A B–Z junction induced by an A … A mismatch in GAC repeats in the gene for cartilage oligomeric matrix protein promotes binding with the hZαADAR1 protein

GAC repeat expansion from five to seven in the exonic region of the gene for cartilage oligomeric matrix protein (COMP) leads to pseudoachondroplasia, a skeletal abnormality. However, the molecular mechanism by which GAC expansions in the COMP gene lead to skeletal dysplasias is poorly understood. Here we used molecular dynamics simulations, which indicate that an A … A mismatch in a d(GAC)6·d(GAC)6 duplex induces negative supercoiling, leading to a local B-to-Z DNA transition. This transition facilitates the binding of d(GAC)7·d(GAC)7 with the Zα-binding domain of human adenosine deaminase acting on RNA 1 (ADAR1, hZαADAR1), as confirmed by CD, NMR, and microscale thermophoresis studies. The CD results indicated that hZαADAR1 recognizes the zigzag backbone of d(GAC)7·d(GAC)7 at the B–Z junction and subsequently converts it into Z-DNA via the so-called passive mechanism. Molecular dynamics simulations carried out for the modeled hZαADAR1–d(GAC)6.d(GAC)6 complex confirmed the retention of previously reported important interactions between the two molecules. These findings suggest that hZαADAR1 binding with the GAC hairpin stem in COMP can lead to a non-genetic, RNA editing–mediated substitution in COMP that may then play a crucial role in the development of pseudoachondroplasia.

hZ␣ ADAR1 (PDB code 2ACJ), NMR chemical shift mapping of hZ␣ ADAR1 , and d(GAC) 7 ⅐d(GAC) 7 titration, a complex model is proposed here. Subsequent MD simulation of the complex model confirms the importance of certain amino acids in hZ␣ ADAR1 recognizing the B-Z/Z-DNA conformation. Based on hZ␣ ADAR1 and d(GAC)⅐d(GAC) binding studies, a model of how hZ␣ ADAR1 can anchor to the Z-philic GAC repeat, facilitate A-to-I editing of the corresponding mRNA transcript of COMP, and lead to pseudoachondroplasia is also proposed.

An A . . . A mismatch amid G . . . C and C . . . G base pairs imposes B-Z junction formation
For MD simulation, the d(GAC) 6 ⅐d(GAC) 6 repeat sequence has been considered to complete one helical turn of a DNA duplex (i.e. 10 bases per turn in a normal B-DNA). Flanking sequences (one GAC repeat) on either sides are added to avoid the end-fraying effect during MD simulation. The effect of the A . . . A mismatch in the d(GAC) 6 ⅐d(GAC) 6 duplex (Fig. 1A) has been investigated at the atomistic level using MD simulations by considering two different starting glycosyl conformations for the mismatch, following earlier studies (7,17). In the first model, both As are chosen to have an anti conformation (anti . . . anti). On the other hand, one of the two As in the second model is chosen to be in an anti glycosyl conformation, and the other is chosen to be in a ϩsyn glycosyl conformation (ϩsyn . . . anti).

Starting model with anti . . . anti glycosyl conformation
Analysis of the 500-ns simulation of the d(GAC) 6 ⅐d(GAC) 6 duplex that comprises six A . . . A mismatches with anti . . . anti glycosyl conformation shows that the nonisostericity of the A . . . A mismatch with respect to canonical G . . . C and C . . . G base pairs induces distortions in the helix. Distortions are seen within ϳ2 ns of the simulation, become prominent ϳ63 ns through the unwinding of the helix, and stay in the same conformation until the end of the simulation ( Fig. 1B and supplemental Movie S1). This eventually reflects in the root mean square deviation (RMSD), whose average value stays at ϳ4.5 Å between 0.5 to 10.5 ns and ϳ6.5 Å between 10.5-60.3 ns before finally reaching the highest value of ϳ9 Å (Fig. 1C).
The conformational features that are associated with such helix unwinding are as follows. First, As take up a high anti glycosyl conformation (61%) and exhibit a preference for -syn (39%) transiently during the simulation (Fig. 1D), and Gs that are engaged in canonical hydrogen bonding with Cs profoundly favor a Ϯsyn glycosyl conformation (92%) after 50 ns (Fig. 1E).
Concomitantly, helical twist angles at GA, AC, and CG steps also exhibit variations. Among the three steps, CG steps exhibit a higher population of low helical twists (64% of helical twists lower than 10°) compared with AC (22%) and GA (20%) steps (Fig. 2E, left panel). Yet another property that can support the formation of Z-DNA in the midst of B-DNA in the d(GAC) 6 ⅐d(GAC) 6 duplex is the angle formed by three adjacent phosphates. For the average structure calculated over the last 10 ns, the angles at the central phosphate in the following steps are below 110°, which further supports the formation of local Z-DNA (ϳ110°) (21): G 4 pA 5 (119°), C 6 pG 7 (96°), G 7 pA 8 (90°), C 9 pG 10 (81°), G 10 pA 11 (99°), C 12 pG 13 (92°), G 13

Starting model with ؉syn . . . anti glycosyl conformation
To further explore the global conformational preference for the A . . . A mismatch, another starting conformation, the ϩsyn . . . anti glycosyl conformation, is considered for the mismatch. This conformation is specifically chosen based on the glycosyl angle preference for the A . . . A mismatch in the CAGcontaining RNA duplex (17,22).
As seen above, the nonisosteric character of the A . . . A mismatch with respect to the flanking canonical G . . . C/C . . . G base pairs triggers unwinding of the helix after ϳ90 ns of simulation ( Fig. 3A and supplemental Movie S2). Time versus RMSD profile, calculated with respect to the initial model ( Fig.  3B), was also indicative of significant deviation from initial model (ϳ5 Å). The associated conformational changes are as follows: Gs predominantly taking the Ϯsyn glycosyl conformation (76%) (Fig. 3C) along with As in anti/ϩsyn conformation (Fig. 3D). Exceptionally, some of the Gs and As briefly take a high-anti and -syn conformation, respectively. Backbone torsion angles such as ⑀, , ␣, and ␥ in GA steps exhibit the characteristics of Z-DNA (35%), whereas AC (64%) and CG (45%) steps favor forming BI and BII conformations (Fig. 2C). CG (37%), GA (52%), and AC (28%) steps are also populated by other conformations such as (t,t,gϪ,gϩ), (gϪ,gϪ,t,t), (t,gϪ,gϩ,t), (t,gϪ,t,t), and (gϪ,t,gϩ,t) (Fig. 2D). As discussed earlier, these unusual conformations may be due to the interaction of counterions with the duplex (supplemental Fig. S1B). As before, CG steps (46% of helical twists less than 10°) exhibit a lower twist angle compared with GA (2%) and AC (2%) steps (Fig. 2E, right panel). The angles at the central phosphate in the  Together, these properties confirm the presence of the B-Z junction in the GAC repeat-containing duplex with a ϩsyn . . . anti starting glycosyl conformation for the A . . . A mismatch. Nonetheless, the preference for the Z conformation is less prominent compared with the anti . . . anti starting glycosyl conformation.

Canonical base pairs containing the d(GAC) 6 ⅐d(GTC) 6 duplex and a T . . . T mismatch containing the d(GTC) 6 ⅐d(GTC) 6 duplex retain B-form geometry
Control simulations carried out for the d(GAC) 6 ⅐d(GTC) 6 duplex with G . . . C and A . . . T canonical base pairs (Fig. 4A) to pinpoint that the B-Z junction formation observed in d(GAC) 6 ⅐d(GAC) 6 is purely due to nonisomorphism of the A . . . A mismatch indicate the dominance of B-form geometry (Fig. 4B). Although Z-DNA characteristics are observed during the simulation by 34% CG steps having helical twists lower than 10°( Fig. 4C) along with 42% of Gs preferring the Ϯsyn glycosyl conformation (Fig. 4D), this is comparatively lower than in the mismatch situations (Figs. 1E and 3C). It is noteworthy that As (96%) prefer predominantly anti/high-anti glycosyl conformations (Fig. 4E). Few GA/GT steps also show Z-DNA backbone conformation (Fig. 4, F and G). In fact, such a minor population can be attributed to the presence of cations in the minor groove, as pointed out in an earlier study (supplemental Fig. S1C) (20).
In summary, only a minor population of Z-DNA is observed in d(GAC) 6 ⅐d(GTC) 6 and d(GTC) 6 ⅐d(GTC) 6 duplexes compared with d(GAC) 6 ⅐d(GAC) 6 duplexes. To further validate that B-Z junction formation is mainly induced by an A . . . A mismatch, CD studies were carried out (see below).

CD confirms B-Z junction formation in the d(GAC) 7 ⅐d(GAC) 7 duplex
At 50 mM NaCl salt concentration, the CD spectrum of d(GAC) 7 ⅐d(GAC) 7 shows a positive peak between 270 -280 nm and a negative peak around 260 nm, a typical characteristic of B-form DNA (Fig. 6A). However, with an increase in NaCl concentration in the range of 0.05 M to 4.2 M, the negative ellipticity around 260 nm moves toward positive ellipticity. Additionally, the spectra start developing two negative peaks (ϳ290 nm and ϳ205 nm) with respect to the increase in salt concentration, which are all Z-DNA signature peaks (Fig. 6A). Nevertheless, canonical base pairs containing the d(GAC) 7 ⅐d(GTC) 7 duplex do not exhibit any B-to-Z transition with respect to the increase in NaCl concentration and stay in B-form, with positive and negative peaks around 285 nm and 260 nm, respectively (Fig. 6B). The d(GTC) 7 ⅐d(GTC) 7 duplex that has seven T . . . T mismatches that also exhibit same tendency as the d(GAC) 7 ⅐d(GTC) 7 duplex (supplemental Fig. S3). Thus, saltdependent CD spectra clearly indicate that the A . . . A mismatch dictates B-Z junction formation, which subsequently converts the duplex to complete Z-form at a higher salt concentration.

The Z␣ domain of human ADAR1 binds with the d(GAC) 7 ⅐d(GAC) 7 duplex
The CD spectra of the d(GAC) 7 ⅐d(GAC) 7 duplex (N) and hZ␣ ADAR1 protein (P) titration clearly show that increasing the concentration of P (viz. increasing the P/N ratio by keeping N as a constant) completely changes the duplex to the left-handed Z-form. As the concentration of P increases, the negative peak at ϳ255 nm gradually diminishes, accompanied by the appearance of a new negative peak at ϳ295 nm and a shift in the positive peak from 280 nm to 275 nm, characteristic features of the Z-DNA conformation (Fig. 6C). In contrast, the d(GAC) 7 ⅐d(GTC) 7 duplex, which contains only canonical base pairs, does not exhibit such a tendency for B-to-Z transition (Fig. 6D). Such a scenario is seen irrespective of the number of repeats in the duplex. For instance, the d(GAC) 6 ⅐d(GAC) 6 duplex, which has 6 A . . . A mismatches, also takes up the Z-form upon increasing the P/N ratio (supplemental Fig. S4A), whereas the corresponding canonical duplex does not exhibit such characteristics (supplemental Fig. S4B). In fact, d((GAC) 3 T 4 (GAC) 3 ) (wherein one GAC in the d(GAC) 7 is replaced by T 4 to facilitate the hairpin formation), which is expected to form a hairpin with three A . . . A mismatches, also exhibits B-to-Z transition upon titration with hZ␣ ADAR1 (supplemental Fig. S5). This situation mimics hairpin formation in d(GAC) 7 by having one GAC repeat in the hairpin loop and six GAC repeats in the stem with three A . . . A mismatches. Although there is a possibility that d(GAC) 7 can take up either an intramolecular hairpin conformation (with three A . . . A mismatches) or an intermolecular duplex conformation (with seven A . . . A mismatches) in solution (supplemental Fig. S6), it is difficult to identify the preferred conformation from CD data. Indeed, both conformations may equally be populated in vitro, unlike in vivo, wherein it can take up only the hairpin conformation. Thus, d((GAC) 3 T 4 (GAC) 3 ) titration with hZ␣ ADAR1 confirms that d(GAC) 7 can adopt a stable hairpin conformation with three A . . . A mismatches, which subsequently facilitates binding with the protein.

hZ␣ ADAR1 binds d(GAC) 7 ⅐d(GAC) 7 with nanomolar affinity
In accordance with the CD results, 1D proton NMR spectra of hZ␣ ADAR1 and d(GAC) 7 ⅐d(GAC) 7 duplex titration also confirm the interaction between these two. Overall, the spectra show a gradual reduction in peak intensity as the concentration of hZ␣ ADAR1 increases. Although it may be difficult to identify the amino acids/nucleotides that are associated with the proton chemical shifts simply from the 1D spectra, the signature chemical shifts around 0 to Ϫ1 ppm, 9.8 ppm, and 9.6 ppm can be assigned to protons corresponding to Thr-191 (23), the H⑀1 proton of Trp-195 and the amide proton of Ala-158 (24), respectively, which are located in and around the binding site of hZ␣ ADAR1 (PDB code 2ACJ). Likewise, the chemical shifts between 5 to 6 ppm belong to backbone protons of the DNA duplex (25). Notably, the protein and DNA chemical shifts in these regions do not overlap with each other (Fig. 7A). MST exhibits a nanomolar binding affinity between hZ␣ ADAR1 and the d(GAC) 7 ⅐d(GAC) 7 duplex, with a dissociation constant (K D ) of 41 nM (Fig. 7B).

The d(GAC) 7 ⅐d(GAC) 7 -hZ␣ ADAR1 complex model
The aforementioned information about hZ␣ ADAR1 amino acids (Thr-191, Trp-195, and Ala-158) that may be involved in interaction with the d(GAC) 7 ⅐d(GAC) 7 duplex (Fig. 7A), along with the readily available complex structure of hZ␣ ADAR1 and a B-Z junction (PDB code 2ACJ), have been used to model the hZ␣ ADAR1 -d(GAC) 6 ⅐d(GAC) 6 complex. Fig. 7C shows the modeled complex derived from X-ray (former) and MD (latter) structures. As more than one hZ␣ ADAR1 can bind to a single duplex, depending on the availability of Z-philic centers (PDB code 2ACJ) (21), d(GAC) 6 ⅐d(GAC) 6 can also accommodate more than one hZ␣ ADAR1 molecule (Fig. 7D).

MD simulation retains the conserved interactions between hZ␣ ADAR1 and the DNA duplex
The modeled hZ␣ ADAR1 . . . d(GAC) 6 ⅐d(GAC) 6 complex has been subjected to 300-ns MD simulations to optimize the interaction between the two. It is noteworthy that the complex has been modeled so that two monomers of hZ␣ ADAR1 interact with two different strands of the duplex (Fig. 7C), as reported earlier (PDB code 2ACJ). Analysis of the MD trajectories reveals that hZ␣ ADAR1 interacts with the duplex through its minor groove (Fig. 8A). The Lys-169, Asn-173, and Arg-174 residues of hZ␣ ADAR1 monomers participate in a hydrogenbonding interaction with the duplex backbone atoms (like O5Ј, O1P, and O2P) either transiently or persistently (Fig. 8, B-D). This is consistent with previous mutagenesis and NMR studies (26,27) that show the importance of the above mentioned residues in facilitating the interaction between the two. However, a minor difference in the nature of interaction is also seen. For instance, Tyr-177, which is involved in a stacking interaction in the crystal structure (PDB code 2ACJ), is engaged in a transient hydrogen-bonding interaction with the sugar-phosphate backbone atoms (Fig. 8E). Similarly, Trp-195 does not participate in any direct hydrogen-bonding interaction with the duplex, although it lies in the proximity of the duplex (Fig. 8E). Thus, the unwinding of the d(GAC) 6 ⅐d(GAC) 6 duplex because of the presence of the A . . . A mismatch (Figs. 1B and 3A) facilitates the interaction of the hZ␣ ADAR1 protein at the minor groove.

Discussion
Left-handed Z-DNA has a higher-energy conformation compared with the canonical B-DNA conformation (28), and in vitro, d(GC) n sequences are shown to choose the Z-form under extreme conditions, like high salt concentrations (29). There is increasing evidence regarding the participation of Z-DNA in gene regulation, the formation of which is believed to relieve stress on the DNA structure through negative supercoiling (30). Proteins that specifically recognize and bind to Z-DNA are also identified: hZ␣ ADAR1 (27), E3L (31), DLM1 (32), and PKZ (33). Interconversion between B-and Z-DNA is believed to take place either through a "stretch-collapse mechanism" or a "zipper mechanism" (34,35), facilitated by base extrusion and base and/or backbone flipping. Intriguingly, the A . . . A mismatch in the hairpin stem of the CAG repeat readily exhibits a preponderance for the Z-DNA conformation through the zipper mechanism (7). As GAC repeats that are responsible for pseudoachondroplasia also contain periodic A . . . A mismatches, we investigate here the ability of the same to adopt a Z-form structure by employing MD simulation, CD, MST, and NMR techniques. Subsequently, its ability to bind with the hZ␣ ADAR1 protein is also explored.  Fig. S1C). This eventually reflects in the CG step taking a low twist in the midst of high twists at the AC and GA steps in d(GAC) 6 ⅐d(GAC) 6 duplexes (64 and 46% in anti . . . anti and ϩsyn . . . anti glycosyl conformations, respectively) (Fig. 2E), leading to a B-Z junction in the vicinity of the mismatch. Such an occurrence of high and low twists in the duplex leads to unwinding of the helix, a typical   Fig. S7). It is noteworthy that the B-Z junction does not show alternating glycosyl (Ϯsyn and anti) and backbone conformations as in the Z-form, wherein alternating glycosyl conformations lead to a zigzag backbone (36,37). Instead, the B-Z junction possesses the characteristics of both B-and Z-forms.

The A . . . A mismatch induces local B-to-Z transition through backbone flipping
In sharp contrast to the d(GAC) 6 ⅐d(GAC) 6 duplex, the d(GAC) 6 ⅐d(GTC) 6 duplex with canonical base pairs and the d(GTC) 6 ⅐d(GTC) 6 duplex with a T . . . T mismatch have a preference for B-form geometry. This finding is further confirmed by CD spectroscopy by titrating NaCl with the d(GAC) 7 ⅐d(GAC) 7 , d(GTC) 7 ⅐d(GTC) 7 , and d(GAC) 7 ⅐d(GTC) 7 duplexes; although d(GAC) 7 ⅐d(GAC) 7 clearly displays B-Zto-Z transition with respect to the increase in salt concentration, the other two do not exhibit such a transition (Figs. 6, A and B, and supplemental Fig. S3). Such an inclination of the A . . . A mismatch toward the Z-form is due to its nonisostericity that is exemplified by a high residual twist and radial difference with the flanking C . . . G/G . . . C base pairs (18). This provides discomfort for the A . . . A mismatch to get accommodated in a B-DNA. Thus, it unwinds the helix to relieve the mechanistic effect arising from the nonisostericity of the A . . . A mismatch with respect to the flanking canonical base pairs as well as to retain the backbone connectivity (18, 38 -40). As seen in the d(CAG) 6 ⅐d(CAG) 6 duplex (7), B-to-Z transition takes place through a zipper mechanism rather than a stretch-collapse mechanism. One can envisage a similar situation in the case of the (GA) n homoduplex, where the nonisostericity between G . . . G and A . . . A may provoke parallel duplex formation (41).

Inclination of the A . . . A mismatch toward Z-DNA leads to passive binding with hZ␣ ADAR1
The mechanism of recognition and binding of hZ␣ ADAR1 protein with the B-Z junction/Z-DNA is still a matter of debate. According to the active mechanism, hZ␣ ADAR1 binds to B-DNA and subsequently converts it into Z-DNA (24). Never-theless, the passive mechanism suggests that hZ␣ ADAR1 traps the transient B-Z junction/Z-DNA and subsequently converts it into Z-DNA (42). The MD simulation (Figs. 1 and 3), CD (Fig.  6C), NMR (Fig. 7A), and microscale thermophoresis data presented here (Fig. 7B) conjointly identify that hZ␣ ADAR1 binds to d(GAC) 7 ⅐d(GAC) 7 in a "passive mechanism" because of the formation of a B-Z junction induced by the A . . . A mismatch.
As discussed above, MD simulation clearly shows the preference for B-Z junction formation in the d(GAC) 6 ⅐d(GAC) 6 duplex (Figs. 1B and 3A), in accordance with the CD spectra of NaCl titration with the d(GAC) 7 ⅐d(GAC) 7 duplex (Fig. 6, A and  B). Although the mismatch-containing duplex (former) has proclivity toward Z-DNA transition (Fig. 6A), the canonical base pair-containing duplex (latter) does not possess such a property (Fig. 6B). In line with this, titration of hZ␣ ADAR1 with d(GAC) 7 ⅐d(GAC) 7 converts the duplex completely to the Z-form irrespective of duplex length ( Fig. 6C and supplemental Figs. S4 and S5). Although some studies have shown that the GAC sequence is prone to form Z-DNA (10,11), the rationale behind such a conformational preference is unknown. For the first time, it has been shown here that the nonisosteric A . . . A mismatch provokes B-Z transition in GAC repeats, which subsequently facilitates binding with the hZ␣ ADAR1 protein through a passive mechanism. This is further confirmed by 1D proton NMR spectroscopy, which indicates tighter affinity between the two (Fig. 7A). Additionally, K D measured by MST also indicates that the d(GAC) 7 ⅐d(GAC) 7 duplex binds with hZ␣ ADAR1 with nanomolar affinity (Fig. 7B). MD simulation carried out on the modeled hZ␣ ADAR1 -d(GAC) 6 ⅐d(GAC) 6 complex (Figs. 7, C and D, and 8A) subsequently confirmed that the protein residues interact with the duplex through the minor groove, in accordance with earlier studies (PDB codes 2ACJ and 3IRQ).

Model for pathogenicity in d(GAC) n expansion disorders through RNA editing mediated by h ADAR1
This study clearly shows that the d(GAC) 7 ⅐d(GAC) 7 duplex is not only prone to form Z-DNA but also binds to the Z-DNA A model for the GAC duplex . . . hZ␣ ADAR1 complex binding domain of the human ADAR1 protein. Intriguingly, expansion in the d(GAC) trinucleotide repeat is shown to cause skeletal dysplasias, such as multiple epiphyseal dysplasia and pseudoachondroplasia (3,6). Hence, based on the results of this study, we propose a model that explains how d(GAC) trinucleotide expansion in the COMP gene may lead to skeletal dysplasia, such as pseudoachondroplasia. According to our model (Fig. 9), d(GAC) 7 , which can form a hairpin structure with the stem possessing a B-Z junction (induced by A . . . A mismatches), facilitates anchorage of the Z-DNA binding domain of h ADAR1 onto the DNA during transcription. Succeeding this event, the double-stranded RNA-specific deaminase domain of h ADAR1 performs A-to-I editing in GAC either in the corresponding nascent RNA duplex (Fig. 9A, top panel) or downstream (Fig. 9A, bottom panel). This eventually codes for Gly instead of Asp in COMP. In fact, samples isolated from pseudoachondroplasia patients show that genomic point mutations in the d(GAC) track of the COMP gene that can code for Gly instead of Asp-473/Asp-482 are among the 70 possible mutations in the COMP gene (2,43). According to the current model, A-to-I editing can lead to such Asp-to-Gly mutation at the protein level during transcription and, thus, can reflect the effect of genomic point mutations, as mentioned above. Such A-to-I editing downstream of the GAC repeat expansion (Asp-482) can also take place (Fig. 9A, bottom panel), resulting in Asp-to-Gly in COMP, which has already been shown to be deleterious (44,45). On the other hand, when the d(GAC) repeat does not undergo expansion, hairpin formation may not take place. Thus, h ADAR1 may not be able to bind to the DNA duplex, and A-to-I editing may not occur (Fig. 9B). Its noteworthy that, although direct evidence for the role of h ADAR1 in pseudoachondroplasia is not well-established, its hyper/altered editing in several neurodegenerative disorders has been well-documented (13,14,16). In line with this, the hypothesis presented here offers new insight into the role of non-genetic A-to-I mutation in pseudoachondroplasia.

A model for the GAC duplex . . . hZ␣ ADAR1 complex
We have shown here that the nonisomorphic nature of the A . . . A mismatch with respect to the flanking base pairs is the underlying factor for the Z-philic nature observed in the d(GAC) 7 ⅐d(GAC) 7 repeat expansion that is found in pseudoachondroplasia. We have shown here, for the first time, that such a structural trait of the A . . . A mismatch facilitates binding of hZ␣ ADAR1 to the d(GAC)) n ϭ 6,7 ⅐d(GAC) n ϭ 6,7 duplex irrespective of the repeat length. A model for the complex of hZ␣ ADAR1 -d(GAC) 7 ⅐d(GAC) 7 duplex and the consequent A-to-I editing during transcription by the double-stranded RNA-specific deaminase domain of h ADAR1 under the disease condition are also presented.

Molecular dynamics simulations
Starting models of d(GAC) 6 ⅐d(GAC) 6 DNA duplexes were manually modeled using the PyMOL suite (57). Subsequently, the models were refined using constrained-restrained molecular geometry optimization using XPLOR-NIH (46). MD sim-ulations of the modeled duplexes (Fig. 1A) were carried out in an explicit solvent environment following the protocol described earlier (7) using the AMBER 12 suite (47). FF99SB forcefield was used during the simulation. The systems were initially equilibrated for 50 ps, and then production runs were extended to 0.5 s for each system using isobaric and isothermal conditions (NPT), 2-fs integration time, and 9-Å cutoff distance for the Lennard-Jones interaction. Following the above procedure, MD simulations of the d(GTC) 6 ⅐d(GTC) 6 and d(GAC) 6 ⅐d(GTC) 6 duplexes were carried out for 0.5 s each. The 3D-NuS web server was used to build these models (48).

Analysis of the trajectories
The Ptraj module of Amber 12 was used to post-process the trajectories corresponding to d(GAC) 6 ⅐d(GAC) 6 , d(GTC) 6 ⅐d(GTC) 6 , and d(GAC) 6 ⅐d(GTC) 6 simulations. RMSD was calculated to acquire quantitative information about the deviation or proximity of the trajectories from the initial structure. Backbone conformational angles and helical parameters A model for the GAC duplex . . . hZ␣ ADAR1 complex were extracted from 3DNA (49) output using in-house programs. PyMOL (57) and VMD (50) were used for visualization, and MATLAB software (The MathWorks Inc., Natick, MA) was used for plotting the graphs. Note that for the analysis, the central 14mer alone was considered.

Docking of the d(GAC) 6 ⅐d(GAC) 6 DNA duplex with the hZ␣ ADAR1 protein
The complex structure of the d(GAC) 6 ⅐d(GAC) 6 DNA duplex and hZ␣ ADAR1 protein was manually modeled by replacing the duplex present in the crystal structure (PDB code 2ACJ) with our MD-derived d(GAC) 6 ⅐d(GAC) 6 duplex. Subsequently, the complex model was subjected to 0.3-s MD simulations using the pmemd.cuda module of the AMBER 16 suite. Analysis was carried out by using the cpptraj module of AMBER 16.

Duplex preparation
HPLC-grade d(GAC) n ϭ 6,7 and d(GTC)) n ϭ 6,7 oligonucleotides were purchased from Sigma-Aldrich. The oligonucleotides were dissolved in 50 mM Tris-HCl and 50 mM NaCl (pH 7.4). The DNA duplex with canonical base pairs was formed by annealing (GAC) 7 and the complementary (GTC) 7 oligonucleotides at 95°C and cooling them down to room temperature for 3 h. On the other hand, only the former was considered for formation of the duplex with the A . . . A mismatch, and the latter was used for formation of the T . . . T mismatch. Subsequently, duplex formation was verified by acquiring the CD spectrum (see below). Likewise, hairpin formation of d((GAC) 3 T 4 (GAC) 3 ) was carried out. It is noteworthy that, to investigate the salt-dependent behavior of the duplex, the

Subcloning of the hZ␣ ADAR1 gene into the pET21a expression vector
The hZ␣ ADAR1 gene cloned in the pMAT cloning vector was acquired from Invitrogen with Ndel and Sal1 restriction sites at the 5Ј and 3Ј ends, respectively. Subsequently, the PCR-amplified, double-digested hZ␣ ADAR1 gene was subcloned into the ampicillin-resistant pDZ1 expression vector, a modified form of the pET-21a vector with a T7 promoter (51-53). The construct was organized in the following order: an N-terminal His 6 tag, GB1 solubility tag, and tobacco etch virus protease cleavage site that were followed by the hZ␣ ADAR1 gene (225 bp).

Protein expression and purification
The pDZ1 expression vector was transformed into Escherichia coli BL21 (DE3) (Bioline) cells for overexpression of the hZ␣ ADAR1 protein. Preinoculum cells grown overnight were transferred into LB medium containing 100 g/ml ampicillin and incubated at 37°C until the optical density reached 0.6 at A 600 . Protein expression was induced by 1 mM isopropyl 1-thio-␤-D-galactopyranoside, followed by overnight incubation at 15°C to attain the optical density at A 600 in the range of 1.4 to 1.6. Cells were then harvested and sonicated in binding buffer containing 20 mM Tris-HCl, 500 mM NaCl, 5 mM imidazole (pH 8.0), and 0.1 mM PMSF. The hZ␣ ADAR1 protein was eluted in a buffer containing 20 mM Tris-HCl, 500 mM NaCl, and 200 mM A model for the GAC duplex . . . hZ␣ ADAR1 complex imidazole (pH 8.0) using nickel-nitrilotriacetic acid affinity column chromatography that was treated with 50 mM NiSO 4 solution.
Purification involved two steps. First, the hZ␣ ADAR1 protein tagged with GB1 protein was purified as described above (supplemental Fig. S8A), followed by removal of the GB1 tag through overnight digestion with tobacco etch virus protease. During the second round of purification, the hZ␣ ADAR1 protein was isolated from the cleaved GB1 tag, and the fractions were collected (supplemental Fig. S8B) in binding buffer. Finally, the protein was dialyzed in NMR buffer (10 mM phosphate buffer and 10 mM NaCl (pH 7.4)). Protein concentration was measured by UV absorption at 280 nm using an extinction coefficient value of 8480 M Ϫ1 cm Ϫ1 .

d((GAC) n ⅐(GAC) n ) n ‫؍‬ 6,7 -hZ␣ ADAR1 complex formation
The d((GAC) n ⅐(GAC) n ) n ϭ 6,7 -hZ␣ ADAR1 complex was prepared by changing the concentration of hZ␣ ADAR1 while retaining the concentration of DNA. For NMR experiments, the following P/N ratios were used by keeping the DNA concentration at 120 M: 0.25, 0.5, 0.75, and 1. For CD experiments, P/N ratios of 0.25, 0.5, 0.75, 1, 1.5, and 2 were used by keeping the DNA concentration at 40 M. The samples were prepared in buffer containing 10 mM sodium phosphate and 10 mM NaCl (pH 7.4), and 10% of D 2 O was added to the NMR sample. The complex was prepared by adding the protein to the DNA sample in fractions of 10 l at 2-min intervals and incubated for 1 h at 25°C.

CD spectroscopy
All CD spectra reported here were acquired in JASCO-1500 and processed by Spectra Manager software. The data were collected in triplicate in the wavelength region of 320 nm to 200 nm, and baseline correction was done with respect to the appropriate buffer. The average of triplicate spectra is reported here.

NMR spectroscopy
1D proton NMR experiments were performed on a Bruker 700-MHz instrument equipped with a room temperature probe. The Zggpw5 pulse sequence (54) was used to acquire data at 25°C. All acquisition parameters were kept identical for all experiments: 768 scans and 32768 1 H complex points. Bruker Top Spin was used for data processing and analysis.

Dissociation constant measurement using microscale thermophoresis
The dissociation constants of hZ␣ ADAR1 binding with the d(GAC) 7 ⅐d(GAC) 7 duplex were estimated using a microscale thermophoresis assay (55,56). The assay was carried out using His 6 -GB1-hZ␣ ADAR1 -tagged protein. The MST assay required one fluorescent binding partner (protein) and one non-fluorescent binding partner (DNA). Prior to titration, NT-647 fluorescent dye was non-covalently attached to the histidine residues of hZ␣ ADAR1. DNA was titrated to hZ␣ ADAR1 in serial dilutions, with concentrations ranging from 0.313 M to 0.000153 M. Subsequently, the assay was carried out in 10 mM phosphate buffer by keeping the concentration of labeled hZ␣ ADAR1 protein as a constant (15 nM). These samples were loaded into Monolith NT.115 MST Premium-coated capillaries, and the MST analysis was performed using 100% light-emitting diode (LED) power and 60% MST power in NanoTemper Monolith NT.115 at 24°C. Using NanoTemper software, K D was calculated using the mass action equation from triplicate experiments.
Author contributions-N. K. carried out molecular dynamics simulations, analyzed the data, and performed salt-dependent CD experiments. Y. A. carried out subcloning, expression, and purification of hZ␣ ADAR1 along with CD, NMR titration, and K D experiments of the hZ␣ ADAR1 -DNA complex, followed by docking and MD simulations of the hZ␣ ADAR1 -DNA complex. T. R. designed and supervised the entire project. N. K., Y. A., and T. R. wrote the manuscript.