Computational modeling of AGO‐mediated molecular inhibition of ARF6 by miR‐145

Inhibition of ADP‐ribosylation factor 6 messenger RNA (ARF6 mRNA) by microRNA‐145 (miR‐145), mediated by Argonaute (AGO) protein, has been found to play essential roles in several types of cancer and cellular processes. This study aimed to model the molecular interaction between miR‐145 and ARF6 mRNA with AGO protein. The sequences of miR‐145 and the 3’ untranslated region (UTR) of ARF6 mRNA were retrieved from miRTarBase, followed by miRNA target‐site and structure predictions were done using RNAhybrid, RNAfold, and simRNAweb, respectively. The interaction between the miRNA‐mRNA duplex and AGOwas further assessed via molecular docking, interaction analysis, and dynamics, using PatchDock Server, PLIP, and VMD/NAMD, respectively. The models between miR‐145, predicted target site of ARF6 mRNA, and AGO protein returned stable thermodynamic variables with negative free energy. Specifically, the RNA duplex had an energy of ‐19.80 kcal/mol, while the docking had ‐84.58 atomic contact energy supported by 70 hydrogen bonds and 14 hydrophobic interactions. However, the stability of the RMSD plot was still unclear due to limited computational resources. Nevertheless, these results computationally confirm favorable interaction of the three molecules, which can be utilized for further transcriptomics‐based drugs or treatments.


Introduction
ADP ribosylation factor 6 (ARF6) is a small GTPase that plays a role in diverse cellular processes, including cell adhesion and migration . Previous stud ies found that ARF6 acts as an oncogene, promoting tu mor cell invasion in several cancer types (Hashimoto et al. 2004; Eades et al. 2015; Xu et al. 2019). According to Sabe (2003), the expression of ARF6 would inactivate the activity of Ecadherin, thus reducing the cell junctions. It is coupled with an increased Ncadherin expression, which allows the cells to attach to collagen, a component of ex tracellular matrix (Janiszewska et al. 2020). It also plays a role in the fibroblast growth factor receptor (FGFR) and Wnt signaling pathways (Mrozik et al. 2018). These pro cesses allow the cells to move to the extracellular matrix, enter the lymphatic/blood systems, and extravasate to form tumors (Oh et al. 2012). ARF6 expression also corre lates with other processes, such as macrophagemediated inflammation . The activity of ARF6 is posttranscriptionally regulated by tumor suppressor miR 145 (Zeinali et al. 2019) via the binding to the 3' UTR of the ARF6 mRNA (Pashaei et al. 2016). However, long noncoding RNA regulator of reprogramming (lincRNA RoR) acts as a natural competitor or sponge for miR145, inhibiting the miRNA activity (Eades et al. 2015). This would lead to the overexpression of ARF6 and eventually the formation of tumors.
RNA silencing process requires the presence of the RNAInduced Silencing Complex (RISC), a ribonucleo protein complex that utilizes small RNA as a template to recognize the complementary sequence of the target mRNA (Zhang 2013). One major constructor of RISC is Argonaute (AGO) protein (Zhang et al. 2018), which acts as an essential effector in posttranscriptional gene silenc ing (Li et al. 2014). Structurally, AGO protein consists of Nterminal, PAZ, MID, and PIWI domains, which are or ganized in a bilobal conformation (Djuranovic et al. 2011). Specifically, the PAZ domain would bind to the 3' end of the miRNA, while the 5' end would anchor to the MID do main (Li et al. 2014). The binding of miRNA to the AGO protein would 'guide' the AGOcentered RISC to bind with the complementary mRNA at 3' UTR region, lead ing to mRNA cleavage (Zhang 2013). As AGO protein plays an important role in RNA silencing, it is crucial to understand the molecular interaction between these three molecules. This can be modeled via molecular docking simulation, which will be followed by molecular dynam ics simulation to assess the stability of the molecules. By mimicking the condition of cytoplasm where the miRNA mediated gene silencing process mainly occurs , we can computationally assess the behavior of the molecules in the real environment. In this study, the molecular simulations were done between AGO protein and miR145ARF6 mRNA duplex, aiming to model the interaction based on in silico point of view for further de velopment of drug and treatments.

Materials and Methods
In this study, the pipeline was constructed based on other similar studies (e.g. Das et al. (2015) and Rath et al. (2016)). Several adjustments were made to match the re quirements of the software with the available computer re sources. The complete pipeline is shown in Figure 1.

Sequence retrieval and structure prediction
The sequences of miR145 and 3' UTR region of ARF6 mRNA were retrieved from miRTarBase under the 'miRNA' and 'Target Gene' tabs, respectively (acces sion ID: MIRT278608) (Chou et al. 2018). After that, the miRNA target site was predicted by using RNAhy brid (Rehmsmeier et al. 2004). The secondary and ter tiary structures of miR145, its predicted target site, and the miRNAmRNA duplex were then visualized by using RNAfold (Lorenz et al. 2011) and simRNAweb (Boniecki et al. 2015; Magnus et al. 2016, respectively.

Molecular docking and dynamics
The miRNA molecule was blindly docked with human AGO2 protein (PDB ID: 4F3T) to locate the binding site by using PatchDock Server (Duhovny et al. 2002; SchneidmanDuhovny et al. 2005, followed by the dock ing between the miRNAmRNA duplex and the protein. PatchDock Server utilized rigid docking based on the ge ometries of the molecules (SchneidmanDuhovny et al.

2005
). As a comparison, the protein was also docked with its native ligand (miR20a) by using the same software. The molecular interaction between the proteinmiRNA and proteinduplex was then assessed by using Protein Ligand Interaction Profiler (PLIP) (Salentin et al. 2015).
Lastly, molecular dynamics of the complex were sim ulated by using VMD/NAMD pipeline under NVT con dition (i.e. constant particle number, volume, and tem perature) by following the NAMD Tutorial file (http://ww w.ks.uiuc.edu/Research/namd/) (Humphrey et al. 1996; Phillips et al. 2005. The complex was first solved into a water box, followed by energy minimization for 1,000 steps at 1 atm (pressure) and 310 K (temperature). The resulting coordinate files alongside the combination of protein, nucleotide, carbohydrate, lipid, water, and CHARMM general force fields were then used for the analysis with 1,650,000 steps. The CHARMM topology and parameter files were taken from MacKerell Jr (2001).
All analysis was done under default parameters of each software (Zhang and Verbeek 2010; Ahirwar et al. 2016; Lorenz et al. 2016; Magnus et al. 2016 in Windows com puter with Intel® Core™ i78750H CPU @2.2GHz and 8GB RAM. The tertiary structures of the simulation were then visualized in the VMD tool and PyMOL (The PyMOL Molecular Graphics System, Version 2.3.0, Schrodinger, LLC), respectively. The 2D/3D structure prediction steps took around 14 days (depends on the online queue), while the molecular docking and dynamics required one and seven days, respectively.

Sequence retrieval and structure prediction
After the sequences of miR145 and 3' UTR of ARF6 was retrieved from the miRTarBase, the miRNA target site was predicted using RNAhybrid. The predicted site turned to match the prediction by miRanda (Betel et al. 2008) that is stored in the miRTarBase entry (accession ID: MIRT278608); there is binding between ARF6 mRNA and miR145 at the 956th position of the UTR region, denoted by 20 pairing nucleotides (Figure 2, yellow highlighted areas). Nucleotide bindings are characterized by hydrogen bonds between AU (two bonds) and CG (three bonds) of the interacting nucleobases. Besides, the minimum free energy (MFE) of the binding is 28.2 kcal/mol, showing favorable interaction. However, the p   (Table 2) were inputted into simRNAweb to pre dict the tertiary structure of the RNA molecules. The 2D and 3D structures of the RNAs were shown in Figure 3.

Molecular docking and dynamics
We did two docking simulations: miR145 & AGO protein and miRNAmRNA duplex & AGO by using PatchDock Server. The bestscored model with negative thermody namic value (Figure 4) was retrieved, with the statistical result shown in Table 3. Also, the docking result of AGO protein with its native ligand is showed in Table 3. The docking between miRNA and AGO protein shows a fa vorable interaction, denoted by the negative value of ACE (532.40 kcal/mol) and supported by 18 intermolecular hydrogen bonds (Table 4a). This also applies to the dock ing between miRNAmRNA duplex and AGO protein ( 84.58 kcal/mol) with 70 intermolecular hydrogen bonds and 14 intermolecular hydrophobic interactions (Table 4b   ( (......))))). -1.20 MiRNA-mRNA duplex . ((.((.(((((((.((((((((....))))))))))))))).)).)) -19.80   Table 5). Further molecular dynamics of the molecule is shown in Figure 5, as well as the free energy and RMSD plot of the protein backbone in Figure 6. As shown in Figure 6a, the dynamics of conforma tional energy (i.e. bond), nonbond energy (i.e. vdW, electrostatic), and other energies (i.e. kinetic, total, tem perature, pressure) of the molecule are stable after a short initial fluctuation. However, the RMSD of the molecule might not have reached equilibrium yet (Figure 6b): it still fluctuates in the range of 2.0 2.5Å until the end of the simulation. The previous study showed that wildtype miRNAmRNA heteroduplex and AGO complexes stabi lized at 1.5Å with larger fluctuations when the number of mismatches increased (Xia et al. 2013). As the inter action between miR145, 3' UTR of ARF6 mRNA, and AGO was not included in the study (Xia et al. 2013), the equilibrium point might be higher. To better evaluate the energyminimized state, the simulation should be done in a better computer to accommodate a longer time duration.

Discussion
According to Table 2, MFE of miR145 and ARF6 mRNA duplex is 19.80 kcal/mol, which is generally lower than the previous study that comprised of 31 miRNAmRNA interactions (Rath et al. 2016). Further docking analysis shows that there is favorable interaction between AGO2 protein and miR145. Interestingly, it has a lower ACE (532.40 kcal/mol) than the native ligand of the protein (miR20a, 235.780 kcal/mol). This might due to the structural similarity between miR20a and miR145. On the other hand, miRregulated ARF6 and AGO have an ACE of 84.58 kcal/mol. This value is higher than the na tive ligand of the protein (miR20a, 235.780 kcal/mol) as the protein conformation would best fit its natural ligand, as well as several siRNAs and miRNAs (Kandeel andKi tade 2013; Rath et al. 2016). Nevertheless, there are 70 hydrogen bonds and 14 hydrophobic interactions between the two molecules which stabilize the complex. ARF6 has been found to promote the development of several types of cancer, for instance, breast cancer (Li et al. 2017). Moreover, several inhibitors have been found to suppress the activity of ARF6, resulting in the suppression of cancer invasion and/or metastasis (Li et al. 2017); this includes ARF6 small interfering RNA (siRNA) (Xu et al. 2015) and miR145 (Eades et al. 2015). This idea was fur ther supported by Ye et al. (2018), showing the inhibition of breast cancer development by overexpression of miR 145. Another study confirmed the low level of miR145 expression among breast cancer patients with different on set of age, ranging from very young (<35 years old) until postmenopausal (>50 years old) patients (Tsai et al. 2018). This study elucidates the regulation of ARF6 by miR 145 by the assistance of AGO protein based on in silico ap proach. The molecular docking shows a strong interaction between AGO and miR145. In addition to negative value of MFE, the presence of hydrophobic aliphatic amino acids, such as alanine and leucine, provides structural sta bility between the molecules, while aromatic amino acids, such as histidine and phenylalanine, further support the complex stability. Furthermore, the interaction between AGO and the RNA duplex is also supported by strong hy drophobic amino acids, namely leucine, isoleucine, ala nine, and phenylalanine. Further molecular dynamics af firms this stability, with stable dynamics and constant RMSD energy between 2.0-2.5 Å. However, there were several limitations that we encountered due to the unavail ability of (i) reliable 3D structures of RNAs, (ii) RNA specific molecular docking and dynamics software, and (iii) highperformance computer (HPC). As a result, the procedure was mostly done by using opensource online software, while the rest was simulated with limited com putational power. Nevertheless, these results affirm the feasibility of molecular inhibition of ARF6 by miR145 with the assistance of AGO protein. In this end, it is also emphasized that the high power GPUbased workstation with high specification of RAM and HDD is sufficient to conduct the whole computational process. This condition is mainly assisted by the availability of the Highresolution GPU in our workstation.

Conclusions
This study shows that there is a strong, favorable interac tion between miR145, 3' UTR of ARF6 mRNA, and AGO protein computationally, affirming results from the previ ous studies. Future studies should incorporate the inter action between ARF6 mRNA and lincRNARoR, also, to complete the RISC molecule, to mimic their interactions in the real environment. The resulting binding affinity and stability of the molecules should be incorporated in further drug development to create a universal drug that mimics the activity of miR145 in controlling ARF6 expression.