.05. A correlation matrix was generated using the two-tailed Pearson correlation estimator with p < 0.05 defined as significant to determine the correlation of different values within a group of aptamers, with the size n = 18 for RNA and n = 24 for DNA aptamers.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Author contributions

B.J.L. and N.K.P. designed research; B.J.L. and A.J.G. performed research; B.J.L analyzed data and B.J.L., U.F.H and N.K.P. wrote the paper. N.K.P supervised research.

Funding

Open access funding provided by Manipal University Jaipur.

Data availability

The data are available from the corresponding author upon reasonable request.

Declarations

Competing interests

The authors declare no competing interests.