Domain II has A-form helical properties
It is well-established that in addition to perfect dsRNA, PKR can bind and be activated by RNAs with complex secondary structural features, including internal loops, bulges, and pseudoknots, as well as single-stranded and non-Watson-Crick motifs.7,12-14,57
It may be the dynamic nature of many RNA structures that enables PKR to accommodate these structural imperfections; for example, A-bulge-induced bends in the middle of RNA helices are straightened upon PKR binding.57
Additionally, noncontiguous helices within RNA can combine to activate PKR.12,57
The unifying principle in activation of PKR by these various RNAs may be maintenance of an overall A-form, or dsRNA-like, topology. However, little direct testing of this idea has been conducted.
Domain II of HCV RNA contains four short helical regions, as well as internal loops, bulges, and mismatches, and yet has the ability to activate PKR (, ). We propose that PKR still binds and is activated by domain II, despite absence of long base paired stretches, because the overall conformation of several of these loop regions are largely A-form. We first make a comparison between L4 of domain II and a loop from another RNA known to facilitate binding of a dsRBD, followed by a comparison between the P2-P4 section of domain II and a model A-form helix.
We found p20-dependent protection from nuclease digestion at several nucleotides in the apical loop (L4) of domain II (). Mutation of this loop to an unstructured U7-loop resulted in decreased activation of PKR (). These data prompted a detailed inspection of the NMR structure of domain II39
, which revealed that the apical loop, L4, is fairly structured, with U80, A81, C84, and A85 turned in towards the helix. In addition, U80 and A85 interact via the 2′-hydroxyl of U80 hydrogen bonding with the N6 of A85, and U86 is extruded (). Thus, L4 resembles a tetraloop having sequence AGCC.
Fig. 10 Structural comparisons of domain II with AGNN loop and 19 bp dsRNA. (a) Secondary structures of domain II L4 and snR47h RNA apical loop employed for rmsd analysis. Numbering for snR47h is from the reference 62. At left, domain II structure contains a (more ...)
There is precedent for other dsRBDs to interact with RNA terminal loops, including those from Staufen, RNase III, Drosha, and ADAR 2.58-61
In particular, Feigon and co-workers found that the dsRBD of Rnt1p RNaseIII binds AGNN tetraloops, wherein the conserved A and G do not form sequence-specific contacts with the protein but rather help orient the remainder of the tetraloop to form non-sequence specific minor-groove contacts primarily via the NN nucleotides.59,62
We thus considered the possibility that L4 of domain II, which conforms to the AGNN motif, binds the dsRBD of PKR in a similar fashion.
A structural comparison between the NMR structures of AGCC from L4 and AGAA from Rnt1p-snR47h RNA complex59,62
was carried out (). We consider first positioning of the bases, followed by positioning of the sugar-phosphate backbone. Overall, the relative positions of the bases in the two tetraloops are quite similar: the A's at the 5′-end of the loop (A81 from domain II and A15 from snR47h) are oriented inwards and stack over the helix below; the G's at position 2 of the loop (G81 from domain II and G16 from snR47h) are flipped out of the loop; and the remaining two bases roughly stack over the 3′-end of the helix. In the AGAA tetraloop, the highly conserved G16 is in the syn
conformation about the glycosidic bond with the rare C2′-endo sugar pucker. These conformational characteristics do not lead the nucleobase of G16 to make direct contact with Rnt1p, but they do allow for exposure of the non-bridging phosphate oxygens between G16 and A17. While the corresponding G in L4 (G82) is anti
in most of the NMR structures entered into the pdb, the sugar pucker of G82 in the majority of the 12 NMR structures is also C2′-endo resulting in similar exposure of the non-bridging phosphate oxygens between G82 and C83 ().
Next, the sugar phosphate backbones of AGCC and AGAA were overlaid. We found an rmsd between the two backbones of just 1.57 Å (). Moreover, visual inspection of the overlay reveals good superposition of the 2′-hydroxyls and phosphates. Given that PKR interacts with 2′-hydroxyls and phosphates rather than base moieties,45
this suggests that the structured nature of the L4 loop may contribute to activation of PKR by domain II through a mechanism similar to that by which AGNN tetraloops are recognized by Staufen and RNase III.
It should be noted that the effects of the L4-L5 mutation to domain II on PKR activation are subtle, suggesting that PKR binding to the L4 tetraloop may not be critically dependent upon specific recognition of the structural details described above. Indeed, for Rnt1p RNase III, binding affinity for an AGNN versus a nonspecific tetraloop was not drastically different.59,62
The stem of the snR47 construct used in the Rnt1p-snR47h NMR structure contains only 15 bp, while in the NMR structure of Staufen dsRBD bound to a RNA tetraloop, the stem is only 14 bp. Given that the dsRBD motif typically requires 16 bp of dsRNA for binding, the minimal nature of the constructs used in these two RNA constructs may in fact “force” the dsRBD of Rnt1p and Staufen to interact with structured loop sequences. This may also be the case for interaction of L4 from HCV IRES domain II with PKR, in which the absence of 33 bp within the stem of domain II required for functional dimerization of PKR necessitates binding to the structured tetraloop. Thus, HCV IRES domain II may be an example of a biological RNA that presents a minimal construct for PKR activation.
Next, we turn our attention to the portion of domain II between P2 and P4. As mentioned, it is possible that activation of PKR by domain II is facilitated by A-form structural mimicry. In fact, the loops in this region of domain II are largely symmetrical—L2 is a 2×2 loop and L3 is a 3×4 loop—opening the possibility for nearly bulge-free non-Watson-Crick base pairing. Indeed, the NMR structure of domain II reveals a stretch of noncanonical base pairs within L2 and L3.39
In particular, L2 consists of two pyrimidine-pyrimidine base pairs, U63•C104 and U64•U103, while L3 contains three non-Watson-Crick base pairs, A73•A93 (AA N7 symmetric), A72•U95 (AU Hoogsteen), and G71•A96 (sheared GA). In fact, the only base in L2 or L3 that does not directly participate in base pairing is G94, which was the only base within L3 that was not protected from p20-dependent in-line cleavage (). These observations suggest that, despite their non-canonical nature, PKR may recognize the base pairs in L2 and L3 as dsRNA. We do note, however, that the identity of the bases in these loops seems to be relatively unimportant, as substitutions in L2 and L3, which maintained their symmetrical geometries, led to only minor effects on activation.
To further explore A-form mimicry within L2 and L3, we overlaid the P2-P4 portion of domain II with a crystal structure of a model 19 bp A-form dsRNA63
(). The backbones of these two structures gave an overall rmsd of 3.03 Å. In particular, the lower portion of these structures, which contains L2, overlays very well. Visual inspection of this region () reveals very similar positioning of the 2′-hydroxyls within L2 and model dsRNA, the only significant difference being the width of the two helices, which is ~ 11 Å between 2′-hydroxyls for the A-form helix and ~9 Å for L2, induced by the smaller pyrimidine-pyrimidine base pairs.64
The deviation between domain II and model dsRNA is greater near L3, likely due to the curvature of domain II induced by L3. Visual inspection of this overlay region, however, still suggests that the 2′-hydroxyl group positions within L3 are highly similar to those of an overall A-form geometry, with only a few exceptions (, lavender residues). One exception in particular, G94 (, arrow), is not protected by p20, as described above. In summary, domain II from P2 through P4, appears to mimic A-form dsRNA, and contribute a potential 22 A-form-like base pairs.
Below P2 are L1 and P1. In contrast to L2 and L3, L1 is a 5-nucleotide bulge, which leads to a kink in the backbone, although this region is likely flexible as is characteristic of large bulges. The role of L1 in activation is unclear. Protections mapped onto the NMR structure in suggest that the dsRBMs of a PKR dimer interact extensively with domain II, with the exception of L1; also, L1 can be deleted with just a slight loss in activation. Thus L1 likely plays little direct role in activation, although it remains somewhat surprising that it can be tolerated in this activating element.
At the base of domain II is the P1 pairing, which contributes 8 Watson-Crick base pairs. Together with the 22 A-form-like base pairs from P2 through P4, this gives 30 base pairs total. Assuming that L4 interacts productively with p20, as proposed above, an effective total number of base pairs is ~33, which is the minimum number needed for activation of PKR.4
Similarly, addition of base-paired segments leading to this value were found for the IFN-γ mRNA pseudoknot.12
Thus the ability of shorter base paired segments and symmetrical loops to sum to a PKR activating total appears to be a common theme.
Lastly, we briefly consider the weaker activation of PKR by domain III-IV of HCV IRES. This domain contains more extensive helical regions than domain II (), yet it is a less potent activator of PKR (). We suggest that this may be due to the branched nature of this domain, which contrasts with the unbranched domain II. In particular, structural motifs within domain III of the IRES appear to resemble motifs in well-studied PKR-inhibitory RNA structures. Specifically, the geometries of the four-way junction from IIIabc and the three-way junction comprised of the branched IIId stem loop from HCV IRES () parallel portions of the inhibitory Epstein-Barr EBER1 RNA and adenovirus VAI RNA.65,66
EBER1 contains a 4-way junction comprised of two stem regions and two branched stem-loops, with one shorter than the other, while the central domain of VAI RNA contains a three-way junction that is critical for PKR inhibition.16,67
Thus, the two multi-junction regions in domain III may cause PKR to adopt a combination of activating and inhibiting binding geometries, leading to its poorer activation.
Possible roles for PKR in HCV replication
It is known that NS5A regulates PKR activation throughout the lifecycle of HCV. Observation that NS5A can equally inhibit activation of PKR by domains II and III-IV suggests possible models for involvement of these IRES domains in PKR activation. In , we present possible modes for regulation of host and virus translation through PKR and NS5A interaction. Early in the viral lifecycle, when NS5A levels are low, PKR could be activated by the IRES leading to phosphorylation of eIF2α and inhibition of cap-dependent host translation. However, viral translation, which has been shown to function independent of PKR phosphorylation,32,68,69
would continue unabated, leading to high levels of viral proteins (, column 2). Later in the viral lifecycle, when NS5A levels have increased, PKR activity would be inhibited, leading to upregulation of host translation but still allowing high levels of viral translation. In this way, early in infection translation would be predominantly of viral proteins, but host proteins would be made available to the virus for later stages of replication.
Mediation of host and viral translation through NS5A-PKR interactions
We demonstrated that NS5A binds tightly to domains III-IV but only weakly to domain II. One possible consequence of this is that low levels of NS5A could be sequestered by binding to domains III-IV, allowing PKR activity to remain high via activation by domain II. In fact, the spacing of the bands in , suggest that domains III-IV can titrate 3 or more NS5A proteins, consistent with the presence of 5 GU-rich elements in this region.50
Only later in infection, when NS5A levels become high enough to exceed the concentration of binding sites, would PKR be bound by NS5A and be inhibited. Weaker binding of NS5A to PKR than RNA is suggested by the 50% values for RNA binding and PKR inhibition ( and ). This model for the interaction of PKR, NS5A, and IRES elements is consistent with observation that domain II is required for both replication and translation, whereas domains III-IV are only required for translation. Indeed, it has previously been suggested that domain II could function as a switch between translation and replication.26
Given the evidence here presented, it is possible that this domain II regulation of HCV translation and replication is mediated by differential interaction with PKR and NS5A.