A bimodular nuclear localization signal assembled via an extended double-stranded RNA-binding domain acts as an RNA-sensing signal for transportin 1

Significance

The double-stranded RNA-binding domain (dsRBD) is an abundant, conserved RNA-binding motif. Besides RNA binding, dsRBDs can serve as protein-interaction domains. In the human RNA-editing enzyme adenosine deaminase acting on RNA (ADAR1), one of its three dsRBDs mediates nuclear import by interacting with the import receptor transportin 1 (Trn1). RNA binding interferes with Trn1 binding, thereby preventing nuclear import. Using NMR spectroscopy and cell biological analysis, we show that the regions flanking this dsRBD form a bimodular Trn1-dependent nuclear localization signal. The dsRBD itself is not involved in Trn1 interaction but properly positions the Trn1 interacting regions. Using molecular modeling, we provide a structural explanation on how dsRNA binding prevents the dsRBD from accessing the interacting cavity of Trn1, thereby preventing nuclear import of RNA-bound ADAR1.

Abstract

The human RNA-editing enzyme adenosine deaminase acting on RNA (ADAR1) carries a unique nuclear localization signal (NLS) that overlaps one of its double-stranded RNA-binding domains (dsRBDs). This dsRBD-NLS is recognized by the nuclear import receptor transportin 1 (Trn1; also called karyopherin-β2) in an RNA-sensitive manner. Most Trn1 cargos bear a well-characterized proline-tyrosine-NLS, which is missing from the dsRBD-NLS. Here, we report the structure of the dsRBD-NLS, which reveals an unusual dsRBD fold extended by an additional N-terminal α-helix that brings the N- and C-terminal flanking regions in close proximity. We demonstrate experimentally that the atypical ADAR1-NLS is bimodular and is formed by the combination of the two flexible fragments flanking the folded domain. The intervening dsRBD acts only as an RNA-sensing scaffold, allowing the two NLS modules to be properly positioned for interacting with Trn1. We also provide a structural model showing how Trn1 can recognize the dsRBD-NLS and how dsRNA binding can interfere with Trn1 binding.

Trafficking of macromolecules between the nucleus and cytoplasm through nuclear pore complexes (NPCs) is a selective and efficient process orchestrated by transport receptors that associate with their cargoes via cognate nuclear localization signals (NLSs) or nuclear export signals (NESs). Transport receptors of the karyopherin-β (Kapβ) family, also known as importins and exportins, account for the vast majority of the cargo flow through the NPC. Transport directionality is primarily driven by the RanGTPase nucleotide cycle, which produces an asymmetric distribution of RanGTP and RanGDP on both sides of the nuclear envelope (1–4).

Unlike importin-β, which forms a heterodimer with the adaptor importin-α to import substrates containing classical NLSs (cNLSs) (1, 5), transportin 1 (also known as Kapβ2; thereafter designated Trn1) binds cargoes in the cytoplasm without adaptors, targets them to the nucleus through the NPC, and releases them in the nucleus upon RanGTP binding (6, 7). In contrast to the well-defined cNLS recognized by importin-α, Trn1 recognizes a diverse set of sequences with poor apparent conservation that cannot be appropriately described by a conventional consensus sequence. Nevertheless, structural and biochemical studies uncovered a set of loose principles common to some Trn1 signals (8–11). These principles can be described as follows: a peptide segment of 15–30 residues with intrinsic structural disorder, an overall basic character, and some weakly conserved sequence motifs, including a relatively conserved proline-tyrosine (PY) dipeptide, leading to the name PY-NLS for this class of signals (8). However, many characterized Trn1 cargoes do not contain such a PY-NLS (4, 12–15). Therefore, Trn1 interacts not only with the well-characterized PY-NLSs, but at least with another class of NLS, the recognition of which is far less understood. Whether these other classes of NLS share some similarities with PY-NLSs remains to be determined.

One Trn1 cargo that lacks a typical PY-NLS is the RNA-editing enzyme ADAR1, a member of the adenosine deaminases acting on RNA (ADARs) family. These enzymes convert adenosines to inosines by hydrolytic deamination in structured and double-stranded RNA substrates (16, 17). Inosines formed by adenosine deamination can alter the coding potential of mRNAs but also can affect splicing, localization, or transport of cellular and viral RNAs (16, 17). ADARs have a common modular domain organization that includes one to three dsRNA binding domains (dsRBDs) in their N-terminal region followed by a C-terminal catalytic domain (SI Appendix, Fig. S1A) (18).

ADAR1 is expressed from two promotors, giving rise to a long, IFN-inducible version (ADAR1-i) and a constitutively expressed shorter version (ADAR1-c) (19). ADAR1-i harbors an NES in its unique N-terminal region (20) and is mostly cytoplasmic whereas ADAR1-c lacking the N-terminal NES is primarily nuclear (SI Appendix, Fig. S1B) (21). Both isoforms shuttle between the nucleus and cytoplasm (22, 23). Nuclear import of both ADAR1 versions is mediated by Trn1 via an NLS that overlaps the third dsRBD and shows no similarity to a PY-NLS (20, 22–24). Importantly, RNA-binding by this dsRBD interferes with Trn1 binding and nuclear import, thus constituting an RNA-regulated NLS (23).

The dsRBD is a 65–70 amino acid domain that shape-specifically recognizes dsRNA (25–27). Structures of different dsRBDs have been determined, revealing a conserved αβββα fold with two α-helices packed against a three-stranded β-sheet (28, 29). dsRBD-mediated nucleocytoplasmic trafficking has been reported for ADAR1 (22, 23), but also for other dsRBD-containing proteins (30–36). Like for ADAR1, interaction with dsRNA has often been described to regulate the trafficking function of these dsRBDs. Importantly, the exact elements involved in the NLS activity of ADAR1-dsRBD3 still remain elusive. Further, how ADAR1-NLS that clearly differs from PY-NLSs interacts with Trn1 and how dsRNA binding competes with nuclear import remain unknown.

To address these questions, we determined the solution structure of ADAR1-dsRBD3 by NMR spectroscopy. Besides a typical dsRBD fold, the structure reveals the presence of an additional N-terminal α-helix that brings the N- and C-terminal extremities in close proximity. We demonstrate that ADAR1-NLS is a bimodular NLS formed jointly by the combination of the N- and C-terminal fragments flanking ADAR1-dsRBD3. The dsRBD acts as a scaffold that helps to properly position the extensions for interaction with Trn1. We also provide a structural model showing how dsRNA binding can inhibit Trn1 binding and therefore prevent nuclear import.

Results

Nuclear Localization of ADAR1 Requires the Region N-Terminal to dsRBD3.

Previously, we have mapped the NLS in human ADAR1 to a region spanning Met708–Arg801. This region contains the third dsRBD of the protein with 18 and 7 amino acid-long extensions at the N- and C-terminal ends, respectively (Fig. 1A). The region is necessary and sufficient for nuclear import of ADAR1 and of unrelated reporter constructs (Fig. 1B) (24). Seeking for the molecular determinants responsible for the NLS activity in ADAR1, we tested whether a construct encompassing only dsRBD3, but lacking the upstream stretch of amino acids (Thr725–Arg801), would retain NLS activity. This shorter construct failed to show NLS activity (Fig. 1B). This behaviour is similar to what we saw for Schizosaccharomyces pombe Dicer dsRBD, where a C-terminal extension is essential to regulate the subcellular localization of the protein (36). Additionally, protein-sequence inspection through hydrophobic-cluster analysis predicts the presence of potential secondary-structure elements outside the dsRBD (SI Appendix, Fig. S2A) (37). We therefore determined the solution structure of ADAR1 Met708–Arg801 that embeds dsRBD3 (Fig. 1 C–E).

Fig. 1.

An N-terminal extension upstream of ADAR1-dsRBD3 is essential for NLS activity. (A) Domain organization of human ADAR1c (Uniprot P55265 entry). The N-terminal Z-DNA binding domain is in green, the three dsRBDs are in blue, and the C-terminal deaminase domain is in yellow. Annotated domain boundaries are reported below each domain. The minimal NLS fragment, as previously identified (23), is indicated (Met708-Arg801). (B) The constitutively expressed, full-length ADAR1c spanning M296 to V1226 is localized to the nucleus. Removal of the third dsRBD Δds3 leads to cytoplasmic accumulation of myc-tagged ADAR1c. The minimal NLS spanning Met708 to Arg801 is sufficient to mediate nuclear transport of a myc-tagged pyruvate-kinase (PK) reporter construct. A fragment starting at the beginning of the canonical dsRBD (Thr725-Arg801) lacks NLS activity. FITC channel shows transfected constructs in confocal sections. Nuclear DNA is stained with DAPI, and cellular outlines are shown in the differential interference constrast (DIC) channel. (Scale bar: 10 µm.) (C) Sequence of ADAR1-dsRBD3 (Met708–Arg801) with the corresponding secondary structure elements indicated (color code is the same as in D and E). Amino acid numbers refer to the full-length human ADAR1 protein (Uniprot P55265). (D) NMR ensemble. Overlay of the 20 final structures with color-coded secondary structure elements: α1 in blue, β-strands 1–3 in red-orange-yellow, α2 in green and the additional N-terminal helix α_N in purple. (E) Schematic representation of the lowest energy structure. The same color code is used for secondary structure elements. See also SI Appendix, Figs. S2 and S3.

ADAR1-dsRBD3 Is Extended by an N-Terminal α-Helix.

We determined the solution structure of human ADAR1-dsRBD3 with 2,165 NOE-derived constraints. The structure is very precise, with a backbone rmsd over the entire domain (residues 716–795) of 0.39 ± 0.10 Å for the ensemble of 20 conformers (Fig. 1D and Table 1). Pro727 to Glu795 form the dsRBD core domain, with the expected αβββα topology (28, 29) (Fig. 1 C–E). The residues immediately preceding helix α1 (Ile716–Asn726) form a well-structured α-helix (thereafter called helix α_N) that folds back on the dsRBD between helix α1 and the C-terminal half of helix α2 (Fig. 1D and SI Appendix, Fig. S2B). Even if helix α_N is closely packed against the dsRBD, it does not participate to the main hydrophobic core of the domain formed by side-chains of α1, α2, and the β-sheet (Fig. 1 D and E and SI Appendix, Fig. S3 A–C). Helix α_N interacts mostly via four hydrophobic residues with helices α1 and α2: (i) Ile716 with Ile793, (ii) Leu719 with Tyr734, Leu789, and Ile793, (iii) Tyr722 with Gly730, and (iv) Leu723 with Leu789, Arg790, and Ile793 (SI Appendix, Fig. S3 A–C). Apart from the remarkable helix α_N, the rest of the domain adopts a canonical dsRBD structure, similar to Xlrbpa-dsRBD2 (26) (SI Appendix, Fig. S3 D and E).

Table 1.

NMR experimental restraints and structural statistics

NMR distance and dihedral constraints
Distance constraints
Total NOE	2,165
Intraresidue	430
Sequential	524
Medium range (|i − j| < 5 residues)	507
Long range (|i − j| ≥ 5 residues)	704
Hydrogen bonds	49
Structure statistics
Distance restraints violations (mean ± SD)
Number of NOE violations > 0.2 Å	1.0 ± 0.7
Maximum NOE violation, Å	0.24 ± 0.06
Deviation from ideal covalent geometry (mean ± SD)
Bond lengths, Å	0.0033 ± 0.0001
Bond angles, °	0.43 ± 0.01
Impropers, °	1.15 ± 0.08
Average pairwise rmsd,* Å
Backbone	0.39 ± 0.10
Heavy atoms	0.74 ± 0.08
Ramachandran analysis
Most favored region	88.1%
Allowed region	11.8%
Disallowed region	0.1%

^*Pairwise rmsd was calculated using residues 716–795 for the ensemble of 20 refined structures.

The fact that helix α_N does not interact with the common hydrophobic core suggested that the canonical dsRBD without helix α_N might still be folded. We therefore expressed and purified Thr725–Arg801; this shorter construct is folded on its own, as shown by the large dispersion of its amide signals in the (¹⁵N,¹H)-heteronuclear single quantum coherence (HSQC) (SI Appendix, Fig. S4A). Additionally, the chemical-shift difference between the constructs, with or without helix α_N, perfectly reflects on the position of this helix as described in the preceeding paragraph (SI Appendix, Fig. S4 B and C). Importantly, the fact that the Thr725–Arg801 construct is folded indicates that the loss of NLS activity (Fig. 1B) is not caused by misfolding of the domain but rather by the loss of essential features present in the segment Met708–Asn724.

Although helix α_N does not impact the overall structure of the dsRBD core, it changes radically the relative position of its N and C termini. In canonical dsRBDs, the N and C termini are quite distant and situated on opposite faces whereas, in the extended ADAR1-dsRBD3, the termini are much closer and located on the same side of the domain (Fig. 1 D and E and SI Appendix, Fig. S3 D and E). This important point will be further considered in Deciphering the Nuclear Localization Elements Within ADAR1-dsRBD3.

In addition, to validate that the structure of the domain is not altered by the lack of further upstream elements, we expressed and purified a construct including most of the linker joining dsRBD2 and dsRBD3: i.e., Met688–Arg801. The NMR footprint of the folded part is absolutely unperturbed (SI Appendix, Fig. S4E), showing that the structure of ADAR1-dsRBD3 is not altered in the presence of further upstream elements. Unfortunately, any longer construct containing dsRBD2 failed to be expressed in a soluble form, and we did not succeed in refolding these protein constructs for structural investigations.

Deciphering the Nuclear Localization Elements Within ADAR1-dsRBD3.

Even though we could show that elements in the upstream segment (Met708–Asn724) are essential for NLS activity (Fig. 1B), the actual molecular determinants controlling nuclear import of human ADAR1 via the interaction with Trn1 remained elusive. We therefore used our NMR structure to uncover these molecular determinants.

As mentioned in the Introduction, well-characterized Trn1 substrates consist of stretches of 15–30 residues with intrinsic structural disorder and poor apparent sequence conservation, but adopting a similar extended conformation upon Trn1 binding (8–10, 38, 39). We therefore focused our attention to the unstructured peptide segments flanking the structured dsRBD3: i.e., Met708–Lys715 and Glu797–Arg801 in the N- and C-terminal parts, respectively (Fig. 1D and SI Appendix, Fig. S2B). To test whether these flexible fragments are important for NLS activity, we mutated residues from the basic or hydrophobic portion of the N-terminal stretch into alanine (K712A, R714A, and K715A and M708A, M709A, P710A, and V713A for the basic and hydrophobic portions, respectively). Similarly, residues from the C-terminal stretch were mutated as well (K798A, E800A, and R801A). Mutations were introduced either individually or in combinations. Their impact was evaluated in vivo by transfection of reporter constructs complemented by nuclear import assays on permeabilized cells (Fig. 2 A and B and SI Appendix, Table S1). Importantly, mutations in both the N- and C-terminal fragments strongly affect the NLS activity of the construct. The most dramatic effect was observed when changing Arg801 to alanine, which completely abolished NLS activity, both in the context of a pyruvate kinase reporter construct and in full-length ADAR1c (Fig. 2A). Similarly, Trn1 was unable to transport recombinantly expressed R801A ADAR1-NLS to the nucleus in import assays on permeabilized cells (Fig. 2B). Pull-down assays with R801A also support that direct interaction with Trn1 is lost in this mutant (Fig. 2C). In the N-terminal fragment, none of the single amino acid substitutions showed a significant loss of nuclear accumulation (SI Appendix, Table S1). Although a triple mutation of the basic amino acids K712A/R714A/K715A did not affect nuclear localization, the double mutation M708A/P710A showed both nuclear and cytoplasmic signals, suggesting that these residues are contributing to Trn1 interaction (Fig. 2A). In addition, deletion of residues Met708–Lys715 clearly abolished nuclear accumulation (Fig. 2A) and the deletion mutant (ΔM708–K715) failed to interact with Trn1, as judged by either import assays on permeabilized cells or pull-down assays with Trn1 (Fig. 2 B and C).

Fig. 2.

Elements flanking ADAR1-dsRBD3 are important for nuclear transport and interaction with transportin 1. (A) Confocal sections of indicated fragments transfected as myc-tagged PK fusions. Met708–Arg801 comprises the minimal NLS. Deletion of the N-terminal fragment ΔMet708–Lys715 abolishes NLS activity. Simultaneous mutation of M708A and P710A in this region strongly reduces NLS activity whereas the basic amino acids are not important for NLS activity (K712A + R714A + K715A). The C-terminal Arg801 is essential for nuclear import, both in the context of the PK fusion construct and in full-length ADAR1c. (B) Nuclear import assays on permeabilized cells using GST fusion constructs. Met708–Arg801 is imported into the nucleus in a Trn1-dependent fashion. An N-terminal deletion ΔMet708–Lys715 abolishes nuclear import. Also, the canonical dsRBD fold alone (Thr725–Arg801) fails to accumulate in the nucleus. The C-terminal R801A mutation abolishes nuclear import. Fusion proteins are visualized in the FITC channel. DAPI shows nuclear DNA; cellular outlines are shown in the DIC channel. (Scale bars: 10 µm.) (C) GST-fusion constructs of dsRBD constructs shown in A and B were tested for their ability to interact with Trn1 in pull-down assays. Input lanes show the proteins used for the interaction studies. GST-coupled wild-type construct (Met708–Arg801) can precipitate full-length Trn1. Deletions of the dsRBD N-terminal fragment (716–801), the N-terminal fragment and helix α_N (725–801), or the R801A mutation abolish interaction with Trn1. Input and precipitated material were run on a single Western blot. After blotting, the membrane was cut at the 70 kDa band. The upper part of the blot was detected with an anti-His antibody, detecting the Trn1 band, whereas the lower part of the blot was detected with a GST antibody, detecting the dsRBD fusions. The black lines indicate parts where the blot was cropped to arrange the marker because it was loaded in the middle of the gel. The lower part of the gel containing GST breakdown products was cropped.

Having shown that both the N- and C-terminal fragments flanking the structured dsRBD3 are essential for NLS activity, we postulated that the extended dsRBD might act as a scaffolding element bringing closely together two distant stretches of residues to form a functional bimodular NLS (Fig. 3A). To test this hypothesis, we first asked whether the relative position of the N- and C-terminal fragments is important for NLS activity. We therefore monitored the localization properties of a construct lacking helix α_N (deletion of residues Ile716–Asn724) (Fig. 3B). In this construct, the two modules are no longer on the same side of the domain. Similarly, a triple mutant that disrupts helix α_N structure without deleting its sequence was generated. We mutated into polar residues three buried hydrophobics making contacts with α1 and α2 to impair the tight packing of α_N onto the rest of the domain (i.e., I716N/L719S/L723N) (SI Appendix, Figs. S3 A–C and S4 C and D). In both mutant proteins, the resulting constructs accumulate in the cytoplasm (Fig. 3 B and C). However, a direct involvement of helix α_N in Trn1 binding and nuclear import cannot be truly excluded. To evaluate whether helix α_N could be directly involved in NLS activity, we mutated the solvent exposed residues of helix α_N into alanines (E718A, R721A, and N724A) individually or in combination. Both the individual and triple mutated constructs fully retained NLS activity (Fig. 3D and SI Appendix, Table S1).

Fig. 3.

The relative orientation of the N- and C-terminal fragments flanking ADAR1-dsRBD3 is essential to build a functional NLS. Schematic representations (Left) of ADAR1-dsRBD3 constructs tested for NLS activity. The dsRBD core domain is in gray, the N- and C-terminal fragments are in blue and in red, respectively, and the additional helix α_N is in purple. Localization of the corresponding constructs (Right). (A) In wild-type ADAR1-dsRBD3 (Met708–Arg801) a functional NLS is formed by the combination of the N- and C-terminal fragments. This functional NLS is highlighted in cyan. Consistently, a PK fusion harboring Met708–Arg801 accumulates in the nucleus as shown in the FITC channel. (B) Construct with a deleted helix α_N (Met708–Lys715 fused to Thr725–Arg801) positions the critical regions for Trn1 interaction on opposite phases of the domain and consistently leads to cytoplasmic accumulation of a reporter construct (ΔI716–N724). (C) Construct with a disrupted helix α_N carrying a triple mutation in the buried residues of helix α_N (I716N, L719S, and L723N) also fails to accumulate in the nucleus. (D) Construct with helix α_N carrying a triple mutation in the exposed residues of helix α_N (E718A, R721A, and N724A) can still properly position the N- and C-terminal fragments and consistently accumulates in the nucleus. Transfected constructs are detected in confocal sections in the FITC channel. DNA is visualized by DAPI, and cellular outlines are shown in the DIC channel. (Scale bar: 10 µm.) See also SI Appendix, Fig. S4.

These data show that the N- and C-terminal modules flanking ADAR1-dsRBD3 are essential elements of the NLS and that their relative spatial positioning is critical for import. The extended dsRDB would act as a scaffolding element essential to position the two modules for a productive interaction with Trn1.

Modeling the Interaction of the N- and C-Terminal Fragments Flanking ADAR1-dsRBD3 with Transportin 1.

Next, we wanted to test whether the proposed molecular model of the ADAR1 bimodular NLS (Fig. 3A) was compatible with the structures of both the extended dsRBD (this work) and Trn1 bound to well-characterized PY-NLSs (8–10, 38, 39). In other words, can the extended dsRBD fit inside the C-terminal arch formed by HEAT repeats 8–20 (8, 40) (see also SI Appendix, Fig. S5 for Trn1 nomenclature and geometry)? We first asked whether the same binding interface is used by Trn1 to bind the ADAR1-NLS and PY-NLSs. Two tryptophans (Trp460 and Trp730) had been shown to play a pivotal role in the binding of PY-NLSs by Trn1 (8, 10). Trp460 belongs to repeat H10, in the N-terminal half of the arch, called site A, whereas Trp730 is located in repeat H16, in the C-terminal half of the arch, called site B (SI Appendix, Fig. S5) (10). Binding of NLS peptides interacting at both sites A and B is strongly impaired when these tryptophans are mutated to alanines (i.e., W460A and W730A), and these Trn1 mutants mediate nuclear import of PY-NLSs very poorly (10). In addition, NLS peptides interacting only at site A (e.g., TAP NLS) are strongly impaired in their ability to bind to the W460A but not to the W730A mutant (10). We thus tested the W460A and W730A mutants for the interaction with ADAR1-dsRBD3 by coprecipitation assays (Fig. 4). Binding of ADAR1-dsRBD3 is abolished by either mutation, demonstrating that ADAR1-NLS is recognized by Trn1 using the same pockets as the ones used to bind PY-NLSs and that ADAR1-NLS is interacting with both site A and site B and therefore spans the entire C-terminal arch.

Fig. 4.

Binding pockets in Trn1 that are required for the interaction with PY-NLSs are also required for the interaction with ADAR1-NLS. (A) GST-fusion protein harboring the ADAR1-NLS (Met708–Arg801) is transferred to the nucleus in permeabilized cells in the presence of wild-type Trn1. Mutation of the hydrophobics W460A or W730A that are essential for the interaction with PY-NLSs fail to transport the ADAR1-NLS to the nucleus. (B) Mutations W460A and W730A in Trn1 fail to interact with ADAR1-NLS in pull-down assays. Western blot shows His-tagged Trn1 and GST-tagged dsRBD input material. GST–dsRBD fusions coupled to GST beads pull down wild-type Trn1 but none of the two mutant versions W460A or W730A.

We next modeled the interaction between Met708–Arg801 ADAR1-dsRBD3 and Trn1 using the six available X-ray structures of Trn1 bound to various NLSs as starting structures. Briefly, the structures were superimposed on the Trn1 proteins, and the different NLSs were then aligned according to their relative position in the structures (Fig. 5A). Next, the sequence of the N- and C-terminal fragments flanking ADAR1-dsRBD3 were manually aligned with the others such as to maximize sequence similarities while keeping a virtually extended peptide chain. The model was then constructed using docking protocols and simulated annealing refinements, as described in Materials and Methods.

Fig. 5.

Modeling the interaction with transportin 1. (A) Structure-based alignment of the six available structures of Trn1 in complex with NLS peptides from the following proteins: hnRNP A1, hnRNP D, hnRNP M, FUS/TLS, TAP/NXF1, and Nab2p (PDB ID codes 2H4M, 2Z5N, 2OT8, 4FDD, 2Z5M, and 4JLQ, respectively). The structures were superimposed on the Trn1 proteins (Right; for clarity, Trn1 proteins are not shown and only the NLS peptides are displayed as sticks), and the various NLS sequences were then aligned according to their relative position in the structures (Left). (B) Superposition of the structural model of the interaction between ADAR1-dsRBD3 and Trn1 with the different NLSs complexes (Left; for clarity, Trn1 proteins are not shown). The position of individual residues from the N- and C-terminal fragments of ADAR1-dsRBD3 are better seen on the Right where only ADAR1-dsRBD3 is shown using the same orientation. (C) Surface representation of the structural model of the complex between ADAR1-dsRBD3 in green and Trn1 in gray. ADAR1-dsRBD3 perfectly fits without any atomic clashes into the Trn1 C-terminal arch formed by HEAT repeats 8–20. (D) Schematic representation of the molecular contacts seen in the structural model between the N- and C-terminal fragments of ADAR1-dsRBD3 in green and Trn1 HEAT repeats 8–18. Within the C-terminal arch, site A (H8–H13) and site B (H14–H18) are shown in red and in blue, respectively. Molecular contacts are shown as yellow dashed lines. See also SI Appendix, Fig. S5.

Although this model is based on little experimental data and should therefore be interpreted with caution, it is of great interest for the understanding of ADAR1-NLS recognition by Trn1. It shows that the extended ADAR1-dsRBD3 fits remarkably well into the Trn1 C-terminal arch without any steric clashes (Fig. 5C), even if ADAR1-dsRBD3 is closely surrounded by Trn1 and cannot be freely rotated in this cavity. In addition, it illustrates how the flexible N- and C-terminal modules, in adopting an extended conformation, can imitate canonical PY-NLSs and interact with the C-terminal arch of Trn1 (Fig. 5B). In our model, part of the N-terminal module (Met708–Lys712) interacts with site B at repeats H18–H14, and Lys712–Arg714 interact with repeats H13–H11 from site A (Fig. 5D). The C-terminal module Glu797–Arg801 is exclusively interacting with site A at repeats H11–H8.

Most importantly, the model strongly supports our hypothesis that the structured part of ADAR1-dsRBD3 (residues Ile716–Asn796) would not be necessary for the formation of the NLS per se but, rather, simply acts as a scaffolding domain to form a functional NLS from distant parts of a polypeptide chain (as schematically illustrated in Fig. 5D). We therefore challenged our structural model further.

ADAR1-dsRBD3 Acts Only as a Scaffolding Domain of the Bimodular NLS.

If the structured part of ADAR1-dsRBD3 is dispensable for NLS activity and acts only as a scaffolding domain, one could in principle replace ADAR1-dsRBD3 with an unrelated dsRBD lacking NLS activity while keeping the N- and C-terminal modules that form the NLS. However, helix α_N should be kept to appropriately position the modules of the bimodular NLS with respect to each other. We therefore constructed a chimeric protein containing a dsRBD of canonical structure that shows no NLS activity (i.e., Xlrbpa-dsRBD2) (SI Appendix, Fig. S6 A and B) (22, 26), flanked by the N- and C-terminal fragments of ADAR1-dsRBD3, including helix α_N. In addition, we mutated three residues in α1 and α2 of Xlrbpa-dsRBD2 (i.e., S116G, T176R, and K179I) (SI Appendix, Fig. S6B) to allow the interaction of α_N with α1 and α2. Although Xlrbpa-dsRBD2 wild-type constructs show no NLS activity (SI Appendix, Fig. S6A), the chimeric construct constitutes a fully functional NLS (Fig. 6B), demonstrating that the sequence of the ADAR1-dsRBD3 is not important in itself for NLS activity. However, this chimeric construct cannot exclude that the sequence of helix α_N present in both constructs also plays a role (Fig. 6 A and B). To definitely prove that the entire ADAR1-dsRBD3, including helix α_N, is dispensable for the formation of a functional NLS, one should design a minimal peptide substituting the dsRBD, where the N- and C-terminal modules are appropriately spaced to mimic their relative position in ADAR1-dsRBD3. To achieve this challenging design, we used our structural model (Fig. 5) and, in particular, the sequence and structural alignment of ADAR1-dsRBD3 with all of the different NLSs (Fig. 5 A and B). Among all NLS peptides, we drew a particular attention to the fused in sarcoma/translocated in sarcoma (FUS/TLS) NLS because its central region adopts an α-helical structure forming approximately two helical turns (Gly515–Arg521; PDB ID code 4FDD) (38). The two helical turns could create a spacer that maintains the N- and C-terminal modules at the same distance as in the original dsRBD context (SI Appendix, Fig. S6D). We therefore designed a minimal putative NLS consisting of the flanking fragments of ADAR1-dsRBD3 linked by the two helical turns of the FUS/TLS NLS (SI Appendix, Fig. S6C). Although the FUS/TLS linker by itself lacks NLS activity (SI Appendix, Fig. S6 E and F), addition of the N- and C-terminal fragments fully restores NLS activity (Fig. 6C and SI Appendix, Fig. S6F), proving that ADAR1-dsRBD3 did not contribute to NLS activity, per se, but instead provided the appropriate spacing between the N- and C-terminal modules. This finding strongly validates our structural model and firmly establishes that ADAR1-NLS is a bimodular NLS formed by the combination of the N- and C-terminal fragments flanking ADAR1-dsRBD3.

Fig. 6.

The structured region of ADAR1-dsRBD3 acts as a scaffolding domain and is dispensable for ADAR1-NLS formation. Schematic representations of constructs tested (Left). ADAR1-dsRBD3 and Xlrbpa-dsRBD2 core domains are represented in gray and in yellow, respectively. The N- and C-terminal fragments are in blue and in red, respectively, the additional helix α_N is in purple, and the small linker from the FUS/TLS PY-NLS is shown in pink. Localization of the corresponding fusion constructs upon transfection visualized in confocal sections (Right). (A) Wild-type ADAR1-dsRBD3 construct (Met708–Arg801) can transfer a PK fusion construct to the nucleus via the bimodular NLS. The functional NLS is highlighted in cyan. (B) Chimeric construct between ADAR1-dsRBD3 N- and C-terminal modules (NTM and CTM, respectively) and Xlrbpa-dsRBD2 core domain can also mediate nuclear transport of a fusion protein. (C) Minimal NLS construct formed by adding a seven-residue linker of the FUS/TLS NLS (FUSlinker) to join the N- and C-terminal modules of ADAR1-dsRBD3 (NTM and CTM, respectively). The exact sequence of the minimal peptide is MMPNKVR GEHRQDR KAER, with the linker shown in italics and the N- and C-terminal modules from ADAR1-dsRBD3 in bold. This minimal NLS construct is also able to transfer a PK fusion construct to the nucleus. Localization of the fusion construct is shown in the FITC channel, DNA is shown in the DAPI channel, and cellular outlines are shown in the DIC channel. (Scale bar: 10 µm.) See also SI Appendix, Fig. S6.

Modulation of Transport by Nucleic Acid Binding.

Binding of ADAR1-dsRBD3 to Trn1 is strongly impaired in the presence of dsRNA, leading to weaker nuclear import (SI Appendix, Fig. S7A) (22, 23). However, the competition between Trn1 and dsRNA may result from: (i) a direct competition for the same residues of ADAR1-dsRBD3; (ii) a structural rearrangement of ADAR1-dsRBD3 upon dsRNA binding, impeding Trn1 binding; or (iii) a mutually exclusive binding of either Trn1 or dsRNA due to steric hindrance. To distinguish these possibilities, we investigated dsRNA binding of ADAR1-dsRBD3.

First, using mutational analysis, we tested whether positively charged residues from the flanking fragments are involved in dsRNA binding with isothermal titration calorimetry (ITC), using a 24-base pair duplex RNA substrate. Mutants in the N- or C-terminal modules retain the capacity to bind dsRNA (e.g., R801A in the C-term, ⁷¹²KVRK⁷¹⁵ to ⁷¹²AVAA⁷¹⁵ or M708A+P710A or a deletion mutant in the N-term (Thr725-Arg801) (Fig. 7A and SI Appendix, Fig. S7B). Conversely, lysine substitutions in the N-terminal tip of helix α2, which strongly impair dsRNA binding, do not affect nuclear localization of the construct (⁷⁷⁷KK⁷⁷⁸ to ⁷⁷⁷AA⁷⁷⁸) (Fig. 7 A and B). These findings indicate that Trn1 and dsRNA are not competing for the same residues of ADAR1-dsRBD3.

Fig. 7.

Two distinct interfaces on ADAR1-dsRBD3 bind Trn1 or dsRNA in a mutually exclusive manner. (A) Isothermal titration calorimetry of wild-type ADAR1-dsRBD3 Met708–Arg801 construct and mutants showing that important residues for Trn1 interaction are not important for dsRNA binding. The name of the construct is indicated above each titration. (B) A double mutation abolishing RNA binding by the dsRBD (K777A + K778A) does not interfere with nuclear localization of a resulting fusion protein and localizes like the wild-type ADAR1-NLS Met708–Arg801. Transfected PK fusion proteins are visualized in the FITC channel, nuclei are stained with DAPI, and cellular outlines are shown in the DIC channel. (Scale bar: 10 µm.) (C) Superposition on ADAR1-dsRBD3 of the structures of the two complexes, namely ADAR1-dsRBD3 bound to dsRNA and the model of Trn1 bound to ADAR1-dsRBD3. The surface of Trn1 is shown in gray, ADAR1-dsRBD3 is shown schematically in green, and the GluR-2 R/G stem-loop dsRNA helix is shown schematically in yellow. The dsRNA molecule extends across and strongly clashes with the C-terminal arch of Trn1. (D) Schematic representation of the two functional binding interfaces of ADAR1-dsRBD3 (Met708–Arg801) and the association of these nonoverlapping surfaces with their respective binding partner (i.e., Trn1 and dsRNA helices). These two binding sites cannot be occupied simultaneously due to the large dimensions of both Trn1 and dsRNA helices. See also SI Appendix, Fig. S7.

Second, to further confirm that the N- and C-terminal fragments of ADAR1-dsRBD3 are not involved in dsRNA binding, we titrated a ¹⁵N-labeled ADAR1-dsRBD3 into a typical ADAR1 substrate, GluR-2 lower stem-loop (LSL) (41), and vice versa (SI Appendix, Fig. S7C). We could thus map the regions of interaction (SI Appendix, Fig. S7D) that revealed a canonical mode of binding for ADAR1-dsRBD3 (26, 27). Importantly, the resonances of the N- and C-terminal fragments are not shifted upon dsRNA binding, confirming that they do not bind RNA (SI Appendix, Fig. S7D). Also, regions in proximity of the N- and C-terminal modules are not shifted, excluding that major structural rearrangements would prevent Trn1 binding. Together, our data suggest that binding of Trn1 and dsRNA by ADAR1-dsRBD3 involves different binding interfaces and that the mutually exclusive binding of either Trn1 or dsRNA might result from steric hindrance.

Indeed, when we model ADAR1-dsRBD3 bound to the GluR-2 R/G stem-loop (41) using our chemical-shift perturbation following a previously used procedure (42) (SI Appendix, Fig. S7E) and superpose this model onto the model of ADAR1-dsRBD3 bound to Trn1, one sees that steric hindrance prevents ADAR1-dsRBD3 from binding both molecules simultaneously (Fig. 7 C and D). These models explain well how dsRNA can inhibit nuclear import.

Discussion

ADAR1-dsRBD3 Is an Extended dsRBD Allowing a Dual Function.

Here, we show that ADAR1-dsRBD3 adopts an extended dsRBD structure. Similarly, other dsRBDs exhibit structural extensions (27), such as the budding yeast RNase III Rnt1p, which accommodates a long C-terminal helix α3 (43, 44), and the fission yeast dicer Dcr1, which accommodates a short C-terminal helical turn and a CHCC zinc-binding domain (36). In contrast to previously reported dsRBD extensions, the one described here occurs upstream of the canonical dsRBD. The additional helix α_N is conserved in all ADAR1 proteins of vertebrates from fish to mammals, as evidenced by the absolute conservation of the pattern of hydrophobic and hydrophilic residues in this region (SI Appendix, Fig. S8).

Helix α_N is positioned between helices α1 and α2, which is similar to what was recently found in Staufen1-dsRBD5 for the Staufen-swapping motif (SSM) that drives dimerization of the protein (45). Globally, the N-terminal helix of the SSM motif adopts a similar position and orientation to helix α_N in ADAR1-dsRBD3 (SI Appendix, Fig. S9). Importantly, it has been proposed that the dynamic nature of the 16-residue-long linker connecting the SSM subdomain to Staufen1-dsRBD5 would allow the SSM to interact both in cis leading to a monomeric form and in trans leading to dimerization (45). Although the structural feature is rather similar, the resulting functions are different, which underline the propensity of dsRBDs to acquire new functions via N- or C-terminal extensions (36, 45).

Although we demonstrated here that the primary role of helix α_N is to act as a scaffolding element for a bimodular NLS to interact with Trn1, one might ask whether helix α_N might also be implicated in RNA binding similarly to the additional helix α3 of Rnt1p (43, 44). Our NMR and ITC titrations (Fig. 7A and SI Appendix, Fig. S7 A–C) indicate that the contribution of helix α_N to RNA binding is very limited. In the structural model of ADAR1-dsRBD3 bound to RNA (SI Appendix, Fig. S7D), only Asn726 in the short kink joining α_N to α1 appears to be in the position to make direct contacts with RNA. Altogether, these points argue against a strong implication of α_N in RNA binding.

Comparison of the ADAR1-NLS with PY-NLSs.

We previously demonstrated that Trn1 acts as the unique nuclear import receptor of human ADAR1 via the segment Met708–Arg801, which encompasses the third dsRBD of the protein (23). In the present study, we identified the molecular determinants responsible for the interaction with Trn1 as the combination of the N- and C-terminal modules flanking the folded domain. These modules do not entirely comply with the collection of physical rules that describe the best-characterized substrates of Trn1, namely PY-NLSs (4, 46). Their similarities and differences are thus of great interest.

First, PY-NLSs are disordered peptide segments, a feature that, to some extent, is fulfilled by ADAR1-NLS. Indeed, the modules flanking the folded dsRBD are structurally disordered in the free protein (Fig. 1D). However, the bimodular organization of ADAR1-NLS is unique, with the two NLS modules separated by a small globular domain of about 9 kDa. The two modules are more than 80 residues apart in the primary structure (Fig. 1C). Second, PY-NLSs have an overall basic character, and, with five positive charges (Arg or Lys) within the two modules, ADAR1-NLS conforms perfectly to this rule. Finally, PY-NLSs display some weakly conserved sequence motifs, with some structural conservation that was classified as three so-called epitopes (9). Epitope 1 is composed of a stretch of either hydrophobic or basic residues interacting with Trn1 site B (Fig. 5). Our structural modeling and in vivo data suggest that ADAR1-NLS holds a hydrophobic epitope 1 composed of residues ⁷⁰⁸MMPN⁷¹¹ (Figs. 2A and 5D). Epitope 2 is composed of a single and well-conserved positively charged residue (Arg/Lys/His) interacting with a negatively charged pocket formed by repeats H11–H12. In ADAR1-NLS, epitope 2 might be constituted by Lys798 because its position is very similar to the conserved Arg of other PY-NLSs (Fig. 5 A, B, and D). Finally, epitope 3 is composed of a relatively well-conserved Pro-Tyr dipeptide interacting with Trn1 site A. There is no PY in the ADAR1 C-terminal module, but our structural model shows that residues ⁸⁰⁰ER⁸⁰¹ could occupy the corresponding Trn1 pocket and therefore represent a rather degenerated equivalent to epitope 3. Interestingly, although the PY energetic contribution is sometimes important and explains the sensitivity to mutations in this motif (e.g., hnRNP M, hnRNP D, and FUS/TLS) (9, 10, 38), the contribution of epitope 3 to the overall binding affinity is sometimes relatively small and epitope 3 is then tolerant to mutations (11). In the case of ADAR1-NLS, however, Arg801 is essential for Trn1 binding and is thus an important constituent for nuclear import.

Therefore, considering the molecular determinants of ADAR1-NLS that we uncovered here, the number of currently predicted Trn1-interacting NLSs is most likely underestimated. Indeed, even though it is assumed that each epitope can accommodate substantial sequence diversity, the limits are still vague. The fact that some characterized Trn1 cargos lack a PY-NLS suggests that the actual variability is larger than currently conceived (12–15). We believe that more bimodular NLSs, such as in ADAR1, will be found in the future and that our study will help identifying them.

Nuclear Import Regulation via an RNA-Sensing NLS.

A bimodular NLS where two active modules are interrupted by a small folded domain is an impressive molecular construction but seems rather complicated to generate a functional NLS. However, this architecture allows the corresponding NLS to be switched on and off depending on the presence of RNA associated with the intervening domain (in this case, a dsRBD) (Fig. 7 C and D) (23). Although the size of the intervening domain is crucial for this mechanism, because it should fit into the Trn1 C-terminal arch, it is conceivable that similar bimodular NLSs interrupted by other domains might exist among Trn1 cargos. Interestingly, in addition to this RNA-sensing NLS, other mechanisms to regulate Trn1-mediated nuclear import have been described. For instance, posttranslational modifications, such as arginine methylation (47) or cysteine oxidation via reactive oxygen species (48), have been recently reported to modulate Trn1-mediated nuclear import. In the case of ADAR1, the bimodular NLS is switched off upon binding to dsRNA substrates, which would prevent ADAR1 from carrying RNAs back into the nucleus. The fact that dsRNA stimulates nuclear export of dsRBD-containing proteins (23, 30, 32) suggests that ADARs might leave the nucleus bound to substrate RNAs: for instance, pre-miRNAs. The binding of the RNA-sensing NLS to Trn1 could then help the dissociation of ADAR1-associated RNAs in the cytoplasm and ensure an efficient reimport of ADAR1 free of RNAs to the nucleus.

Materials and Methods

Cloning, Expression, and Protein Purification for Structural Studies.

The DNA sequence encoding the third dsRBD of human ADAR1 (residues 708–801) (Uniprot entry P55265) was subcloned by PCR amplification from FLAG-His-tagged full-length ADAR1-c plasmid (23) between NdeI and XhoI cloning sites in Escherichia coli expression vector pET28a. The construct contains an N-terminal tag whose sequence MGSSHHHHHHSSGLVPRGSH includes a six-histidine stretch used for protein purification.

Mutagenesis experiments were performed using the Quikchange Kit (Stratagene) following the manufacturer’s instructions. All mutant proteins were checked to be properly folded by running (¹⁵N,¹H)-HSQC spectra.

Proteins were overexpressed in BL21(DE3) Codon-plus (RIL) cells in either LB media or M9 minimal media supplemented with ¹⁵NH₄Cl and ¹³C-labeled glucose. The cells were grown at 37 °C to OD600 ∼0.4, cooled down at 30 °C, and induced at OD600 ∼0.6 by adding isopropyl-β-d-thiogalactopyranoside to a final concentration of 0.5 mM. Cells were harvested 16 h after induction by centrifugation. Cell pellets were resuspended in lysis buffer [50 mM Tris⋅HCl (pH 8.0), 1 M NaCl, 20 mM imidazole, 1 mM DTT] and lysed by sonication. Supernatant was loaded on an Ni-NTA column on an ÄKTA Prime purification system (Amersham Biosciences), and the protein of interest was eluted with an imidazole gradient. The fractions containing the protein were pooled, dialyzed against the NMR Buffer [20 mM NaPi (pH 7.0), 100 mM KCl, 1 mM DTT], and concentrated to ∼0.9 mM using Vivaspin 5000 MWCO (Sartorius Stedim Biotech). All other ADAR1-dsRBD3 constructs and mutants were expressed and purified similarly.

NMR Spectroscopy.

All NMR spectra were recorded at 313 K on Bruker AVIII-500 MHz, AVIII-600 MHz, AVIII-700 MHz, AVIII-750 MHz, and Avance-900 MHz spectrometers (all equipped with a cryoprobe except for AVIII-750) with Bruker shaped tubes or 5-mm Shigemi tubes. The data were processed using TOPSPIN 3.0 (Bruker) and analyzed with Sparky (www.cgl.ucsf.edu/home/sparky/). Protein resonances were assigned using 2D (¹H,¹⁵N)-HSQC, 2D (¹H,¹³C)-HSQC, 3D HNCA, 3D HNCACB, 3D CBCA(CO)NH, 3D (H)CCH-TOCSY, 3D H(C)CH-TOCSY, 3D NOESY-(¹H,¹⁵N)-HSQC, and two 3D NOESY-(¹H,¹³C)-HSQC optimized for the observation of protons attached to aliphatic carbons and to aromatic carbons, respectively. In addition, the assignment of aromatic protons was conducted using 2D (¹H,¹H)-TOCSY and 2D (¹H,¹H)-NOESY measured in D₂O; histidine protonation and tautomeric form were determined from a long-range (¹H,¹⁵N)-HSQC spectrum (49). We recorded all NOESY spectra with a mixing time of 100 ms.

Protein Structure Determination.

Automated NOE cross-peak assignments (50) and structure calculations with torsion-angle dynamics (51) were performed using the macro noeassign of the software package CYANA 3.0 (52). Peak lists of the four NOESY spectra were generated as input with the program ATNOS (53) and manually cleaned to remove artifact peaks. The input also contained 49 hydrogen bonds. Hydrogen-bonded amides were identified as slowly exchanging protons in the presence of D₂O. Their bonding partner was identified from preliminary structures as well as from analysis of the characteristic NOE pattern found in α-helices and β-sheets. We calculated 100 independent structures that we refined in a water shell with the program CNS 1.21 (54, 55) including distance restraints from NOE data and hydrogen-bond restraints as previously described (36). The 20 best energy structures were analyzed with PROCHECK-NMR (56) and the iCING web server (57) (https://nmr.le.ac.uk/icing/). Overall structural statistics of the final water-refined structure are shown in Table 1. Structures were visualized, and figures were prepared with the program PYMOL (www.pymol.org).

Import Assays on Permeabilized Cells.

HeLa cells were grown on coverslips and treated with digitonin at a final concentration of 30 µg/mL for 5 min on ice. Cells were then incubated with recombinantly expressed GST-tagged protein and His-tagged purified Trn1 protein (both at 50 nM), 1 mM GTP, 1 mM ATP, 20 U/mL creatine phosphokinase, 5 mM creatine phosphate, 0.1% BSA, and 1 mM PMSF, in import buffer (58). After 30 min at room temperature, cells were washed in import buffer three times for 5 min and immunostained using polyclonal anti-GST antibody as described in Transfection and Immunolocalization and imaged on a Zeiss LSM710.

Pull-Down Assays.

Recombinant dsRBDs, mutant and wild-type, were expressed as GST fusion proteins from pET24d. Recombinant proteins were purified from E. coli CS41 using glutathione Sepharose as described (23). His-tagged Trn1 was cloned as described earlier (23). Purified GST–dsRBD constructs and His–Trn1 constructs were incubated with glutathione beads (Sigma) at a concentration of 75 nM in 500 µL of binding buffer [20 mM Hepes (pH 7.9), 110 mM potassium acetate, 2 mM magnesium acetate, 1 mM EGTA, 1 mM DTT, 1 mM PMSF, 20 mM NaCl, 5% (vol/vol) glycerol, 1% Triton X-100, 14 mM 2-mercaptoethanol] for 1 h at room temperature. Subsequently, the beads were washed three times in binding buffer and resuspended in SDS loading buffer. The coprecipitate was loaded on a 9% SDS/PAGE, blotted onto nitrocellulose. The blot was cut in half at the 70-kDa band. The upper part of the blot was detected with an anti-His antibody, detecting the Trn1 band whereas the lower part of the blot was detected with a GST antibody, detecting the dsRBD fusions.

Transfection and Immunolocalization.

Constructs of interest were cloned into pcDNA3 into which an AUG codon was engineered upstream of a unique HindIII site, followed by a pyruvate kinase 6× myc fusion (24). Fragments were inserted into the HindIII site generating a translational fusion with pyruvate kinase-myc. For a complete list of used primer pairs see SI Appendix, Table S2. Resulting plasmids were transfected into HeLa cells using Ca-phosphate (59) or polyethyleneimine-mediated transfection (Polyplus). Cells were fixed and stained 48 h after transfection using a monoclonal anti-myc antibody in combination with a secondary Alexa 488 secondary antibody and counterstained with DAPI as described (24). Microscopic confocal sections were taken on a Zeiss LSM710. Images were processed using Adobe Photoshop CS7.

Modeling the Interaction with Transportin 1.

The interaction between ADAR1-dsRBD3 and Trn1 was modeled using the six available X-ray structures of Trn1 bound to various NLS sequences as initial structures (PDB ID codes 2H4M, 2OT8, 2Z5K, 2Z5N, 4FDD, and 4JLQ) (8–10, 38, 39). The six structures were superimposed on the Trn1 proteins, and the various NLS sequences were then aligned according to their relative position in the structures (see Fig. 5A for the alignment). Then, the sequence of the N- and C-terminal fragments flanking ADAR1-dsRBD3 were manually aligned with the other NLSs to maximize sequence similarities while keeping a virtually extended peptide chain. We used this alignment, together with the structures of Trn1 bound to the various NLSs, to construct a set of highly ambiguous distance restraints to loosely constrain the N- and C-terminal tails to adopt an extended conformation along the path adopted by the NLS peptides in all crystal structures. The model building was then done as follow: (i) the bundle of 20 NMR structures of ADAR1-dsRBD3 (Met708–Arg801) was manually placed into the Trn1 canonical binding groove with the structure visualization program PYMOL to roughly position the N- and C-terminal fragments in proximity with the canonical NLS binding pockets (coordinates from the 2H4M structure); (ii) the structures were energy-minimized with a conjugate-gradient minimization and subsequently a rigid-body minimization with two rigid groups defined as one for Trn1 and one for ADAR1-dsRBD3; (iii) each of these minimized structures was subjected five times to a restrained simulated annealing protocol in implicit water using different seeds to generate initial velocities. The protocol consisted of 4 ps of dynamics at 1,000 K, followed by cooling to 25 K over 22 ps. Different types of restraints were applied for the different parts of the molecules: (i) the N- and C-terminal modules of ADAR1-dsRBD3 (i.e., residues Met708–Lys715 and Asn796–Arg801) were restrained only by the set of highly ambiguous intermolecular distance restraints described above; (ii) the backbone of the structured part of ADAR1-dsRBD3 (i.e., Ile716–Glu795) as well as the entire Trn1 atoms were treated as rigid groups allowing relative movements of those rigid blocks; and (iii) the side chains of the structured part of ADAR1-dsRBD3 (i.e., Ile716–Glu795) were harmonically restrained to their initial position, with a weak force constant allowing small side-chain rearrangement in the dsRBD. The 100 resulting complexes were finally energy-minimized, and the 10 best energy structures were pooled and analyzed with PYMOL. A representative structure was then selected for figure preparation (Fig. 5).

RNA Samples and Isothermal Titration Calorimetry.

To assess the nucleic acid binding properties of dsRBD constructs, we produced RNA substrates by in vitro transcription with T7 polymerase, namely a dsRNA duplex of 24 bp as previously described (36), and the GluR-2 R/G editing site stem-loop (GluR-2 LSL), a typical ADAR substrate used previously for structural studies with ADAR2 (41). RNAs were purified by anion-exchange HPLC under denaturing conditions as previously described (42). DNA templates were purchased from Microsynth AG.

ITC experiments were performed on a VP-ITC instrument (MicroCal) calibrated according to the manufacturer's instructions. The samples of protein and nucleic acids were prepared in and dialyzed against ITC buffer [25 mM NaPi (pH 7.0), 100 mM NaCl, 2 mM 2-mercaptoethanol]. The concentration of protein and nucleic acid was determined using OD absorbance at 280 and 260 nm, respectively. The sample cell (1.4 mL) was loaded with 2 μM of the dsRNA duplex; the concentration of ADAR1-dsRBD3 wild type and variants in the syringe was 80 μM. Titration experiments were done at 25 °C and typically consisted of 34 rounds of 8-μL injections. Data were plotted and analyzed using MicroCal Origin software version 7.0, using equations for a single-binding-site model.