Influenza Polymerase Can Adopt an Alternative Configuration Involving a Radical Repacking of PB2 Domains.

Influenza virus polymerase transcribes or replicates the segmented RNA genome (vRNA) into respectively viral mRNA or full-length copies and initiates RNA synthesis by binding the conserved 3' and 5' vRNA ends (the promoter). In recent structures of promoter-bound polymerase, the cap-binding and endonuclease domains are configured for cap snatching, which generates capped transcription primers. Here, we present a FluB polymerase structure with a bound complementary cRNA 5' end that exhibits a major rearrangement of the subdomains within the C-terminal two-thirds of PB2 (PB2-C). Notably, the PB2 nuclear localization signal (NLS)-containing domain translocates ∼90 Å to bind to the endonuclease domain. FluA PB2-C alone and RNA-free FluC polymerase are similarly arranged. Biophysical and cap-dependent endonuclease assays show that in solution the polymerase explores different conformational distributions depending on which RNA is bound. The inherent flexibility of the polymerase allows it to adopt alternative conformations that are likely important during polymerase maturation into active progeny RNPs.

Introduction

The eight single-stranded RNA segments of the influenza virus genome (vRNA) are individually packaged in rod-shaped ribonucleoprotein particles (RNPs). Within the RNP, the conserved 3′ and 5′ ends of each vRNA segment (the promoter) are bound to the viral-RNA-dependent RNA polymerase, and the rest of the vRNA is coated with nucleoprotein (NP). The polymerase is a heterotrimer composed of subunits PA, PB1, and PB2, and, in the context of the RNP, it performs both transcription and replication in the infected cell nucleus using the same template vRNA (Fodor, 2013, Ortín and Martín-Benito, 2015, Resa-Infante et al., 2011). Transcription of viral mRNA occurs through “cap snatching” (Plotch et al., 1981), whereby short capped oligomers, derived from host pre-mRNA, are bound by the PB2 subunit (Guilligay et al., 2008), cleaved by an endonuclease in the PA subunit (Dias et al., 2009), and then used to prime mRNA synthesis by the PB1 subunit. In contrast, replication involves unprimed synthesis of a full-length copy of the vRNA into cRNA and subsequently the inverse process back to progeny vRNA. Nascent replicates are co-transcriptionally packaged with incoming, newly synthesized polymerase and NP into progeny vRNPs or cRNPs. Recent crystal structures of bat influenza A (bat FluA; Pflug et al., 2014) and human influenza B (FluB; Reich et al., 2014) polymerases gave the first structure-based insight into how the vRNA promoter is specifically bound and how RNA synthesis is performed. In particular, comparison of the FluA and FluB structures suggested a mechanism for cap snatching whereby in situ rotation of the PB2 cap-binding domain could direct cap-bound, host pre-mRNA first toward the endonuclease for cleavage and then into the polymerase active site to prime transcription (Reich et al., 2014). However, there are many open questions. For instance, what is the conformation of the polymerase in the active transcription initiation or elongation state? Is the polymerase in a different state for replication and does replication require accessory polymerases, as has been proposed (Jorba et al., 2009, York et al., 2013), and in what conformation are they? Is there a difference between cRNA- and vRNA-bound polymerase, and how are incoming apo-polymerase and NP incorporated into a nascent RNP? Here, we use a combination of X-ray crystallography, biochemistry, and solution biophysical techniques to characterize alternative functional configurations of the multifunctional influenza polymerase. Our findings highlight the high flexibility of the polymerase, particularly the PB2-C domains, and suggest that the polymerase can exist in a number of alternative states, each of which may be important during different steps of transcription, replication, and progeny RNP assembly.

Results

Structure of Isolated PB2-C from Influenza A/H5N1 Polymerase

One of the original FluB polymerase structures (denoted FluB2) lacked electron density for the entire PB2-C region (residues 250–757), whereas it was visible in the FluB1 structure (Reich et al., 2014), suggesting that PB2-C interacted less strongly with the core of the polymerase (PA/PB1/PB2-N). Consequently, we decided to express the entire PB2-C domain, which comprises the mid, cap, cap-627 linker, 627, and nuclear localization signal (NLS) domains, to determine whether this behaved as a functional unit with perhaps an alternative structure. PB2 residues 247–736 of A/Vietnam/1203/2004(H5N1) (i.e., lacking the presumed flexible NLS peptide 737–759) (Figure 1A), were expressed in E. coli, purified, and co-crystallized with or without the cap analog m7GTP in two different crystal forms diffracting to 3.2 Å and 2.4 Å resolution, respectively (Table 1). Both structures are essentially the same and show continuous electron density for the mid, cap, cap-627 linker, and 627 domains but none for the NLS domain. A sequence alignment of PB2-C from influenza A, B, and C polymerases with superposed secondary structure is given in Figure S1.

Figure 1

Structure of the H5N1 PB2-C Multi-domain Construct

(A) Flu A/H5N1 PB2-C construct used for structural studies. Here and throughout the PB2 mid, cap-binding, cap-627 linker, 627 and NLS domains are colored magenta, orange, wheat, deep-salmon and firebrick, respectively. The region 737–759, which contains the bipartite NLS sequence (firebrick checkered), was excluded from the construct.

(B) Ribbon diagram of the structure of Flu A/H5N1 PB2-C colored as in (A). The mid and cap-627 linker domains form one rigid unit (mid-link). The bound m7GTP cap analog is shown in the cap-binding domain in sphere representation. The entire NLS domain is missing from the electron density.

(C) The equivalent PB2-C domain as observed in the FluB promoter complex (the bat FluA PB2-C structure is similar but the cap-binding domain is rotated). The structures in (B) and (C) are orientated equivalently after superposition of the mid-link module.

See also Figure S1.

Table 1

Crystallographic Data Collection and Refinement Statistics, Related to Experimental Procedures

	H5N1 PB2-C + m7GTP Form 1	H5N1 PB2-C + m7GTP Form 2	Flu B Polymerase c5′ End: 1–12	Flu B Polymerase v5′ End: 1–12	FluB Endonuclease PB2 NLS
Data collection
Space group	P3221	H32	P1	P21	P21
Cell dimensions
a, b, c (Å)	97.3	123.0	193.7	126.6	35.4
97.3	123.0	210.0	200.5	80.8
126.2	229.9	210.6	133.1	37.9
α, β, γ (°)	90.0	90.0	117.7	90.0	90.0
90.0	90.0	92.8	107.7	97.1
120.0	120.0	113.7	90.0	90.0
Number of crystals	1	1	1	1	1
Resolution (Å)	50.0-2.40	50.0-3.20	50-4.1	50-3.40	50-1.76
Outer shella	2.48-2.40	3.30-3.20	4.3-4.1	3.55-3.40	1.76-1.70
Rmerge	0.154 (1.31)	0.389 (1.66)	0.093 (0.626)	0.259 (1.38)	0.043 (0.559)
I/σI	12.2 (2.04)	8.21 (2.02)	5.49 (1.23)	7.83 (1.66)	12.68 (1.84)
Completeness (%)	100 (100)	99.9 (100.0)	92.9 (94.9)	99.6 (99.6)	98.2 (93.0)
Redundancy	11.11 (10.48)	10.80 (10.98)	1.75 (1.77)	6.37 (6.11)	2.72 (2.71)
Refinement
Resolution (Å)	50.0-2.40 (2.46-2.40)	50.0-3.20 (3.28-3.20)	50.0-4.10 (4.15-4.10)	50.0-3.40 (3.48-3.40)	40.4-1.70 (1.78-1.70)
No. of reflections work/free	26,428/1,145	10,884/467	188,032/7,760	83,996/2,595	21,181/1,098
Rwork/Rfree	21.2/23.4 (30.5/29.2)	27.0/29.4 (36.9/35.2)	25.7/28.7 (37.5/42.1)	25.2/27.5 (38.2/34.9)	22.0/26.9 (33.4/37.2)
No. of atoms (total)	3,559	3,376	106,314	34,637	1,669
Protein	3388	3347	104724 (6 trimers)	34113 (2 trimers)	1819
Ligand	33 (m7GTP)	29 (m7GDP)	1590 (RNA)	524 (RNA)	150 (NLS)
Solvent	138	–	–	–	47
B-factors (Å2)	76.1	70.5	193.5	111.3	32.7
Protein	77.3	70.4	193.1	111.8	32.6
RNA	–	–	202.3	76.6	–
Ligand	45.5 (m7GTP)	83.1 (m7GDP)	–	–	33.9 (NLS)
Solvent	49.3	–	–	–	31.8
RMSDs
Bond lengths (Å)	0.007	0.006	0.0025	0.007	0.013
Bond angles (°)	1.230	1.046	0.539	0.963	1.549
Validationb
Ramachandran (%)
Favored	97.0	95.3	92.3	94.7	98.6
Outliers	0.23	0.47	1.29	0.6	0
Clash score	0.14	0.29	1.16	0.53	3.87
MolProbity score	0.75	1.04	1.55	1.29	1.53

Outer-shell statistics in brackets.

MolProbity: http://molprobity.biochem.duke.edu/.

While each of the mid, cap-binding, cap-627 linker, and 627 domains are individually little changed in structure, there is a major difference in the packing arrangement of the four domains in the isolated H5N1 PB2-C structure (Figure 1B) compared to that observed in the full promoter-bound polymerase (Figure 1C). Analysis of the relative domain movements highlights that the mid (251–322) and cap-627 linker domains (490–536) form a rigid unit, denoted mid-link module. The integrity of this module is maintained by the short anti-parallel β strand interactions between residues 287–289 (mid) and 528–530 (cap-627 linker), the packing of the three-stranded anti-parallel β sheet (496–514, cap-627 linker) on helix 306–316 (mid), and the hydrophobic packing of Met535, Met536 (cap-627 linker) with Met283 and Ile266 (mid). Relative to the mid-linker module, there are hinges in the regions 321–324 and 484–498, which allow the cap-binding domain to rotate (as previously described; Reich et al., 2014), and another hinge at 536–541, which allow the 627 domain to rotate by 62° (Table 2; Figure S1). The net result is that whereas PB2-C forms an arc in the promoter-bound FluA and FluB structures, the isolated PB2-C structure is straighter. In addition, the NLS domain, although not visible in the electron density, must have separated from the 627 domain (Figures 1B and 1C). The 424-loop (residues 420–427) of the cap-binding domain is well ordered in the H5N1 PB2-C structure with its tip packing on the interface between the mid and linker domain and Arg423 making a salt bridge with Glu520, both absolutely conserved residues. This helps to stabilize this particular orientation of the cap-binding domain.

Table 2

PB2-C and PA Inter-domain Hinge Rotations and Translations Calculated by Comparing v3′-5′- or c5′-Bound Crystal Structures of FluB Polymerase

Domain 1	Domain 2	Hinge Residues between Domain 1 and 2	Rotation Angle around Hinge (°)a	Center of Mass Translation of Domain 2 (Å)b
PB2-N (1–251)	mid-link (252–322, 498–536)	PB2: 250–256	136	24
Mid-link	cap-binding (323–497)	PB2: 321–324	62	27
Cap-binding	mid-link	PB2: 484–498	−62	–
Mid-link	627 (537–677)	PB2: 536–541	59	67
627	NLS (696–740)	PB2: 677–700	133	93
PA-C (200-726)	PA endonuclease (1–195)	PA: 190–200	137	2

Calculated using DynDom (Taylor et al., 2014).

Center of mass translation of domain 2 between the v3′-5′- and c5′-bound conformations after superposing the polymerase core (PA-C, PB1, and PB2-N).

Structures of H5N1 PB2-C have been obtained with and without m7GTP. The m7GTP is bound in the cap-binding site as expected (Guilligay et al., 2008), but the triphosphate is in an unusually bent conformation enabling the γ-phosphate to interact with His357, Lys339, Arg355 and the ribose 2′OH (Figure S2A). Due to the particular rotation of the cap-binding domain with respect to the mid-link module, the ribose and phosphates are juxtaposed to residues Asn510-Val511 of the last strand of mid-link three-stranded anti-parallel β sheet (496–514). This environment makes a much less accessible cap-binding site than that observed in previous structures of the full-length polymerase (Figure S2B). To test whether this precludes capped RNA binding, we attempted co-crystallization of H5N1 PB2-C with the dinucleotide cap-analog m7GpppG but no electron density for this ligand was found in the resultant map. Modeling, based on observed RNA binding to the cap-binding site in the full FluB1 polymerase structure (Reich et al., 2014), suggests that whereas a straightened triphosphate could be accommodated, the second nucleotide would clash with the last strand of the three-stranded anti-parallel β sheet. Thus, it is likely that the observed H5N1 PB2-C conformation sequesters the cap-binding site against the three-stranded β sheet in a way that prevents capped RNA binding.

Structure of FluB Polymerase with Bound 5′ cRNA

FluB polymerase was crystallized with a cRNA 5′ end 12-mer (5′-pAGCAGAAGCAGA-3′) giving a P1 crystal form diffracting to 4.1 Å resolution (Table 1). To solve the structure, the polymerase core (PA-C, PB1, and PB2-N) and then the PA endonuclease were placed by automatic molecular replacement. Subsequently, the PB2 mid and cap-binding domains could be unambiguously positioned manually in residual positive difference density. There are six polymerases in the asymmetric unit, arranged as a hollow spherical particle with 32-point symmetry (Figures S3A and S3B). Despite the moderate resolution, using map averaging and map sharpening, an essentially complete model of the FluB polymerase-c5′ complex could be obtained (Table 1).

The structure of the FluB-c5′ complex is remarkable in that, whereas the polymerase core (PA-C, PB1, and PB2-N) has relatively minor changes overall (see below), the PB2-C mid, cap-binding, cap-627 linker, and 627 domains are dramatically rearranged compared to the full FluB polymerase-promoter complex (Reich et al., 2014) (Figures 2A–2C). Furthermore, the altered organization of the PB2-C domains is essentially identical to that observed in the crystal structure of the equivalent part of FluA/H5N1 PB2-C in isolation (Figures 2A and 2B). The conservation of this alternative PB2-C multi-domain structure across FluA and B strains (and FluC, see below) suggests that it is of functional importance. Comparing the two FluB polymerase conformations confirms that the mid and cap-627 linker constitute a single rigid unit (root mean square deviation [RMSD] = 0.88 Å for 109 aligned Cα positions in the mid-link module, residues 252–322 and 498–536). It also reveals an additional hinge around PB2 residues 250–256 between PB2-N and PB2-C, about which the mid-link module rotates by 135.8° between its position in the new c5′-bound structure compared to the promoter-bound FluB1 structure (Table 2; Figure S1). The net consequence of all the relative rotations between the PB2 domains is that the 627 and NLS domains are translated by remarkable center of mass displacements of 67 Å and 93 Å respectively from their positions in the promoter-bound structures, the mid-link domain by 24 Å and the cap-binding domain by 27 Å (Figure 2D; Table 2). A striking example of the repacking of domains between the c5′ and promoter-bound conformations concerns the endonuclease and NLS domains, which intimately interact in the c5′-bound structure (see below). Another example is the complete change in interface mediated by the PB1 palm domain helix α10 (residues 280–297). In the promoter-bound conformation, this helix interacts with the β sheet of the 627 domain, whereas in the c5′ conformation, the same helix, slightly displaced, makes extensive interactions with the cap-binding domain (Figure 2E). Indeed, in the c5′ conformation, the cap-binding domain appears to be immobilized by several inter-domain interactions, whereas in the promoter-bound conformation, it can clearly rotate in situ consistent with the proposed cap-snatching mechanism (Reich et al., 2014). The 627 domain is exposed on the periphery of the complex (Figures S3A and S3B) with host-specific residue 627 being highly accessible.

Figure 2

Structure of the FluB Polymerase-c5′ RNA Complex

(A) Ribbon diagram of the FluB polymerase c5′ RNA complex structure with PA-Nter (endonuclease) colored forest green, the rest of PA (green), PB1 (cyan), PB2-N (red) and the PB2-C domains as in Figure 1A. The c5′ RNA 12-mer is in violet. The unobserved linker between the 627 and NLS domains is shown dotted.

(B) View of the A/H5N1 PB2-C structure (with bound m7GTP) after superposition on the equivalent domains of the FluB polymerase-c5′ RNA complex. The root-mean-square difference in Cα positions is 1.68 Å for 310/424 aligned Cαs over residues 254–670 in the mid, cap-binding, cap-627 linker and 627 domains, showing that the domain arrangement is essentially identical.

(C) Structure of the FluB polymerase-promoter complex (with the 3′ and 5′ vRNA respectively yellow and violet) after superposing via the PB1 subunit on the FluB polymerase-c5′ RNA complex. Comparison of (A) and (C) shows that the polymerase core (PA-C, PB1 and PB2-N) is largely unchanged apart from near the promoter binding site (see Figure S2C).

(D) Diagram comparing the positions of the PB2-C and PA endonuclease domains in the FluB c5′ RNA (left) and promoter (right) complexes relative to the conserved PA-C, PB1 and PB2-N core (dotted ellipse). Table 2 gives the relative rotations and displacements of the domains.

(E) Involvement of PB1 helix α10 (residues 275–295, cyan) in interacting either with the cap-binding domain (orange) or the 627 domain (deep salmon) in respectively the c5′ (left) or full promoter bound (right) FluB polymerase conformations. The two structures were superposed via the invariant polymerase core, here represented by PB2-N and the PB1 helix. In the c5′ bound structure the cap-binding domain is immobilized by inter-domain interactions whereas in the promoter bound structure the cap-binding domain (represented by an ellipse) is free to rotate.

See also Figures S2–S4.

Interaction of the PA Endonuclease with the PB2 NLS Domain

Another remarkable feature of the FluB-c5′ complex is that the PA-Nter endonuclease is repositioned and interacts directly with the PB2 NLS domain (Figure 2A). Instead of being packed against the PB1-Cter/PB2-Nter helical bundle, via endonuclease helix α4 (PA residues 84–98), as observed in the promoter-bound structures (Pflug et al., 2014, Reich et al., 2014), the endonuclease is rotated in situ by 137°. This results in a different but less extensive interface with the PB1-PB2 helical bundle, which, however, is compensated by a completely novel interface between the endonuclease and the PB2 NLS domain with a substantial total buried surface area of 2,950 Å2 (Figure 2A). As in the H5N1 PB2-C structure, the FluB NLS domain has separated from the 627 domain, the two domains being connected by an extended linker (677–693), most of which lacks electron density. In the FluB-c5′ complex, the compact, globular part of the NLS domain packs against endonuclease helix α6 (FluB PA residues 164–179) and the beginning of strand β7 (150–152) on the side of the nuclease. Most interestingly, residues 745–770 of the bipartite NLS containing, extreme C-terminal peptide of PB2 (Tarendeau et al., 2007) (for FluB, 740-KRKRYSALSNDISQGIKRQRMTVESMGWALS-770, bipartite NLS underlined) form a long α helix that packs on the endonuclease domain (Figure S1). The helical conformation of the NLS containing peptide contrasts with the extended structure the same peptide makes when bound to α-importin (Pumroy and Cingolani, 2015, Tarendeau et al., 2007).

To study this interaction further, we co-crystallized a synthetic FluB PB2 29-mer NLS peptide (residues 742–770) with the FluB endonuclease domain PA (residues 1–197). The high-resolution (1.7 Å) structure so determined (Table 1) shows the same interaction between the endonuclease and NLS peptide as observed in the lower-resolution FluB-c5′ complex (Figure 3A). The amphipathic helical NLS peptide runs roughly perpendicular to PA helix α3, strand β3 (in the vicinity of PA Glu81, corresponding to active site residue Glu80 in FluA), and helix α4 and makes a number of hydrophobic and specific polar interactions, resulting in a total buried surface area of 1,500 Å2 (Figure 3A). There are charged polar interactions between PA residues Asp50, Glu78, and Arg169 with and PB2 residues Arg757, Asp750, and Thr761, respectively. Of particular note is the side chain of PB2 Gln758, which is completely buried and reaches inward to make multivalent hydrogen bonds with the main-chain carbonyl oxygens of PA Val79 and Glu81 and the hydroxyl of Tyr46, just behind the nuclease active site (Figure 3B). The normal two metal coordination of divalent cations (in this case magnesium ions, since no manganese was added) is observed in the endonuclease active site. The endonuclease C-terminal residues 190–194 fold back toward the active site, with the side chain of Glu193 approaching within hydrogen-bonding distance of metal coordinating His41. The electron density in the c5′-bound polymerase is compatible with this and indeed shows that there is only a short solvent exposed linker of approximately five residues connecting the endonuclease to the rest of PA (from residue Ile200, PA is bound to PB1), unlike in the promoter bound structure, where this folding back does not occur and the solvent-exposed linker is approximately nine residues. The conformation of this linker region may affect the accessibility of the endonuclease active site to substrate RNA.

Figure 3

High-resolution Structure of FluB Endonuclease-NLS Peptide Complex

(A) Ribbon diagram of the FluB PA-Nter endonuclease (forest green) and PB2 NLS-peptide (yellow) with key residues as sticks and secondary structure elements labeled. The amphipathic helical NLS peptide runs roughly perpendicular to PA helix α3 (e.g., PB2 Ile751 and Ile755 stack on PA Tyr46), strand β3 (in the vicinity of PA Glu81, corresponding to active site residue Glu80 in FluA) and helix α4 (e.g., PB2 Val762 and Met765 make contacts with PA Met83, Ile87 and Val91). Two magnesium ions (magenta spheres) are present in the nuclease active site.

(B) Detail of the PB2 NLS peptide interactions with the endonuclease in the vicinity of the active site. Residues coordinating the two magnesium ions are shown as well as the hydrogen bonding interactions (red dotted lines) of PB2 Gln758 with PA Tyr46 and the main-chain carbonyls of Val79 and Glu81. These interactions are likely to be FluB specific, due to genera dependent sequence differences.

(C) Isothermal calorimetry data and curve fit to derive the affinity of the FluB NLS peptide for the endonuclease domain.

(D) Effect of different bound RNAs on FluB polymerase cap-dependent endonuclease activity. 7 M urea gel showing the endonuclease activity of FluB polymerase on 32P-capped 20-mer RNA with time (minutes) in the presence of no vRNA (panel 1), 1.2 μM 3′ vRNA 1-18 (panel 2), 1.2 μM 5′ v RNA 1–18 (panel 3), 1.2 μM 5′ and 3′ vRNA (panel 4) or 1.2 μM 5′ cRNA 1–18 (panel 5). RNA markers are shown in the first and last lanes (lanes L).

See also Figure S5.

Structure of Influenza B Polymerase with Bound 5′ vRNA

Co-crystallization of FluB polymerase with only a 5′ vRNA 12-mer (5′-pAGUAGUAACAAG-3′) gave a new monoclinic crystal form diffracting to 3.4 Å. There are two complexes in the asymmetric unit and both are in the conformation very like the FluB1 form originally described (Reich et al., 2014) and thus quite different from the structure determined with the c5′ bound (Figure S3C). This is surprising given that the v5′ or c5′ RNAs used, at the resolution of the structures obtained, appear to have the same stem-loop structure bound in the same way as previously described (Pflug et al., 2014). The two RNAs differ at 5/12 positions, two of which alter one of the stem base pairs (3-U:A-8 in vRNA to C:G in cRNA): one is a substitution in the loop (6-U to A), and two (11-AG to GA) are distal to the stem-loop and, at least for the vRNA, engage in base pairs with the 3′ end of the promoter when present. The higher-resolution v5′ structure clearly shows that bases 11–12 maintain stacking with PA His506 in the absence of the 3′ end, and the lower resolution c5′ density is compatible with this. On the other hand, the consequence of the absence of the 3′ end in both these structures, as well as the short 5′ end (12-mer), results in movements or refolding of regions of the polymerase that interact with the promoter, such as the PB1 β-ribbon (residues 185–207) and PB1 residues 670–681, and PB2 residues 82–91 also become disordered (Figure S3D). These changes are clear in the higher-resolution v5′ structure but are also compatible with the electron density in the c5′ structure.

Comparison with Apo-FluC Polymerase Structure

The crystal structure of the complete influenza C (FluC) heterotrimeric polymerase has recently been determined in the absence of any bound RNA (Hengrung et al., 2015). FluC polymerase PA, PB1, and PB2 subunits are 25.6%, 40.8%, and 25.2% identical to their FluB counterparts (with very similar numbers when compared to FluA). The apo-FluC polymerase structure is remarkably similar to the FluB polymerase-c5′ complex, with not only the core polymerase (PA-C, PB1, and PB2-N) being in the same overall configuration but also the position of the mid, cap-binding, and cap-627 linker domains being quasi-identical (Figures S4A and S4B). The access to the cap-binding site is similarly obscured by residues from the cap-627 linker. The only differences concern the orientations of the 627 domain and the PA endonuclease-PB2 NLS domain unit (Figures S4C and S4D). In the apo-FluC structure, the NLS domain packs against the endonuclease in the same way as in the FluB c5′ structure. However, the FluC NLS peptide is two residues shorter than in FluA and eight residues shorter than in FluB (Figure S1), so the helix bound to the endonuclease is correspondingly shorter (Figure S4D). Also, the entire apo-FluC endonuclease-NLS unit is rotated in situ by 94° compared with FluB c5′ structures (Figure S4CD). Furthermore the FluC 627 domain is packed closer to the polymerase core and to the NLS domain than in the FluB c5′ structure, thus making the apo-FluC structure slightly more compact (Figure S4D). However, the close juxtaposition of the 627 and NLS domains in the apo-FluC structure is quite different to that observed in the FluA and FluB promoter-bound structures (which is the same as in the crystal structure of the isolated double domain; Tarendeau et al., 2008), due to a large rotation about the flexible linker joining them. A priori, it is not clear whether these differences reflect the absence of any RNA in the FluC structure, sequence divergence between FluC and FluB polymerases, or different crystal-packing constraints.

Context-Dependent Endonuclease Activity

As the PB2 NLS peptide interacts with the endonuclease in both the FluB c5′ and apo-FluC structures, with, in the case of FluB, Gln758 making intimate interactions with active-site proximal residues, we studied biochemically the effect that this binding might have on nuclease activity. We first measured the affinity of the FluB synthetic NLS peptide to the endonuclease by isothermal calorimetry (ITC) (Figure 3C). The derived Kd of 16.5 μM is only moderate, but in the context of the heterotrimeric polymerase, this would be effectively enhanced due to the intra-molecular nature of the interaction. One question that arises is whether the same kind of interaction could occur in FluA polymerase? In avian and human FluA, the NLS peptide is invariant with sequence 736-KRKRDSSILTDSQTATKRIRMAIN-759 (residues conserved in human/avian FluA and FluB underlined) (Figure S1). This peptide is not only shorter but also there are many substitutions compared to FluB, notably Gln758 is replaced by Ile754 in FluA, and similarly the PA endonuclease has diverged in sequence, so it is not clear whether a similar interaction is made. Indeed, no clear interaction was detected between a synthetic H3N2 FluA PB2 22-mer NLS peptide (residues 738–759) and the FluA endonuclease in corresponding experiments, nor was it possible to co-crystallize a complex, so the question remains open.

We next examined the effect of NLS peptide binding on the activity of the isolated endonuclease domain. Using a quantitative fluorescence resonance energy transfer (FRET)-based solution method (Kowalinski et al., 2012) (Figure S5), a small increase in the FluB endonuclease domain activity was detected upon titrating FluB NLS peptide, with peptide concentrations up to several times the Kd (Figure S5D). This increase was not observed with FluA endonuclease domain (whose intrinsic nuclease activity is estimated to be ∼40 times higher than for FluB; Figures S5A and S5B), although as indicated above, the Kd for NLS peptide binding is not known in the FluA case, but it is likely to be higher than that for FluB.

These results show that NLS peptide binding in trans to the isolated domain does not have a marked effect on the domain nuclease activity, consistent with the metal-binding active site not being perturbed in the structure. However, depending on the relative disposition of the cap-binding and endonuclease domains and the accessibility of their RNA binding sites, the nuclease activity, in the context of the complete heterotrimer, could vary markedly. Therefore, we assayed the cap-dependent endonuclease activity of the full polymerase as a function of which kind of viral RNA is bound, reasoning that this might be a good probe of the polymerase solution conformation. Time courses of FluB polymerase nuclease activity, with either no RNA or bound to v3′, v5′, c5′, or v3′+v5′ RNAs, were monitored over a 2-hr period using a radioactively labeled capped 20-mer as substrate. The results show that with no RNA, or just the v3′ end bound, there is very low activity in contrast to the maximal activity exhibited when both v3′ and v5′ (i.e., the full promoter) are bound (Figure 3D). In the latter condition, the polymerase cap-dependent nuclease activity is far higher (∼100 times) than for the isolated endonuclease domain and furthermore only requires magnesium and not manganese, which is essential in the case of the domain (Datta et al., 2013) (Figure S5E). Both v5′ and c5′ alone activate the cap-dependent endonuclease activity (although less than the full promoter), but the c5′ less so than the v5′ (Figure 3D). As shown below, these observations correlate with biophysical results showing that the polymerase conformation is very different without bound RNA or with only the v3′ end bound as compared to when bound to v5′, c5′, or v3′+v5′. We propose that the endonuclease activity reflects the fraction of time spent by the polymerase with the cap-binding domain and nuclease domain suitably positioned to maximize efficient capped-RNA binding and cleavage (see Discussion).

Small-angle X-Ray Scattering and Multi-angle Laser Light Scattering

To further probe the solution structure of FluB polymerase with different bound RNAs, we undertook small-angle X-ray scattering (SAXS), multi-angle laser light scattering (MALLS) and crosslinking mass spectrometry experiments.

SAXS measurements were made on solutions of FluB polymerase with either no bound RNA (apo) or with only v3′, only v5′, or both v3′ and v5′ RNAs, as they eluted from an on-line size-exclusion column. There was a clear distinction between the elution time and profile of the different samples. The apo- and v3′-bound samples behaved similarly, eluting earlier but with an asymmetric profile, whereas the v5′-only- or v3′- and v5′-bound samples eluted at a later time with a symmetric profile (Figure 4A). Equivalent results were obtained by MALLS experiments with the apo-FluB sample eluting before the v5′- or c5′-only-bound samples, which behave equivalently (Figure 4B). The MALLS measurements allowed estimation of the molecular weight of the particles in the two elution peaks giving very similar values for each (237 kDa for apo and 226 kDa for v5′ or c5′ bound, somewhat less than the actual value of 262.7 kDa), indicating monomeric complexes and no aggregation in either peak. In contrast, calculation of the radius of gyration at each time point of the elution during the SAXS measurements showed that whereas the v5′-only or v3′- and v5′-bound samples had a radius of gyration of ∼44 Å, uniform across the peak, the radius of gyration of the apo- or v3′-only-bound samples varied across the elution profile between ∼48 Å and 56 Å (Figure 4A). Taken together, these results show that apo- or v3′-only-bound polymerase has a much more extended conformation in solution but with conformational heterogeneity likely due to flexibility, whereas binding of either the v5′ or c5′ RNA ends leads to compaction of the polymerase, which is not measurably changed upon further addition of the v3′ end.

Figure 4

Solution Conformation of FluB Polymerase by SAXS and MALLS

(A) Radius of gyration derived from Guinier plots of SAXS data as a function of gel filtration retention time for FluB polymerase with no RNA or 3′ vRNA only (left peak) and 5′ or 3′+5′ vRNA (right peak). Colors for each sample are as indicated.

(B) SEC-MALLS elution profiles together with derived molecular weights for FluB polymerase with no RNA (left peak) and bound c5′ or v5′ RNA (right peak). Colors for each sample are as indicated.

See also Figure S6.

The question arises as to whether the two distinct solution states observed by SAXS and MALLS, apo/3′ bound or v5′/v5′-3′ bound, correspond to any of the known crystal structures. The radii of gyration calculated from the apo-FluC, apo-FluB (modeled on apo-FluC), c5′-bound, or v5′-3′-bound structures are respectively 41.5, 41.7, 43.3, and 41.9 Å. Thus, all crystal structures are slightly more compact than the v5′/v5′-3′-bound solution state and do not correspond at all to the apo/3′-bound solution state. Furthermore, at the level of the scattering curves, as expected by the closer correspondence of the radius of gyration, the c5′-bound structure fits best the v5′/v5′-3′-bound scattering curve at low resolution. However, none of the models (or even an average of the three) fit satisfactorily the data at higher resolution as highlighted by the clear secondary maximum at ∼0.12 Å−1 in the theoretical scattering curves of all models, which is smeared out in the experimental curve (Figure S6). These observations indicate that there is a greater variation in the conformations present in solution than those sampled in crystal structures, not only for the highly flexible apo- and v3′-bound forms but also for the more compact 5′-RNA-bound forms.

Crosslinking Mass Spectrometry of Influenza Polymerase Conformation in Solution

Recombinant polymerase complexes with no bound RNA (Bat FluA and FluB) or 5′-3′ vRNA bound (Bat FluA) were crosslinked using two amine-reactive, homo-bifunctional reagents of different length, disuccinimidyl-suberate (DSS) or disuccinimidyl-glutarate (DSG). Subsequently, crosslinked peptides were subjected to mass spectrometric analysis and stringently identified using the xQuest/xProphet software (Walzthoeni et al., 2012). High-confidence crosslinks with a linear-discriminant (ld) score (Walzthoeni et al., 2012) higher than 25 were selected for the analysis, resulting in a total of 234 unique crosslinks (see Figure S7 and Tables S1, S2, and S3 for more details on the datasets). Pooled crosslinks from the different samples were then mapped onto four different FluB polymerase structures: the crystallographically determined c5′- or v5′-3′-bound conformations, the v5′-3′-bound conformation with the alternative cap-binding domain orientation modeled on the bat FluA structure (Pflug et al., 2014), and a modeled apo-FluB conformation, obtained by superposing the FluB 627, NLS, and endonuclease domains on to the positions of the equivalent domains in the apo-FluC structure. Crosslinks were deemed to be satisfied (or violated) if the corresponding lysine-lysine distance in the structures was shorter or longer (c.f. or violated) than 35 Å (see Experimental Procedures). Out of the 234 crosslinks, 231 mapped into structured regions and 197 (85%) crosslinks were satisfied in at least one of the apo, 5′ cRNA or 5′-3′ vRNA structures (Table S2). The percentage of crosslinks not satisfied by any structure (15%) is significantly higher than the expected false-positive rate of crosslink identification implemented in xProphet (5%). In agreement with the SAXS results, these data suggest that the high-scoring crosslinks observed as “violated” might be explained by uncharacterized alternative conformations co-existing in solution (Figures S7B–S7G).

Although most crosslinks were satisfied in more than one conformation (Table S2; Figure 5A), certain subsets of crosslinks uniquely support individual apo-, c5′-bound, or v5′-3′-bound conformations, suggesting that these conformations coexist in solution. These crosslinks involve the PA endonuclease, PB2-627, and PB2-NLS domains, i.e., the domains that show the most pronounced differences between the observed crystal structures. In particular, seven crosslinks from PB2-627 to PB2-mid and PB2-NLS domains are violated in v5′-3′-bound conformations but become satisfied in the c5′-bound or apo conformation due to the large displacements of the PB2-627 and PB2-NLS domains (Figure 5B). On the other hand, one crosslink from the PB2-627 to the PB2-NLS satisfied in v5′-3′-bound conformations gets violated in the c5′-bound and apo conformations. Likewise, the three different orientations of the PA endonuclease lead to satisfaction of different set of crosslinks (Figure 5C). Interestingly, the different conformations coexist in the sample of FluB reconstituted with vRNA: the crosslinks obtained from this sample corresponded not only to the 5′-3′ vRNA conformation but also to the 5′ cRNA and apo conformations.

Figure 5

Cross-links Confirm the Crystallographically Observed Polymerase Conformations

(A) Crosslinks satisfied in the four different conformations of FluB polymerase (v5′-3′ bound as in the FluB1 structure, v5′-3′ bound with rotated cap-binding domain as observed in the bat FluA structure, c5′-bound and modeled apo-FluC conformation). Satisfied crosslinks are colored blue. The PB1 subunit and the PA-C domain are colored gray. The PA endonuclease and PB2-C domains are color-coded according to the domain structure.

(B) Crosslinks originating from the PB2-627 domain that are satisfied in at least one of the apo, c5′-bound, or v5′-3′ bound conformations. Satisfied crosslinks are colored blue, violated are red.

(C) Crosslinks originating from PA nuclease that are satisfied in at least one of the apo, c5′-bound, or v5′-3′ bound conformations.

See also Figure S7.

Discussion

During influenza replication in the infected cell nucleus, active polymerase complexes are bound to and operate on vRNA (or cRNA) within the context of ribonucleoprotein particles. Newly synthesized polymerase subunits are imported into the nucleus as PA-PB1 heterodimers by RanBP5 (Deng et al., 2006) or separate PB2 monomers by importin α (Tarendeau et al., 2007), which then heterotrimerize in the nucleus (Deng et al., 2005, Huet et al., 2010). It is plausible that polymerase trimers (or possibly PA-PB1 heterodimers, before PB2 associates) first encounter viral RNA upon binding to the nascent c5′ or v5′ ends that emerge from replicating vRNPs or cRNPs, respectively. These ends can bind with high affinity as a stem loop to the allosteric 5′ binding site formed by the PA and PB1 subunits (Pflug et al., 2014). Polymerases so bound might then nucleate assembly of progeny RNPs by directing sequential addition of incoming nucleoproteins on to the elongating replicate RNA. This model is consistent with results showing that interactions between polymerases are required to promote efficient replication (Jorba et al., 2008, Jorba et al., 2009, York et al., 2013). A detailed mechanistic understanding of all these processes requires structural information on the polymerase in its various functional states. So far, structures of bat FluA and FluB polymerase bound to the v3′-v5′ promoter show a similar conformation, which has been interpreted to be the pre-transcription initiation configuration, competent for cap snatching and cap-dependent priming (Reich et al., 2014). However, there are indications both from crystallography (Reich et al., 2014) and from low-resolution electron microscopy of native RNPs (Arranz et al., 2012, Moeller et al., 2012) that the polymerase, particularly the modular PB2 subunit, is highly flexible and can adopt a variety of conformations. Here, we have used a variety of structural (crystallography and SAXS), biochemical (endonuclease activity), and biophysical (MALLS and crosslinking mass spectrometry) methods to characterize the conformation of influenza polymerase when it is RNA free or bound to v3′, v5′, c5′, or v3′+v5′ RNAs, all of which are likely to be relevant at some stage in the maturation or activity of the polymerase.

The most remarkable result is the concordance between crystal structures of FluA/H5N1 PB2-C and c5′ RNA bound FluB polymerase that reveals a completely different stable packing of the PB2 mid, cap-binding, cap-627 linker, and 627 domains compared to the full promoter-bound structures. Furthermore, a similar arrangement of PB2 domains is observed in the apo-FluC polymerase structure (Hengrung et al., 2015). An unexpected common feature of the c5′ FluB and apo-FluC polymerase structures is that the PB2 NLS domain packs against the PA endonuclease, with the C-terminal NLS peptide forming an α helix that participates in this interaction. We also report a high-resolution crystal structure of the FluB endonuclease-NLS peptide binary complex, which reveals the details of this interaction. However, biochemical assays show that this interaction in itself does not inhibit the activity of the isolated endonuclease domain. We also show that the cap-dependent nuclease activity of the trimeric polymerase is considerably enhanced when v5′, c5′, or v3′+v5′ RNA are bound (in line with previous observations, e.g. Li et al., 2001) and correlates with the compact solution conformation of these states as detected by MALLS and SAXS. In contrast, both apo- and v3′-RNA-bound polymerases have poor cap-dependent nuclease activity, comparable to that of the isolated nuclease domain, again correlating with the extended conformation of these states in solution. Finally, both SAXS and crosslinking mass spectrometry results show that in solution, a variety of states coexist not restricted to the known crystal structures.

Our interpretation of these results is as follows. First, they suggest that whatever the RNA bound (or not), multiple polymerase conformations exist in equilibrium in solution, although apo/v3′-bound polymerases are preferentially extended, whereas v5′-, c5′-, or v3′+v5′-bound polymerases are preferentially compact. The extended apo/v3′-bound states presumably correspond to the situation where all flexibly linked domains (notably those of PB2-C and the endonuclease) are dangling in solution on their extended linkers, whereas the more compact states involve significant interfaces between these domains and with the polymerase core as observed in the various crystal structures. Second, crystallization generally favors more compact, less flexible conformations that are stabilized by crystal contacts and possibly oligomerization as well (as in the case of the c5′-bound FluB and apo-FluC structures). Crystal conformations may thus represent minority species in solution, as seems particularly likely for the apo-FluC structure, since MALLS and SAXS both suggest that the apo state should be highly extended. The cap-dependent endonuclease activity of the polymerase correlates well with the solution behavior in that apo/v3′-bound polymerases have low activity and v5′-, c5′-, or v3′+v5′-bound polymerases have much higher activity. Interestingly, the apo-FluC and c5′-bound FluB both have conformations in which the environment of the cap-binding site (including the position of the linker connecting the endonuclease to the rest of PA) disfavors capped RNA binding. This is likely representative of conformations in which the relative disposition of the cap-binding and endonuclease domains does not allow favorable coupling of strong capped RNA binding with nuclease cleavage; the nuclease activity then diminishes toward that of the isolated domain, which is weak mainly because of poor intrinsic RNA binding and a fast RNA off-rate (Datta et al., 2013). Interestingly, we find that c5′-, v5′-, or v3′+v5′-RNA-bound polymerases are increasingly active in cap-dependent endonuclease activity, respectively. This can be explained, assuming that the v3′+v5′-RNA-bound crystal structure is the optimal one for cap-dependent endonuclease activity, by an increasing propensity of the c5′-, v5′-, or v3′+v5′-RNA-bound polymerases, respectively, to be in this state. This would be consistent with the fact that c5′-bound polymerase preferentially crystallizes in a different conformation than v5′- or v3′+v5′-RNA-bound polymerase. Exactly why the conformation of the distal PB2-C region, which is distant from the promoter binding site, depends on which RNA is bound is still obscure. There is no striking overall difference in the core polymerase (PA-C, PB1, and PB2-N) between the FluB c5′-, v5′-, or v3′+v5′- RNA-bound polymerase crystal structures, except local perturbations around the RNA binding site as described above. Furthermore, the total buried surface area of the mobile domains (PB2-C and endonuclease) with the polymerase core (PA-C, PB1, and PB2-N) is very similar for the promoter or c5′-bound structures (6,621 Å2 or 5,995 Å2, respectively). However, the NLS domain/endonuclease interaction provides an additional 2,950 Å2 apparently favoring the c5′-bound conformation, although it must be remembered that these numbers do not directly reflect binding energy. We propose that differential RNA binding more or less rigidifies the core polymerase to promote interfaces either between PB1 and the 627 domain and the endonuclease with the PB1-Cter/PB2-Nter helical bundle, in the case of v5′ or v5′+v3′, or PB1 and the cap-binding domain and the endonuclease with the NLS domain, but always allowing some probability of interconversion. With no RNA or only v3′ RNA bound (i.e., without any 5′ RNA bound), increased flexibility/instability around the promoter binding site could propagate to more distal regions of the core and destabilize both types of potential interface with PB2-C, allowing flexibly linked domains to be mobile in solution. In this connection, we point out that recent studies of the structurally and functionally related La Crosse bunyavirus polymerase have revealed considerable disorder in the apo state and also highlighted the role of vRNA 5′ end binding in stabilizing the polymerase active site (Gerlach et al., 2015).

Finally, it is interesting to consider the functional significance of the c5′-bound polymerase conformation. The conservation of an alternative PB2-C domain arrangement across FluA, B, and C genera reinforces the conclusion that it is functionally important. The conformation is characterized by immobilization of the cap-binding domain against PB1 and packing of the NLS domain against the endonuclease such that cap snatching is disfavored. The most likely possibility is that this conformation corresponds to an incoming apo-polymerase binding to a nascent c5′ end emerging from a replicating vRNP. Such a polymerase would not be required to do cap snatching but could be involved in initiating and guiding progeny cRNP formation by interacting with incoming apo-NP. Only at the end of replication would the polymerase bind the c3′ end to complete progeny RNP formation, at the same time activating the polymerase by a switch in conformation. A mechanistic model of this kind has recently been proposed to describe RNP replication by the related bunyavirus polymerase (Gerlach et al., 2015). Alternatively, this conformation could represent that of the replicase rather than of the transcriptase, which again would not be expected to do cap snatching, but could receive a 3′ end template donated by another RNP, as suggested in some models of replication (Jorba et al., 2009). In this scenario, the parental and incoming polymerase heterotrimers could initially form an asymmetric dimer, with the incoming polymerase possibly being activated by binding to small virus-generated RNAs (svRNAs) that correspond to truncated 5′ end vRNAs and that accumulate in infected cells (Perez et al., 2012, Umbach et al., 2010). Clearly, further careful experiments are required to clarify the role of these alternative polymerase conformations. Finally, it is possible that particular interacting host factors (or indeed a second interacting polymerase, NP or NEP) could stabilize one or other polymerase conformation, thus preferentially promoting for example transcription, vRNA to cRNA replication, or the reverse. This is plausible given that several well-known host-specific variants are located in PB2-C domains, particularly the 627 and NLS domains (Cauldwell et al., 2014). Related to this is the intriguing possibility that the intra-molecular, inter-subunit binding of the PB2 NLS peptide to the PA endonuclease could play a role in the release of PB2 from the nuclear import factor importin α (which also binds the NLS peptide) and thus promote assembly of the heterotrimer through the interaction of PB2 with the PA-PB1 heterotrimer. The specific, but genera-dependent, interaction between the PB2 NLS and the endonuclease could also partly explain why transplanting of an unrelated NLS does not rescue the replication defect of mutations in the PB2 NLS (Resa-Infante et al., 2008).

Experimental Procedures

Protein Production, Crystallization, and Structure Determination

Residues 247–736 from the A/Vietnam/1203/2004(H5N1) PB2 subunit were expressed in E. coli. Influenza B/Memphis/13/03 polymerase, purified as described previously (Reich et al., 2014), was mixed with either nucleotides 1–12 of the vRNA 5′ end (5′-pAGUAGUAACAAG-3′) or cRNA 5′ end (5′-pAGCAGAAGCAGA-3′). Residues 1–197 of FluB PA were expressed in E. coli, and purified endonuclease was mixed with synthetic FluB PB2 NLS peptide (residues 742–770). Crystallization trials were performed with a Cartesian robot and diffraction data measured on European Synchrotron Radiation Facility (ESRF) MX beamlines. Structures were solved by molecular replacement with PHASER and refined using REFMAC5 within the CCP4i package (Winn et al., 2011).

Endonuclease Activity Assays

FluA or FluB endonuclease domain activity was assayed as a function of bound NLS peptide using a FRET-based method (Kowalinski et al., 2012) with substrate 6-FAM-5′-CUCCUCAUUUUUCCCUAGUU-3′-BHQ1 (IBA). For cap-dependent endonuclease activity, FluB polymerase was mixed with a 32P-labeled capped 20-mer RNA (5′-AAUCUAUAAUAGCAUUAUCC-3′) and no RNA or v3′, c5′, v5′, or v3′+v5′ RNA and the time course of degradation monitored on a 20% acrylamide-7 M urea gel.

Biophysical Methods

The Kd for FluB NLS peptide binding to the endonuclease was derived from ITC measurements performed at 25°C using an ITC200 Micro-calorimeter (MicroCal). SEC-MALLS was performed on FluB polymerase with or without 12-mers of v5′ or c5′ RNA bound. SAXS measurements were performed on ESRF beamline BM29 on FluB polymerase with either no bound RNA or with only v3′, only v5′ or both v3′ and v5′ RNAs as they eluted from an online size-exclusion column. Automatic data processing used the ATSAS package (Petoukhov and Svergun, 2007).

Crosslinking Mass Spectrometry

Crosslinking was performed using recombinant FluA or FluB polymerase or reconstituted FluA-vRNA polymerase complex by addition of isotope-labeled DSS or DSG as described previously (Leitner et al., 2014). Protein digestion was performed at 37°C using LysC for 4 hr and trypsin for 12 hr, and crosslinked peptides were enriched using gel filtration. Fractions were analyzed by liquid-chromatography-based mass spectrometry using a nanoAcquity ultraperformance liquid chromatography column connected to a LTQ Orbitrap Velos Pro instrument (Thermo Scientific). Mass spectrometry data were processed using the xQuest/xProphet. Identified crosslinks were mapped to known polymerase structures and analyzed using Xlink Analyzer (Kosinski et al., 2015).

Author Contributions

E.T., supervised by D.H., expressed and crystallized the H5N1 PB2-C and solved the structure with the help of A.P. D.G. crystallized FluB polymerase with c5′ or v5′ RNA; performed endonuclease assays with full-length polymerase; cloned, expressed, and first crystallized the FluB endonuclease; and provided samples for biophysical experiments. S.G. crystallized FluB endonuclease with the NLS peptide and performed ITC and FRET assays. A.R. performed and analyzed SAXS measurements. N.H. and K.E.O provided FluC polymerase co-ordinates prior to publication. F.B. and T.B. performed crosslinking mass spectrometry with samples provided by D.G., A.P., and F.B. Crosslinking data were analyzed by J.K., T.B., F.B., and M.B, who supervised the crosslinking mass spectrometry. S.C. did crystallographic analysis of the FluB c5′, v5′, and endonuclease-NLS complexes, overall directed the project, and wrote the paper with input from the other authors.