Structural Analysis of dsRNA Binding to Anti-viral Pattern Recognition Receptors LGP2 and MDA5

1Homologous double-stranded RNA (dsRNA) dependent ATPases RIG-I, MDA5, and LGP2 (RIG-I-like helicases [RLHs]) are key cytosolic pattern recognition receptors in the vertebrate innate immune response against RNA viruses. RIG-I senses primarily 5′ppp blunt-end dsRNA (5′ppp-dsRNA), whereas MDA5 is activated by long dsRNA. Consequently, the two sensors respond to different but overlapping sets of viruses. Both activated receptors trigger the same downstream signaling pathway, leading to interferon (IFN) induction. RIG-I and MDA5 possess tandem N-terminal caspase activation and recruitment domains (CARDs), a central DECH-box helicase domain, and a C-terminal domain (CTD). LGP2 differs in lacking CARDs (Figure 1A) and thus independent signaling activity.

2Extensive studies of RIG-I and MDA5 have elucidated their mode of RNA binding and activation and signaling mechanisms. RIG-I binding to short 5′ppp-dsRNA via its CTD and helicase domains releases the CARDs from an auto-inhibitory state, in an ATP-dependent manner. This allows them to interact with and oligomerize, in a poly-ubiquitin-dependent fashion, downstream signaling partner MAVS. By contrast, MDA5 binds co-operatively to long dsRNA to form protein-coated filaments; the resulting oligomerization of MDA5 CARDs activates MAVS.

3LGP2 is reported to be both a positive and negative regulator of the anti-viral response. A positive role for LGP2 in anti-viral signaling is supported by the higher susceptibility of LGP2 knockout mice to certain RNA viruses and co-operative activity with MDA5. Indeed, small amounts of LGP2 enhance MDA5-mediated signaling, and a picornavirus-derived MDA5 agonist was found through its interaction with LGP2. Furthermore, LGP2, like MDA5, is specifically targeted for inactivation by paramxyovirus V protein. A negative role for LGP2 emerged from inhibitory activity observed upon overexpression of LGP2 (,, and references therein).

4To gain further insight into the role of LGP2 in the anti-viral response and its co-operative role in MDA5 signaling, we performed structural, biochemical, and cell-based studies on chicken (ch) LGP2 and MDA5. Interestingly, chicken and another Galliforme, turkey (according to its draft genome), both lack a RIG-I gene, unlike other vertebrates, including many birds. Nevertheless, chicken cells express MDA5, LGP2, and MAVS and produce type I IFN in response to highly pathogenic avian influenza virus, most likely via chMDA5 in co-operation with chLGP2. We present the crystal structure of chLGP2 bound to 5′ mono-phosphate (5′p), 5′ tri-phosphate (5′ppp), and 3′ overhang dsRNA at, respectively, 1.5, 2.2, and 2.0 Å resolution and characterize the RNA binding and ATPase activity of chLGP2 and human (h) LGP2. We also report crystal structures of 1:1 and 2:1 CARD-deleted chMDA5-dsRNA complexes at, respectively, 2.60 and 2.75 Å resolution. We demonstrate by electron microscopy (EM) that both chMDA5 and chLGP2 make filaments with dsRNA with the same axial repeat. Finally, functional studies reveal that LGP2 enhanced poly(I:C)-dependent MDA5 signaling, in both chicken and human cells, requires an intact RNA binding site on both the LGP2 helicase and CTD domains.

5We determined three co-crystal structures of full-length chLGP2 with the ATP transition state analog adenosine 5′-diphosphate:aluminum fluoride (ADP:AlF4) and either a 10-mer palindromic 5′p dsRNA (chLGP2_10p) or 5′ppp dsRNA (chLGP2_10ppp) or a 5′ppp and 3′ two nucleotide (GG) overhang hairpin RNA duplex (chLGP2-3ovg) at, respectively, 1.5, 2.2, and 2.0 Å resolution. A fourth structure of chLGP2 bound to a 12-mer palindromic dsRNA has no bound nucleotide. See Table 1 for crystallographic details.

6Overall, chLGP2 resembles other dsRNA-bound RLHs with the two RecA-like helicase domains (Hel1, Hel2), helicase insertion domain (Hel2i), pincer domain (P), and CTD wrapping around the dsRNA stem. Hel1 contains the conserved helicase motifs Q, I, Ia, Ib, Ic, II, IIa, and III, whereas Hel2 contains motifs IV, IVa, V, Va, and VI (Figures 1A and 1B; see also Figure S1 for the secondary structure of chLGP2 and sequence alignment with other RLHs). The unusually high resolution of the structures reveals details of the ADP:AlF4 binding site and the highly hydrated chLGP2-dsRNA interface. One distinctive feature of LGP2 is that the second pincer domain helix (α19) has eight turns and connects directly to the CTD, whereas in RIG-I, α19 has only six turns, followed by an extended, proline-rich connecting peptide (Figure 1B). Thus in LGP2, the CTD-pincer linkage appears to be more constrained and lacking the functionally important flexibility observed in RNA free RIG-I. The structure of chLGP2-dsRNA-ADP:AlF4 complex is in a highly ordered, closed conformation, mimicking the transition state for ATP hydrolysis, and thus most closely resembles the duck RIG-IΔCTD-dsRNA-ADP:AlF4 complex (dRIG-I; Protein Data Bank [PDB]: 4A36) in the disposition of the domains, except that the CTD is present as well. In the chLGP2 structure without nucleotide, the helicase is in a semi-closed state with Hel2 separated from Hel1 and partially disordered. See the Discussion for an analysis of the different nucleotide-dependent conformations observed for RLHs.

7ADP:AlF4 is tightly bound at the interface between the chLGP2 Hel1 and Hel2 with conserved motifs Q, I, and II from Hel1 and motifs Va and VI from Hel2 engaged in the interaction (Figures 1C and S2A). The adenosine base is partially stacked between His4 (motif Q) and LGP2 conserved Arg32 (motif I), and base-specific interactions are provided by the main-chain carbonyl of Glu2 and side chain of Gln7 (motif Q). This differs from RIG-I, where the base stacks between motif Q Arg244 (hRIG-I) and motif I Phe272 (equivalent to His4 and Arg32 in LGP2, respectively). The ADP ribose hydrogen bonds with conserved Glu67, which itself is stabilized by a salt bridge with Arg32 (Figures 1C and 1D). Glu67 emerges from a kink in helix α3 (Figure 1D), a feature that is conserved in MDA5 (see below; Figure S1), but not RIG-I, where the equivalent helix is straight and not involved in nucleotide binding (Figure 1D). Motif I (25-PTGAGKTR-32) wraps around the α- and β-phosphates of the ADP, making several hydrogen bonds. Motif Va (440-EEGLD-444) and motif VI (467-QGRARA-472) directly interact with ADP:AlF4, a characteristic of the closed form. Motif Va Asp444 interacts with the ADP ribose and stabilizes Arg471 of motif VI. Arg471 interacts with the α-phosphate and two fluorines of AlF4. Arg469, stabilized by Gln465, interacts with a third fluorine, mimicking stabilization of the transition state (Figures 1C and S2B). The high-resolution structure reveals the complex hydrogen bonding network involving Asp131 and Glu132 (motif II), AlF4, Mg2+, and water molecules in the presumed transition state. The octahedrally coordinated Mg2+ ion ligates one oxygen atom of the β-phosphate, two fluorine atoms, and three water molecules. Two magnesium coordinated water molecules form direct contacts with the side chains of Thr31 (motif I) and Asp131 (motif II) (Figures 1C and S2B). One water molecule coordinated by Glu132 (motif II, DECH) and Gln465 (motif VI) likely represents the attacking nucleophilic water, which catalyzes ATP hydrolysis (Figures 1C and S2B).

8According to, hLGP2 has significant basal (dsRNA-independent) ATPase activity. We find that both chLGP2 and hLGP2 ATPase activity is strictly dsRNA dependent (Figure 1E). Unlike MDA5, the activity is not dsRNA length dependent (Figures S2C and S2D). Moreover the ATP catalytic efficiency, as indicated by kcat/Km, is much lower for both chLGP2 and chMDA5 than for dRIG-I (Figures 1E and S2C). The importance of the interaction of Glu67 with the nucleotide ribose is underlined by the abolished ATPase activity of a chLGP2 K66A/E67A mutant (Figure 1E). Within the highly conserved motif I sequence (25-PTGAGKTR-32, chLGP2) the fourth position differs, being C/S in RIG-I, S in MDA5, and A/G/S in LGP2 (Figure S2E), and an A28C mutation abolishes the ATPase activity of chLGP2 (Figure 1E). Glu132 (motif II) and Gly468 (motif VI; Figure S2F) are absolutely conserved and crucial for ATP hydrolysis by RIG-I and MDA5, correlating with their mutation underlying genetic disease. Accordingly, E132Q or G468S mutations abolish the ATPase activity of chLGP2 (Figure 1E).

9Because of the full closure of all domains around the dsRNA (Figure 2A), many residues of chLGP2 make direct interactions to the dsRNA. These are detailed in Figure 2B, in which numerous water-mediated protein-RNA or direct water-RNA interactions are omitted for clarity. The protein-RNA interface buries a total solvent accessible surface of 4,700 Å2 (10p) or 4,786 Å2 (10ppp), as compared with 3,109 Å2 for hRIG-I in the semi-closed state (PDB: 5E3H [3TMI]). RNA binding residues come from the CTD and the conserved motifs Ia, Ib, Ic, and IIa from Hel1 and IV, IVa, and V from Hel2. In addition, Hel2i and the pincer domain interact with the dsRNA. Hel2i residues from helix α10 (Gln256, Gln260, Arg261, Glu264, and Asn267) as well as Arg285 from helix α11 interact with both dsRNA strands via the minor groove. Arg486 and Arg490 from the long helix α18 of the pincer domain also interact directly with the phosphate backbone of the dsRNA 5′ strand. The above-cited interacting residues are conserved or conservatively substituted polar residues in LGP2 from different organisms (Figure S3).

10Within the full-length chLGP2 structure, the CTD interacts with the dsRNA extremity in a similar fashion to previously reported for the isolated hLGP2 CTD (PDB: 3EQT), with the terminal base pair (5′-G1:C1∗-3′) making extensive hydrophobic interactions with aromatic residues Phe595, Phe599, and Trp602 from the 591–602 “capping loop” between strands β18 and β19 (Figure 2C). This loop also makes significant contacts with Hel1 in the vicinity of motif Ib, thereby reinforcing the closure of LGP2 over the RNA. The 3′ end interacts mainly with the turn between strands β16 and β17 (the 571–574 “3′ end-binding loop”; Figure S1), with fully conserved Glu571 hydrogen bonding with both hydroxyls of C1∗ and His574 bridging between the C1∗ and G2 riboses (Figure 2C). These interactions would appear to block extension of the 3′ strand of the dsRNA (but see below). The 5′ end interacts with both the CTD and the extended 402–413 “Hel2-loop” (Figure S1), which emerges between strand β10 and helix α16 and crosses the major groove of the dsRNA (Figure 2C). In chLGP2, the G1 base is imperfectly stacked between His406 and tilted Phe595, whereas the equivalent residues in hLGP2 are aliphatic, Asn408 and Ile597, respectively. Interestingly, an aromatic residue (His or Tyr) is found at position 406 in the Hel2-loop only in bird, reptile, and amphibian LGP2s (i.e., not in mammals or fish, in which it is usually an Asn), whereas only birds (with a few exceptions) and frogs have an aromatic residue at position 595 (Figure S3). The deletion of the Hel2-loop significantly affects ATP hydrolysis by hMDA5. We found that the H406A mutation in chLGP2 also reduced ATP hydrolysis but did not significantly change the affinity to RNA (Figures 1E and 3B). The CTD of LGP2 contains a lysine-rich motif (644-KKKYKKWS-651), with highly conserved Lys648 and Lys649 contacting phosphates of both strands across the major groove (respectively A7∗, U8∗, and C5) and conserved Trp650 to U3 and A4 phosphates (Figure 2B). The K648/K649E double mutation reduces chLGP2 affinity to dsRNA 56 times compared with wild-type (WT). A quadruple mutant with the additional glutamate substitutions of the conserved helicase residues Lys138 and Arg490 that interact with consecutive 5′ strand phosphates (Figure 2B) has ∼1,500 times less affinity, while the helicase mutations alone reduce the affinity by only 2.7 times (Figure 3B). This confirms the strong direct binding of the CTD to the RNA binding compared to the helicase domain and reveals the cooperativity of dsRNA binding to the two domains.

11An intriguing observation is that the blunt end of the dsRNA is located 1 bp deeper into the chLGP2 CTD than occurs in RIG-I (Figure 3A). This is not because the entire CTD is raised up but because the capping loop of LGP2 is flat, whereas that of RIG-I bends downward placing the side chain of Phe853 at the same level as the blunt-end base pair in LGP2 (Figure 3A). This shifts all the canonical contacts between helicase motifs and backbone phosphates 1 bp down from the blunt end in LGP2 compared with RIG-I, with motifs Ib and Ic contacting phosphate 1 and phosphate 2 of the 3′ strand in RIG-I and LGP2, respectively (Figure 3A). Consequently, the overall footprint of LGP2 on dsRNA is 10 bp for LGP2 rather than 9 bp for RIG-I, consistent with the larger surface area of protein-RNA contact for LGP2.

12To further characterize RNA binding to chLGP2, we determined structures with the 5′ppp form the same 10-mer dsRNA (i.e., the cognate RIG-I ligand) and a 5′ppp hairpin dsRNA with 3′ GG overhang at 2.2 Å (chLGP2-10ppp) and 2.0 Å (chLGP2-3′ovg) resolution, respectively (Table 1). Note that the hairpin stem has some base pair differences from the 10-mer dsRNA beyond the second base pair (Figure S4). We also measured the binding affinity of chLGP2 to different RNAs with and without various nucleotides and compared the results with hRIG-I (Figure 3B).

13In the chLGP2-10ppp structure, the 5′ end has a unique conformation (corresponding to one of the chLGP2-10p conformations), with the 5′ppp being accommodated without structural changes. The α- and β-phosphates interact with His406, Ser407, and Asn408 from the Hel2-loop, and the β- and γ-phosphates interact with Lys634 (Figure S4A). An octahedrally coordinated magnesium ion directly interacts with the α- and β-phosphates as well as the phosphate of G2 (Figure 2C). The 5′ppp conformation in the chLGP2-3′ovg structure is different and resembles the other 5′ end conformation observed in the chLGP2-10p structure. The tri-phosphate makes alternative interactions with Asn408, Arg607, and Lys634, and there is no bound magnesium (Figure S4B).

14The chLGP2-3ovg structure is overall similar to the previous structures, the major difference being a shift of the first G1-C1∗ base pair toward Hel1, a slight shift in the opposite direction of the CTD (including the capping and 3′ end-binding loops), and a significant rearrangement of the β3-α4 loop (residues 79–90), which would otherwise clash with the additional bases (Figure 2D). The net result is to disengage Glu571 and His574 from the ribose of C1 allowing 3′ extension of the RNA by extrusion of overhang nucleotide G-1∗ between the capping loop and the refolded β3-α4 loop. The G-1∗ base is stacked between the side chains of Arg600 and Lys87, and the resulting conformation of the β3-α4 loop permits G-1∗ and G-2∗ to make numerous interactions with residues 79–90, including a number of base-specific interactions (Figures 2E and 2F). This together with the burial of the G-2∗ ribose, whose 3′ OH makes a hydrogen bond to the phosphate of C1∗, suggests that the observed structure may be specific to this particular 3′-GG overhang but does show how relatively minor structural perturbations allow extrusion of a 3′ overhang, while extension of the double helical stem is prevented by maintenance of the stacking of the capping loop over the first base pair.

15The affinities of LGP2 to various dsRNA ligands were determined by fluorescence anisotropy (Figure 3B). chLGP2 binds to a blunt-end 12-mer dsRNA with an affinity independent of whether the 5′ is a hydroxyl or a tri-phosphate (Figures 3B and S5B), consistent with the lack of significant additional interactions to the tri-phosphate (and its variable conformation) observed in the crystal structures. By comparison, hRIG-I, as expected, binds to 5′ppp more tightly than to 5′-OH dsRNA, a result of the interactions of the 5′ppp with three lysines and a histidine, all strictly conserved. In addition, while hRIG-I’s affinity to 5′ppp dsRNA depends on which nucleotide is also bound, being highest for ADP:AlF4 (Kd ∼ 1 nM) and 20 times weaker with ADP or no nucleotide (Figure 3B), the affinity of chLGP2 for dsRNA is independent of bound nucleotide (Figure 3B). Finally, consistent with the 3′-ovg structure, the affinity of chLGP2 for 3′ and 5′ overhangs is only marginally reduced by a factor of two. In comparison, hRIG-I can also bind 3′ overhangs with high affinity (provided there is a 5′ppp), but 5′ overhangs are far less well accepted.

16In the absence of endogenous RIG-I, it is thought that chMDA5, perhaps in co-operation with chLGP2, is responsible for the RNA dependent anti-viral response to viruses that are usually detected by RIG-I. We therefore determined the crystal structure of chMDA5 (see Figure 4A for the domain structure) and investigated its ability, as well as that of chLGP2, to form filaments with dsRNA.

17For co-crystallization, we used 5′p palindromic duplex RNAs, the ATP analog AMPPNP, and a construct denoted chMDA5ΔCARD-Q, comprising the helicase and CTD (residues 298–994) with the mutation E436Q and lacking seven residues at the C terminus. The E436Q mutant in conserved motif II (i.e., DECH becomes DQCH) virtually abolishes ATPase activity and for another DEAD-box helicase, VASA, permitted trapping of the helicase substrate RNA. Unlike for the hMDA5 crystal structure the long, acidic loop between helixes α12 and α13 of Hel2i, which is 17 residues shorter in chMDA5, was not deleted (Figure S1). The structure of a 1:1 complex of chMDA5ΔCARD-Q with ADP:Mg2+ and 10 bp dsRNA (chMDA5-10) and a 2:1 complex with 24 bp dsRNA (chMDA5-24) were solved by molecular replacement at 2.60 and 2.75 Å, respectively (Table 2), as well as a 2:1 complex with 27 bp dsRNA at 7.2 Å resolution.

18Overall, the chMDA5-dsRNA structures resemble that of hMDA5 bound to a 12-mer dsRNA (PDB: 4GL2; root-mean-square deviation of 1.16 Å for all Cαs) (Figure 4) but with considerably higher resolution (hMDA5 is at 3.56 Å), and the model is more complete. Interestingly, the hMDA5, chMDA5-10 (Figure 4B), chMDA5-24 (Figure 4E), and chMDA5-27 structures all capture the dsRNAs bound in different longitudinal positions. The chMDA5-24 and chMDA5-27 structures reveal two different 2:1 complexes with two MDA5 molecules bound to the same dsRNA. Following, looking perpendicular to the RNA axis, we denote the head end of MDA5 as the pincer domain containing face and the tail end as that containing the CTD (see Figures 4B and S1 for the secondary structure of chMDA5 and sequence alignments).

19In all chMDA5-dsRNA structures, the helicase is in the semi-closed conformation (Figures 4B and S6; Table S1), but unlike the semi-closed forms of RIG-I and LGP2, Hel2 is well ordered. In the two higher resolution structures, the electron density in the nucleotide binding site clearly corresponds to ADP:Mg2+, presumably because of slow AMPPNP hydrolysis. The ADP:Mg2+ is bound in canonical fashion by motifs Q and I, with the adenosine stacked between Arg330 (motif I) and Arg302 (motif Q) (Figure 4C). This arrangement is more similar to that in chLGP2 than RIG-I, with helix α3 being kinked toward the nucleotide binding cleft in the same way, allowing Glu369, stabilized by Arg330, to hydrogen-bond to the 3′ OH of ADP (corresponding to Glu67 and Arg32 in chLGP2; compare Figures 1D and 4D).

20MDA5 binds to the dsRNA stem using the canonical helicase motifs Ia, Ib, Ic, IIa, IV, IVa, and V as well as conserved glutamines Gln568 and Gln572 from α10 of Hel2i, in a similar manner to other RLHs (Figure 5). The distinguishing feature of MDA5 is that its CTD interacts intimately with Hel2i and is displaced relative to the position in LGP2 or RIG-I to allow a dsRNA helix to pass through the molecule. Thus, the domain arrangement in MDA5, at the tail end, resembles an open horseshoe rather than a closed circle (compare Figures 5A and 2A). The Hel2-loop (729–740 in chMDA5, notably residues His733, Asn734, and Lys738; Figure S1), senses the major groove of the dsRNA, as in LGP2 and RIG-I, but without making any specific interactions (Figure 5C). The CTD 3′ end-binding loop (898-ENMH-901 in chMDA5) contacts the backbone of the 5′ strand, without blocking extension of the 3′ end, and the conserved 977–980 turn, toward the end of the CTD, also makes backbone interactions to both strands in the major groove (Figure 5C). Another major distinctive feature of MDA5 is the position and nature of the CTD “capping loop” (918–927 in chMDA5; Figure S1). While in all MDA5 crystal structures with dsRNA this loop is disordered, it is well structured in the isolated hMDA5 CTD structure (PDB: 3GA3), revealing it to be more compact than in LGP2 or RIG-I. Moreover, the MDA5 loop, unlike those of LGP2 and RIG-I, lacks the bulky hydrophobic residues that would favor interaction with a blunt end base pair. Modeling of the loop from the superposed hMDA5 CTD structure suggests that when binding to a continuous dsRNA the capping loop and N-terminal end of α16 of Hel2 could favorably interact with the minor groove and extended 5′ strand backbone, respectively (Figure 5D).

21In the chMDA5 24-mer structure, two MDA5 molecules are stacked head to head on one dsRNA, with a two-fold symmetry axis perpendicular to the dsRNA and between the two molecules (Figures 4E and 6A); additionally there is a 35° bend in the dsRNA axis (Figures 6A and 6B). In the chMDA5 10-mer structure, there are two 1:1 complexes in the asymmetric unit, which are arranged exactly as in the 24-mer structure except that the dsRNA is not continuous, lacking one base pair between the molecules (Figures 6A and 6B). The hMDA5 12-mer structure has two 1:1 complexes in the asymmetric unit, but with no pseudo-continuous RNA helix running through multiple MDA5 molecules. Compared with the chMDA5 10-mer structure, the hMDA5 12-mer structure has three extra base pairs at the tail end but one fewer at the head end (Figures 6A and 6B). By comparison, the chMDA5 24-mer structure gains 1 bp at the head end (making the dsRNA continuous between the two molecules) and 1–2 bp at the tail end (because the 24-mer structure probably corresponds to a superposition of two structures with a shift of 1 bp). More base pairs at the tail end correlate with a slight movement of the CTD toward the dsRNA. The bend in the dsRNA axis in the 1:1 and 2:1 chMDA5 structures allows two MDA5 molecules to pack closely together in head-to-head fashion, with a center-of-mass to center-of-mass distance of 43.1 Å and total buried protein-protein surface of 1,961 and 1,907 Å2, respectively. The head-to-head interface involves two-fold symmetric interactions between the pincer domain of one molecule and two protruding loops of the other molecule (Figure 6C). The beginning of the pincer domain (812-SGSAVER-817) interacts with the loop preceding Hel1 helix α9 (489-RSNS-492), which protrudes from the surface of the neighboring molecule. The end of the pincer domain (852-LQSI-855) interacts with the loop (562-KSE-564) between helices α11 and α12 of the second molecule. Note that residues V816 and E817 in chMDA5 are equivalent to I841 and E842, implicated in the head-to-tail interaction in hMDA5.

22In the chMDA 27-mer complex, the dsRNA is straight. Moreover, compared with the 24-mer complex, the two MDA5 head-to-head molecules are more distantly separated by 48.7 Å, and their relative rotation differs, with the second molecule of the 27-mer dimer being rotated a further 106° around and translated 10.6 Å further along the dsRNA (Figures 6A and 6B). Only regions 852–855 and 562–564 are close enough to make contact, thus reducing the total buried surface area to 484 Å2. Thus, whereas in the case of the 24-mer 2:1 complex, stronger protein-protein interactions are able to distort the path of the dsRNA, the 27-mer structure paradoxically suggests that chMDA5 prefers to bind at the end of a short dsRNA, even if this means reducing protein-protein interactions. However, it cannot be ruled out that the observed arrangement is dependent on crystal contacts.

23To test the ability of chLGP2 to form filaments on continuous dsRNA, complexes with Φ6 2,948, 4,063, and 6,374 bp long dsRNA were studied by EM. When mixed with a protein/RNA ratio of 0.2:1 in the presence of 2 mM ATP and imaged using negative stain, chLGP2 forms short and discrete multimeric clusters on Φ6 dsRNA (Figure 6D). At a ratio of 1:1, corresponding to one protein per 15 bp, as in hMDA5 polar filaments, more extensive and regular dsRNA coating interspersed with naked dsRNA is observed (Figure 6D). The presence of ATP or ADP:AlF4 favors the formation of filamentous regions. In comparison, at a 1:1 ratio, full-length chMDA5 completely coats dsRNA, independently of bound nucleotide, and at a ratio of 0.2:1, long clusters are observed on a few dsRNA molecules (Figure 6E). Thus, chLGP2 binding to dsRNA is less co-operative than chMDA5 (this work) and hMDA5. Imaging of chLGP2 and full-length chMDA5 filaments by cryo-EM shows that they are qualitatively similar, with a subunit axial translation of ∼44 Å (Figures 6D and 6E), identical to that reported for hMDA5.

24The mechanism of chLGP2 stimulation of chMDA5 function was investigated using the RNA binding-deficient chLGP2 mutants in the CTD, helicase, or both, which we characterized above and monitoring IFN-β activation in chicken DF1 cells upon stimulation with poly(I:C). Exogenously provided chLGP2 readily augments the response of endogenous chMDA5 to poly(I:C) (Figures 7A and S7A), as also found in the human system (Figures S7B and S7C), whereas LGP2 itself lacks transduction ability (Figures 7B and S7D). Unlike the human system, in which exogenous hMDA5 exhibits a constitutive activity not enhanced by exogenous hLGP2 (Figure S7D, no RNA), exogenous chMDA5 lacks constitutive activity, but co-transfection with chLGP2 results in significant activation of the chIFNβ promoter (Figure 7B, compare WT and Ø, blue bars) suggesting that chLGP2 overexpression can stabilize the binding of chMDA5 to cellular RNAs (as found with ATP hydrolysis mutants of MDA5; ). Co-transfection of chLGP2 has a strong enhancing effect on the activation of limiting amounts of exogenous chMDA5 by poly(I:C) (Figure 7B, compare green and blue bars for wild-type chLGP2), similar to the dose-dependent enhancement observed in the human system (Figures S7D and S7E). Enhancing activities were most evident when either MDA5 or poly(I:C) was in limited amounts. Furthermore there appears to be a species restriction, as hLGP2 readily enhances hMDA5 activation in human Huh7.5 cells (Figures S7C–S7E) but is inactive toward chMDA5 in DF1 cells (Figure 7B). RNA binding-deficient chLGP2 mutants in the CTD, helicase, or both show poor enhancing activity toward chMDA5 (Figure 7B, CTD°, hel°, and CTD°-hel° constructs), also observed in the human system (Figure S7F). Finally, hLGP2 also exhibited a potent enhancing effect on the poly(I:C)-mediated activation of the hMDA5 IE/KR mutant, which has been reported to be deficient in signaling because of a defect in homopolymerization (Figure 7C), this property being abrogated with RNA binding-deficient hLGP2 mutants (Figure S7G).

25We describe unusually high-resolution crystal structures of chLGP2 bound to dsRNA and ADP:AlF4, which show chLGP2 to be in the fully closed, ATP hydrolysis transition state. chLGP2 binds to dsRNA with its CTD capping the blunt end, similarly to RIG-I, but differs in having a longer overall footprint and by accommodating 5′ or 3′ overhangs with relatively minor perturbations in structure and little loss in affinity.

26We also present several structures of chMDA5, all of which are in the semi-closed state with ADP:Mg2+ bound. chMDA5 seems to prefer head-to-head packing when bound to short dsRNA (24-mer to 27-mer) in contrast to the polar head-to-tail packing described for long, signaling active hMDA5-dsRNA filaments. chLGP2 also forms filaments on dsRNA structurally resembling those of MDA5, although chLGP2 dsRNA coating is less co-operative.

27Functionally, conserved RNA binding determinants in both the CTD and helicase are required for LGP2-mediated enhancement of MDA5 signaling in both chicken and human, suggesting a conserved underlying mechanism for this phenomenon.

28Structures of RIG-I and MDA5 can be classified into four states depending on the degree of closure of the Hel1 and Hel2 domains, and this is usually correlated to the type of bound nucleotide (Table S1; Figure S6A). RNA free structures of full-length RIG-I (in which the CARDs are bound to Hel2i) or the helicase domain alone are in the very flexible “open” conformation with Hel1 and Hel2 well separated and Hel2 poorly ordered. Most hRIG-I helicase-CTD-dsRNA structures with no nucleotide, sulfate, or ADP are in the “semi-open” conformation with the poorly ordered Hel2 not contacting the RNA. In one “semi-closed” state structure, co-crystallized with ATP analog ADP:BeF3, Hel2 is better ordered, but motifs V and VI do not engage with the nucleotide. The only fully “closed” RIG-I structure in which all motifs are correctly positioned for ATP hydrolysis, is that of the isolated dRIG-I helicase domain bound to dsRNA and ADP:AlF4. The closed conformation of the chLGP2-dsRNA-ADP:AlF4 complexes reported here thus closely resembles the latter dRIG-I structure. In contrast, the chLGP2 12-mer dsRNA structure without nucleotide is semi-closed with Hel2 poorly ordered, except where it contacts the RNA. Intriguingly, a closed structure of RIG-I including the CTD and an ATP analog has not yet been reported.

29To characterize quantitatively these different conformations, we superposed, via the Hel1 domain, all RLH structures on the chLGP2 closed state as reference. The rotation angle required to superpose the Hel2 domain of the test structure on that of the reference structure varies systematically in concordance with the qualitative description above, being 0° to 3° for the closed state, 7° to 13° for the semi-closed state, ∼40° for the semi-open state, and 50° to 60° for the open state (Table S1).

30The two different chLGP2 conformations described here provide a unique opportunity to compare the closed and semi-closed states of the same RLH. The ∼10° rotation of Hel2 between the closed state and semi-closed states corresponds to a shift of the backbone interactions made by Hel2 (mediated by motifs IVa and V) by one phosphate down the 3′ strand of the dsRNA, whereas the interactions of Hel1 (mediated by motifs Ia, Ib, Ic, and IIa) with the dsRNA are unchanged (Figures S6B and S6C). The movement of Hel2 is coupled to that of Hel2i, which pivots around its contact with the CTD (which itself does not shift much), while maintaining similar contacts with the RNA (Figure S6B). This comparison highlights an ATP-dependent structural transition, which in one sense can correspond to a co-operative tightening of dsRNA interactions upon ATP binding and in the other sense the relaxation of the grip upon the dsRNA upon ATP hydrolysis. This structural transition needs to be taken into account in understanding the role of ATP binding and hydrolysis in RLH function, including filament formation.

31The stem binder MDA5 and end-capping LGP2/RIG-I critically differ in the position of the CTD and the nature of the residues on the CTD “capping loop” (Figure S1). In LGP2 and RIG-I “capping loop” aromatic hydrophobic amino acids (Phe595, Phe599, and Trp602 in chLGP2; Phe853 and Phe856 in hRIG-I) contact the blunt-end base pair of the dsRNA by edge-on or stacking interactions, essentially preventing binding to a continuous stem. In MDA5, the loop contains no bulky hydrophobic residues and is disordered in all structures of MDA5 bound to dsRNA so far. However, it is ordered in the structure of the hMDA5 CTD alone (PDB: 3GA3), where it forms an additional β strand to the main β sheet of the CTD rather than bulging out as in RIG-I and LGP2 (Figure 5D). As pointed out previously, MDA5 CTD is significantly displaced away from the dsRNA compared with RIG-I and is able to pack closer to Hel2i because of the 10 residues shorter α12 in MDA5 Hel2i. In chMDA5, the pining of the CTD to Hel2i is reinforced by the wrapping of the extended loop between α14 and α15 of Hel2i (residues 629–646) over the CTD (e.g., Asp635 interacts with His914 and Arg916 of CTD domain). In chMDA5, this α14-α15 loop is shorter than that of hMDA5 (638–670) but longer than in RIG-I or LGP2. The close CTD-Hel2i interaction combined with the less hydrophobic and less intrusive CTD capping loop allows MDA5 to bind internally to dsRNA.

32Interestingly, chLGP2 resembles a chimera, combining a MDA5-like helicase domain and a RIG-I like CTD. The similarity of LGP2 to MDA5 is reflected in common structural features, not found in RIG-I, such as the mode of nucleotide binding (Figures 1D and 4D) and the length of the Hel2i domain helices (Figure S1) and susceptibility to protein V. However, the hydrophobic nature of the LGP2 CTD capping loop is clearly RIG-I like. Corresponding to these mixed structural features, LGP2 has the strong blunt-end binding characteristic of RIG-I yet is also able to make MDA5-like filaments, albeit less efficiently. Perhaps the chimeric nature of the LGP2 structure allows its CTD to adopt a different position closer to Hel2i, allowing LGP2 to form filaments similarly to MDA5, whereas RIG-I does not form well-ordered, extended filaments. The similarity in overall appearance and axial subunit translation of LGP2 filaments to MDA5 filaments suggests that LGP2 could also form head-to-tail packing on long dsRNA. Furthermore, modeling suggests that in MDA5, the capping loop could not only play a role in dsRNA binding (Figure 5D) but also be involved in the tail-to-head protein-protein interface. The same could be true in the case of LGP2, although it is unclear what alternative interactions the aromatic residues on the capping loop of LGP2 might make. This and related questions will be answered only by high-resolution EM reconstructions of LGP2 and MDA5 filaments.

33The head-to-head arrangement we observe for dimers of MDA5 bound to short dsRNA contrasts with the intermolecular packing arrangement derived from EM reconstructions of mouse MDA5-coated dsRNA filaments, which exhibit a polar head-to-tail packing. When the chMDA5 structure was fitted into the EM map of the helical filament, the superposition also showed that the head-to-tail configuration fits significantly better (correlation coefficient 0.87 compared with 0.78 for head to head). However, the chMDA5-24 structure indicates that on short dsRNA, and in the absence of stabilizing co-operative interactions forming an extended multimer, head-to-head stacking appears to be the preferred mode of interaction, and head-to-tail packing has not yet been observed in any crystal form. Thus it is possible that a head-to-head dimer nucleates filament formation. Interestingly, Dicer-related helicase 3, an ortholog of the Dicer and RIG-I that is essential for secondary small interfering RNA production in Caenorhabditis elegans, binds and recognizes 22G-RNA as a dimer, which has been modeled as a head-to-head dimer.

34Various hypotheses exist as to how LGP2 might play positive and negative roles in RLH signaling. The high affinity of LGP2 end binding to short dsRNA supports that it could potentially compete for cognate RIG-I ligands and thus exert a negative effect, as proposed by others. However, the affinity for 5′ppp-dsRNA is slightly lower than for RIG-I, suggesting that an excess of LGP2 over RIG-I might be required, as observed by others. Such a high unbalance between the two proteins, both of which are similarly transcriptionally upregulated by type I IFN, remains to be demonstrated in natural infection conditions. LGP2 has a considerably higher affinity for non-phosphorylated blunt-end dsRNA than RIG-I and thus could act as sponge to prevent misactivation of RIG-I by such RNAs, which may exist in the host cell. This would be an additional mechanism to promote self-discrimination or non-self-discrimination by RIG-I in parallel with the recently recognized role of the RIG-I ATPase activity and the CARD2-Hel2i interface to prevent deleterious constitutive activation of RIG-I by more weakly binding, non-cognate cellular RNAs.

35The mechanism of enhancement of MDA5 signaling by LGP2 in human cells has been proposed to depend on a direct effect of LGP2 on MDA5 filament formation. It was concluded that LGP2 attenuates MDA5 filament length, giving rise to shorter complexes that more efficiently stimulate anti-viral signaling than longer MDA5 filaments. In the chicken system, LGP2-stimulated MDA5 signaling could work by a similar mechanism. Additionally, our results suggest that LGP2 could potentially compensate for the lack of RIG-I through high-affinity end binding to a RIG-I like ligand, which could then nucleate, by RNA-dependent protein-protein interactions, signaling-competent MDA5 oligomer formation. Such a mechanism is supported by the observed strong RNA end binding mode of LGP2 (which is more permissive in terms of 3′ and 5′ modifications than RIG-I) and the functional requirement of an intact RNA binding site on both the CTD and helicase. Interestingly, the enhancing effect is observed only when RNA ligand and/or MDA5 are in limited amounts, as is likely to be the case early in infection.

36In conclusion, our structural and functional results lay the basis for further studies to elucidate the exact mechanistic role of LGP2 in regulating MDA5 signaling by, for instance, determining the composition and structure of mixed filaments and/or oligomers containing both MDA5 and LGP2.

37Purified chMDA5 (full-length, residues 1–1,001), chMDA5ΔCARD (residues 298–994), chMDA5ΔCARD-E436Q constructs, and a SUMO fusion of full-length chLGP2 (1–674), expressed in Escherichia coli, were mixed with dsRNA of various lengths, prepared by in vitro T7 transcription, and either AMPPNP or ADP:AlF4 and initially screened for crystallization using a Cartesian robot.

38Optimized crystals were flash frozen and diffraction data collected on various beamlines at the European Synchrotron Radiation Facility (ESRF). Data were processed with XDS and further analyzed using the CCP4 suite. Structures were determined by molecular replacement and refined with REFMAC5.

39Full-length chMDA5 or chLGP2 complexes with Φ6 bacteriophage dsRNA (Thermo Scientific), with or without ATP analogs, were examined by negative stain EM. For cryo-EM, 1:1 mixtures (one MDA5/LGP2 molecule per 15 bp of dsRNA) were frozen on glow-discharged grids (Quantifoil Micro Tools) and data collected with an FEI Polara microscope equipped with a K2 summit detector (Gatan). Power spectra were calculated from masked 2D class averages.

40chMDA5, chLGP2, or hRIG-I was titrated into dsRNA solutions made by annealing 5′-FAM-labeled 12-mer RNA with unlabeled cRNA. Anisotropic measurements were made with excitation wavelength 495 nm and emission wavelength 515 nm during 100 s and Kd values derived by curve fitting.

41dsRNA-dependent ATPase reactions were performed using a Malachite green assay kit (Bioassays) over 0.5–30 min, as detailed previously.

42IFN-β promoter activation upon transient expression of expression vectors coding for chicken or human MDA5 and LGP2 and poly(I:C) stimulation in chicken DF1 and human Huh7.5 cells was determined as previously described

43E.U. designed experiments and performed biochemistry, crystallization, data collection, and structural analysis. M.L. helped with protein production. H.M. with the help of E.U. performed the EM analysis. J.B. and D.G. designed, performed, and/or analyzed cellular experiments. S.C. directed the project, performed structural analysis, and wrote the paper with input from the other authors.