Structural Determinants of Slippage-mediated Mutations by Human Immunodeficiency Virus Type 1 Reverse Transcriptase*

Single-base deletions at nucleotide runs or -1 frameshifting by human immunodeficiency virus type 1 (HIV-1) reverse transcriptase (RT) result from template slippage during polymerization. In crystal structures of HIV-1 RT complexed with DNA-DNA template-primer, the palm subdomain in the template cleft contacts the template backbone near the proposed site of slippage via the Glu89 side chain. We investigated the role of Glu89 in frameshifting by perturbing this interaction. Substitutions with Asp, Gly, Ala, Val, Ser, Thr, Asn, or Lys were created in recombinant HIV RT, and frameshift frequencies of the resulting mutant RTs were measured. All substitutions led to reduced -1 frameshifting by HIV-1 RT (2-40-fold). Interestingly, the suppression of -1 frameshifting frequently coincided with an enhancement of +1 frameshifting (3-47-fold) suggesting that Glu89 can influence the slippage of both strands. Glu89 substitutions also led to reduced rates of dNTP misincorporation that paralleled reductions in -1 frameshifting, suggesting a common structural mechanism for both classes of RT error. Our results reveal a major influence of Glu89 on slippage-mediated errors and dNTP incorporation fidelity. The crystal structure of HIV-1 RT reveals a salt bridge between Glu89 and Lys154, which may facilitate -1 frameshifting; this concept is supported by the observed reduction in -1 frameshifting for K154A and K154R mutants.

Human immunodeficiency virus, type 1 (HIV-1) 5 reverse transcriptase (RT), which converts HIV-1 genomic RNA to a doublestranded DNA form, displays an intermediate level of fidelity among DNA polymerases (1)(2)(3). The relative error proneness of HIV-1 RT is recognized as one of the factors responsible for high genetic variation in HIV, which ultimately leads to the rapid emergence of drug-resistant HIV variants. The broad structural basis of HIV-1 RT fidelity is reasonably well understood. The first structural insights into HIV-1 RT fidelity came from biochemical studies on nucleoside analog RT inhibitor-resistant mutants, in which the residues affected tend to be located at or near the polymerase active site. For example, it was shown that the (Ϫ)-2Ј-deoxy-3Ј-thiacytidine-resistance mutation M184V, affecting a residue located near the incoming dNTP, conferred an increased fidelity of dNTP incorporation on HIV-1 RT (4 -6). The first crystal structure of HIV-1 RT complexed with double-stranded DNA template-primer showed specific structural elements termed primer grip and template grip, which correctly position the templating base and the primer 3Ј terminus for the insertion of incoming dNTP ( Fig. 1) (7). The structure of HIV-1 RT complexed with template-primer and dNTP showed that the ␤3-␤4 hairpin of the fingers subdomain provides key contacts for the templating nucleotide and the incoming dNTP (8). The importance of this region for dNTP selection and fidelity is further supported by reports of several mutations in the ␤3-␤4 hairpin that affect RT fidelity (9 -12).
Whereas residues proximal to the dNTP-binding pocket can directly influence polymerase fidelity, the contribution of residues that contact template-primer duplex to fidelity is presumably indirect. Such residues primarily influence processes related to template-primer interaction such as the RT-template-primer affinity, polymerase translocation along the template during synthesis, and processivity (13)(14)(15). Mutations at such residues can also affect fidelity. For example, two well studied structural elements, ␣H of the thumb subdomain and the template grip ␤5a in the palm subdomain of HIV-1 RT, which are known to affect polymerase processivity (15,16) also affected fidelity (16,17). Alanine substitutions of residues in the ␣H helix that form a part of the minor groove-binding track in the thumb have been shown to alter frameshift fidelity (16,18). The E89G variant in ␤5a of the template grip increases dNTP incorporation fidelity of HIV-1 RT (17).
Glu 89 is of particular interest because of its unique location in the RT template grip (see below) and its established importance in HIV-1 replication and drug resistance. The E89G variant was first identified in an in vitro phenotypic screen of a random library of RT mutations (19,20). HIV-1 virions containing the E89G mutation are replication competent and exhibit multidideoxynucleoside analog resistance (20). Although the E89G mutation is rarely observed alone in natural settings, it has been observed in association with M184V in (Ϫ)-2Ј-deoxy-3Ј-thiacyti-dine-resistant isolates from HIV-infected individuals (21). In the context of SIV replication in monkeys, the E89G mutation rendered the virus less fit forcing it to acquire secondary mutations (22). Mansky and colleagues (23) measured the mutation rate of HIV-1 containing the E89G mutation in a single-cycle replication assay and found that it led to a small reduction (20%) in mutation rate (23). Detailed analysis of mutation patterns in vitro suggests that the reduction in one type of RT error is compensated for by an increase in other types of errors (24).
The crystal structures of HIV-1 RT complexed with the doublestranded DNA show Glu 89 to be a key anchor residue in the ␤5a template grip (7,8). Glu 89 is in close proximity to the sugar-phosphate backbone of the template strand near the penultimate base pair. Thus, substitution at Glu 89 may affect polymerase fidelity by changing the conformation of the template-primer duplex, thereby altering the geometry of the dNTP-binding site (25). Additionally, it is also possible that Glu 89 substitutions affect the positioning of the template-primer rather than its conformation.
A high frequency of single-base deletion errors, or Ϫ1 frameshift errors, is a hallmark of mutations by HIV-1 RT. This type of error occurs at rates 90-fold higher in homopolymeric runs of nucleotides than at non-runs (26). HIV-1 RT also generates ϩ1 frameshifting at nucleotide runs, although at a ϳ5-fold lower frequency than Ϫ1 frameshifting (26). The Ϫ1 frameshifting is thought to be mediated by template slippage events, whereas the ϩ1 frameshifting is caused by primer strand slippage during polymerization (27). The location of Glu 89 is approximately where template slippage might occur, suggesting a role for the Glu 89 in (Ϫ1) frameshifting. In addition, of the 5-6 residues in the ␤5a strand, only the Glu 89 side chain is in close proximity to template strand.
Considering its biological importance in drug resistance (20,21), we wished to study the effect of altering this interaction between template grip and template-primer on slippage-mediated events. Employing the previously reported Glu 89 mutants (28), we investigated the influence of limited vertical scanning mutagenesis of Glu 89 on the frequency of frameshifting by HIV-1 RT. We show here that all eight substitutions of Glu 89 tested display decreased frequencies of Ϫ1 frameshifting. Unexpectedly, the decrease in Ϫ1 frameshifting was also accompanied by increased frequencies of ϩ1 frameshifting. Crystal structures of HIV-1 RT complexed with duplex DNA show contacts between Lys 154 and Glu 89 , thus predicting a possible role for Lys 154 in Ϫ1 frameshifting. By using site-directed mutants of Lys 154 to measure Ϫ1 frameshifting, we confirm this prediction. Furthermore, substitutions at Glu 89 increased the fidelity of dNTP incorporation showing the influence of Glu 89 on multiple types of replication errors.

EXPERIMENTAL PROCEDURES
Enzymes, dNTPs, and Oligonucleotides-HIV-1 RT is a heterodimer of p66 and p51 subunits. RT mutants were prepared by introducing the mutation in the catalytic subunit, p66. The mutant p66 RTs were expressed from the expression plasmid pRT and reconstituted with the wild-type RT p51 expressed from p6HRT51, followed by purification as described before (29). The purified RTs were found to be free of DNase activity even after incubation of excess RT with radiolabeled DNA substrate for up to 8 h. Purification of K154A, K154R, and P157S mutants was as described previously (30,31). Deoxynucleotide triphosphates (dNTPs) were purchased from Amersham Biosciences. The oligonucleotides used in this study are listed in Table 1.
Measuring Frameshift Mutagenesis-Frameshift fidelities were measured via a phage-based gapped duplex assay described by Bebenek et al. (26). The M13mp2 DNA used as a substrate for the assay contained a single-base (T) insertion in the run of Thr at position 70 of lacZ␣ for scoring Ϫ1 frameshifts (5T-1 substrate) and a single-base deletion for scoring ϩ1 frameshifts (7Tϩ1 substrate) (32). Both M13mp2-gapped DNA substrates were generous gifts of T. A. Kunkel (National Institutes of Health, NIEHS, Research Triangle Park, NC). M13-gapped DNA substrates were filled in by wild-type and mutant RTs as described previously (26). Complete filling in was ensured by agarose gel electrophoresis and the successfully filled-in gapped duplex DNA was electroporated into Escherichia coli strain MC1061. Electroporations were performed on separate days for each of three trials for each DNA synthesis reaction. Cells were allowed to recover for 10 min, then mixed with E. coli CSH50 cells (⌬(pro-lac), thi, ara, strA/FЈ(proAB, lacI q Z_M15, traD36) in the presence of 0.25 mM isopropyl ␤-thiogalactopyranoside (LabScientific) and soft agar containing 0.2 mM X-Gal (5-bromo-4-chloro-3-indolyl ␤-D-galactopyranoside; LabScientific), and plated on M9 plates. Plates were incubated at 37°C for 15 h and the plaques were screened for ␤-galactosidase activity by looking for those displaying the dark blue reversion phenotype. Such revertant plaques were picked for sequencing and stored in 1 ml of 0.9% saline at 4°C. Reversion frequencies were determined by dividing the number of revertants by the total number of plaques screened. All reversion frequencies were corrected for background reversion frequency of the substrates (1.30 ϫ 10 Ϫ5 for the 5T-1 substrate and 8.3 ϫ 10 Ϫ5 for the 7Tϩ1 substrate (32)). The isolation of single-stranded DNA from revertant plaques and the determination of nucleotide sequences were as described previously (33). The forward mutation frequencies were determined using a wild-type lacZ␣-gapped duplex DNA as described previously (24).
Single dNTP Exclusion Assay-The "single dNTP exclusion" or the "minus dNTP" assay were performed on DNA-DNA template (55-mer) and primer (28-mer) oligonucleotides (Table 1). A 5Ј-32 P-labeled primer oligonucleotide was extended by purified wild-type or the eight Glu 89 mutant RTs in the presence of all four combinations of three dNTPs.

TABLE 1 Sequences of template and primer oligonucleotides used in fidelity assays
In each case, the top strand represents the primer and the bottom strand, the template.

Mispair extension
The DNA template-primers were prepared by annealing the template (at a molar ratio of 2 template to 1 primer) to the end-labeled primer. Reactions were initiated by combining equal volumes (10 l) of enzymetemplate-primer solution and dNTP salt solution in which all four dNTPs were present, or in which one of the dNTPs was omitted. Each RT was used at two different concentrations (1ϫ and 2.5ϫ). The 1ϫ concentrations for wild-type, E89A, and E89V were, respectively, 0.68, 0.09, and 0.18 nM. Enzyme concentrations for wild-type and mutant RTs were selected to be at equal levels of DNA-dependent DNA polymerase activities on the template-primer used here. Thus, the 1ϫ concentration for each enzyme is different in terms of protein input or unit activity as measured on poly(rA)⅐oligo(dT) template-primer, but similar with respect to their enzyme activity on the DNA-DNA template-primer used. After mixing, reactions contained 100 nM template-primer, increasing concentrations of wild-type, and mutant RTs as described above, 250 M dNTPs, 50 mM Tris-HCl, pH 8, 60 mM NaCl, 20 mM dithiothreitol, 0.05% IGEPAL (surfactant), and 10 mM MgCl 2 . The reactions were terminated after 6 min by the addition of 20 l of stop solution (95% formamide, 20 mM EDTA). Eight microliters of boiled reaction mixture were subjected to electrophoresis in a 10% urea polyacrylamide gel for 2.5 h at 100 W. Autoradiograms were analyzed by densitometry, and band intensities were quantitated via ImageQuant. Total band intensity was calculated as the total intensity of the band immediately below the barrier site plus all the bands above it. Extension was calculated as the sum of all bands at and above the barrier site. Percentage extension as a fraction of the total band intensity was then calculated from total band intensity and extension.
Mispair Extension Reactions-Primer extension reactions were performed using four different 5Ј-32 P-labeled DNA 31-mers (G:C, G:T, G:G, G:A; see Table 1) annealed to a 55-mer DNA template using wildtype and mutant RTs. As before, two different concentrations of each RT (selected to be equal in DNA-dependent DNA polymerase activity between different enzymes as tested on a DNA-DNA template-primer) were tested for each enzyme at 1ϫ and 2.5ϫ concentrations. The 1ϫ concentrations used for the wild-type, E89A, E89V, E89D, and E89G were 0.68, 0.12, 0.18, 0.09, and 2.3 nM, respectively.
Reactions and buffer conditions were as described above for the dNTP exclusion assay, but in the presence of dGTP, dATP, and dTTP, each at 25 M. Reactions were terminated after 10 min and analyzed as previously described for dNTP exclusion assay.

RESULTS
Single-base Deletion Errors-Glu 89 of the ␤5a strand in the palm subdomain of HIV RT is located near the site of template strand slippage.
Thus, it is well positioned to either directly or indirectly influence Ϫ1 frameshifting. To investigate whether the presumptive interaction between the template grip Glu 89 and the template strand influences frameshift mutagenesis by HIV-1 RT, we perturbed this interaction by substituting Glu 89 with Asp, Gly, Ala, Val, Ser, Thr, Asn, or Lys. We then determined the frequency at which polymerizing mutant RTs lead to Ϫ1 frameshifting, using an M13-based reversion assay that employs a lacZ␣ variant as a reporter gene (26). In this assay, deletion of a base in a run of 5 thymidines causes reversion to the wild-type phenotype (dark blue plaques). Interestingly, all substitutions at Glu 89 led to a decrease in the frequency of the Ϫ1 frameshift errors by HIV-1 RT (Table 2). Substitutions to Val, Lys, and Thr resulted in the greatest decreases (40-, 28and 8-fold, respectively) in Ϫ1 frameshift frequency. Other substitutions (Asp, Gly, Ala, Ser, and Asn) led to smaller effects, ranging from 2to 5-fold reduction from the wild-type frequency. It is possible that revertants can also result from ϩ2 events because of primer slippage, which can also restore the reading frame. However, sequencing a sampling of 12 revertants from each of the 9 reactions revealed only Ϫ1 events. These results suggest that wild-type Glu 89 leads to one of the highest rates of single-base deletion mutations, whereas all substitutions tested lead to reductions. The fingers, palm, and thumb subdomains are shown in blue, red, and green (RNase H domain not shown), whereas the template strand is in light green and the primer strand in yellow. Secondary structure elements, labeled in white, are numbered according to the notation in Jacobo-Molina et al. (7). The various structural elements near the active site, such as the ␤3-␤4 hairpin (in fingers), primer grip (palm), which is made of ␤12 and ␤13 strands, the template grip element ␤5a (palm), which contains the Glu 89 , and the ␣H (thumb) are labeled. Illustration was produced using Molscript (39) and Raster3D (40).

TABLE 2 Frequencies of frameshift errors by HIV-1 RT Glu 89 mutants on lacZ␣ template
The frequencies of Ϫ1 or ϩ1 mutation by WT, E89D, E89G, E89A, E89V, E89S, E89T, E89N, or E89K reverse transcriptases were measured in three separate trials, and the mean frequency and S.D. are shown after subtracting the background frequency (as described under "Experimental Procedures"). -Fold increases in Ϫ1 frameshift frequency and -fold decreases in ϩ1 frameshift frequency versus wild-type RT are shown. Single-base Insertion Errors-Although HIV-1 RT can also produce ϩ1 frameshifts, which are mediated by the primer slippage events, the residue Glu 89 does not contact the primer, which is located on the other side of the point of Glu 89 contact with template-primer duplex. Therefore, it was of interest to determine whether Glu 89 would have any influence on primer slippage events. We measured the ϩ1 frameshifting frequency of Glu 89 -substituted RT mutants using a different M13 substrate reporter gene, which contained a single-base deletion in lacZ␣ such that the insertion of a base within a run of 7 Ts results in the reversion phenotype. Although both ϩ1 and Ϫ2 (deletion of 2 bases via template slippage) mutations could restore a functional lacZ␣ gene, previous analysis of a large number of clones has revealed no Ϫ2 mutations in this construct. 6 In contrast to the results obtained for Ϫ1 frameshift frequency measurements, all Glu 89 substitutions (except the Glu to Val substitution) led to an increase in the frequency of ϩ1 frameshift errors ( Table 2). The wild-type ϩ1 frameshift frequency was about five times lower than the Ϫ1 frameshift frequency. The substitutions led to a 2.6 -47-fold increase in ϩ1 frameshift frequency over that of wild-type (Table 2). Thus, the majority of frameshift mutations by wild-type RT were of the Ϫ1 type, whereas the majority of frameshift mutations by all of the variant RTs were of the ϩ1 type. Some of the substitutions, such as Ser, Thr, and Asn caused increases in total frameshifting (7-, 2-, and 6-fold, respectively) while maintaining the inverse ratio of ϩ1 to Ϫ1 frameshifting compared with wild-type RT (Table 1).

Enzyme
Base Substitution Fidelity-It was surprising that the wild-type enzyme displayed the highest Ϫ1 frameshift mutation frequency, as one would expect the wild-type enzyme to have minimal slippage-mediated mutations, which have little evolutionary benefit for protein coding sequences. Therefore, we investigated the influence of wild-type Glu 89 on base substitution fidelity. Base substitution errors (or misincorporation) result from two events during polymerization. First the insertion of an incorrectly paired dNTP (misinsertion) must occur. Such events result in a mispaired primer terminus and are generally not preferred by most DNA polymerases. For the misinserted base to become a misincorporated base, the mispaired primer terminus must be extended by correct insertion of a base opposite the next template base (mispair extension).
To investigate the effect of Glu 89 on both misinsertion and mispair extension, we utilized a gel-based single dNTP exclusion assay or minus dNTP assay (1), which provides a gross estimate of the efficiency of base substitution for a polymerase without proofreading function. This assay involves copying a heteropolymeric DNA template using a 5Ј-end-labeled primer in the presence of only 3 dNTPs (absence of a single dNTP). Primer extension past a template site for which the complementary dNTP is missing depends on the ability of the polymerase to misinsert dNTPs as well as to extend the mispaired primer. Thus the extent of primer extension under these conditions allows a gross estimate of the ability of the polymerase to generate misincorporations (the combined result of both misinsertion and mispair extension events).
The assay was set up as described before (34). A 22-nucleotide 5Ј-end-labeled DNA primer was annealed to a 55-nucleotide oligonucleotide DNA. Absence of one dNTP in the reactions creates a barrier to primer extension at the template position where RT would need to incorporate the missing dNTP. For polymerases (e.g. RT), which lack exonucleolytic proofreading activity, one can obtain an estimate of efficiency of misincorporation, a combined sum of changes in the misinsertion and mispair extension efficiencies, by measuring the amounts of products extended past the barrier. For the results of this assay to be meaningful, it was critical that the enzyme inputs in terms of activity levels be nearly identical in all reactions. To this end, we ensured the presence of equivalent levels of RT in control reactions in which all four dNTP were included. As shown in Fig. 2A, the wild-type and mutant RTs synthesized similar amounts of full-length and intermediate sized products (Fig. 2A, panel labeled All dNTPs), indicating that similar inputs of all RTs, based on DNA-dependent DNA polymerase activity, were used in the reactions. When increasing amounts of enzyme were included in these reactions, the amount of full-length product increased with all RTs tested, indicating that the reactions were not rate-limiting. Wild-type HIV-1 RT efficiently extended the primer past the first barrier site when dATP, dCTP, or dGTP were missing ( Fig. 2A, panels labeled Minus dATP, dCTP, dGTP, respectively). In each of the remaining cases of excluding a single dNTP, the wild-type enzyme was capable of passing through several barrier sites and performing DNA synthesis. In reactions without dATP or without dGTP, wild-type enzyme also synthesized a limited amount of full-length product ( Fig. 2A,   The panel on the left shows the extension products from reactions containing all four dNTPs, whereas reactions in which dATP, dCTP, dGTP, or dTTP are missing in are shown in the panels to the right. Reactions with each RT were carried out at two RT concentrations (see "Experimental Procedures"). The band at the barrier site (for which the complementary dNTP is missing) is indicated by a large dot to the left. The nucleotide sequence of the DNA product synthesized is indicated to the left and the major band at the bottom of panels represents unused primer. B, quantitative plot of extension products in dNTP exclusion assay for all the mutants for reactions missing dATP, dCTP, and dGTP. Because reactions missing dTTP produced no extension, we were unable to plot data for dTTP exclusion. Percentage extension as a fraction of total band intensities are shown for wild-type and all Glu 89 RT mutants.
We then compared primer extension by wild-type and mutant RTs in reactions missing each of the three dNTPs. In reactions where one dNTP is missing, little or no full-length products were generated as expected, but different RTs extended primers to different extents ( Fig.  2A). Fig. 2A shows the results of a dNTP exclusion assay for three of the RTs; wild-type, E89A, and E89V. As seen in the second panel, which displays products of the reactions lacking dATP (Minus dATP panel), E89V RT does not extend the primers beyond the polymerization block imposed by the missing dNTP as readily as does wild-type RT, whereas E89A RT polymerizes to an intermediate extent. E89A and E89V RTs therefore display higher levels of fidelity with respect to dNTP incorporation and extension than does wild-type RT, with E89V displaying the highest fidelity. This trend in fidelity was also apparent in reactions in which dCTP or dGTP were excluded. We note some exceptions. E89A for example, displayed site-specific increase in misincorporation at one site ( Fig. 2A, ϪdATP panel). However, in general, there is a decreased misincorporation compared with wild type. Extension reactions by other Glu 89 mutants revealed similar increases in fidelity (gels not shown). We do not believe that synthesis above the barrier is because of contamination with the correct dNTP for two reasons. One, the dNTPs we used were high pressure liquid chromatography purified by the manufacturer (Amersham Biosciences). Two, in primer extension experiments, addition of different dNTPs leads to a differential mobility shift (data not shown). Therefore, variations in extension synthesis beyond the barrier site, despite the use of equal inputs of activity for different mutant RTs, suggests changes in the efficiency of misincorporation.
To estimate differences in base substitution fidelities of the Glu 89 mutant panel, we quantified the DNA synthesis above the first barrier for all mutants. The sum of intensities of all products above the "no substrate" barrier for which the complementary dNTP is missing (presented as a fraction of the sum of the intensities of the barrier band plus those above it) shows that all substitutions at Glu 89 decrease the efficiency of base misinsertion by RT (Fig. 2B). Quantitation of extension products representing polymerization beyond the barrier revealed that the trend in increased fidelity observed for misinsertion frequency mimics the trend for Ϫ1 frameshift fidelity (Table 2 and Fig. 2B). For example, E89V substitution resulted in the largest decrease in both Ϫ1 frameshifting and base substitution errors (Fig. 2), whereas E89A, E89S, and E89N displayed fidelities most similar to wild-type in both cases. In fact, graphical representations of changes in fidelity in comparison to wildtype for the eight Glu 89 substitutions for Ϫ1 frameshifting and base substitution are superimposable (compare Table 2 and Fig. 2B). Although, at first, this seems to suggest that the decreased efficiency of misincorporation may be because of slippage-mediated synthesis, we believe this to be unlikely because of the fact that none of the barrier sites being studied (the first in each case) are in the nucleotide runs in the heteropolymeric DNA template used in this assay.
In addition to its error proneness for internal misinsertions and frameshifting events, HIV-1 RT has been shown to be capable of inserting non-templated dNTPs at blunt ends of both RNA-RNA and DNA-DNA template-primer (upon reaching the end of copying of a long template) at a high efficiency (35). Interestingly, we observed that some of the Glu 89 mutants also display an absence of non-templated addition of nucleotides. For example, E89V showed a complete lack of nontemplated addition (as shown by the light band at the top of the lane, above the dark band representing the full-length product), but E89A retained this activity. Similarly, E89D, E89G, E89K, and E89T all showed a complete lack of non-templated dNTP incorporation, whereas E89N and E89S retained levels of non-templated dNTP incorporation that were comparable with wild-type (data not shown).
Mispair Extension Frequency-Because the level of substitution determined via dNTP exclusion assay is a combined result of both misinsertion and mispair extension events, it was of interest to determine the contribution of mispair extension alone to the increases in base substitution fidelity. We thus employed a gel-based mispair extension assay in which purified RT enzymes are used to extend 5Ј-end-labeled primers with mismatched termini to compare the relative efficiencies with which wild-type and Glu 89 mutant RTs perform mispair extension events (Fig. 3). The primer extension assay was designed such that the absence of dCTP from reactions would both decrease the chance of observing primer extension resulting from nucleolytic removal of the mispaired primer terminus, as well as allow for the formation of a product that was maximally 10 bases longer than the primer. Thus, any primer extension observed must result from true mispair extension events.
Neither wild-type nor mutant RTs were able to efficiently extend G:G or G:A mispairs, therefore, only the results of extension from the G:T-mispaired terminus and the correctly paired G:C terminus are described. All substitutions at Glu 89 led to an increase in mispair extension fidelity (except E89A; see Fig. 3), although quantitation of extension products revealed that the trend in mispair extension fidelity did not completely mimic the trend in the Ϫ1 frameshift fidelity (Figs. 1  and 3). Substitutions with Asp, Val, Ser, or Thr reduced mispair extension frequency to similar degrees (about 50% of wild-type mispair extension), whereas the E89G and E89K substitutions displayed maximal reductions and Asn substitution had a minimal effect. In all, it appears that the increased fidelity of Glu 89 mutants is because of reductions in both misinsertion and mispair extension events.
Overall Mutation Frequencies-It was of interest to ask whether the large changes observed in Ϫ1 and ϩ1 frameshifting frequencies affect The RTs were incubated with template-primer containing a 32 P-5Ј-endlabeled primer for 10 min before stopping the reactions. The RTs, at two different concentrations, were incubated with correctly paired template-primer (G:C) and the G:T mispaired template-primer. Bottom, quantitative plot of mispair extension products. Band intensities for G:T mispair extension reactions (run in triplicate) with all Glu 89substituted mutants were quantified by densitometry. Total intensity was calculated as primer band plus all bands above, whereas extension was calculated as the sum of all bands above the primer band. Percentage extension as fraction of total band intensity is shown here for wild-type (wt) and mutant RTs.
the overall mutation frequencies of RT mutants. To examine this, we selected two mutants, E89K and E89V, displaying the largest decreases in Ϫ1 frameshifting and mutant E89S, which showed the largest increase in ϩ1 frameshifting. Using a forward mutation frequency assay, we measured the overall mutation frequency of these mutants and compared it to that of wild-type HIV-1 RT. As shown in Table 3, the mutation frequencies of E89K, E89V, and E89S were altered minimally, bringing about reductions of 1.2-, 1.3-, and 1.6-fold from that of wildtype RT, respectively (which correspond to reductions of 16, 25, and 38%, respectively).
Other Template Grip Residues Interacting with Glu 89 or the Template Strand-We wished to investigate the role of other amino acid residues in the vicinity of Glu 89 in frameshifting. Pro 157 in ␣E is also part of the template grip. Interestingly, this residue contacts the same sugar moiety of the template as Glu 89 approaching it from the other side of the same deoxyribose ring. A P157S mutation in HIV-1 RT has been shown to confer resistance to inhibition by nucleotide analogs such as 3TCTP ((Ϫ-)-2Ј-deoxy-3Ј-thiacytidine triphosphate) and FTCTP (2Ј,3Ј-dideoxy-5-fluoro-3Ј-thiacytidine triphosphate) (30). More importantly, the sequence specificity in sensitivity to inhibition by chain terminators displayed by the wild-type HIV-1 RT was altered by the P157S substitution, further confirming its ability to modulate template-primer conformation via interaction with the template strand. Therefore, we tested the effect of the P157S mutation on Ϫ1 frameshift fidelity. Our results show that the effect was minimal with a 1.8-fold reduction in frameshifting (Table 4). When compared with the effect of Glu 89 , revealed by Glu 89 substitution mutants (2.5-40-fold decrease in Ϫ1 frameshifting), the role of Pro 157 appears to be minor. However, one cannot exclude the possibility that more pronounced effects could result from substitutions with residues other than serine.
In HIV-1 RT complexed with DNA and dNTP, in addition to its contact with the sugar-phosphate backbone, Glu 89 forms a salt bridge with the side chain of Lys 154 , also part of the template grip element ␣E. This contact between Glu 89 and Lys 154 is likely to be transient and may contribute to structural constraints that may directly or indirectly affect the stability of frameshifted complexes of RT and template-primer. To investigate this hypothesis, we tested the influence Lys 154 on Ϫ1 frameshifting. Lys 154 is the only positively charged residue in the VLPQGWK motif in the ␤8 -␣E loop at the junction of the fingers and palm subdomains. Among several substitutions of Lys 154 that were previously tested for the effect on HIV-1 RT function (31), we selected two that retained wild-type-like polymerase activities, K154A and K154R substituted mutant RTs, and tested them for Ϫ1 frameshifting. Interestingly, as seen in Table 4, both mutations led to significant reductions in Ϫ1 frameshifting, with 10-and 7.5-fold reductions, respectively (Table 4). These results suggest that Lys 154 is involved in the template-slippagemediated frameshifting events.

DISCUSSION
Our results demonstrate an important role for the Glu 89 of template grip ␤5a in frameshift mutagenesis by HIV-1 RT. In the wild-type enzyme, Glu 89 is positioned to allow a relative high frequency of Ϫ1 frameshifting (i.e. template slippage) with a 5 times lower frequency of ϩ1 frameshifting (primer slippage). Substitutions at Glu 89 appear to disturb this ratio resulting in greater ϩ1 frameshifting events. Our analysis of wild-type and eight different substitutions at position 89 reveals a key role for Glu 89 in many aspects of polymerase fidelity tested, including Ϫ1 and ϩ1 frameshifting, dNTP misincorporations, and mispair extensions. It appears that all types of RT errors except ϩ1 frameshifting are at a maximal level when the wild-type Glu residue is present.
Because of the proximity of Glu 89 with the template backbone, it is not surprising that substitution of this residue leads to a change in frameshifting because of either direct or indirect effects on template slippage. However, the decreased frequency of Ϫ1 frameshifting for all Glu 89 substitutions was unexpected.
The characteristic high frequency of Ϫ1 frameshift errors in homopolymeric runs may be because of the ability of HIV-1 RT to stabilize intermediate nucleic acid structures that result in non-copying of template nucleotides that would yield slipped, mismatched, or unpaired bases. Such structural aberrations are likely to disrupt the positioning of the ribose and phosphate moieties of one or more template base(s). The relatively high Ϫ1 frameshifting frequency of wild-type RT suggests that Glu 89 may directly or   indirectly facilitate the proper conformation of the template grip leading to stabilization of aberrant template-primer duplexes in the template-primer cleft of RT. In a typical, heteropolymeric template sequence, transient misalignments between template and primer nucleotides can be resolved, at some probability, by sequence-specific Watson-Crick base pairing, resulting in correct copying of the sequence. However, within a run of identical nucleotides, the partial misalignment could resolve incorrectly with base pairing to the previous or following primer base. An opposite effect was observed for ϩ1 frameshift fidelity; all substitutions led to increased ϩ1 frameshifting. It appears that the template-and primer-slippage events are inter-related and are apparently mutually exclusive. For example, wild-type Glu 89 appears to facilitate template slippage (Ϫ1 frameshifting), but not primer slippage (ϩ1 frameshifting). In contrast, all substitutions tested here appear to disfavor the template-slippage but promote primer-slippage events. The structural aberration(s) of nucleic acid during primer slippage will be different from that during template slippage and by extension, contacts with the enzyme should also be different. The effect of Glu 89 mutations on the ability of RT to accommodate non-canonical nucleic acids is also likely to be different. Structural studies with the relevant mutants and/or relevant RT-DNA complexes may provide insights into the precise role of residue 89 in ϩ1 frameshifting. It is also possible that suppression of template slippage leading to an increase in the frequency of primer slippage may be because of an innate property of the homopolymeric nucleic acid duplex, such as an unusual conformation, that forces slippage events at these sites. HIV-1 RT is reportedly sensitive to minor groove compression caused by homopolymeric runs of adenosine (36). The fact that several substitutions at Glu 89 cause a reduction in Ϫ1 frameshifting suggests that whether the homopolymeric stretches of sequence leads to template or primer-slippage events is in part determined by the nature of the interaction of the residue 89 or the surrounding residues of template grip ␤5a with the nucleic acid.
Structural Insights into Frameshifting Determinants-To rationalize the influence of RT mutations on frameshifting, we modeled the various substitutions into the structure of HIV-1 RT complexed with DNA and dNTP. As mentioned before, the crystal structure of HIV-1 RT reveals an interaction between Glu 89 and Lys 154 residues (Fig. 4, A and B). Glu 89 is located in the template grip, contacting the template sugar phosphate backbone near the n-2 base (where the templating base is denoted n). Lys 154 is located within the ␤8 -␣E loop forming the junction between the fingers and palm subdomains. Residues other than lysine or arginine at that position are known to significantly affect RT-DNA-dNTP complex formation (31). Interestingly, Lys 154 is in a position to form a hydrogen bond with Glu 89 (8) (Fig. 4B). We hypothesize that Lys 154 and Glu 89 act in concert to maintain the integrity of the template grip region and by extension the polymerase active site.
We hypothesize that ϩ1 and Ϫ1 frameshifting rates can be explained by movements that are influenced by the nature of the protein-DNA contacts present along the two strands. Furthermore, contacts on the template and primer strand are correlated, in the sense that modifying the contacts along one strand may influence the precise pattern of contacts along the other strand because of the necessity for the duplex DNA to pass between the template and primer grips. Thus, substitutions at position 89 would affect interactions with the template strand.
It is common to visualize the precise change in DNA conformation during slippage events as a bulge in the duplex. However, for frameshifting to occur, RT must accommodate this bulge. Because an extra nucleotide can occur in either strand, specific structural "pockets" should be present in both template and primer grip elements, neither pockets are observed. In fact, when we attempted modeling a slippage intermediate, a bulge could not be accommodated near the Glu 89 contact, but only 5-6 bases away from the primer 3Ј terminus near the end of the polymerase module part of the template cleft. Furthermore, if there was an extrahelical pocket, frameshifts mediated by HIV-1 RT should not be exclusively limited to nucleotide runs.
An alternate possibility is suggested by base sharing observed in other nucleic acid-protein interactions. For example, in MutS complexed with mismatched DNA (37), a strand containing an extra nucleotide can overwind around the other strand thus keeping the extra base within the duplex and base pairs shear. Base pair sharing (or shearing) can occur most readily in a single nucleotide repeat and for both insertion and extension steps. A similar base sharing is observed in the case of the RNA/DNA duplex containing the polypurine tract sequence when bound to the ribonuclease H domain of HIV-1 RT (38). This possibility is shown schematically in Fig. 5. Thus, base sharing would imply that Glu 89 mutants vary in their ability to accommodate the kinked backbone of the template or primer DNA for Ϫ1 and ϩ1 frameshifting events, respectively.
Multiple Effects of Glu 89 Substitutions-All tested substitutions at Glu 89 led to an increase in base substitution fidelity, as demonstrated by the results of dNTP exclusion assay results (Fig. 2). The pattern of changes observed among various mutants for Ϫ1 frameshifts was similar to that of base substitutions. It has been proposed that subtle varia- The duplex at the top shows a short template-primer duplex DNA containing a run of 5 nucleotides near the primer 3Ј terminus. The position of Glu 89 at the penultimate base pair is shown near the template strand by an upward arrow. The two diagrams below the normal duplex show a nucleotide bulge on the template or primer strand that accompany template or primer slippage, respectively. There are no receiving pockets on the template-primer binding cleft to accommodate these extrahelical projections. Therefore, the slippage events can be better explained by base sharing shown in the bottom two diagrams. tions in the conformation and/or position of the template-primer duplex can influence the dNTP-binding pocket, as shown by altered nucleoside analog sensitivity of the E89G variant (25) and the impact of P157S mutant on sequence-specific susceptibility to chain terminators (30). Our results from the dNTP exclusion assay further corroborate that Glu 89 , in addition to affecting frameshifting, also influences the efficiency of dNTP misinsertion consistent with the previous report of a E89G mutant (17). These results are in agreement with the notion that template-contacting residues influence the geometry of the dNTP binding pocket. The effects of Glu 89 substitutions on mispair extension were milder and did not correlate with the pattern of variation observed for Ϫ1 frameshifting. Thus, the contacts by Glu 89 or the nearby residues that apparently affect DNA conformation and or positioning, do not appear to influence mispair extension frequencies to a considerable degree.
The fact that Glu 89 substitutions led to decreases in both Ϫ1 frameshift frequency and base substitution frequency, and that the trends in fidelity for Glu 89 mutants were strikingly similar suggests that these distinct mutational events are linked and probably result in a particular enzyme-template-primer conformation. It appears that the status of the duplex DNA that favors primer slippage also results in lower rates of misinsertion and mispair extension and that the wild-type RT (with a Glu at this position) seems to perform more template-slippage events as well as higher rates of misinsertion and mispair extension events. It is possible that the interaction of non-Glu amino acid residues with the template backbone affords a more stable (i.e. less flexible) DNA conformation within the dNTP-binding pocket. As the template strand forms a key surface of the dNTP binding pocket, this results in greater fidelity of dNTP incoporation.
It is not surprising that Glu 89 is crucial for many aspects of polymerization, as it directly contacts the template in the template grip of HIV-1 RT (Fig. 4A). Its influence on nucleoside analog sensitivity and RT fidelity suggest that alteration of Glu 89 imparts changes to the dNTP binding pocket that alter the ability of RT to discriminate between dNTPs and ddNTPs (dideoxynucleoside triphosphates). Here we have shown a new, specific role of Glu 89 in polymerase fidelity of HIV-1 RT; namely, that Glu 89 is a determinant of the high frequency of Ϫ1 frameshifting at nucleotide runs. Second, our results confirm the previous conclusion (17) that Glu 89 determines the efficiency of correct dNTP incorporation and mispair extension fidelity and suggest that these effects are a consequence of the proximity of Glu 89 to the template backbone.