FluPol is a highly flexible protein complex; however, the conformational states it can adopt are uncharacterized. Understanding the nature of these conformational states is central to determining the regulatory mechanisms of this enzyme. To this end, we have determined the structure of FluPol from influenza C virus5 (FluPolC), in the absence of promoter RNA. We expressed all three individual subunits of FluPolC in insect cells by infection with a single baculovirus construct. FluPolC purified from this system was active in both replication and transcription initiation (Extended Data Fig. 1). We crystallized apo-FluPolC in two different crystal forms (Extended Data Table 1), and solved its structure at 3.9 Å (Extended Data Fig. 2) and 4.3 Å resolution.

Our model of FluPolC (Fig. 1) comprises 711 of the 754 residues of PB1 (94.3%), 762 out of 774 for PB2 (98.4%) and 693 out of 709 for P3 (97.7%). FluPolC forms a relatively compact structure (Fig. 1a, b). P3 folds into two domains connected by a long linker (Fig. 1c): an amino-terminal endonuclease domain (P3endo) and a C-terminal domain (P3C), which sandwiches PB1 at the heart of the molecule. PB1 has the canonical right-hand-like polymerase fold, possessing palm, fingers and thumb subdomains with additional N- and C-terminal extensions (PB1N-ext and PB1C-ext) that facilitate interactions with the other subunits (Fig. 1d). The thumb of PB1 is reinforced by P3C. The priming loop of PB1, believed to facilitate de novo replication initiation4, is not visible in our structure and is probably disordered. PB2 stacks against one face of PB1, contacting both domains of P3. PB2 comprises 9 domains: the N-terminal PB1 interaction domain (PB2N-ter), PB2N1, PB2N2, PB2lid and PB2mid domains, a cap-binding domain (PB2cap), a linker domain (PB2cap-627 linker), the 627 domain (PB2627) and a C-terminal nuclear localization signal (NLS) (PB2NLS) domain (Fig. 1e).

Figure 1: Structure of FluPolC.
figure 1

a, b, Two views of the structure of the FluPolC heterotrimer, coloured according to subunit (PB1, orange; PB2, green; P3, blue). The cap-binding pocket and endonuclease active site are shown as blue and red spheres, respectively. In a, the PB1 catalytic aspartates, residues 446 and 447, are also highlighted red. The position of PB2 residue 649 (equivalent to PB2 residue 627 in FluPolA) is marked by a purple sphere. ce, Structures of FluPolC subunits P3 (c), PB1 (d) and PB2 (e), coloured and labelled by domain. f, Domain maps of each FluPolC subunit.

PowerPoint slide

The fold of each FluPolC domain is very similar to its counterpart in FluPolA and FluPolB, even though the sequence identity between these polymerases is only ~30% (Extended Data Table 2 and Supplementary Fig. 1). The average root mean squared deviation (r.m.s.d.) values of Cα atoms between equivalent superposed domains of FluPolC and FluPolA, or of FluPolC and FluPolB are 1.6 Å or 1.5 Å, respectively, demonstrating that the FluPol fold is conserved across influenza A, B and C viruses. All key active site residues within FluPolC are structurally conserved, and we confirmed, by mutation, that FluPolC shares common mechanisms with FluPolA (Extended Data Fig. 3a). The PB1 subunits of FluPol A, B and C belong to a structural grouping that most closely resembles the polymerases of Reoviridae and Cystoviridae/Flaviridae (Extended Data Fig. 3b).

However, there are substantial differences between apo-FluPolC and the activated structures for promoter-bound FluPolA and FluPolB. Most striking are the position of P3endo and the arrangement of the C-terminal domains of PB2 (Fig. 2, Supplementary Video 1 and Extended Data Table 3). Thus, PB2627, which in FluPolA houses a crucial polymorphism (Glu627Lys) for the determination of viral host range and pathogenicity6, lies level with the endonuclease domain in the apo structure, (Fig. 2a), whereas in the activated structures it lies close to PB1palm (Fig. 2b). The PB2mid and PB2cap-627 linker domains are rearranged en bloc, by a rotation of 140° and a translation of 30 Å, between the apo and activated conformations. PB2cap also changes; in the apo structure, it is tucked in between the PB1palm and PB2cap-627 linker, while in the activated structures it does not extensively contact other domains. Finally, PB2NLS packs between PB1C-ter helix α23 and P3endo helix α7 in apo-FluPolC (Fig. 3a), but is near the base of the PB1palm in the activated structures (Fig. 2b). The movement of PB2NLS amounts to a rotation of 130° and a translation of 90 Å. The regions rearranged within PB2 match those that are disordered in the FluB2 structure4, lying immediately downstream of a conserved glycine (PB2 residue 255 in FluPolC).

Figure 2: Comparison of apo-FluPolC with promoter-bound FluPolA.
figure 2

a, Apo-FluPolC, depicted in the same orientation and colouring as in Fig. 1b, but with the C-terminal domains of PB2 coloured as in Fig. 1e. Domains that do not change between apo and promoter-bound conformations are depicted as semi-transparent. b, Promoter-bound FluPolA, shown as in a. c, d, The PB2 subunits of apo-FluPolC (c) and promoter-bound FluPolA (d), depicted as in a and b.

PowerPoint slide

Figure 3: Critical changes between apo-FluPolC and promoter-bound FluPolA.
figure 3

a, Equivalent views of FluPolC (left) and FluPolA (right), showing domain arrangement differences. The inset shows the arrangement of P3endo within the FluPolC structure. b, Close-up of the FluPolC (left) and FluPolA (right) cap-binding domains. PB2 residues 520–535 in FluPolC are coloured dark orange. The cap-binding pocket is shown with blue spheres. c, d, Two views of a superposition of apo-FluPolC and FluPolA, with FluPolC coloured as in Fig. 1 and FluPolA in lighter colours. 5′ and 3′ promoter RNAs in the FluPolA structure are coloured pink and yellow, respectively. Arrows highlight differences between the two conformations.

PowerPoint slide

The buried area between PB2 and P3 (5,000 Å2) is more extensive in the apo than the activated, promoter-bound structures, reflecting new contacts between PB2 and P3endo (Fig. 3a and Extended Data Fig. 4a, b). Additionally, the extreme C terminus of PB2 is visible in our apo structure (Extended Data Fig. 4b), forming a helix (α30) that packs against the back face of P3endo (near P3 helices α2, α3 and α7). There are also many new contacts between PB2 and PB1 (Fig. 2 and Extended Data Fig. 4c).

One important consequence of the arrangement seen in apo-FluPolC is that the cap-binding pocket is occluded (Fig. 3b). PB2cap is folded in on the rest of the subunit, facing residues 520–535 from the PB2cap-627 linker domain. This is consistent with the observation that promoter RNA is required for FluPol cap-binding and endonuclease activity7,8. Thus, the structure we observe represents a closed, pre-activation state of FluPol and suggests that the viral RNA (vRNA) promoter causes the rearrangements necessary to form the activated structure.

Alternatively, the structure of FluPolC could indicate a fundamental conformational difference between FluPols from different influenza viruses. To clarify this, we performed small angle X-ray scattering (SAXS) experiments with FluPolC, as these allowed us to distinguish between closed and activated FluPol conformations (Extended Data Fig. 5a). The observed scattering profiles from FluPolC were similar to a profile calculated from the FluPolA crystal structure, indicating that FluPolC can adopt the same activated conformation (Extended Data Fig. 5b). Thus, the change we see in the FluPolC crystal is not an influenza virus type difference. However, promoter RNA was not required for the activated conformation to be detected, indicating that changes between apo and promoter-bound structures need not exclusively be caused by RNA binding. This suggests that the energy barrier between different FluPol conformations is low. Indeed, when placed into a phosphate-based buffer, FluPolC adopted a currently uncharacterized conformation that was even more open than that of the promoter-bound structures (Extended Data Fig. 5c). These results suggest that FluPol may be poised between several different conformations, with only subtle environmental changes needed for a particular conformation to be favoured.

In line with this assessment, differences around the promoter-binding site between the apo-FluPolC and promoter-bound structures are small. Minor changes are evident around the pocket that binds the intra-base paired hook structure of the 5′ strand of the vRNA promoter (Extended Data Fig. 6); however, sequence alignments suggest that these differences are influenza-virus-type-specific. More interesting are the differences around the binding site for the 3′ strand of the vRNA promoter. In the apo structure, PB2 helix α4, PB2N1 and the associated region of PB1thumb lie ~5 Å further away from the polymerase core than in the promoter-bound structures (Fig. 3c). This change is transmitted to the neighbouring PB1C-ext–PB2N-ter interaction domain through PB1 helix α22, resulting in a 20° rotation of this domain between the apo and vRNA promoter-bound conformations (Fig. 3d). Since the PB1C-ext–PB2N-ter domain lies next to PB2NLS in apo-FluPolC (Fig. 3a), this rotation could trigger the movement of PB2NLS from its apo position, leading to the subsequent massive reorganization of FluPol after vRNA–promoter binding (Supplementary Video 2).

Notably, only one currently reported FluPol structure (FluB2) contains a fully ordered vRNA promoter (in the others, the 3′ vRNA strand is either truncated or partially disordered)4. However, this does not display a stable activated conformation, as the C-terminal two-thirds of PB2 are not resolved. We suggest that this is because initial binding of the vRNA promoter, into a resting position away from the active site, generates a dynamic equilibrium between closed and activated conformations. The activated structure is only seen when the 3′ end of the 3′ vRNA promoter strand is either not present or disordered3,4. Hence, the activated conformation might only be fully stabilized when this 3′ end is released from its resting position to enter the polymerase active site. To test this hypothesis, we compared the ability of a full-length or truncated (lacking four nucleotides at the 3′ end of the 3′ strand) vRNA promoter to stimulate FluPolC cap-dependent cleavage activity (Fig. 4). We reasoned that stabilization of an activated over a closed conformation would enhance capped-RNA cleavage, as the cap-binding pocket in PB2 becomes more accessible. In line with this, we observed a significant enhancement in capped RNA cleavage in the presence of the truncated promoter RNA (Fig. 4). The relative inefficiency of the full-length vRNA promoter to stimulate cleavage supports our assertion that initial promoter binding results in a closed/activated equilibrium. The mechanism behind this may involve the PB1 β-ribbon (177–212 in FluPolA)3, which is disordered in the apo-FluPolC structure, but adopts different conformations in the activated and FluB2 structures.

Figure 4: Capped-RNA cleavage assays with FluPolC.
figure 4

a, Representative autoradiograph of a capped-RNA cleavage assay. In each reaction, radiolabelled capped RNA was incubated at 30 °C for 2 h with FluPolC and the indicated strands of the vRNA promoter. b, Quantification of cleavage, expressed as the percentage of cleaved to total RNA, from three replicates of this assay, performed with the same polymerase preparation. Mean cleavage percentage is plotted. Error bars show s.d. Asterisk indicates a significant difference between cleavage with full or truncated vRNA promoter (n = 3, P = 0.0003, two-tailed t-test).

PowerPoint slide

In summary, we have solved the structure of the RNA polymerase from an influenza C virus in the absence of RNA, uncovering a closed conformation accessible to FluPol. Our structure explains the observation that FluPol in the absence of promoter RNA is unable to perform cap-snatching7,8, and we propose a mechanism for how vRNA promoter might bring about FluPol activation. However, the closed conformation captured here may have a wider functional relevance, because it could still be accessible to FluPol bound to a fully ordered vRNA promoter that does not enter the active site. Therefore, in the context of a non-transcribing viral ribonucleoprotein complex (RNP), containing FluPol, RNA and nucleoprotein, FluPol may well adopt this closed conformation. In addition, dependent on stabilization of the PB1 priming loop, the closed conformation that we observe might still allow de novo initiation, as this is not dependent on cap-snatching. Thus, the conformation that we observe, in addition to being a transcription pre-activation state, could be relevant during genome replication initiation. This would allow the activity of FluPol within an RNP to be regulated by other viral factors and host proteins9,10,11,12.

Our work underlines the tremendous flexibility of this protein complex. This flexibility offers an explanation for the differences between several low-resolution electron microscopy reconstructions of RNP-associated FluPol, as well as explaining why the promoter-bound structures do not fit well into these reconstructions13,14,15,16. Furthermore, since negative-sense RNA virus polymerases share a common organization, with a central polymerase core surrounded by various functional modular appendages1,17, the conformational flexibility revealed here might be a theme among all of these polymerases and not just particular to FluPol18.


Protein expression and purification

The three subunits of the influenza C/Johannesburg/1/1966 virus polymerase (PB1: AAF89738, PB2: AAF89739, P3: AAF89737) were co-expressed in Sf9 cells from codon-optimized genes (GeneArt) cloned into a single baculovirus using the MultiBac system19. Expression and purification of FluPolC proceeded as previously described for FluPolA12, except that the gel filtration buffer used was 0.5 M NaCl, 25 mM HEPES-NaOH, pH 7.5 and 10% (v/v) glycerol. For crystallization and storage, protein purified in this buffer was supplemented with 0.5 mM TCEP, and 10 mM MgCl2 or 10 mM CaCl2.

Crystallization, data collection and structure determination

Crystals of FluPolC, belonging to two different space groups, grew from sitting-drop vapour-diffusion experiments at 20 °C20,21, set up using a protein:precipitant ratio of 2:1. In these experiments, 5 mg ml−1 protein was mixed with either 70% (v/v) Morpheus G2 (Molecular Dimensions), supplemented with 0%–1% 1 M NaOH, to generate P43212 crystals; or with crystal-seeds and 0.2 M NaCl, 0.1 M Na-HEPES, pH 7.5 and 25% (w/v) PEG 4000, for P212121 crystals. For heavy atom derivatization, P43212 crystals were soaked in a solution of gold(i) potassium cyanide dissolved in mother liquor, for 2–3 h at 20 °C. Crystals were cryo-protected using 25% (v/v) glycerol in crystallization buffer, before flash-cooling in liquid nitrogen, and data collection on beamlines I03 and I04 at the Diamond Light Source, Didcot, UK. The beam size was matched to the crystal size and data were collected on a Pilatus 6M detector at a wavelength of 0.9763 Å (tetragonal native), 1.0350 Å (tetragonal derivative) and 0.9795 Å (orthorhombic native). Data collection statistics are shown in Extended Data Table 1. Data were processed using Xia2 (ref. 22) and HKL2000 (ref. 23). Initial phases were obtained by single isomorphous replacement with anomalous scattering (SIRAS), using data from native P43212 and gold-derivatized crystals. The P43212 data used at this stage was collected earlier (at a wavelength of 0.8634 Å) than that subsequently used in refinement. Heavy atoms were located with SHELX24 and phases improved by two-fold non-crystallographic averaging (the crystallographic asymmetric unit contained two heterotrimers) and solvent flattening (solvent content 76%) using Phenix.autosol25. The tetragonal and the orthorhombic data were sharpened to 40 Å2 and 36 Å2, respectively. The P212121 crystals were solved by molecular replacement (program Phaser26), using the P43212 structure as the search model. As expected the orthorhombic crystals possessed four heterotrimers in the crystallographic asymmetric unit, allowing phase improvement using non-crystallographic symmetry (NCS) averaging and solvent flattening using general averaging program (GAP) (D.I.S. and J.M.G., unpublished observations). The published fragments of FluPolA (PDB accessions 4IUJ, 4AWH, 4CB4, 3A1G and 2VY7) were fitted by eye using Coot, which was used for all model building27. Comparison with the complete FluPolA and FluPolB structures (4WSB and 4WSA, respectively), aided by the anomalous scattering from the sulphur atoms as markers, allowed us to build and refine complete models for FluPolC. This provided a total of six independent views of the polymerase. Performing superpositions of these demonstrated that the molecule adopts a virtually identical conformation across all copies from both crystal forms (mean pairwise r.m.s.d. in Cα was 0.94 Å between all pairs of molecules across both space groups). Refinement (Extended Data Table 1) used BUSTER28 aided by NCS and initially phase restraints, and REFMAC29 with secondary structure restraints using PROSMART30.

SAXS experiments

SAXS measurements were performed on beamline B21 at Diamond Light Source, Didcot, UK. Samples were prepared onsite using a Shodex Kw-403 size exclusion column and Agilent HPLC. Approximately 40–60 μl of sample were collected for SAXS at 20 °C using a sample to detector distance of 3.9 m and X-ray wavelength of 1 Å. Samples were exposed for 300 s in 10 s acquisition blocks. Images were corrected for variations in beam current, normalized for exposure time and processed into 1D scattering curves using GDA and DAWN. Buffer subtractions and all other subsequent analysis were performed with the program ScÅtter ( Samples were checked for radiation damage by visual inspection of the Guinier region as a function of exposure time.

Data analysis

Figures and videos were prepared using PyMOL ( and Chimera31. Structural comparisons used SHP32.

Polymerase activity assays

For the cap-dependent cleavage and transcription assays, FluPolC (400 ng per reaction) was incubated for 2 h at 30 °C with or without (as indicated) NTPs (1 mM ATP, 0.5 mM each CTP/UTP/GTP), radiolabelled capped 20-nucleotide or 11-nucleotide RNAs, 0.6 μM each 5′ and 3′ vRNA promoter strands, in a reaction buffer containing 7.5 mM MgCl2, 1.0 mM TCEP, 2 U μl−1 RNasin (Promega), 20 mM HEPES-NaOH, pH 7.5, 100 mM NaCl and 5% (v/v) glycerol. For the de novo initiation and elongation assays, FluPolC (400 or 800 ng per reaction, as indicated) was incubated for 2–3 h at 30 °C with 2.5 mM adenosine and 0.075 μM [α-32P]GTP or 1 mM ATP, 0.5 mM each CTP/UTP, 0.1 mM GTP, 0.3 μM [α-32P]GTP and (as indicated) 0.6 μM each 5′ or 3′ vRNA promoter strands, in the same reaction buffer as above. The reaction volume was 4 μl for all reactions. Products were denatured by boiling (98 °C, 5 min) after the addition of formamide (4 μl) and separated on a denaturing 20% polyacrylamide gel, with the indicated size markers. Products were visualized by autoradiography. For all activity assays except the cleavage assays, the sequences of the promoter RNA oligonucleotides used were: 5′-AGCAGUAGCAAGGAG-3′ (5′ vRNA) and 5′-CUCCUGCUUCUGCU-3′ (3′ vRNA). The sequences of the RNAs used in the capped-RNA cleavage assays were 5′-AGCAGUAGCAAGGGG-3′ (5′), 5′-UAUACCCCUGCUUC-3′ (3′ truncated) or 5′-UAUACCCCUGCUUCUGCU-3′ (3′ full length).

Capped and radiolabelled RNA was produced by incubating 5′ diphosphate synthetic 20-nucleotide (5′-ppAAUCUAUAAUAGCAUUAUCC-3′)3,4 or 11-nucleotide (5′-ppGAAUACUCAAG-3′)33,34 RNA (Chemgenes), with [α-32P]GTP, vaccinia virus capping enzyme (NEB) and 2′-O-methyltransferase (NEB), following the manufacturer’s instructions. The resulting RNAs were gel purified before use in the above assays.