RNA virus RNA-dependent RNA polymerases (RdRps) catalyse the formation of the first phosphodiester bond of the nascent strand either de novo or in a primer-dependent manner1. A key structural feature that differentiates these two mechanisms is the priming loop, which is a loop structure that protrudes from the palm5 or thumb subdomain6,7 towards the active site of de novo-initiating RdRps. Strictly primer-dependent RdRps lack such a structure8. An important function of the priming loop is to act as stacking platform for the initiating NTP during de novo initiation on the 3′ terminus of the bound single-stranded RNA (ssRNA) template6,9, i.e. to ensure terminal initiation. The interaction between the NTP and loop is typically mediated by a conserved aromatic residue, which buttresses the sugar–base of the initiating NTP6,9,10. A second role of the priming loop is to provide selection against double-stranded RNA (dsRNA) and ensure that only an ssRNA terminus is positioned into the active site10–12. This is consistent with the observation that the priming loop undergoes a conformational change, either retracting like a compressed spring9 or folding into the minor groove of the product duplex5, to allow egress of a product duplex.

Some RdRps do not follow the above paradigm and use a combination of de novo and primer-dependent initiation to replicate and transcribe their genomes4,13. One of the best-studied examples of this type of polymerase is the RdRp of the influenza A virus. The genome of this virus consists of eight negative-sense viral RNA (vRNA) segments. Each segment is bound by multiple nucleoprotein (NP) molecules and one copy of the RdRp, which yields a complex called the viral ribonucleoprotein (vRNP)4. It is in the context of this vRNP that the RdRp replicates vRNA via a complementary RNA (cRNA) intermediate and transcribes vRNA into messenger RNA (mRNA)4. All activities depend on the binding of the RdRp to conserved 5′ and 3′ terminal sequences present in each vRNA and cRNA. These terminal sequences constitute the vRNA and cRNA promoters, respectively4.

The influenza A virus RdRp is formed by polymerase basic protein 1 (PB1), polymerase basic protein 2 (PB2), and polymerase acidic protein (PA) (Supplementary Fig. 1a). Together these three subunits form a central core, which carries out viral RNA synthesis, and multiple flexible domains at the periphery of this core, including cap-binding and endonuclease domains14 (Supplementary Fig. 1a). During transcription, the cap-binding and endonuclease domains enable the RdRp to snatch capped oligonucleotides from nascent cellular RNA polymerase II (Pol II) transcripts. After alignment of a capped oligonucleotide to the 3′ terminus of the vRNA template, the RdRp initiates transcription at position 2 or 3 of the vRNA template, depending on the sequence of the primer (Supplementary Fig. 1b)4. In contrast, during replication the RdRp initiates de novo at positions 1 and 2 of the 3′ terminus of the vRNA template (terminal initiation) to copy the vRNA into a cRNA (Supplementary Fig. 1c)15. When the RdRp next copies this cRNA back into vRNA, it initiates de novo at positions 4 and 5 of the 3′ terminus of the cRNA template (internal initiation) and produces pppApG (Supplementary Fig. 1d). This dinucleotide is subsequently moved to bases 1 and 2 of the cRNA 3′ terminus, where it acts as a primer for vRNA synthesis (Supplementary Fig. 1d)15.

To identify functional RdRp elements required for the initiation of influenza A virus transcription and replication, we superposed the P2 initiation complex (Supplementary Fig. 2) with the bat influenza A virus RdRp. The superposed structure places the tip residues of a PB1 β-hairpin (PB1 residues 642–658) close to the 3′ sugar–base of the template and the sugar–base of the initiating nucleotide (Fig. 1a). This indicates that this β-hairpin may act as a stacking platform for terminal initiation on U at position 1 and C at position 2 of the vRNA promoter (Fig. 1b). The β-hairpin may also act as a stacking platform for internal initiation on U at position 4 and C at position 5 of the cRNA promoter (Fig. 1c), but only if the 3′ terminus of the cRNA promoter can pass through the active site. A superposed structure of the bat influenza A virus RdRp with the poliovirus 3Dpol elongation complex (Supplementary Fig. 3) shows that the PB1 β-hairpin does not appear to obstruct egress of the template strand and that the PB1 β-hairpin might thus support internal initiation in its modelled position. Most of the PB1 β-hairpin is unresolved in the structures of the RdRps of influenza B and C viruses16,17, but a sequence alignment of PB1 indicates that the β-hairpin is present in all influenza virus RdRps (Supplementary Fig. 4).

Figure 1: The PB1 β-hairpin is essential for influenza A virus RNA synthesis in cell culture and terminal de novo replication initiation in vitro
figure 1

a, View of the influenza A virus RdRp active site with superposed φ6 RdRp P2 initiation complex. For clarity only the conserved PB1 motifs A–F are shown, coloured yellow, orange, pink, red, cyan and pale green, respectively. The PB1 β-hairpin (abbreviated ‘PL’) is shown in blue. Magnesium ions that coordinate the GTPs from the φ6 complex are depicted as grey spheres and the 3′ end of the template strand is indicated. b,c, Model of the influenza A virus RdRp active site and dsRNA channel (labelled ’DC’) during terminal de novo initiation on the 3′ end of a vRNA template (dark grey; b) or during internal de novo initiation on the 3′ end of a cRNA template (c). The PB1 β-hairpin is shown as a blue dotted line. Black dotted lines indicate hydrogen bonds and grey spheres represent the magnesium ions that are coordinated by motifs A (yellow) and C (pink). The template and nascent strand exit channels are labelled ‘TC’ and ‘NSC’, respectively. d,e, Analysis of the steady-state RNA levels in RNP reconstitution assays (Supplementary Fig. 5) with an NA vRNA (d) and NA cRNA template (e). Ribosomal 5S rRNA was used as loading control. Asterisks denote an uncharacterized product. Graphs show the mean RNA levels with error bars indicating s.d. (n = 3). f, Terminal pppApG synthesis on a vRNA promoter and g, internal pppApG synthesis on a cRNA promoter. The structures of the promoter strands are as in Supplementary Fig. 1 and based on published sources14,30, with the radiolabelled GTP shown in red and the unlabelled ATP in blue. The gel images show the main product after treatment with SAP to aid the separation of the radioactive product from the [α-32P]GTP substrate and radioactive contaminants (see Supplementary Fig. 8 for not-treated reactions), and detection by autoradiography. Graphs in f and g show the mean percentages of polymerase activity relative to that of the wild-type using three independent RdRp preparations. Error bars represent s.d. (n = 3).

Full size image

To investigate the role of the PB1 β-hairpin in influenza A virus RNA synthesis, we engineered a mutant in which the four conserved tip residues A648, H649, G650, and P651 (Supplementary Fig. 4) of the influenza A/WSN/33 virus PB1 were deleted (Δ648–651). We first tested the activity of this mutant in cell culture by measuring the synthesis of neuraminidase (NA)-encoding viral RNA in RNP reconstitution assays (Supplementary Fig. 5). As negative control for viral RNA synthesis, we used a PB1 mutant in which the two conserved aspartates of motif C were mutated to alanine (DD445–446AA)18. Examination of the replication (vRNA and cRNA) and transcription (mRNA) products revealed that Δ648–651 had failed to efficiently replicate and transcribe the NA vRNA (Fig. 1d). No replication or transcription was observed in the DD445–446AA control. Similar results were observed when we substituted the plasmid providing the NA vRNA with a plasmid expressing an NA cRNA (Fig. 1e). Together, these results indicate that the PB1 β-hairpin is critical for efficient influenza A virus RNA synthesis.

The RNP reconstitution assay above did not allow us to distinguish whether transcription or replication were differentially affected due to their interdependency in cell culture. To study the role of the PB1 β-hairpin in more detail, mutant and wild-type RdRps were purified from 293T cells. All three recombinant enzymes formed hetrerotrimeric complexes with the three RdRp subunits present at an approximately 1:1:1 ratio (Supplementary Fig. 6a). Moreover, wild-type and Δ648–651 showed similar promoter binding behaviour in a single-molecule Förster resonance energy transfer (sm-FRET) assay19 (Supplementary Fig. 7), which indicates that deletion of the tip of the β-hairpin does not affect the overall conformation of the RdRp complex. We next tested their ability to support terminal and internal pppApG synthesis using [α-32P]GTP and ATP on influenza A virus RNA promoters. Alkaline phosphatase was added after the reaction to aid the separation of the radioactive product from the [α-32P]GTP substrate and radioactive contaminants (see Supplementary Fig. 8 for non-treated reactions). As shown in Fig. 1f, Δ648–651 exhibited a tenfold reduced terminal initiation activity on a vRNA promoter, but internal initiation activity on a cRNA promoter remained unaffected (Fig. 1g). No pppApG synthesis was observed in the absence of RdRp or with DD445–446AA. Using uridine to adenosine substitutions at positions 4 (4U→A) or 1 (1U→A) of the cRNA promoter we confirmed that initiation on the cRNA promoter occurred internally as expected (Supplementary Fig. 9).

A explanation for the finding described above that absence of a full-length PB1 β-hairpin results in a tenfold decrease in terminal, but not internal, initiation is that the energy requirements for these two initiation modes are different. Phosphodiester bond formation depends on multiple steps that affect the activation energy of initiation, including template and substrate binding, active site rearrangements, and the stabilisation of the initiating NTP by a priming loop. To investigate whether the 3′ UCA overhang upstream of the cRNA initiation site had an effect on the activation energy, we calculated the free energy difference (ΔΔG°) between the template-dinucleotide duplexes that are formed during internal and terminal initiation (see Methods and Supplementary Table 1 for details) and found a ΔΔG° of −1.5 kcal mol−1. This difference in thermodynamic stability means that the 3′ UCA overhang upstream of the cRNA initiation site lowers the activation energy of internal de novo initiation 11-fold compared with terminal de novo initiation, which may, at least in part, explain why the RdRp does not need a full-length priming loop to stabilize the initiating ATP during internal initiation.

De novo initiation represents only the first nucleotide incorporation event in influenza A virus RNA replication. To explore the role of the β-hairpin in nascent strand extension, we assayed the ability of Δ648–651 to synthesize pApGpC from 5′ radiolabelled pApG and CTP on a vRNA promoter. As shown in Fig. 2a, we observed no significant difference in activity between the wild-type and Δ648–651, while no extension of the pApG was observed in the control reactions with DD445–446AA or without RdRp. Similar results were observed during trinucleotide synthesis on a cRNA promoter using pApG and UTP (Fig. 2b). However, when we measured progressive elongation of an ApG in the presence of [α-32P]GTP, ATP, and CTP on a vRNA promoter, Δ648–651 exhibited significantly higher activity than the wild-type enzyme (Fig. 2c). We also measured the extension of ApG, in the presence of [α-32P]GTP, ATP, and UTP on a cRNA promoter, mimicking progressive pppApG-extension after primer realignment according to the prime–realignment model (Supplementary Fig. 1d). As shown in Fig. 2d, we observed that the activity of the Δ648–651 mutant was also higher than that of the wild-type on the cRNA promoter. This implies that a complete PB1 β-hairpin limits progressive extension during replication. When we superposed the bat influenza A virus RdRp with the poliovirus 3Dpol elongation complex, the PB1 β-hairpin clashed with the product duplex (Supplementary Fig. 10a). This indicates that the PB1 β-hairpin may have to undergo a conformational change to allow template–product duplex egress (Supplementary Fig. 10b), which might be a rate-limiting step in influenza A virus RNA synthesis.

Figure 2: The PB1 β-hairpin loop is not essential for extension during replication.
figure 2

a,b, Extension of a 5′ radiolabelled pApG with CTP on a vRNA promoter (a) and with UTP on a cRNA promoter (b). Top schematics show two-dimensional models of the reactions with the radiolabelled pApG shown in red and the unlabelled CTP and UTP in blue. The promoter structures are depicted as in Fig. 1. c,d, Elongation of ApG on a vRNA promoter in the presence of ATP, CTP and [α-32P]GTP (c), or on a cRNA promoter in the presence of ATP, UTP and [α-32P]GTP (d). The schematics show models of the reaction with colours as in Fig. 1. The major RNA products are labelled with ‘P’. The graphs in c and d show the mean percentages of polymerase activity relative to wild-type using three independent RdRp preparations. Error bars represent s.d. (n = 3).

Full size image

In contrast to replication, transcription of influenza A virus relies on a primer-dependent initiation mechanism (Supplementary Fig. 1b). To examine the role of the PB1 β-hairpin in this process, we used a radiolabelled, capped 11 nt RNA as primer that can directly anneal with the template (Supplementary Fig. 1b). As shown in Fig. 3a, we observed more efficient elongation in the presence of Δ648–651 compared with the wild-type RdRp. A similar result was obtained when we made the transcription reaction dependent on endonuclease activity and used rabbit β-globin mRNA as cap donor in our assay (Fig. 3b). No transcription initiation was observed in the negative controls (Fig. 3a,b). Interestingly, we found an increase in the cleavage activity of Δ648–651 compared with the wild-type RdRp using a radiolabelled, capped 20 nt RNA in the absence of NTPs (Supplementary Fig. 11). Cleavage was also observed for DD445–446AA, but not in the buffer-only control or in a reaction containing a PA endonuclease active site mutant (PA D108A) (Supplementary Fig. 11). Together, these results indicate that the full-length PB1 β-hairpin is not required for transcription initiation and that a trade-off may exist between PB1 β-hairpin-dependent terminal de novo initiation and primer-dependent transcription. To visualize this, we superposed the bat influenza A virus RdRp with the nascent strand of poliovirus 3Dpol elongation complex and found that the PB1 β-hairpin clashes with the 3′ end of the capped-oligonucleotide if it enters the RdRp via the path of the nascent strand (see Supplementary Fig. 10c). This indicates that the PB1 β-hairpin needs to undergo a conformational change to allow unobstructed capped-oligonucleotide entry and base pairing with the template. This conformational change may be similar to the change that facilitates template–product duplex egress (Supplementary Fig. 10d, line ii) or different (Supplementary Fig. 10d, line iii).

Figure 3: The PB1 β-hairpin is not essential for transcription initiation.
figure 3

a, Extension of a capped 11 nt RNA. The major product of the reaction is labelled with ‘P’. b, The schematic shows a two-dimensional model of a transcription reaction with β-globin mRNA as primer donor. The promoter structure and colours are as in Fig. 1. Sequence and priming site of the β-globin primer are based on previously published results31. Graphs represent average activity of three independent experiments using three independent RdRp preparations. Error bars represent s.d. (n = 3).

Full size image

To identify the amino acid at the tip of the PB1 β-hairpin that could act as stacking residue during terminal de novo initiation on the vRNA promoter, we examined the position of the tip residues of the PB1 β-hairpin (Supplementary Fig. 4) in our structural model (Fig. 1a). P651 resides closest to the 3′ sugar-base of the template. Although not considered an aromatic residue, prolines have been found to form stacking interactions with aromatic residues20. An example can be observed in the bat influenza A virus RdRp between PB1 H363 and PA P371 (see PDB entry 4WSB). The position of P651 relative to the modelled position of the template and initiating NTP is, however, not compatible with stacking and more similar to that of the intercalating P65 of the Escherichia coli integration host factor (IHF)21 (Supplementary Fig. 12). H649, the other residue that might stack the initiating NTP, is unresolved in the current bat influenza A virus RdRp structure and putatively oriented towards the triphosphate of the initiating NTP. Consequently, the β-hairpin will have to undergo a conformational change if one or both of these putative stacking residues is involved in supporting the initiating NTP.

To investigate the role of the amino acids at the tip of the PB1 β-hairpin, we engineered recombinant RdRps with alanine substitutions at fully conserved PB1 positions H649 (H649A), G650 (G650A), P651 (P651A), and the partially conserved P647 (P647A). All mutants were able to form a trimeric RdRp complex (Supplementary Fig. 6b) and all mutants of the three conserved tip residues showed a promoter-binding behaviour similar to that of the wild-type (Supplementary Fig. 7d; P647A was not tested), indicating that point mutations in the tip of the β-hairpin did not affect the overall conformation of the RdRp complex. However, individual mutation of all four residues did impair terminal de novo initiation on a vRNA promoter with P651A having the strongest effect (Fig. 4a). The DD445–446AA control mutant showed no dinucleotide formation.

Figure 4: PB1 P651 is essential for terminal de novo initiation during replication.
figure 4

a, Synthesis of pppApG or b, ApG on a vRNA promoter. The upper panels show the major product (labelled ‘pppApG+SAP’ or ’ApG’) after detection by autoradiography. Graphs in the lower panels show the mean polymerase activity relative to the wild-type using three independent RdRp preparations. Error bars represent s.d. (n = 3). c, Analysis of the level of RNA synthesis on an NA vRNA template as described in Supplementary Fig. 5. Ribosomal 5S rRNA was used as a loading control. Graphs show the mean steady state RNA levels of three independent experiments plotted on a logarithmic scale. Error bars represent s.d. (n = 3).

Full size image

To differentiate between amino acid requirements for sugar–base and triphosphate binding, we replaced ATP with adenosine in the reactions. We found that Δ648–651 produced little ApG in comparison to the wild-type RdRp (Supplementary Fig. 13), in agreement with the idea that the PB1 β-hairpin is important for supporting the initiating sugar–base. Next, we tested the point mutants in this assay and found that the P651A mutation severely reduced activity, while the other point mutations resulted in activities that were at least 50% of that of the wild-type RdRp (Fig. 4b). This indicates that P651 in the tip of the PB1 β-hairpin is the most critical for supporting the initiating sugar–base. To further investigate the role of P651, we replaced it with phenylalanine (P651F). This mutant, which formed a trimeric complex and could bind promoter RNA (Supplementary Fig. 6b and 7d), showed increased activity (∼57% of wild-type) relative to P651A (Fig. 4a). This indicates that P651 is involved in stacking interactions, which is consistent with the position of P651 close to the initiating NTP in the structural model (Fig. 1a).

H649 is an alternative amino acid in the PB1 β-hairpin that may support the initiating NTP14. Although, H649A showed stronger de novo initiation activity than P651A with ATP and adenosine as substrate, we investigated its potential involvement in stacking by replacing it with phenylalanine (H649F) or arginine (H649R, Supplementary Fig. 14a). H649F showed decreased activity compared with H649A, while H649R was as active as H649A (Supplementary Fig. 14b). Both mutants formed trimers and showed endonuclease activities similar to that of wild-type RdRp, indicating that their overall structure is not affected by the mutations (Supplementary Fig. 14a and 14c). These results imply that H649 is unlikely to be involved in supporting the initiating NTP through stacking.

To investigate the role of the conserved tip residues in the context of an RNP, we performed an RNP reconstitution assay in 293T cells using an NA vRNA as template (see also Supplementary Fig. 5). Apart from P647A all point mutations greatly affected viral RNA synthesis (Fig. 4c). However, P651A showed the strongest reduction in cRNA synthesis compared with all the other mutants. Moreover, a phenylalanine at position 651 had a less dramatic effect on RNA synthesis than an alanine, a result consistent with those of our in vitro experiments (Fig. 4a,b). This thus indicates that P651 is more important for the coordination of the initiating NTP than H649, but the RNP reconstitution data do demonstrate that H649 is important for the functioning of the RdRp.

In summary, the best interpretation of our results is that the PB1 β-hairpin is essential for terminal de novo initiation during replication (Fig. 1) and that P651 is the primary amino acid involved in the coordination of the initiating NTP during de novo initiation (Fig. 4). Dissection of the steps involved in viral RNA synthesis using in vitro assays revealed that the PB1 β-hairpin is not important for capped-oligonucleotide-dependent initiation on a vRNA template and, surprisingly, also not for internal de novo initiation on a cRNA template (Fig. 1). We believe that the differential requirement for a priming loop during terminal de novo initiation versus internal de novo initiation can be explained by the higher thermodynamic stability of the cRNA template–dinucleotide duplex versus the vRNA template–dinucleotide duplex, which lowers the activation energy of internal de novo initiation compared with terminal de novo initiation. However, in the absence of a ternary structure of the influenza A virus RdRp with a completely resolved PB1 β-hairpin, we cannot fully exclude the possibility that P651 is an important determinant of the structural conformation of the PB1 β-hairpin and that its mutation affects terminal de novo replication initiation indirectly. High-resolution structural data of an influenza virus RdRp in complex with initiating NTPs and template RNA will be therefore needed to confirm our observations and shed further light on the role of the PB1 β-hairpin in the initiation of viral genome replication.

Methods

Plasmids and RNA oligonucleotides

Plasmids pcDNA-NP, pcDNA-PB1, pcDNA-PA, pcDNA-PB2-TAP, and pcDNA-PB1a have been described previously18,22,23. To construct plasmids expressing mutant PB1 proteins, plasmid pcDNA-PB1 was subjected to site-directed mutagenesis using the primers listed in Supplementary Table 2. RNA oligonucleotides used for smFRET were synthesized by IBA, and labelled, purified, and annealed as described previously19.

Sequence alignment and structural modelling

PB1 amino acid sequences from influenza A virus A/WSN/33 (H1N1) (A/WS), influenza B virus B/Michigan/22687/09 (B/MI), and influenza C virus C/JJ/50 (C/JJ) were aligned using ClustalX24 and visualized using ESPript25. To incorporate the φ6 RdRp P2 initiation complex (1HI0) into the model of the bat influenza A virus RdRp crystal structure (4WSB), we aligned active site residues 449–458 of the P2 enzyme with residues 442–449 of the bat influenza A virus RdRp PB1 subunit in Pymol 1.3. The poliovirus 3Dpol elongation complex (3OL7) was aligned to the bat influenza A virus RdRp PB1 subunit using active site residues 324–332.

Purification of recombinant influenza virus RdRp

To purify influenza A/WSN/33 virus RdRp, human embryonic kidney 293T cells (stocks were originally sourced from the ATCC, stored in the Dunn School cell bank at the University of Oxford, and mycoplasma tested, but not authenticated prior to our experiments) were transfected with plasmids expressing PB1 or mutant PB1, PA, and TAP-tagged PB2 using Lipofectamine 2,000 (Invitrogen) according to the manufacturer's instructions. Cells were grown in DMEM (Sigma) supplemented with 10% FCS at 37 °C and harvested 48 h after transfection. TAP-tagged RdRps were isolated through IgG sepharose chromatography (GE Healthcare) and tobacco etch virus (TEV) protease cleavage as described previously23. TAP-tagged RdRps for smFRET experiments were concentrated to ∼50 nM in 50 mM HEPES (pH 7.5), 10% glycerol, 500 mM NaCl, 0.05% n-Octyl β-d-thioglucopyranoside (OTG), and 0.5 mM tris(2-carboxyethyl)phosphine (TCEP) before storage at −80 °C in small aliquots. Wild-type and mutant RdRps were purified side-by-side and analysed through SDS–PAGE and silver staining with a SilverXpress kit (Invitrogen). All experiments were performed with at least three batches of wild-type and mutant RdRps.

RNP reconstitution assays

This assay utilizes the assembly of active vRNPs from influenza virus A virus RdRp, NP and an NA-encoding vRNA expressed from a cellular RNA polymerase I (Pol I) promoter to measure the synthesis of viral RNA in cell culture using primer extension in the presence of a reverse transcriptase (RT) (Supplementary Fig. 5). To compare the activity of mutant RdRps in cell culture, plasmids expressing the PB1, PA, PB2, NP, and the NA vRNA or cRNA of influenza A/WSN/33 virus were transfected into 293T cells using Lipofectamine 2,000 (Invitrogen) according to the manufacturer's instructions. Twenty-four hours after transfection, cells were harvested, the RNA extracted using Trizol (Invitrogen), and the RNA levels assessed using reverse transcription with 32P-labelled oligonucleotides against NA vRNA-derived RNA species and ribosomal 5S rRNA (Supplementary Table 3). 32P-derived signals were resolved using 6% denaturing PAGE, imaged using phosphorimaging on a FLA-5000 scanner (Fuji), and analysed using AIDA (RayTek) and Prism 6 (GraphPad). In all experiments, the apparent RNA levels were normalized to the 5S rRNA loading control and background corrected using the PB1 active site mutant (PB1 DD445–446AA) signal.

sm-FRET promoter binding assays

To assess promoter binding we used a custom-built confocal microscope that allowed alternating-laser excitation (ALEX) of donor and acceptor fluorophores19,26. By placing a Cy3 donor dye on U17 of the 3′ strand and an Atto647N acceptor dye on U6 of the 5′ strand of the promoter, we minimized the possible interference of the dyes with RdRp–promoter complex formation (Supplementary Fig. 7a), while keeping the expected change in distance upon promoter binding at a measurable ∼10 Å (Supplementary Fig. 7b). In a typical experiment, ∼100 nM RdRp was pre-incubated with 1 nM double-labelled promoter RNA in buffer A (50 mM Tris-HCl (pH 8.0), 5% glycerol, 500 mM NaCl, 10 mM MgCl2, 100 µg BSA ml−1) for 15 min at 28 °C. Samples were diluted tenfold in buffer A before the measurements were taken at average excitation intensities of 250 µW at 532 nm and 60 µW at 635 nm. A custom LabView code was used to detect and register signals. The output was analysed using a custom-written Matlab code as described previously19. The apparent energy transfer efficiency (E*) values were plotted as one-dimensional distributions and fitted with a single Gaussian to obtain the mean E* and the associated standard deviation.

In vitro dinucleotide synthesis assays

We performed 3 µl reactions essentially as described previously15. Briefly, for pppApG synthesis assay we incubated a reaction mixture containing 0.05 µM [α-32P]GTP (3,000 Ci mmole−1, Perkin-Elmer), 1 mM ATP, 5 mM MgCl2, 1 mM DTT, 1 U RNasin µl−1 (Promega), 0.7 µM promoter, 5% glycerol, 0.05% NP-40, 75 mM NaCl, 10 mM HEPES pH 7.5, and 5 ng RdRp µl−1. The reactions were incubated at 30 °C for 2 h and depending on the experiment then either heated to 95 °C for 2 min, treated with 0.5 U shrimp alkaline phosphatase (SAP, Promega) for 30 min at 37 °C, and subsequently stopped with 3 µl stopping solution (90% formamide, 10 mM EDTA, 0.01% bromophenol blue and 0.01% xylene cyanol), or directly stopped with 3 µl stopping solution. All reactions were next heated to 95 °C for 2 min and resolved by 23% denaturing PAGE. The level of [α-32P]GMP incorporation was visualized using phosphorimaging as described above. To assess ApG synthesis activity, we replaced ATP in the pppApG synthesis reaction with 1 mM adenosine.

Calculation of activation energies

To calculate the fold difference in activation energy between terminal and internal de novo initiation, we first calculated the ΔΔG° between the internal template-dinucleotide duplex ( A G ⟶ U C A U C ⟵ − − − − − − ) and the terminal duplex ( A G ⟶ U C ⟵ ) from melting curve analyses27,28. Since no melting data was available for 3 nt 3′ overhangs, we used ( A G ⟶ C A U C ⟵ − − − − ) to calculate the ΔΔG° (Supplementary Table 1). Using Δk = e−ΔΔG°/RT (ref. 29), where R is the gas constant, T is 303 K, and ΔΔG° is −1.5 kcal mol−1, this gives an 11-fold difference in activation energy.

In vitro pApG extension assays

Dinucleotide extension assays were performed as described above for the dinucleotide synthesis assays with the key modifications that [α-32P]GTP and ATP were replaced with radiolabelled pApG and 1 mM CTP or UTP, and that no alkaline phosphatase treatment was performed at the end of the reaction. To prepare 5′ radiolabelled pApG, we set up a 10 µl kinase reaction containing 1 mM ApG, 0.17 µM [γ-32P]ATP (3,000 Ci mmole−1, Perkin-Elmer), 1× polynucleotide kinase buffer (Invitrogen), and 1 U T4 polynucleotide kinase µl−1 (Invitrogen). Kinase reactions were incubated at 37 °C for 1 h. The radiolabelled product was next analysed on 23% denaturing PAGE gels, excised, eluted overnight in deionized water, and desalted using NAP-10 columns (GE Healthcare).

In vitro ApG extension assays

To test the extension of an ApG dinucleotide, we set up 3 µl reactions containing 0.05 µM [α-32P]GTP (3,000 Ci mmole−1, Perkin-Elmer), 500 µM ApG, 1 mM ATP, 500 µM CTP or 500 µM UTP, 5 µM MgCl2, 1 mM DTT, 1 U RNasin µl−1, 0.7 µM promoter, 5% glycerol, 0.05% NP-40, 75 mM NaCl, 10 mM HEPES pH 7.5, and 5 ng RdRp µl−1. Reactions were incubated at 30 °C for 1 h and analysed on 20% denaturing PAGE gels.

Capped-oligonucleotide cleavage assays

To analyse the capped-oligonucleotide cleavage activity of the RdRp, first a synthetic 20 nt RNA with 5′ diphosphate (5′-ppAAUCUAUAAUAGCAUUAUCC-3′) (Chemgenes) was capped with a radiolabelled cap-1 structure in 20 µl reactions containing 1 µM RNA, 0.25 µM [α-32P]GTP (3,000 Ci mmole−1, Perkin-Elmer), 0.8 mM S-adenosylmethionine, 0.5 U vaccinia virus capping enzyme µl−1 (NEB), and 2.5 U 2′-O-methyltransferase µl−1 (NEB) at 37 °C for 1 h. The product was analysed on 16% denaturing PAGE gels, excised, eluted overnight in deionized water, and desalted using NAP-10 columns (GE Healthcare). Capped-oligonucleotide cleavage assays were performed as 4 µl reactions that contained 5 mM MgCl2, 1 mM DTT, 1 U RNasin µl−1, 0.7 µM promoter, ∼1,500 c.p.m. capped RNA primer, 5% glycerol, 0.05% NP-40, 75 mM NaCl, 10 mM HEPES pH 7.5, and 5 ng RdRp µl−1. Reactions were incubated at 30 °C for 1 h and analysed by 20% denaturing PAGE. The background was corrected using the signal of the PA active site mutant (PA D108A).

In vitro capped-oligonucleotide extension assays

To test the transcriptional activity of the influenza A virus RdRp, we set up 3 µl reactions containing 0.05 µM [α-32P]GTP (3,000 Ci mmole−1, Perkin-Elmer), 1 mM ATP, 1 ng rabbit β-globin mRNA µl−1, 500 µM CTP, 5 µM MgCl2, 1 mM DTT, 1 U RNasin µl−1, 0.7 µM promoter, and 5 ng RdRp µl−1. Reactions were incubated at 30 °C for 1 h. Products were analysed by 20% denaturing PAGE. To test the ability of the influenza A virus RdRp to extend a capped-oligonucleotide, a synthetic 11 nt RNA with 5′ diphosphate (5′-ppGAAUACUCAAG-3′) (Chemgenes) was capped as described above for the capped-RNA cleavage assay. Extension assays were next performed as 4 µl reactions containing 5 mM MgCl2, 1 mM DTT, 1 U RNasin µl−1, 0.7 µM promoter, 500 µM ATP, 500 µM CTP, 500 µM GTP, ∼1,500 c.p.m. capped RNA primer, 5% glycerol, 0.05% NP-40, 75 mM NaCl, 10 mM HEPES pH 7.5, and 5 ng RdRp µl−1. Reactions were incubated at 30 °C for 1 h and analysed by 20% denaturing PAGE.