Adenosine deaminase acting on RNA (ADAR) is a class of enzymes to process the adenosine-to-inosine (A-to-I) RNA editing through deamination reaction, making it the most prevalent type of post-transcriptional nucleotide modification in humans [1, 2]. Since inosine can form Watson-Crick base pairing with cytidine, the resulted inosine behaves like guanosine and, as a result, the A-to-I editing on RNA significantly modifies RNA function, such as miRNA biosynthesis and target recognition, splicing patterns, gene regulation, and the long-non-coding RNA functions [2–4].
There are three ADARs in humans, namely ADAR1, ADAR2, and ADAR3 [5]. Both ADAR1 and ADAR2 are catalytically active and expressed in almost all tissues, whereas ADAR3 is only expressed in brain and has no detectable catalytic activity. It has been reported that ADAR3 can compete with ADAR2 to inhibit RNA editing by directing its binding to GRIA2 pre-mRNA in glioblastoma [6]. Because ADAR1’s editing functions on dsRNA inhibits innate immunity and the interferon-mediated response, dysregulation of ADAR1 activity causes upregulated interferon and autoinflammatory diseases such as Aicardi-Goutieres syndrome (AGS) and dyschromatosis symmetrica hereditarian (DSH) [7–9]. AGS is a rare childhood autoimmune disorder that primarily affects the brain, immune system, and skin. [10] Currently it has been reported that seven genes, including ADAR1, confirmed to be associated with AGS [11]. The severity of the disease caused by the loss-of-function ADAR mutations may depend on the overall impact of these mutations on A-to-I RNA editing activity [12, 13]. A decrease in editing activity below a certain threshold likely triggers the development of a neurological disorder [14].
Molecular dynamics (MD) simulations have long been a useful tool in providing structural and dynamic details of proteins [15–19]. In this work, we carried out MD simulations to study two AGS-causing mutations in ADAR1 deaminase domain, R892H and Y1112F, identifying their structural deviation from the wild-type (WT). From the N-terminus to the C-terminus, ADAR1 includes two Z-DNA binding domains (ZBDs), three dsRNA binding domains, and deaminase domain. Currently, the solved structure of ADAR1 is limited to the ZBDs [20]. We used homology modeling to build one apo ADAR1 deaminase domain, employing the solved human ADAR2 deaminase domain structure as a template. To enhance our conformational sampling, we also generated another ADAR1 structure using AlphaFold2, a method known for its high precision in protein structure prediction [21]. For each of the two model structures, the R892 residue was replaced by histadine to form two R892H strucures, and the Y1112 residue was replaced by phenealanine to form two Y1112F strucures. All of the six structures underwent MD simulations to refine and collect conformations, ultimately helping us deduce the structural criteria for RNA substrate recognition. Additionally, we retrieved a wild-type conformation from the AlphaFold2 trajectory, superimposed it with the RNA-bound ADAR2 structure, replaced the RNA sequence by HER1 RNA [22], and submitted it for MD simulations.
Sequence alignment of ADAR1 and ADAR2 deaminase domains (Figure S1) indicates 39% sequence identity and 59% sequence similarity, suggesting that the available ADAR2 structural data provides useful guidance in envisioning how ADAR1 associates with a dsRNA substrate. The ADAR2 deaminase domain is composed of 20 loops, four of which come into contact with the dsRNA. As indicated in Fig. 1a and Figure S1, these loops are referred to as the base-flipping loop (E485’-G489’), the 5'-binding loop (A454’-R477’), the 3'-binding loopI (A347’-L352’), and the 3’-binding loopII (L584’-N597’) [23]. For ease of reading, ARAD2 residues are marked with primes to distinguish them from the unprimed ADAR1 residues. As shown in Fig. 1b, ADAR2 catalytic site is enclosed by E396’, E488’, and K594. The E396’ (on the N-terminal end of α2 helix in the core structure) secures the flipping-out adenine (on the substrate RNA strand) base while the E488’ (on the base-flipping loop) inserts into the minor groove, filling the space left by the flipping-out adenine, and contacts the unpaired, orphaned base located on the complementary RNA strand. Taking the substrate RNA strand as a viewpoint, from top to bottom in Fig. 1a, the 3’ end and 5’ end regions relative to the flipping-out base, which is to be edited, are in contact with the 3’-binding loops and 5’-binding loop, respectively. Because of its high flexibility, the 24-base-long 5’-binding loop structure was not resolved in the apo ADAR2 structure (PDB: 1ZY7) but was resolved in the RNA-ADAR2 complex (PDB: 5HP2) [24, 25]. The R348’ (in the 3’-binding loopI) and K594’ (in the 3’-binding loopII) both stabilize the dsRNA. Except R348’ not conserved by ADAR1 G865, the catalytic site’s E396’, E488’, and K594’ are conserved by ADAR1’s E912, E1008, and K1120. The E912-E1008 and E1008-K1120 separation distances were adopted to categorize the collected WT, R892H, and Y1112F conformations. We hypothesized that if the E1008-K1120 distance exceeds the tolerance level, it would discourage the dsRNA recognition process, ultimately reducing the A-to-I editing activity.
<Insert Fig. 1>
Numerous AGS-causing mutations are on the ADAR1 deaminase domain, including A870T (A353’ in ADAR2), I872T (V355’), R892H (K376’), K999N (Q479’), G1007R (G487’), Y1112F (A587’), and D1113H (E588’), as summarized in Figure S2. By analyzing these corresponding residues on the ADAR2 deaminase domain structure, we foresee the possible consequence of ADAR1’s mutations. For example, A353’ and V355’ are in vicinity of E396’ in ADAR2 and accordingly either A870T or I872T mutation impairs the interaction between E912 and the flipping base in ADAR1. Because G487’ is next to E488’ and K376’ holds salt bridges to secure the 3’-binding loopII, either G1007R or R896H mutation surely deforms the active site. Q479’ is in the C-terminus of the 5’-binding loop and thus the correlated K999N mutation may interfere the induced-fit process when associating the dsRNA substrate. A587’, E588’, and K594’ are located in the 3’-binding loopII; thus Y1112F and D1113H mutations may pull K1120 away from the active site. Since ADAR1’s Y1112 and ADAR2’s A587’ possess dissimilar sizes and hydrophilicities, it is necessary to learn the role played by Y1112 before investigating Y1112F mutation. Our objective is to explore the susceptibility of ADAR1's active site. To achieve this, we selected an orthosteric site R892H mutation and an allosteric site Y1112F mutation. Our conformational analysis suggests that in R892H variant, the substituted H892 is not able to maintain the R892-E1123 ionic linkage. On the other hand, the hydrophobic F1112 in Y1112F mutant reduces its solvent-accessible surface area (SASA) and causes a contraction movement of the 3’-binding loopII toward the α1 helix. Both mutations rearrange K1120 position and interfere the substrate RNA recognition.