Dissecting the fidelity of bacteriophage RB69 DNA polymerase: site-specific modulation of fidelity by polymerase accessory proteins

Bacteriophage RB69 encodes a replicative B-family DNA polymerase (RB69 gp43) with an associated proofreading 3' exonuclease. Crystal structures have been determined for this enzyme with and without DNA substrates. We previously described the mutation rates and kinds of mutations produced in vivo by the wild-type (Pol(+) Exo(+)) enzyme, an exonuclease-deficient mutator variant (Pol(+) Exo(-)), mutator variants with substitutions at Tyr(567) in the polymerase active site (Pol(M) Exo(+)), and the double mutator Pol(M) Exo(-). Comparing the mutational spectra of the Pol(+) Exo(-) and Pol(+) Exo(+) enzymes revealed the patterns and efficiencies of proofreading, while Tyr(567) was identified as an important determinant of base-selection fidelity. Here, we sought to determine how well the fidelities of the same enzymes are reflected in vitro. Compared to their behavior in vivo, the three mutator polymerases exhibited modestly higher mutation rates in vitro and their mutational predilections were also somewhat different. Although the RB69 gp43 accessory proteins exerted little or no effect on total mutation rates in vitro, they strongly affected mutation rates at many specific sites, increasing some rates and decreasing others.     

The roles of Tyr391 and Tyr619 in RB69 DNA polymerase replication fidelity

In the family-B DNA polymerase of bacteriophage RB69, the conserved aromatic palm-subdomain residues Tyr391 and Tyr619 interact with the last primer-template base-pair. Tyr619 interacts via a water-mediated hydrogen bond with the phosphate of the terminal primer nucleotide. The main-chain amide of Tyr391 interacts with the corresponding template nucleotide. A hydrogen bond has been postulated between Tyr391 and the hydroxyl group of Tyr567, a residue that plays a key role in base discrimination. This hydrogen bond may be crucial for forcing an infrequent Tyr567 rotamer conformation and, when the bond is removed, may influence fidelity. We investigated the roles of these residues in replication fidelity in vivo employing phage T4 rII reversion assays and an rI forward assay. Tyr391 was replaced by Phe, Met and Ala, and Tyr619 by Phe. The Y391A mutant, reported previously to decrease polymerase affinity for incoming nucleotides, was unable to support DNA replication in vivo, so we used an in vitro fidelity assay. Tyr391F/M replacements affect fidelity only slightly, implying that the bond with Tyr567 is not essential for fidelity. The Y391A enzyme has no mutator phenotype in vitro. The Y619F mutant displays a complex profile of impacts on fidelity but has almost the same mutational spectrum as the parental enzyme. The Y619F mutant displays reduced DNA binding, processivity, and exonuclease activity on single-stranded DNA and double-stranded DNA substrates. The Y619F substitution would disrupt the hydrogen bond network at the primer terminus and may affect the alignment of the 3' primer terminus at the polymerase active site, slowing chemistry and overall DNA synthesis.     

Biochemical characterization of interactions between DNA polymerase and single-stranded DNA-binding protein in bacteriophage RB69

The organization and proper assembly of proteins to the primer-template junction during DNA replication is essential for accurate and processive DNA synthesis. DNA replication in RB69 (a T4-like bacteriophage) is similar to those of eukaryotes and archaea and has been a prototype for studies on DNA replication and assembly of the functional replisome. To examine protein-protein interactions at the DNA replication fork, we have established solution conditions for the formation of a discrete and homogeneous complex of RB69 DNA polymerase (gp43), primer-template DNA, and RB69 single-stranded DNA-binding protein (gp32) using equilibrium fluorescence and light scattering. We have characterized the interaction between DNA polymerase and single-stranded DNA-binding protein and measured a 60-fold increase in the overall affinity of RB69 single-stranded DNA-binding protein (SSB) for template strand DNA in the presence of DNA polymerase that is the result of specific protein-protein interactions. Our data further suggest that the cooperative binding of the RB69 DNA polymerase and SSB to the primer-template junction is a simple but functionally important means of regulatory assembly of replication proteins at the site of action. We have also shown that a functional domain of RB69 single-stranded DNA-binding protein suggested previously to be the site of RB69 DNA polymerase-SSB interactions is dispensable. The data from these studies have been used to model the RB69 DNA polymerase-SSB interaction at the primer-template junction.     

Protein determinants of RNA binding by DNA polymerase of the T4-related bacteriophage RB69

DNA polymerase (gp43) of phage T4 plays two biological roles, one as an essential DNA binding replication enzyme and the other as an mRNA-specific autogenous translational repressor. Binding of T4 gp43 to its mRNA target (translational operator RNA) interferes with gp43-DNA interactions, but it is unclear how the protein determinants for binding DNA are affected by the dynamics of gp43-mRNA interactions. We have used RB69 gp43, a natural variant of the T4 enzyme whose crystal structure has been determined to identify protein sites that respond to the interaction with specific RNA. We used protein phosphorylation markers, photocross-linking studies, protease sensitivity assays, and mutational analyses to examine the effects of operator RNA on the enzyme's five structural domains (N, exo, palm, fingers, and thumb). Our studies suggest that this RNA affects gp43-DNA interactions through global effects on protein structure that occlude DNA-binding sites but leave the enzyme accessible to interactions with the sliding clamp (RB69 gp45) and possibly other polymerase accessory proteins. We discuss the possible biological significance of putative RNA-binding motifs in the N and palm domains of RB69 gp43.     

Processivity clamp gp45 and ssDNA-binding-protein gp32 modulate the fidelity of bacteriophage RB69 DNA polymerase in a sequence-specific manner, sometimes enhancing and sometimes compromising accuracy

Numerous studies of the impact of accessory proteins upon the fidelity of DNA synthesis have provided a complex and sometimes discordant picture. We previously described such an analysis conducted in vitro using various bacteriophage RB69 gp43 mutator DNA polymerases with or without the accessory proteins gp32 (which binds single-stranded DNA) plus gp45/44/62 (processivity clamp and its loaders). Mutations were scored at many sites in the lacZalpha mutation reporter sequence. Unexpectedly, the accessory proteins sometimes decreased and sometimes increased fidelity at a handful of specific sites. Here, we enlarge our analysis with one particular mutator polymerase compromised in both insertion accuracy and proofreading and also extend the analysis to reactions supplemented only with gp32 or only with gp45/44/62. An overall 1.56-fold increase in mutation frequencies was produced by adding single or multiple accessory proteins and was driven mainly by increased T(template)*G(primer) mispairs. Evidence was found for many additional sites where the accessory proteins influence fidelity, indicating the generality of the effect. Thus, accessory proteins contribute to the site-specific variability in mutation rates characteristically seen in mutational spectra.     

Structure and enzymatic properties of a chimeric bacteriophage RB69 DNA polymerase and single-stranded DNA binding protein with increased processivity

In vivo, replicative DNA polymerases are made more processive by their interactions with accessory proteins at the replication fork. Single-stranded DNA binding protein (SSB) is an essential protein that binds tightly and cooperatively to single-stranded DNA during replication to remove adventitious secondary structures and protect the exposed DNA from endogenous nucleases. Using information from high resolution structures and biochemical data, we have engineered a functional chimeric enzyme of the bacteriophage RB69 DNA polymerase and SSB with substantially increased processivity. Fusion of RB69 DNA polymerase with its cognate SSB via a short six amino acid linker increases affinity for primer-template DNA by sixfold and subsequently increases processivity by sevenfold while maintaining fidelity. The crystal structure of this fusion protein was solved by a combination of multiwavelength anomalous diffraction and molecular replacement to 3.2 A resolution and shows that RB69 SSB is positioned proximal to the N-terminal domain of RB69 DNA polymerase near the template strand channel. The structural and biochemical data suggest that SSB interactions with DNA polymerase are transient and flexible, consistent with models of a dynamic replisome during elongation.     

Sensitivity of bacteriophage RB69 DNA polymerase to N2-(p-n-butylphenyl)-2'-deoxyguanosine nucleotides

The response of bacteriophage RB69 DNA polymerase to N2-(p-n-butylphenyl)-2'-deoxyguanosine 5'-triphosphate (BuPdGTP), related nucleotides and non-nucleoside inhibitors was measured and compared to values obtained for the closely related DNA polymerase from bacteriophage T4. Both enzymes showed similar responses to inhibitors in terms of Ki values and the ability to utilize BuPdGTP as a substrate. These results provide the relevance of using the recent crystal structure of RB69 DNA polymerase for analysis of BuPdGTP/B family DNA polymerase interactions.     

Probing minor groove hydrogen bonding interactions between RB69 DNA polymerase and DNA

Minor groove hydrogen bonding (HB) interactions between DNA polymerases (pols) and N3 of purines or O2 of pyrimidines have been proposed to be essential for DNA synthesis from results obtained using various nucleoside analogues lacking the N3 or O2 contacts that interfered with primer extension. Because there has been no direct structural evidence to support this proposal, we decided to evaluate the contribution of minor groove HB interactions with family B pols. We have used RB69 DNA pol and 3-deaza-2'-deoxyadenosine (3DA), an analogue of 2-deoxyadenosine, which has the same HB pattern opposite T but with N3 replaced with a carbon atom. We then determined pre-steady-state kinetic parameters for the insertion of dAMP opposite dT using primer/templates (P/T)-containing 3DA. We also determined three structures of ternary complexes with 3DA at various positions in the duplex DNA substrate. We found that the incorporation efficiency of dAMP opposite dT decreased 10(2)-10(3)-fold even when only one minor groove HB interaction was missing. Our structures show that the HB pattern and base pair geometry of 3DA/dT is exactly the same as those of dA/dT, which makes 3DA an optimal analogue for probing minor groove HB interactions between a DNA polymerase and a nucleobase. In addition, our structures provide a rationale for the observed 10(2)-10(3)-fold decrease in the rate of nucleotide incorporation. The minor groove HB interactions between position n - 2 of the primer strand and RB69pol fix the rotomer conformations of the K706 and D621 side chains, as well as the position of metal ion A and its coordinating ligands, so that they are in the optinal orientation for DNA synthesis.     

Thermus aquaticus DNA polymerase I mutants with altered fidelity. Interacting mutations in the O-helix

Phe(667) in the conserved O-helix of Thermus aquaticus (Taq) DNA polymerase I (pol I) is known to be important for discrimination against dideoxy-NTPs. We show here that Phe(667) is also important for base selection fidelity. In a forward mutation assay at high polymerase concentration, wild type pol I catalyzed frequent A --> T and G --> T transversions and -1 frameshifts at nonreiterated sites involving loss of a purine immediately downstream of a pyrimidine. The mutants F667L and A661E,I665T,F667L exhibited large decreases in A --> T and G --> T transversions, and the triple mutant displayed reduction in the aforementioned -1 frameshifts as well. Kinetic analysis showed that the F667L and A661E,I665T,F667L polymerases discriminated against synthesis of A:A mispairs more effectively and catalyzed less extension of A:A mispairs than the wild type enzyme. These data indicate that Phe(667) functions in maintaining the error frequency and spectrum, and the catalytic efficiency, of wild type pol I. We also found that the strong general mutator activity conferred by the single A661E substitution was entirely suppressed in the A661E, I665T,F667L polymerase, exemplifying how interactions among O-helix residues can contribute to fidelity. We discuss the mutator and anti-mutator mutations in light of recently obtained three-dimensional structures of T. aquaticus pol I.     

Correlation of the kinetics of finger domain mutants in RB69 DNA polymerase with its structure

We have estimated pre-steady-state kinetic parameters for the addition of a single nucleotide residue by a set of RB69 DNA polymerase mutants in which four highly conserved residues in the fingers domain have been replaced by Ala. The relationship between the kinetic constants exhibited by the mutants and the structure of the ternary complex [Franklin, M., Wang, J., and Steitz T. (2001) Cell 105, 657-667] was consistent with the following sets of interactions between the conserved residues and oxygen atoms in the triphosphate portion of the incoming dNTP: (i) the epsilon-amino group of K560 contacts oxygen atoms of the alpha- and gamma-phosphates, (ii) the amide side chain of Asn 564 forms a hydrogen bond via a water molecule with the nonbridging oxygen of the beta-phosphate, and (iii) the epsilon-amino and delta-guanidino groups of K486 and R482, respectively, contact the nonbridging oxygens of the gamma-phosphate. We have also determined the pre-steady-state kinetic parameters for the addition of both dCTP and dCDP onto a 13/20mer primer/template with an exo(-) derivative of RB69 DNA polymerase and have shown that the deoxynucleoside diphosphate can be incorporated, in contrast to the behavior of the Klenow fragment which cannot use dCDP as a substrate. We have shown that, with RB69 DNA polymerase, in contrast to the Klenow fragment, there is no inhibition of the primer-extension reaction by incoming NTPs having either noncomplementary bases or ribo- instead of a deoxyribose moieties. This implies that the mode of recognition of incoming dNTPs and triggering of the conformational change, which is thought to occur prior to the chemical step, differs between these two enzymes.     

Bidentate and tridentate metal-ion coordination states within ternary complexes of RB69 DNA polymerase

Two divalent metal ions are required for primer‐extension catalyzed by DNA polymerases. One metal ion brings the 3′‐hydroxyl of the primer terminus and the α‐phosphorus atom of incoming dNTP together for bond formation so that the catalytically relevant conformation of the triphosphate tail of the dNTP is in an α,β,γ‐tridentate coordination complex with the second metal ion required for proper substrate alignment. A probable base selectivity mechanism derived from structural studies on Dpo4 suggests that the inability of mispaired dNTPs to form a substrate‐aligned, tridentate coordination complex could effectively cause the mispaired dNTPs to be rejected before catalysis. Nevertheless, we found that mispaired dNTPs can actually form a properly aligned tridentate coordination complex. However, complementary dNTPs occasionally form misaligned complexes with mutant RB69 DNA polymerases (RB69pols) that are not in a tridentate coordination state. Here, we report finding a β,γ‐bidentate coordination complex that contained the complementary dUpNpp opposite dA in the structure of a ternary complex formed by the wild type RB69pol at 1.88 Å resolution. Our observations suggest that several distinct metal‐ion coordination states can exist at the ground state in the polymerase active site and that base selectivity is unlikely to be based on metal‐ion coordination alone.     

Steady-state kinetic characterization of RB69 DNA polymerase mutants that affect dNTP incorporation.

The function of six highly conserved residues (Arg482, Lys483, Lys486, Lys560, Asn564, and Tyr567) in the fingers domain of bacteriophage RB69 DNA polymerase (RB69 gp43) were analyzed by kinetic studies with mutants in which each of these residues was replaced with Ala. Our results suggest that Arg482, Lys486, Lys560, and Asn564 contact the incoming dNTP during the nucleotidyl transfer reaction as judged by variations in apparent Km and kcat values for dNTP incorporation by these mutants compared to those for the exonuclease deficient parental polymerase under steady-state conditions. On the basis of our studies, as well as on the basis of the crystal structure of RB69 gp43, we propose that a conformational change in the fingers domain, which presumably occurs prior to polymerization, brings the side chains of Arg482, Lys486, Lys560, and Asn564 into the vicinity of the primer-template terminus where they can contact the triphosphate moiety of the incoming dNTP. In particular, on the basis of structural studies reported for the "closed" forms of two other DNA polymerases and from the kinetic studies reported here, we suggest that (i) Lys560 and Asn564 contact the nonbonding oxygens of the alpha and beta phosphates, respectively, and (ii) both Arg482 and Lys486 contact the gamma phosphate oxygens of the incoming dNTP of RB69 gp43 prior to the nucleotidyl transfer reaction. We also found that Ala substitutions at each of these four RB69 gp43 sites could incorporate dGDP as a substrate, although with markedly reduced efficiency compared to that with dGTP. In contrast in the parental exo- background, the K483A and Y567A substituted enzymes could not use dGDP as a substrate for primer extension. These results, taken together, are consistent with the putative roles of the four conserved residues in RB69 gp43 as stated above.     

Characterization of a replicative DNA polymerase mutant with reduced fidelity and increased translesion synthesis capacity

Changing a highly conserved amino acid in motif A of any of the four yeast family B DNA polymerases, DNA polymerase alpha, delta, epsilon or zeta, results in yeast strains with elevated mutation rates. In order to better understand this phenotype, we have performed structure-function studies of homologous mutants of RB69 DNA polymerase (RB69 pol), a structural model for family B members. When Leu415 in RB69 pol is replaced with phenylalanine or glycine, the mutant polymerases retain high-catalytic efficiency for correct nucleotide incorporation, yet have increased error rates due to increased misinsertion, increased mismatch extension and inefficient proofreading. The Leu415Phe mutant also has increased dNTP insertion efficiency opposite a template 8-oxoG and opposite an abasic site. The 2.5 A crystal structure of a ternary complex of RB69 L415F pol with a correctly base-paired incoming dTTP reveals that the phenylalanine ring is accommodated within a cavity seen in the wild-type enzyme, without steric clash or major change in active site geometry, consistent with retention of high-catalytic efficiency for correct incorporation. In addition, slight structural differences were observed that could be relevant to the reduced fidelity of L415F RB69 pol.     

Reversal of a mutator activity by a nearby fidelity-neutral substitution in the RB69 DNA polymerase binding pocket

Phage RB69 B-family DNA polymerase is responsible for the overall high fidelity of RB69 DNA synthesis. Fidelity is compromised when conserved Tyr567, one of the residues that form the nascent polymerase base-pair binding pocket, is replaced by alanine. The Y567A mutator mutant has an enlarged binding pocket and can incorporate and extend mispairs efficiently. Ser565 is a nearby conserved residue that also contributes to the binding pocket, but a S565G replacement has only a small impact on DNA replication fidelity. When Y567A and S565G replacements were combined, mutator activity was strongly decreased compared to that with Y567A replacement alone. Analyses conducted both in vivo and in vitro revealed that, compared to Y567A replacement alone, the double mutant mainly reduced base substitution mutations and, to a lesser extent, frameshift mutations. The decrease in mutation rates was not due to increased exonuclease activity. Based on measurements of DNA binding affinity, mismatch insertion, and mismatch extension, we propose that the recovered fidelity of the double mutant may result, in part, from an increased dissociation of the enzyme from DNA, followed by the binding of the same or another polymerase molecule in either exonuclease mode or polymerase mode. An additional antimutagenic factor may be a structural alteration in the polymerase binding pocket described in this article.     

Alteration in the cavity size adjacent to the active site of RB69 DNA polymerase changes its conformational dynamics

Internal cavities are a common feature of many proteins, often having profound effects on the dynamics of their interactions with substrate and binding partners. RB69 DNA polymerase (pol) has a hydrophobic cavity right below the nucleotide binding pocket at the tip of highly conserved L415 side chain. Replacement of this residue with Gly or Met in other B family pols resulted in higher mutation rates. When similar substitutions for L415 were introduced into RB69pol, only L415A and L415G had dramatic effects on pre-steady-state kinetic parameters, reducing base selectivity by several hundred fold. On the other hand, the L415M variant behaved like the wild-type. Using a novel tC^o-tC_nitro Förster Resonance Energy Transfer (FRET) assay, we were able to show that the partition of the primer terminus between pol and exonuclease (exo) domains was compromised with the L415A and L415G mutants, but not with the L415M variant. These results could be rationalized by changes in their structures as determined by high resolution X-ray crystallography.     

The L561A substitution in the nascent base-pair binding pocket of RB69 DNA polymerase reduces base discrimination

Several variants of RB69 DNA polymerase (RB69 pol) with single-site replacements in the nascent base-pair binding pocket are less discriminating with respect to noncomplementary dNMP incorporation than the wild-type enzyme. To quantify the loss in base selectivity, we determined the transient-state kinetic parameters for incorporation of correct and all combinations of incorrect dNMPs by the exonuclease-deficient form of one of these RB69 pol variants, L561A, using rapid chemical quench assays. The L561A variant did not significantly alter the k(pol) and K(D) values for incorporation of correct dNMPs, but it showed increased incorporation efficiency (k(pol)/K(D)) for mispaired bases relative to the wild-type enzyme. The incorporation efficiency for mispaired bases by the L561A variant ranged from 1.5 x 10(-)(5) microM(-)(1) s(-)(1) for dCMP opposite templating C to 2 x 10(-)(3) microM(-)(1) s(-)(1) for dAMP opposite templating C. These k(pol)/K(D) values are 3-60-fold greater than those observed with the wild-type enzyme. The effect of the L561A replacement on the mutation frequency in vivo was determined by infecting Escherichia coli harboring a plasmid encoding the L561A variant of RB69 pol with T4 phage bearing a mutant rII locus, and the rates of reversions to rII(+) were scored. The exonuclease-proficient RB69 pol L561A displayed a weak mutator phenotype. In contrast, no progeny phage were produced after infection of E. coli, expressing an exonuclease-deficient RB69 pol L561A, with either mutant or wild-type T4 phage. This dominant-lethal phenotype was attributed to error catastrophe caused by the high rate of mutation expected from combining the pol L561A and exo(-) mutator activities.     

DNA structure and polymerase fidelity.

The accuracy of DNA replication results from both the intrinsic DNA polymerase fidelity and the DNA sequence. Although the recent structural studies on polymerases have brought new insights on polymerase fidelity, the role of DNA sequence and structure is less well understood. Here, the analysis of the crystal structures of hotspots for polymerase slippage including (CA)n and (A)n tracts in different intermolecular contexts reveals that, in the B-form, these sequences share common structural alterations which may explain the high rate of replication errors. In particular, a two-faced "Janus-like" structure with shifted base-pairs in the major groove but an apparent normal geometry in the minor groove constitutes a molecular decoy specifically suitable to mislead the polymerases. A model of the rat polymerase beta bound to this structure suggests that an altered conformation of the nascent template-primer duplex can interfere with correct nucleotide incorporation by affecting the geometry of the active site and breaking the rules of base-pairing, while at the same time escaping enzymatic mechanisms of error discrimination which scan for the correct geometry of the minor groove.In contrast, by showing that the A-form greatly attenuates the sequence-dependent structural alterations in hotspots, this study suggests that the A-conformation of the nascent template-primer duplex at the vicinity of the polymerase active site will contribute to fidelity. The A-form may play the role of a structural buffer which preserves the correct geometry of the active site for all sequences. The detailed comparison of the conformation of the nascent template-primer duplex in the available crystal structures of DNA polymerase-DNA complexes shows that polymerase beta, the least accurate enzyme, is unique in binding to a B-DNA duplex even close to its active site. This model leads to several predictions which are discussed in the light of published experimental data.     

Bacillus subtilis tau subunit of DNA polymerase III interacts with bacteriophage SPP1 replicative DNA helicase G40P

Genetic evidence suggests that the Bacillus subtilis dnaX gene only encodes for the τ subunit of both DNA polymerases III (Pol IIIs). The B.subtilis full-length protein and their mutant derivatives τ(373– 563) (lacking the N-terminal, domains I–III or amino acid residues 1–372) and τ(1–372) (lacking the C-terminal region or amino acids 373–563) have been purified. The τ protein forms tetramers, τ(373– 563) forms dimers, whereas τ(1–372), depending on the ionic strength, forms trimers or tetramers in solution. In the absence of single-stranded (ss) DNA and a nucleotide cofactor, τ interacts with the SPP1 hexameric replicative G40P DNA helicase in solution or with G40P-ATP bound to ssDNA, with a 1:1 stoichiometry. G40P(109–442), lacking the N-terminal amino acid residues 1–108, interacts with the C-terminal moiety of τ. The data indicate that the interaction of G40P with the τ subunit of Pol III, is relevant for the loading of the Pol IIIs into the SPP1 G38P-promoted open complex.     

Base selectivity is impaired by mutants that perturb hydrogen bonding networks in the RB69 DNA polymerase active site

To investigate the molecular basis for the selective utilization of nucleoside triphosphates complementary to templating bases, by RB69 DNA polymerase (RB69 pol), we constructed a set of mutants that we predicted would perturb the "floor" of the nascent base-pairing interface in the enzyme. We then determined the pre-steady-state kinetic parameters for the incorporation of complementary and noncomplementary dNTPs by the exo(-) form of RB69 pol and its mutants. We found that the Y567A mutant had the same K(d) and k(pol) values for incorporation of C versus G as the wild-type exo(-) enzyme; however, the k(pol)/K(d) ratio for G versus G incorporation with the Y567A mutant was 10 times higher than the k(pol)/K(d) efficiency of G versus G incorporation using the exo(-) RB69 pol. The reduced level of discrimination by the Y567A mutant against incorporation of mismatched bases was also seen with the Y391A mutant. Stopped-flow fluorescence was also employed to monitor rates of putative conformational changes with the exo(-) RB69 pol and its mutants using a primer-template complex containing 2-aminopurine. The rates of fluorescence changes were equal to or greater than the rates of the rapid chemical quench, indicating that we were monitoring a process occurring before or during the phosphoryl transfer reaction. We have interpreted our results within the context of the crystal structure of the RB69 pol ternary complex [Franklin, M. C., et al. (2001) Cell 105, 657-667].