Identification of the viral RNA promoter stem loop A (SLA)-binding site on Zika virus polymerase NS5

Zika virus has recently emerged as an important human pathogen that has spread to more than 60 countries. Infection of a pregnant woman with Zika virus can cause severe brain malformations in the child such as microcephaly and other birth defects. Despite the medical importance of Zika virus infection, the mechanism of viral replication, a process commonly targeted by antiviral therapeutics, is not well understood. Stem-loop A (SLA), located in the 5′ untranslated region of the viral genome, acts as a promotor for viral replication and thus is critical for recognition of the viral genome by the viral polymerase NS5. However, how NS5 engages SLA is not clear. We have quantitatively examined the intrinsic affinities between Zika virus SLA and NS5, and identified the SLA-binding site on NS5. Amino acid substitutions in the thumb subdomain of the RNA-dependent RNA polymerase (RdRp) and the methyltransferase (MTase) domain reduced SLA-binding affinity, indicating that they each are part of the SLA-binding site. Furthermore, stopped-flow kinetic analysis of Zika NS5-, RdRp- and MTase–SLA interactions identified distinct intermediates during NS5 and SLA complex formation. These data suggest a model for SLA recognition and the initiation of flaviviral replication by NS5.


Results
Full-length Zika NS5 and the individual RdRp and MTase domains interact with Zika SLA. Flavivirus replication is initiated when the viral polymerase NS5 identifies the viral genome via specific recognition of the SLA [10][11][12] . Accordingly, flavivirus NS5 has been shown to bind SLA with high affinity 10,30,32 . We have previ-   www.nature.com/scientificreports/ ously determined that DENV NS5 and SLA interact with 83 nM affinity using fluorescence measurements 30 . However, the roles each of the NS5 domains play in recognizing and binding SLA are not well understood. Here, we used ZIKV NS5 and the individual RdRp and MTase domains, and examined their binding affinities using the ZIKV SLA (80 nt, Fig. 1a) fluorescein-labeled at the 3′ end (SLA-F). Sequence identities between DENV and ZIKV NS5 and SLA are 67% and 64%, respectively. Fluorescence titrations of SLA-F with ZIKV NS5, RdRp domain and MTase domain (Fig. 1b) were carried out in buffer B1 (50 mM Tris adjusted to pH 8.0 with HCl at 20 °C, 150 mM NaCl, 1 mM MgCl 2 , 2 mM β-mercaptoethanol, and 10% glycerol). The formation of the complex between the ZIKV NS5 proteins and SLA-F causes a decrease in nucleic acid fluorescence (Fig. 1c). The maximum observed value of fluorescence quenching ranged from 15% for the full-length NS5, 9% for the RdRp domain, to 7% for the MTase domain. The solid lines in Fig. 1c are the nonlinear least-squares fits of Eq. (11) to the experimental data (see "Methods" section). The binding constants are listed in  19 to look for positively charged patches that would be complementary to the negatively charged SLA (Fig. 2). The RdRp domain is composed of three subdomains, fingers, palm and thumb subdomains, and positively charged residues were selected from all three subdomains for substitutions in NS5. Figure 2 shows the locations of targeted residues within NS5. The selected NS5 mutants were generated by site-directed mutagenesis and purified similar to the wild-type protein (Fig. 3a). The binding constants for NS5 mutants were then measured using SLA-F in buffer B1 (Fig. 3b-c). The binding constants are listed in Table 1. The binding assay showed that mutations in the fingers and palm subdomains did not affect SLA binding (Fig. 3b). For example, a double mutant within the fingers subdomain (R640A-R641A) and a triple mutant in the palm subdomain (K388A-K390A-R391A) bind to SLA with indistinguishable affinities compared to the wide-type NS5. Thus, these regions in the fingers and palm subdomains are not directly involved in interactions with SLA. The residues K388, K390, and R391 in the palm subdomain are located near the product exit site of the RdRp domain likely involved in the dsRNA product binding (Fig. 2) In contrast, several mutants in the thumb subdomain showed reduced binding affinities and/or altered the magnitude of the fluorescence signal (Fig. 3c). The thumb subdomain contains a large area of positively charged surface that include R771, R772, R775, K843, and R844 (Fig. 2). Thus, we generated six mutations in the area (Fig. 3a,c). The triple mutant, R771A-R772A-R775A showed the binding affinity of 2.0 × 10 6 M −1 (K d = 0.50 μM), a 1.8-fold reduction from the wild-type affinity ( Fig. 3c and Table 1). Since K843 and R844 are located in the positively charged patch near the mutated R771, R772, and R775, we tested if additional mutations within the patch reduce the binding to SLA even further. NS5 mutant containing R771A, R772A, R775A, K843A, and R844A indeed displayed a weaker affinity to SLA with K = 1.4 × 10 6 M −1 (K d = 0.71 μM) ( Fig. 3c and Table 1). To determine if the reduced affinity of the R771A-R772A-R775A-K843A-R844A mutant is caused by the structural changes of the mutant, we compared the CD spectra for wild-type NS5 and the mutant (Fig. 3d). The CD spectra were similar, indicating that the secondary structure of NS5 has not been affected by mutations. Thus, the reduced www.nature.com/scientificreports/ affinity of the R771A-R772A-R775A-K843A-R844A mutant is caused by the loss of its direct interaction with SLA. We also generated single and double mutations in the region, but they did not significantly affect SLA binding, suggesting that other positively charged residues within the patch may compensate for the loss of their positively charged side chains (Fig. 2). For example, R775A shows the same binding affinity as the wild-type NS5 ( Fig. 3c and Table 1). A single mutant of R858A showed decreased magnitude of the fluorescence signal, although the binding affinity remained the same as for wild-type NS5 ( Fig. 3c and Table 1). This indicates that Ala substitution in R858A affects the conformation of the bound SLA. A double mutant K843A-R844A also showed a decreased magnitude of the fluorescence signal suggesting that K843 and R844 also induce changes in the nucleic acid structure, although the mutations alone were not able to significantly reduce SLA binding ( Fig. 3c and Table 1). We also tested if the C-terminal region of NS5 is involved in interactions with SLA. A single mutation of R888 to Ala in DENV NS5 was shown to produce a completely non-viable virus 34 . In the crystal structure of DENV NS5, R888 in the C-terminal helix is located near R772 of the positively charged patch 16 , suggesting that R888 could form a part of the SLA-binding surface. The corresponding residue R891 in ZIKV NS5 is disordered in the crystal structure 19 . However, the R891A mutant in ZIKV NS5 binds SLA with the wild-type affinity, suggesting that R891 at the C-terminal end of NS5 is not involved in the interactions with SLA ( Fig. 3c and Table 1).
To determine if the SLA-binding site on the thumb subdomain is specific for SLA interaction, we next measured NS5 protein interactions with a nonspecific single-stranded RNA (ssRNA). Because the identified SLA-binding site does not overlap with the template-binding site (where ssRNA binds, Fig. 2), the SLA-binding site mutant should be able to bind nonspecific ssRNAs. We thus determined the binding affinity of NS5 proteins (wild-type and the R771A-R772A-R775A-K843A-R844A mutant) with etheno-labeled A(pA) 19 , εA(pεA) 19 (Fig. 3e). Wildtype and mutant NS5 bind εA(pεA) 19 with a similar binding affinity of K wt = 2.0 × 10 6 M −1 (K d = 0.50 μM) and K mut = 1.5 × 10 6 M −1 (K d = 0.67 μM). Thus, the mutations that affect SLA binding in NS5 do not significantly change nonspecific ssRNA interaction, indicating that the positively charged patch in the thumb subdomain (R771, R772, R775, K843, and R844) specifically interacts with SLA.  (Fig. 1c). Crystal structures of flavivirus MTases co-crystallized with GTP, cap analog GpppA, cofactor S-adenosyl-L-methione (SAM) or its product S-adenosyl-L-homocysteine (SAH) identified the catalytic site and cofactor binding sites 23,35 . The structures also suggest a putative RNA-binding site, consisting of K28, K29, R37, R41, R42, R57, and R213 (Fig. 4a). Several of these residues (K28, K29, R57, and R213) in DENV MTase have previously been mutated to Ala and their interactions with a short RNA examined 36 . The reduction in RNA binding varied by one to five-fold, with greatest effect observed for the K29A mutation. This putative RNA-binding site with a positively charged patch is located on the same surface as the positively charged patch in the RdRp domain ( Fig. 2, right). To determine whether these MTase residues play a role in the NS5-SLA complex formation, we introduced a four-residue mutation, K28E-K29E-R41E-R42E, in the MTase domain construct and measured its interaction with SLA-F by fluorescence. However, the change in fluorescence signal for the MTase mutant was too small to accurately determine the binding constant. To increase the resolution of the analysis, we utilized the anisotropy signal to examine MTase and SLA interactions (see    S1). The affinity of the mutant MTase K28E-K29E-R41E-R42E measured by anisotropy was K mut = 0.8 × 10 6 M −1 (K d = 1.25 μM), i.e., it decreased fourfold compared to the wild-type MTase (Fig. 4b). The reduced SLA-binding affinity of the mutant was not caused by structural changes induced by the mutations, since the comparison of the CD spectra between the wild-type and mutant MTase show identical curves (Fig. 4d). These results indicate that the mutated residues are most likely involved in NS5-SLA interaction.
To determine if the decreased affinity is specific for SLA compared to non-specific ssRNA, the associations of the wild-type and mutant MTase were measured by fluorescence anisotropy with the fluorescently labeled ssRNA, εA(pεA) 19 in buffer B2 (Fig. 4c). The wild-type MTase binds the ssRNA with K wt = 1.1 × 10 6 M −1 (K d = 0.91 μM). The binding constant of the mutant MTase K28E-K29E-R41E-R42E for the ssRNA is K mut = 0.4 × 10 6 M −1 (K d = 2.5 μM), a 2.8-fold reduction compared to the wild-type. The SLA-binding site identified in the MTase (K28, K29, R41, R42) partially overlaps with the capped RNA-binding site (K14, L17, N18, L20, F25, K29) determined by the crystal structures of MTase in complex with capped RNA 35,37 . Thus, it is likely that the SLAbinding site mutant in the MTase domain also reduces the ssRNA interaction. Nevertheless, the mutant MTase (K28E-K29E-R41E-R42E) reduced the SLA binding affinity more significantly than the ssRNA binding affinity (four-fold vs. 2.8-fold), indicating that the positively charged patch containing K28, K29, R41, and R42 is involved in SLA binding.
Stopped-flow kinetics of ZIKV NS5, RdRp, and MTase binding to SLA. Next, we examined the kinetic mechanism of ZIKV SLA binding to full-length NS5 and the isolated RdRp and MTase domains by fluorescence stopped-flow methods. All stopped-flow experiments were performed under pseudo-first-order conditions with respect to the protein concentration by mixing SLA-F with a large excess of ZIKV NS5, the RdRp domain or the MTase domain. An example of the stopped-flow kinetic trace of SLA-F fluorescence after mixing with ZIKV NS5 is shown in Fig. 5a. The best fit for the NS5-SLA kinetic traces was obtained using a triple-exponential function, indicating that the simplest mechanism that accounts for the observed change of relaxation times is a minimum three-step binding process. The SLA-F alone showed no fluorescence changes when mixed  Fig. 5b and c. The plot of the reciprocal relaxation time of the first exponential phase, 1/τ 1 versus NS5 concentrations showed a linear dependence, indicating a bimolecular association (Fig. 5b). The reciprocal relaxation times 1/τ 2 and 1/τ 3 increased in a hyperbolic manner with NS5 concentrations, suggesting intramolecular transitions of the NS5-SLA complex (i.e., conformational changes) in these steps (Fig. 5c). The dependencies of the individual  12)). The dependence of reciprocal relaxation times 1/τ 1 (red) and 1/τ 2 (blue) upon the MTase concentration are shown in (i,j), respectively. The dependence of amplitudes A 1 (red) and A 2 (blue) upon the logarithm of the MTase concentrations are shown in (k). The solid lines in (i-k) are the global fits of the curves with a two relaxationstep process. To globally fit the kinetic data at various NS5 concentrations and obtain rate constants, the stopped-flow kinetic data were also analyzed using the matrix projection operator technique 38,39 . The three-step binding process is described by Eq. (1): The partial equilibrium constants for each step of the above reaction are related to the K overall as, where K 1 = k 1 /k −1 , K 2 = k 2 /k −2 , and K 3 = k 3 /k −3 . The six kinetic rate constants for the three kinetic steps, as well as fluorescence changes characterizing individual intermediates, are related through the overall binding constants and the maximum fluorescence change ΔF max . Using the overall binding constants (K overall = 3.5 × 10 6 M −1 ) and the maximum fluorescence change (ΔF max = 0.15) that are independently determined (Fig. 1c), partial equilibrium constants and rate constants were determined ( Table 2). The rate constants indicate that the formation of the NS5-SLA complex proceeds first by the fast bimolecular association of NS5 and SLA (k 1 = 9.5 × 10 7 M −1 s −1 and k-1 = 360 s −1 ) followed by two structural rearrangement steps (k 2 = 55 s −1 , k −2 = 12 s −1 and k 3 = 4 s −1 , k −3 = 2.5 s −1 ).
The stopped-flow kinetic traces of SLA-F fluorescence after mixing with ZIKV RdRp were analogously analyzed. The best fit for the RdRp-SLA kinetic traces were obtained using a double-exponential function (Fig. 5e). However, the binding process is likely more complex because the sum of the amplitudes of the fitted relaxation steps is significantly smaller than the observed total amplitude, i.e., the fluorescence of free SLA-F is significantly higher than the fluorescence of the first observed data point (Fig. 5e). Thus, there is a fast step which is beyond the resolution of the stopped-flow instrument, which precedes the two observed relaxation steps. Therefore, in the case of the RdRp-SLA interaction, the simplest minimum mechanism is a three-step binding process, where initial recognition occurs through very fast bimolecular association of SLA and RdRp.
Even though we cannot determine the value of the reciprocal relaxation time for the first fast relaxation process, 1/τ 1 , we can determine its amplitude, A 1 by Eq. (4), where A T is the experimentally observed total amplitude of the kinetic trace at a given total concentration of the protein 38 .
The dependences of the reciprocal relaxation times (1/τ 2 and 1/τ 3 ) and normalized amplitudes, A 1 , A 2 , and A 3 of three observed relaxation processes on the concentration of the ZIKV RdRp domain are shown in Fig. 5f and g. At low RdRp domain concentrations, the total observed amplitude, A T , is dominated by A 2 . With increasing protein concentrations, the contribution of A 1 to A T increases while A 3 remains negative throughout the process. The partial equilibrium constants and kinetic rate constants characterizing the intermediates for the three steps are related to the overall binding constant K overall and the maximum fluorescence change ΔF max , as described above (Table 2). In the RdRp-SLA complex formation, the initial recognition occurs through a very fast bimolecular association of SLA and RdRp (k 1 = 1.0 × 10 9 M −1 s −1 and k-1 = 2000s −1 ), which is followed by two structural rearrangement steps (k 2 = 140 s −1 , k −2 = 70 s −1 and k 3 = 1.0 s −1 , k −3 = 15.0 s −1 ). Note the initial bimolecular association and dissociation (k 1 and k −1 ) of RdRp and SLA are 10 times faster than the initial association of NS5 and SLA.  Fig. 5h. The solid red line is a nonlinear least-squares fit of the data using a double-exponential function, indicating that the simplest mechanism that can account for the data is a two-step binding process (Eq. 5).
The dependencies of the reciprocal relaxation time 1/τ 1 and 1/τ 2 , characterizing the observed kinetic steps as a function of the total MTase domain concentration are shown in Fig. 5i and j, respectively. The plot of the reciprocal relaxation time of the first exponential phase (1/τ 1 ) versus MTase concentrations showed a linear dependence, indicating a MTase-SLA complex is formed after the bimolecular event (Fig. 5i). The dependence of the normalized amplitudes, A 1 and A 2 upon the total concentration of the ZIKV MTase domain is shown in Fig. 5k. The global fit of the kinetic data at various MTase concentrations provides rate constants for the two steps ( Table 2). The kinetic constants indicate that a fast bimolecular association of SLA and the MTase domain (k 1 = 9.0 × 10 7 M −1 s −1 and k-1 = 125 s −1 ) is followed by only one structural rearrangement step (k 2 = 15 s −1 and k −2 = 14 s −1 ) ( Table 2).

Discussion
The SLA-binding site on NS5 is located in the thumb and MTase domains. Replication of the viral genome is a complex process that involves the viral RNA and polymerase, as well as other viral and cellular proteins. In the case of flaviviruses, the SLA at the 5′ end of viral RNA plays a critical role by functioning as a promoter of the replication process [10][11][12] . In order to understand how ZIKV NS5 recognizes the SLA promotor for initiation of RNA synthesis, we examined the roles that each of the NS5 domains play in binding to SLA. Both RdRp and MTase domains bind to SLA. Moreover, the SLA-binding affinities for both domains were similar to each other and were approximately two-fold lower than the affinity of the full-length NS5 for SLA. This indicates that ZIKV NS5 recognizes SLA using both RdRp and MTase domains. We have thus identified the SLA-binding sites within the RdRp and MTase domains. In the RdRp domain, only the thumb subdomain mutants show reduced SLA binding. Alanine substitutions of R771, R772 and R775, and additional K843 and R844 substitutions in ZIKV NS5 significantly reduced the SLA binding affinity from 280 nM (wild-type NS5) to 500 and 710 nM, respectively ( Fig. 3c and Table 1). This suggests that a large positively charged patch consisting of R771, R772, R775, K843 and R844 in the thumb subdomain engages SLA and mediates the formation of the NS5-SLA complex. Interestingly, the single NS5 mutations (R775A and R891A) or a double mutation (K843A-R844A) did not significantly affect the ZIKV NS5-SLA interactions ( Table 1), suggesting that the SLA-binding surface on NS5 encompasses a cluster of residues, and other positively charged residues within the patch can compensate for the loss of one or two positively charged residues. The SLA-binding site identified here does not overlap with the template-binding site or the dsRNA exit site in the RdRp domain, suggesting that NS5 has two RNA binding sites, one for ssRNA template, and the other for SLA. Consistent with this idea, wild-type and the SLA-binding site mutant NS5 (R771A-R772A-R775A-K843A-R844A) bind ssRNA with similar affinities (Fig. 3e). In the MTase domain, the mutations of K28E-K29E-R41E-R42E reduced SLA binding affinity from 310 nM (wild-type MTase) to 1.25 µM, suggesting that this four-residue mutation within the MTase domain eliminates specificity for SLA. An NS5 construct containing mutations in both RdRp and MTase patches was expressed, but the protein was insoluble upon E. coli expression preventing further investigations. The identified SLA-binding residues in MTase and RdRp domains are highly conserved in flavivirus NS5 (Fig. S2), and residues in the thumb subdomain have previously been shown to be important for viral replication. Individual mutations of R770A and R773A, and a double mutation of K840A-R841A in a DENV2 replicon (corresponding to R772, R775, K843 and R844 in ZIKV NS5) delayed or abolished viral replication 32 . These residues are not directly involved in the catalysis of RNA synthesis, and thus these mutants may abolish viral replication by diminishing interactions with SLA.
The two positively charged patches in the thumb subdomain of RdRp and MTase domains are located on the same surface of NS5, strongly suggesting that the SLA-binding site encompasses both domains. During the viral replication cycle, NS5 interacts with SLA at least twice. First, NS5 must recognize the SLA at the 5′ end of the genome to initiate negative-strand RNA synthesis at the 3′ end (i.e. to carry out the polymerase function of NS5) [10][11][12] . Second, NS5 must again bind SLA to methylate the cap at the 5′ end of the nascent positive-strand RNA (i.e., to carry out the MTase function of NS5) 29 . These two functionally different NS5 and SLA interactions would engage the same binding site, since it seems unlikely that NS5 possesses two separate SLA-binding sites. In such a case, both MTase and RdRp domains are likely involved in SLA binding. The positively charged patches in RdRp and MTase are separated by ~ 55-60 Å. Based on the secondary structure prediction (Fig. 1a), ZIKV SLA is a long molecule consisting of at least 23 base pairs (~ 60 Å) without counting the bulge between the top and bottom stems or the top loop. Thus, SLA could simultaneously bind both the RdRp and MTase patches.

Zika NS5 initially recognizes SLA via the RdRp domain.
We have examined the dynamics of the NS5-SLA interaction using stopped-flow techniques. The NS5 recognition of SLA occurs through a three-step mechanism. The first step is a fast, bimolecular association of NS5 and SLA, which is followed by two steps in which structural rearrangements of the NS5-SLA complex occur (Fig. 6). Interestingly, the RdRp domain interactions with SLA also occur through a three-step process, but the first step is significantly faster than for the full-length NS5 (k 1−RdRp = 1.0 × 10 9 M −1 s −1 vs k 1−NS5 = 9.5 × 10 7 M −1 s −1 ). On the other hand, the RdRp has a large reverse rate constant (k −1−RdRp = 2000s −1 ), indicating that the RdRp-SLA complex dissociates more easily than the NS5-SLA complex where the reverse rate constant is much lower (k −1−NS5 = 360 s −1 ). This indicates that the MTase domain in the intact NS5 slows the recognition step but increases the stability of the formed complex.
Scientific RepoRtS | (2020) 10:13306 | https://doi.org/10.1038/s41598-020-70094-y www.nature.com/scientificreports/ Accordingly, binding of the MTase domain to SLA is characterized by only a two-step mechanism and the initial recognition step is much slower than the RdRp domain (k 1−MTase = 9.0 × 10 7 M −1 s −1 vs. k 1−RdRp = 1.0 × 10 9 M −1 s −1 ). Therefore, NS5 likely recognizes SLA first through the RdRp domain. Taken together, we propose an NS5-SLA interaction model wherein the first fast bimolecular association between NS5 and SLA is carried out via the RdRp domain involving the thumb subdomain (Fig. 6). This fast recognition step is then followed by two steps corresponding to two conformational changes, during which NS5 locks itself onto the SLA through the RdRp and MTase domain to form a stable NS5-SLA complex. Crystal structures of flavivirus NS5s show two different relative arrangements of the RdRp and MTase domains, suggesting that interaction between the RdRp and MTase domains is not fixed spatially [16][17][18][19][20]40 . Thus, it is likely that NS5 adopts several conformations by rearrangement of RdRp and MTase domains to accommodate different RNA molecules such as structured SLA, ssRNA, and dsRNA during viral replication.

Reagents and buffers.
A Zika SLA construct containing the first 80 nucleotides with 5′ triphosphate and a 3′ fluorescein tag was obtained from Midland Certified Reagents (Midland, TX). SLA concentration was determined using the extinction coefficient of SLA ǫ 260 = 682,500 cm −1 M −1 . The etheno-modified A(pA) 19 , εA(pεA) 19 was obtained by modification of the oligomer with chloroacetaldehyde as described previously 31 . The concentration of εA(pεA) 19 was determined using the extinction coefficient, ǫ 257 = 3,700 cm −1 M −1 (nucleotide) 31 .  Table S1. Mutations were confirmed by DNA sequencing at the molecular genomics facility at UTMB. Mutant proteins were purified similarly to wild-type proteins. Purified wild-type and mutant proteins were analyzed by SDS-PAGE (Figs.1b, 3a, 4d). Uncropped images of SDS-PAGE gels are shown in Fig. S3.  Titration curves were fit using Kaleida Graph software (Synergy Software, PA). In these experiments, the relative fluorescence change (ΔF obs ) is defined as (F i -F 0 )/F 0 , where F i is the fluorescence of the RNA at a given titration point and F 0 is the initial value of the fluorescence of the sample. Both F 0 and F i were corrected for background fluorescence at the applied excitation wavelength. Interactions between the NS5 proteins (wild-type and R771A-R772A-R775A-K843A-R844A) and ssRNA were measured with etheno-labeled εA(pεA) 19 in buffer B1 (50 mM Tris adjusted to pH 8.0 with HCl at 20 °C, 150 mM NaCl, 1 mM MgCl 2 , 2 mM β-mercaptoethanol, and 10% glycerol). The εA(pεA) 19 samples were excited at λ ex = 325 nm and emission was recorded at λ em = 410 nm 31 .

Expression and purification of Zika
Interactions between the MTase domain proteins (wild-type and K28E-K29E-R41E-R42E) and fluoresceinlabeled SLA were analyzed using fluorescence anisotropy (r), because fluorescence intensity changes were too small to accurately determine the binding constant for the mutant. Buffer B2, containing 50 mM Tris adjusted to pH 8.0 with HCl at 20 °C, 100 mM NaCl, 1 mM MgCl 2 , 2 mM β-mercaptoethanol, and 10% glycerol, was used for anisotropy measurement. Fluorescence anisotropy (r) is defined as: where I VV and I VH are fluorescence intensities where the first and second subscripts refer to vertical (V) polarization of the excitation and vertical (V) or horizontal (H) polarization of the emitted light. The factor, G = I HV / I HH , corrects for the different sensitivity of the emission monochromator for vertically and horizontally polarized light. The samples in the fluorescence anisotropy experiments were excited at 480 nm and the emission recorded at 520 nm (fluorescein peaks for excitation and emission, respectively). Each experiment was carried out in triplicate. Interactions between the MTase domain proteins (wild-type and K28E-K29E-R41E-R42E) and etheno-labeled εA(pεA) 19 were measured in buffer B2 using fluorescence anisotropy. The fluorescence signal of εA(pεA) 19 were monitored with λ ex = 325 nm and λ em = 410 nm 31 . Each experiment was carried out in triplicate.
Fluorescence signal analysis. The binding constant, K 1 , characterizing the SLA association with NS5, is defined as where [C 1 ] F is the concentration of the formed complex.
The observed fluorescence of the sample at any point of the titration is defined as where F F and F C are the molar fluorescence intensities of the free SLA and the formed complex C 1 , respectively. Thus, The mass conservation equation for the total SLA concentration, [SLA] T, in the sample is Using Eqs. (9) and (10), the relative observed change of the SLA fluorescence, ∆F obs , is then where ∆F max = F C /F F , is the maximum value of the observed relative fluorescence quenching. Fluorescence titration curves were fitted using Eq. (11) to determine binding constants. The goodness of the fits were determined using standard chi-square test, which provides the confidence level for the selected binding model of ≥ 0.98. , and the fluorescence signal upon complex formation was monitored as a function of time with λ ex = 480 nm and λ em = 520 nm. Generally, 10-12 traces were collected and averaged for each concentration of proteins. The kinetic curves were fitted to extract relaxation times and corresponding amplitudes using nonlinear least-squares fitting software provided by the manufacturer, with the exponential function defined as www.nature.com/scientificreports/ where F(t) is the fluorescence intensity at time t, F(∞) is the fluorescence intensity at t = ∞, A i is the amplitude corresponding to ith relaxation process, λ i is the time constant (reciprocal relaxation time) characterizing the relaxation process, and n is the number of relaxation processes. Global analyses of the stopped-flow kinetic data were performed using the matrix projection operator technique 38,39 . All analyses of the data were performed using Mathematica (Wolfram, Urbana, IL) and Kaleida Graph (Synergy Software, PA) as previously described 38,39 .
Circular dichroism (CD) spectrometry. The far-UV circular-dichroism measurements for wild-type and mutant proteins were performed on a model No. J-815 spectrometer (JASCO, Oklahoma City, OK). The protein concentration was ~ 0.1 mg/ml in 50 mM Tris buffer (pH 7.4) containing 150 mM NaCl, 10% glycerol and 1 mM MgCl 2 . CD spectra were collected at 20 °C in a 0.2-cm path-length cuvette from 195 to 260 nm. Three scans were averaged in the range of 200-250 nm. Data were converted from millidegrees to mean residue ellipticity (MRE) using the formula: where [θ] is the molar residue ellipticity, θ is the CD signal in millidegrees, c is the protein concentration in µM, l is the path length in mm, and n is the number of amino acid residues. The CD data analysis program BeStSel was used to analyze the data and estimate the secondary structure 41 .