Assessment of Mechanisms of HDO-AntimiR by Molecular Dynamics Simulation

Recently, we found DNA/RNA heteroduplex oligonucleotide-based antimiR (HDO-antimiR) can more eciently inhibit the target miRNA than conventional antimiR after its cellular uptake. But the mechanism of HDO-antimiR about the target-silencing is unknown. We here tried to elucidate the interaction mechanism of HDO-antimiR to miRNA using molecular dynamics (MD) simulation. When interaction of the conventional antimiR or HDO-antimiR and the target miRNA was simulated, they combined with each other in various forms. In the hydrogen bond analyses, base site of the antimiR formed hydrogen bond with miRNA. On the other hand, phosphate site of the HDO-antimiR formed hydrogen bond with miRNA. These results suggested that there were differences about the binding mechanisms between antimiR and HDO-antimiR to the target miRNA. In particular, there was a difference in the binding site between antimiR and HDO-antimiR. Additionally, it was found that guanine in the miRNA is mainly involved in the binding to the antimiR or HDO-antimiR. MD simulation method is useful in understanding the mechanism of oligonucleotide therapeutics. direction. The results are presented as mean ± SEM (n = 3). N.D., not detected. d) HDO-antimiR122; -5 Å shifted from miR122 in the sagittal direction. The results are presented as mean ± SEM (n = 3). e) antimiR122; 20 Å shifted from miR122 in the horizontal section, Two-point interaction pattern. The results are presented as mean ± SEM (n = 3). N.D., not detected. f) HDO-antimiR122; 20 Å shifted from miR122 in the horizontal section, Multiple-point interaction pattern. The results are presented as mean ± SEM (n = 5). g) antimiR122; -5 Å shifted from miR122 in the sagittal direction. The results are presented as mean ± SEM (n = 3). h) HDO-antimiR122; -5 Å shifted from miR122 in the sagittal direction. The results are presented as mean ± SEM (n = 3).


Introduction
MicroRNA (miRNA), which is one of the short endogenous RNAs, regulates various biological phenomena by suppressing more than 100 types of messenger RNA (mRNA) 1 . In recent years, it has been reported that abnormalities in miRNA expression and function are associated with various diseases such as cancer and intractable neurological diseases 2,3 . Therefore, miRNA as a biomarker for diagnosing various diseases centering on cancer has already been clinically applied, and miRNA itself has been given particular importance as a therapeutic target molecule. One of the most promising methods for controlling miRNA is the use of antimiR, a nucleic acid drug that speci cally binds to miRNA via Watson-Crick base pairing by hydrogen bond and suppresses its function 4 . AntimiR consists of a single-stranded oligonucleotide which is chemically modi ed to improve its stability and a nity to its target miRNA.
However, the silencing e cacy in vivo of antimiR is inadequate for clinical application and molecular design of antimiR for e cient silencing in vivo remains limited.
We have developed a novel molecular design of nucleic acid drugs, heteroduplex oligonucleotide (HDO) consisting of antisense oligonucleotide (ASO) targeting mRNA and its complementary RNA (cRNA) 5 , and HDO greatly improves the silencing activity of conventional single-stranded ASO at the hepatocyte 6 , brain microvascular endothelial cells 7 and brain 8, 9 in vivo. Next, we designed double-stranded HDO-antimiR which comprises an antimiR strand and its cRNA-strand 10,11 . HDO-antimiR also showed the improvement of its target miRNA control ability and the avoidance of toxicity. Interestingly, while the conventional single-stranded antimiR binds directly to the target miRNA and suppresses its function on a one-to-one basis, HDO-antimiR not only binds to but also degrades the therapeutic target miRNA. However, the mechanism by which HDO-antimiR affects target miRNAs has not been elucidated. Also, traditional single-stranded antimiR is thought to form Watson-Crick base pairs with the target miRNA, whereas HDO-antimiR already forms such base pairs with the complementary RNA strand. Therefore, it is unclear in Loading [MathJax]/jax/output/CommonHTML/fonts/TeX/fontdata.js detail how antimiR binds the target miRNA since it is not straightforward to observe such a dynamic process directly from experimental approach. To this end, molecular dynamics (MD) simulation is a powerful tool, which can handle the motion of all atoms at nite temperature directly 12 . Especially, it is an advantage of the MD simulation over static energy calculations that the steric effect can be treated inherently in the simulation since it is one of key factors to determine the interaction between nucleic acids 13 including miRNA and antimiR.
In this study, we attempted to elucidate the binding mechanism of antimiR by introducing a molecular dynamics simulation method together with the target miRNA and analyzing its operation for antimiR and HDO-antimiR.

Methods
Building and generation of models AntimiR and HDO-antimiR containing chemically modi ed nucleotides were designed to target miRNA (miR122 or miR21) based on a previous report 10 and built by using Nucleic Acid Builder (NAB) tool of AMBER 14,15 . The force eld (FF) parameters for the two modi ed nucleic acids, locked nucleic acid (LNA) and phosphorothioate (PS) bond, were available on a previous report 16 . Although each PS linkage is chiral and the chirality is known to in uence the potency and safety of ASO 17,18 , we used only Sp-PS linkage in this study. The index for each atom assigned by AMBER is shown in Supplementary Figure 7. The individual modi cations and sequences of the nucleic acids are shown in Table 1.
The frcmod les and lib les have been generated and have been used to build the antimiR/HDO-antimiR and target miRNA sequences for running MD simulations. Distance and angle between antimiR/HDO-antimiR and target miRNA were set using Python program.

Molecular dynamics protocol
The MD simulations were carried out using AMBERTools20 program. All the systems were neutralized by adding K + counter ions and solvated with TIP3P water box. All the systems were minimized by using the steepest descent method for 2000 steps. Then the systems were heated from 0 to 310 K in 20 psec. Initial structure of miRNA and antimiR or HDO-antimiR are shown in Figure 1 and Supplementary Figures 1 and   2. Followed by these steps, MD simulations were carried out for 5 nsec. MD integration was carried out using a 1 fsec time step. The pairlist was updated at every 1000 steps. Constant pressure (1 atm) and temperature (310 K) were maintained throughout the production simulation run.
All the data was collected on 1-picosecond snapshots of simulation data. The trajectories and structures were visualized using VMD. The PTRAJ module of AmberTools20 was used to measure the distance between geometric center of the sequence of miRNA and antimiR. The number of hydrogen bonds was calculated using VMD program. A distance shorter than 3 Å indicates that a hydrogen bond is formed Loading [MathJax]/jax/output/CommonHTML/fonts/TeX/fontdata.js between the atoms of the antimiR or HDO-antimiR and miRNA are considered to be paired. All data represent mean ± s.e.m.

Results
Binding of antimiR or HDO-antimiR and miRNA from 20 Å away At rst, the distance between the antimiR or HDO-antimiR and the miRNA was xed at 20 Å in initial state, and the angles were set in 8 directions at horizontal section of helix at 45° interval to consider the effect of initial con guration on the interaction (Figure 1, Supplementary Figures 1 and 2). Molecular simulations were performed 8 times (miR122) or 4 times (miR21) at each angle. We categorized the patterns of interaction between antimiR and miRNA as follows; Separated (no interaction), One-point interaction, Two-point interaction, Multiple-point interaction (more than two-point interaction).
As a result of simulation, all the four patterns were observed in both antimiR and HDO-antimiR ( Figure 2). Then we investigated how the bonds occurred in patterns of Two-point or Multiple-point interaction. When a nucleic acid binds to a target one, hydrogen bonds are the main component, therefore we evaluated hydrogen bonds between the antimiR or HDO-antimiR and miRNA ( Figure 3). From these results, there was no difference in the frequency of binding between antimiR and HDO-antimiR. So, we analyzed the number and location of hydrogen bonds formed between antimiR and HDO-antimiR about Two-point or more interaction pattern to investigate the mechanism of the difference in effect between antimiR and HDO-antimiR.
In the Two-point interaction pattern about antimiR with miRNA, hydrogen bonds were formed mainly through the atoms in the base site of the antimiR and miRNA near both ends, and no hydrogen bonds were observed in the phosphate site and the sugar site of antimiR. Similar to antimiR, hydrogen bond was mainly observed at the base site in the miRNA ( Figure 4). It is unlikely that miRNA will t in the minor groove of HDO-antimiR, which is a narrow and shallow groove, so we considered the pattern that ts in the deep major groove. In the case of antimiR and miRNA, most We further analyzed these results about the Two-point interaction pattern and Multiple-point interaction pattern when antimiR or HDO-antimiR is 5 Å shifted from miRNA in the vertical axis direction (both miR122 and miR21). At rst, to nd out which sites contribute to nucleic acid binding, we calculated the number of hydrogen bonds con rmed in this simulation for each binding site in the nucleic acid ( Figure  7a-d and Supplementary Figure 5). In these results, there are several types of interaction patterns between HDO-antimiR or antimiR and miRNA. The atoms in the base part mainly form hydrogen bonds in antimiR, whereas in HDO-antimiR, the atoms in the phosphate bond site are mainly involved. On the miRNA side, the atoms at the base were mainly bound either to antimiR or HDO-antimiR. It was also found that the appearance rate of the Multiple-point interaction pattern, which seems to be the strongest binding form, differs depending on the initial conditions between HDO-antimiR and antimiR and miRNA. In the time course about Multiple-point interaction pattern of HDO-antimiR and miRNA showed hydrogen bonds were not xed, and the atom that once formed a hydrogen bond with another atom, then HDO-antimiR bound to the target miRNA gradually. By repeating these movements, it was expected that the miRNA would entwine in the major groove of HDO-antimiR. On the other hand, when HDO-antimiR and miRNA were placed 5 Å away to the vertical axis direction, miRNA was placed in the major groove of HDO-antimiR in the initial state. In the result of this simulation, HDO-antimiR and miRNA formed hydrogen bonds via the atoms at the base of HDO-antimiR, not the phosphate site of the nucleic acid.
Next, in order to investigate which bases affect the binding, we calculated the number of hydrogen bonds for each base in the nucleic acid, then it was corrected by dividing by the number of each base in the sequence in order to eliminate the variation in the number of bases since the number of bases differs depending on the sequence (Figure 7e-h and Supplementary Figure 6). In the results about shifted 5 Å in the vertical axis direction, antimiR showed the number of hydrogen bonds between guanine and cytosine is larger than them between thymine and adenine. These results partly can re ect the fact that the number of hydrogen bonds between guanine and cytosine is 3 and the number of hydrogen bonds between thymine and adenine is 2 when antimiR and miRNA form a complete double strand. On the other hand, there was no difference in the number of hydrogen bonds depending on the base in HDO-antimiR. Binding site analyses showed HDO-antimiR and miRNA formed hydrogen bonds via the atoms at the base of HDO-antimiR, not the phosphate site of the HDO-antimiR when HDO-antimiR and miRNA were placed 5 Å away to the vertical axis direction, and these data suggested HDO-antimiR is bound to miRNA via a base, but it is bound by a mechanism different from that of antimiR and miRNA. These results indicated it is not affected by the base sequence in HDO-antimiR because not base site, but phosphate site affects the bond in HDO-antimiR.

Discussion
In MD simulation, all atoms that make up a nucleic acid molecule can be displayed in coordinates in XYZ space. MD simulation of HDO-antimiR or antimiR and the target miRNA at a distance of 20 Å separation on the axial section of helix con rmed that HDO-antimiR and antimiR bind to the target miRNA actually. It was also found that hydrogen bonds are important for binding, and computational studies have been useful in predicting and understanding molecular phenomena with H-bond interactions 19 .
Our results show the simulations of 20 Å separation on the axial section of helix, representing the very early changes leading to the coupling, and the simulations of 5 Å separation to the vertical axis direction, representing the situation after the coupling has progressed. Nucleobases formed hydrogen bonds in the simulations about antimiR and miRNA both 20 Å separation and 5 Å separation. On the other hand, in the simulation about HDO-antimiR and miRNA, phosphate groups were mainly involved in the bond at 20 Å separation, while atoms at the base site were mainly involved in the bond at 5 Å separation. The result that the phosphate site of HDO-antimiR formed a hydrogen bond in 20 Å separation suggests that it binds in a sequence-independent manner. Since the base part formed a hydrogen bond in 5 Å separation, it was suggested that the bond gradually changed to a sequence-dependent bond. This is a different Loading [MathJax]/jax/output/CommonHTML/fonts/TeX/fontdata.js result from the simulation in the single-strand antimiR, and it is considered that it is a phenomenon peculiar to HDO-antimiR having a double-strand structure. In the sequence of miRNA, guanines had the largest effect on hydrogen bonds between antimiR or HDO-antimiR and miRNA. Compared to miR122, miR21 has a lower number of guanines in the sequence and it may be associated with a lower number of the bindings more than two-point interaction between antimiR and HDO-antimiR. The difference in the number of potential hydrogen bond acceptor sites and donor sites between guanine and other bases may have an effect.
There have been few reports about simulations of nucleic acid and target genes using molecular dynamics. Shen et al. 20 described the potential of estimating melting temperature for oligonucleotides including short interference RNA (siRNA) and ASO by using the standardized molecular dynamics protocol. In another report, Uppuladinne et al. 16 also described structural properties of ASO using molecular dynamics simulations in the duplex of ASO and its complementary RNA, and they evaluated binding energy of the duplex. But they mentioned about the in uence of the several chemically modi cations about ASO, and had only performed the simulation with ASO and its targeted RNA from the binding status. Therefore, unlike our reports, they did not evaluate how ASO binds to a target gene from a distant. Galindo-Murillo et al. 21 reported antisense inhibition by acyclic analogs of nucleic acids using molecular dynamics simulations, and they estimated the potentially identify target ASO analogs by rootmean-square (RMS) values. As just described, there are increasing reports on the function of antisense oligonucleotide using molecular dynamics because it is useful for functional evaluation of antisense nucleic acids.
Several proteins including plasma protein and intracellular protein affect pharmacokinetics of the phosphorothioate modi ed ASO 22 , especially in the RNase H1-dependent ASO 23 . Recently we reported the identi cation of RNase H-dependent tocopherol-conjugated HDO interacting proteins 24 . Probably HDO-antimiR also have the interacting protein to affect to target miRNA, our results suggested that HDO-antimiR alone can bind to the target gene.
Our simulations were only in very short time, which is insu cient as a simulation of therapeutics oligonucleotide showing e cacy every few days, but new ndings on how the nucleic acid binds to a target gene. In particular, it was suggested that the initial interaction with the target miRNA differs between the antimiR and the HDO-antimiR. In the future, if it becomes possible to simulate multiple and long-term oligonucleotide therapeutics with various proteins by using higher-performance computers including supercomputers, it is expected that more precise functions of therapeutic oligonucleotides will be veri ed.