Purifying Selection Masks the Mutational Flexibility of HIV-1 Reverse Transcriptase*

DNA and RNA polymerases share a core architecture composed of three structurally conserved motifs: A, B, and C. Although the amino acid sequences of these motifs are highly conserved between closely related organisms, variation across broader evolutionary distances suggests that only a few residues in each motif are indispensable for polymerase function. To test this, we constructed libraries of human immunodeficiency virus type-1 (HIV-1) containing random single amino acid replacements in motif B of reverse transcriptase (RT), and we used selection in culture to assess RT function. Despite the nearly absolute constancy of motif B in vivo, virus replicating in culture tolerated a range of conservative and nonconservative substitutions at 10 of the 11 amino acid positions examined. These included residues that are invariant across all retroviral subfamilies and highly conversed in diverse retroelements. Several mutants retained wild type infectivity, and serial passage experiments revealed replacements that were neutral or even beneficial to viral fitness. In addition, a number of the selected variants exhibited altered susceptibility to the nucleoside analog inhibitors AZT and 3TC. Taken together, these data indicate that HIV-1 tolerates a range of substitutions at conserved RT residues and that selection against slightly deleterious mutations (purifying selection) in vivo masks a large repertoire of viable phenotypic variants. This mutational flexibility likely contributes to HIV-1 evolution in response to changing selection pressures in infected individuals.

Replicative DNA and RNA polymerases catalyze the transfer of genetic information from parent to progeny. In response to diverse biological requirements, polymerases have evolved with varied substrate specificities, mechanisms of initiation, and associations with other replication proteins (1). This is clearly evident in bacterial and animal viruses, which exhibit wide ranging host environments and replication strategies. Thus, different viral polymerases initiate synthesis from RNA, DNA, or protein primers; require viral or host accessory factors; and preferentially incorporate dNTP, rNTP, or nucleotide analog substrates (1).
Despite these biochemical differences, viral polymerases are structurally very similar (2)(3)(4)(5)(6). The active site motifs A, B, and C are virtually superimposable in all known structures of replicative DNA-and RNA-dependent polymerases (Fig. 1A). Conservation of this core architecture presumably reflects a fundamental requirement for template-directed nucleotidyl transfer. Motifs A and C form antiparallel ␤ sheets that position catalytic aspartates at the active site, and motif B packs against this structure (2,3). In viral reverse transcriptases (RTs) 1 and RNA-dependent RNA polymerases (RDRPs), the amino-terminal portion of motif B forms a loop that contacts both the incoming dNTP and the template strand proximal to the polymerase active site (Fig. 1A) (2)(3)(4)(5)(6). This loop is absent in structures of DNA-dependent polymerases, which instead possess a single, continuous ␣-helix in this region (3). The carboxyl-terminal portion of the motif B helix is common to both RNA-and DNA-dependent polymerases, although these two classes of polymerases share little sequence similarity in this region (3).
The strong structural conservation of motif B in RTs and RDRPs suggests that amino acid residues in this region are critical for polymerase function and viral replication. Two lines of evidence support this suggestion. First, biochemical analyses of mutant RTs and RDRPs have identified a number of amino acid replacements in motif B that severely compromise polymerase activity (7)(8)(9)(10)(11)(12)(13)(14)(15). Second, the primary sequence of motif B is strictly conserved in independent virus isolates; in human immunodeficiency virus type-1 (HIV-1), this region is nearly invariant (16). Motif B sequences are also highly conserved between members of a particular viral family (e.g. retroviruses) (Fig. 1B, top). However, alignments of RT sequences from distantly related retroelements reveal variation in this region (Fig. 1B, middle), and further variability is apparent when RTs are compared with RDRP sequences, since only a single glycine is absolutely conserved across all families (Fig. 1B, bottom). A similar pattern of sequence variation occurs in motifs A and C (17,18).
Taken together, these data suggest that the majority of residues in conserved motifs can accept mutations that preserve essential polymerase functions and, by extension, viral replication. In this study, we used a biological approach to determine the functional importance of individual amino acid sites in motif B of HIV-1 RT. Pools of HIV-1 variants with random single amino acid replacements in motif B were constructed and then passaged in culture. Replication-competent viruses selected from these pools included mutants with substitutions at highly conserved positions, and several variants exhibited altered susceptibility to nucleoside analog inhibitors. We also identified a variant that outgrew the wild type strain during the course of serial passage, demonstrating that mutations in a conserved polymerase element are occasionally adaptive in character. Our data, together with recent studies of other poly-merases (19 -22), indicate that mutational flexibility is a general property of DNA polymerases and that negative natural selection (i.e. purifying selection) (23,24) conceals a number of viable RT variants with unique phenotypes. The potential impact of this plasticity on the evolution of HIV-1 variants in vivo is discussed.

EXPERIMENTAL PROCEDURES
Plasmids, Cells, and Viruses-All mutant strains and random virus pools were derived from pR9, a full-length HIV-1 proviral clone containing NL4 -3 (gag, pol, and env) and HXB2 (5Ј-and 3Ј-LTRs) sequences in which all open reading frames are intact (25). To facilitate manipulations of the pol gene, we used a modified version of pR9 that lacks an ApaI site in the plasmid backbone (kindly provided by Dr. Uta von Schwedler, University of Utah). pR9⌬pol was created from this modified pR9 by replacing the central ApaI/EcoRI fragment (HIV-1 NL4 -3 nt 2010 -5743) with an ApaI-ATCGATGCGGCCGC-EcoRI synthetic linker (unique ClaI and NotI sites are underlined and italicized, respectively). pBS-pol was created by subcloning the same ApaI/ EcoRI fragment into pBluescript II KS(Ϫ) (Stratagene). This fragment includes the entire HIV-1 pol gene (encoding RT) and portions of gag and env. pBS-pol am (containing an amber stop codon substitution at RT position 154) and pBS-pol ClaI (containing a unique ClaI site insertion, GATCGAT, at RT codon 152) were generated from pBS-pol by sitedirected mutagenesis (Muta-Gene phagemid mutagenesis kit; Bio-Rad).
Alignment of Motif B Sequences-GenBank TM accession numbers for the sequences aligned in Fig. 1B  represents A/T/G/C; D is A/G/T; B is C/G/T; H is A/C/T; V is A/C/G; R is A/G; Y is C/T; and S is C/G). These resulted in the exclusion of wild type amino acid codons, as well as variants L149F, Q151H, K154M, S156C, S156W, and Q161H, from the oligonucleotides used for random mutagenesis.  (17,18). Three lentiviral sequences (bracket) and a single member of the other retroviral genera (␣-⑀, spuma) are shown (top) along with a single member of each retroelement group (middle) and three RDRPs for which structures have been solved (bottom). The ␣-helical (␣) and loop regions (dash) are indicated on the top line. The numbers are the codon positions for HIV-1 RT. RT sites with Ն90% sequence identity are highlighted in red, Ն65% identity in yellow, and Ն65% chemical similarity in gray. The following groups of conservative amino acids were considered: nonpolar aliphatic (Gly, Ala, Val, Leu, Ile, and Pro); polar uncharged (Ser, Thr, Cys, Met, Asn, and Gln); positively charged (Arg, Lys, and His); negatively charged (Asp and Glu); and aromatic (Phe, Tyr, and Trp). The boxes outlined in black indicate residues that are invariant in retroviral RT sequences. *, sequences obtained from infectious molecular clones. †, sequences from biochemically active RT clones. See "Experimental Procedures" for sequence sources. EIAV, equine infectious anemia virus; MLV, murine leukemia virus; HTLV-1, positions Gly 152 , Lys 154 , Pro 157 , and Gln 161 , separate mutagenesis reactions were conducted for each mutant oligonucleotide preparation, and the products of these reactions were maintained as separate libraries and pools in subsequent cloning and virus production steps. For the remaining amino acid positions, single mutagenesis reactions were conducted by mixing equal amounts of each oligonucleotide preparation prior to mutagenesis. Products from the mutagenesis reactions were electroporated into ElectroMAX DH10B E. coli (Invitrogen), plasmids were isolated from pools of Ͼ10 4 independent transformants, and ApaI/ EcoRI fragments from the pools were cloned into pR9⌬pol ( Fig. 2A). The resulting full-length pR9 mutant libraries were purified from pools of Ͼ10 4 independent transformants using the Endo-Free Maxiprep kit (Qiagen). Mutant pR9 libraries derived from pBS-pol ClaI were additionally digested with ClaI, transformed back into E. coli, and reisolated. This step increased the proportion of mutants in the final plasmid pools from 5-40 to Ն90%.
Transfections-To prepare wild type virus stocks and random virus pools, CaPO 4 -pR9 DNA coprecipitates were prepared with 10 g of plasmid DNA as described (41) except that the 2ϫ BBS buffer was replaced with 2ϫ HEPES-buffered saline (270 mM NaCl, 10 mM KCl, 1.5 mM Na 2 HPO 4 ⅐2H 2 0, 11 mM dextrose, 40 mM HEPES, pH 7.05). 293T cultures were seeded into 10-cm plates and grown to ϳ25% confluence prior to the addition of the CaPO 4 -DNA coprecipitates. Following a 3-h incubation, culture supernatants were aspirated, 2 ml of 10% glycerol in phosphate-buffered saline (PBS) were added, and the cells were incubated at 37°C for 2 min. Cells were then rinsed twice with 6 ml of PBS, 10 ml of DMEM were added to each plate, and the cultures were returned to the incubator. Culture supernatants were harvested ϳ42 h after transfection, filtered through 0.4-m syringe filters, and stored in 1-ml aliquots in the vapor phase of liquid nitrogen for subsequent analysis. Stocks of individual HIV-1 mutants were produced by transfection of 293T cells as described above, except that the glycerol shock step was omitted. Cultures were instead treated with chloroquine at a final concentration of 25 M in DMEM immediately before adding the DNA/CaPO 4 mixtures. Supernatants were aspirated and replaced with fresh medium 1 day later and then harvested and frozen the following day as described above. The titers of wild type stocks produced by either protocol typically ranged from 2 ϫ 10 5 to 5 ϫ 10 5 focus-forming units (FFU)/ml for frozen stocks and 1 ϫ 10 6 to 4 ϫ 10 6 FFU/ml for fresh preparations.
Infectivity Assay-Single-round infectivities were determined using HeLa-P4 indicator cells (27). HeLa-P4 cells were seeded into 96-well plates at 0.5 ϫ 10 4 cells/well 1 day prior to infection. Serial dilutions of virus were prepared in DMEM containing 20 g/ml DEAE-dextran (Sigma), and 25 l of each dilution were added to separate culture wells. Following a 3-h incubation at 37°C, 175 l of DMEM were added to each well, and the plates were returned to the incubator. Stocks containing low titers were assayed in 24-well plates to increase the accuracy of the assay. In this case, each well received 2 ϫ 10 5 cells, 200 l of virus stock, and 1 ml of medium following the 3-h incubation period.
After 40 h of growth, cultures were fixed and stained with 5-bromo-4-chloro-3-indolyl-␤-D-galactopyranoside (X-gal; Promega). In this time frame, infected cells appeared as isolated groups of 2-5 contiguous blue cells, indicating that the majority of foci were derived from a single cycle of virus replication. Titers from the focal assay were normalized against HIV-1 capsid p24 concentration (DuPont HIV-1 enzyme-linked immunosorbent assay) to determine the infectivity of mutants relative to wild type virus. The infectivity of the wild type virus was 880 Ϯ 180 FFU/ng p24 (means Ϯ S.E.) for frozen stocks and 4500 Ϯ 1100 FFU/ng p24 for fresh supernatants. A variant containing the RT-inactivating D185A mutation served as a negative control and yielded 0.014 Ϯ 0.004 FFU/ng p24 from frozen stocks.
Serial Passage Protocol-Supernatants from 293T cultures transfected with pR9 mutant libraries were used to initiate each serial passage experiment (Fig. 2B, input virus pools). HeLa-P4 cells were seeded at 10 5 cells/10-cm plate in DMEM and incubated at 37°C overnight. Infection was initiated by the addition of 10 4 infectious particles (in 2 ml of DMEM containing 20 g/ml DEAE-dextran), followed by incubation at 37°C for 2-4 h. An additional 6 ml of DMEM was then added to each plate, and incubation was continued overnight. The monolayers were then washed three times with 6 ml of phosphatebuffered saline, and cultures were replenished with 8 ml of fresh medium, which was changed again after 2 days of incubation. On the fifth day after infection, culture supernatants were harvested, passed through 0.4-m filters, and treated with 420 units of DNase I (Worthington) at 37°C for 20 min after adding MgCl 2 to a final concentration of 10 mM. Viral particles were concentrated by layering 1 ml of culture supernatant onto a 250-l cushion of 20% sucrose in phosphate-buff-ered saline, followed by centrifugation at 15,500 ϫ g for 90 min at 4°C. Pelleted virions were resuspended in 100 l of lysis buffer (7 M urea, 0.35 M NaCl, 4 mM EDTA, 10 mM Tris⅐HCl, pH 7.5, 1% SDS) and stored at Ϫ20°C for subsequent RNA extraction and RT-PCR amplification. The remaining unconcentrated culture supernatants were frozen in 1-ml aliquots in the vapor phase of liquid nitrogen. Titers were determined as described above, and the process was repeated to initiate the next passage. Thus, a "passage" is defined as one cycle of infection, incubation, and harvest. The 5-day incubation period corresponds to ϳ5-10 cycles of wild type HIV-1 replication (assuming a replication rate of 1-2 days Ϫ1 ) (42).
Viral RNA Extraction-Viral lysates (100 l; see above) were thawed on ice, and 100 l of TE, pH 8, was added to each tube. Samples were extracted twice with equal volumes of acidic phenol (pH 4) and backextracted at each step by the addition of 100 l of TE, pH 8, to the organic phase. The resulting aqueous phases were then extracted twice with equal volumes of chloroform/isoamyl alcohol (24:1). Viral RNA was precipitated with 300 l of 100% isopropyl alcohol followed by incubation on ice for 30 min. Samples were pelleted by centrifugation at 15,500 ϫ g, washed with 300 l of 70% EtOH, resuspended in 20 l distilled H 2 O, and stored at Ϫ20°C.
If products were observed in control RT-PCRs that lacked RT (see below), the corresponding RNA samples were subjected to a second DNase I treatment to remove contaminating plasmid or proviral DNA. Samples (8 l) were digested with 1 unit of amplification grade DNase I (Invitrogen) for 15 min at room temperature in 2 mM MgCl 2 , 50 mM KCl, 20 mM Tris-HCl, pH 8.4. Reactions were quenched by adding EDTA to a final concentration of 2.5 mM and then heat-inactivated at 65°C for 10 min. The resulting products (1 l/reaction) were then added to RT-PCR mixtures and amplified as described below.
RT-PCR-RT-PCR was performed in thin-walled tubes (Robbins) using the Access RT-PCR system (Promega) in 50-l reactions containing a 200 M concentration of each dNTP, 1 mM MgSO 4 , 5 units of avian myeloblastosis virus reverse transcriptase, 5 units of Tfl DNA polymerase, 10 l of 5ϫ avian myeloblastosis virus/Tfl buffer, 50 pmol each of primers H3 and BH1 (see below), and 1 l of extracted viral RNA. Primers BH1 (5Ј-TATGGATCCCTTTTAGAATCTCCCTGTTTTCTGC-C-3Ј, BamHI site underlined) and H3 (5Ј-AGTCAAGCTTGGATGGCC-CAAAAGTTAAACAATGGCC-3Ј, HindIII site underlined) amplified a 0.8-kb fragment corresponding to nucleotides 2597-3483 of HIV-1 NL4 -3 . Thermocycling conditions in an MJ Research PTC-100 thermocycler were as follows: 48°C for 45 min, 94°C for 2 min; then 40 cycles of 94°C for 30 s, 60°C for 1 min, and 68°C for 2 min; and ending with 68°C for 7 min. Control reactions lacking RT were performed and analyzed in parallel to ensure that amplification was RNA-dependent. When required, viral RNA samples were subjected to a second DNase I treatment to remove contaminating plasmid or proviral DNA (see above). RT-PCR products were sequenced either directly (to quantify fitness) or after subcloning (to identify individual mutants) using primer pNL1 (5Ј-GACTTCAGGAAGTATACTGC-3Ј) and Big Dye Terminator chemistry (Applied Biosystems).
Molecular Cloning and Sequencing-RT-PCR products were digested with BamHI and HindIII (Promega), ligated into pBluescript II KS(Ϫ) (Stratagene), and electroporated into ElectroMAX DH10B E. coli (Invitrogen) for blue/white screening. Individual white colonies were picked and mixed into 25 l of RT-PCR mixture as described above, but without avian myeloblastosis virus RT, and the 0.8-kb pol fragment (nt 2597-3483) was amplified using the Access RT-PCR system (Promega) and primers H3 and BH1. Thermocycling conditions were as follows: 94°C for 3 min; then 30 cycles of 94°C for 30 s, 60°C for 1 min, and 68°C for 2 min; and ending with 68°C for 7 min. The resulting PCR products were purified using the Qiaquick PCR purification kit (Qiagen), resuspended in a final volume of 50 l of distilled H 2 O, and sequenced using primer pNL1.
Quantitation of Relative Fitness-RT-PCR products derived from culture supernatants were directly sequenced as mixed pools without subcloning, and the resulting chromatogram tracings were used to determine the relative frequencies of mutant and wild type genomes at different passage intervals. Peaks that unambiguously represented the wild type sequence were identified at specific nucleotide positions for each motif B codon (except Gly 152 , Trp 153 , and Gln 161 ; see legend to Fig.  4B). Wild type and mutant peak heights were measured at the informative nucleotide position of each codon and used to calculate relative fitness (43) using the expression f n ϭ M n /(W n ϩ M n ), where f n is the relative frequency of mutants at passage n, W represents the peak height for the wild type nucleotide, and M is the sum of all mutant peak heights at the informative nucleotide position. f n /f 1 ratios were plotted as a function of passage number, and the slopes of the resulting fitness vectors were calculated by linear regression (Statview 4.0). The F test was used to determine whether these slopes were significantly different from the neutral case (slope ϭ 0).
Drug Susceptibility Assays-Virus suceptibilities to 3Ј-azido-3Ј-deoxythymidine (AZT; Sigma) and (Ϫ)-␤-2Ј,3Ј-dideoxy-3Ј-thiacytidine (3TC; kindly provided by Dr. Raymond Schinazi, Emory University) were determined by measuring the dose-dependent reduction of infected foci in 96-well HeLa-P4 microcultures. HeLa-P4 cells were seeded into 96-well plates at a density of 0.5 ϫ 10 4 cells/well and incubated overnight. The following morning, culture wells were dosed with varying concentrations of AZT or 3TC, and plates were then returned to the incubator for an additional 3 h. During this period, virus stocks were diluted to 4000 FFU/ml in DMEM containing 20 g/ml DEAE-dextran. Supernatants in the microtiter plates were aspirated, and 25 l of the virus dilutions were added directly to the monolayer in each well. Plates were then returned to the incubator for 3 h. After this time, an additional 175 l of medium were added to each well, a second dose of drug was added (at the same concentration as the first dose), and incubation was continued for an additional 2 days. Culture monolayers were fixed and stained as described above, and blue foci were counted. Control cultures incubated in the absence of drug typically yielded 70 -250 foci/well. Concentrations of drug required to inhibit focus formation by 50% of the untreated control value (EC 50 ) were calculated by linear regression of the resulting dose-response data.

Generation of HIV-1 Populations with Random Motif B Mutations-
To examine the range of mutations in a conserved region of RT that support viral replication, we generated pools of HIV-1 mutants, each randomized at one of 11 amino acid sites in motif B (Fig. 2). Site-directed mutagenesis was performed using degenerate oligonucleotides designed to introduce all 19 possible amino acid substitutions at each motif B position (with a few exceptions; see "Experimental Procedures"). To minimize wild type sequences in the mutant pools, we started with RT subclones containing either a stop codon at position 154 (pBS-pol am ; Fig. 2A, top left) or a unique ClaI restriction site insertion at position 152 (pBS-pol ClaI ) ( Fig. 2A, bottom left) and used mutagenesis primers that corrected these lethal mutations while simultaneously introducing random nucleotides at the target codon sites. Each library of RT subclones was then ligated en masse into an RTdeleted HIV-1 construct (pR9⌬pol) to generate the corresponding full-length HIV-1 pR9 libraries ( Fig. 2A, right). These full-length mutant libraries were separately transfected into 293T cells to produce mutant HIV-1 populations, each randomized at a single codon site in motif B (Fig. 2B). 293T cells do not express the HIV-1 CD4 receptor and therefore cannot be reinfected by progeny virions, enabling us to examine the infectious potential of the virus pools prior to biologic selection (see below). All cloning steps and library manipulations were conducted with Ͼ10 4 independent clones or viruses to minimize sampling bias and ensure population diversity.
Sequence analysis of 180 arbitrarily isolated clones from the Gly 152 , Lys 154 , Pro 157 , and Gln 161 mutant libraries confirmed that the targeted motif B positions contained a diverse array of amino acid substitutions ( Fig. 5 and data not shown). Of the 20 possible amino acids, codons for 12, 16, 17, and 14 were observed in this sampling of the Gly 152 , Lys 154 , Pro 157 , and Gln 161 mutant pools, respectively (average ϭ 15). Our approach limited but did not exclude wild type codons from the libraries, which were present at frequencies of 5-12%. Less extensive sequencing of the other libraries revealed similar mutant diversity and wild type levels (data not shown). Thus, we estimate that the 11 mutant libraries together represent a minimum of 150 different mutations in motif B.

HIV-1 Tolerates Mutations at Conserved Residues in Motif B-To determine the effects of motif B mutations on virus infectivity, each of the mutant HIV-1 pools were titered on
HeLa-P4 indicator cells (Fig. 3A). Randomization of positions Pro 150 or Gly 152 resulted in a dramatic reduction in infectivity (0.2 and 0.1% of wild type, respectively), indicating that these residues are essential for HIV-1 replication. In contrast, pools randomized at positions Leu 149 , Gln 151 , or Trp 153 showed low but significant levels of infectivity (1% of wild type), and the remaining six mutant populations exhibited 15-70% of wild type infectivity, indicating that these pools contained a high proportion of replication-competent variants.
Replication-competent mutants were selected from the random virus pools by subjecting the pools to a single passage in HeLa-P4 cells (Fig. 2B). The spectra of infectious variants in these selected populations were then determined by sequencing individual clones derived from each mutant pool (Fig. 3B). Among 368 clones sequenced from the passage 1 pools, 345 (95%) were replication-competent as evidenced by their persistence in subsequent passages and/or infectivity as purified mutant virus (see below). Altogether, 53 different amino acid substitutions in motif B supported virus replication. Two categories of sites were identified based on the range of substitu- Relative Fitness of Motif B Variants-To further quantify the effect of motif B mutations on HIV-1 replication, we performed serial passages of each mutant virus pool (Fig. 2B, right) and determined the relative frequencies of variant and wild type progeny virus at several passage intervals (Fig. 4). This provides a measure of viral "fitness" (44), here defined as the overall ability of HIV-1 variants to survive and replicate in HeLa-P4 cells relative to wild type HIV-1.
Substitutions in motif B were generally deleterious to viral fitness, as evidenced by the progressive loss of sequence diversity (Fig. 4A) and emergence of wild type virus in most of the FIG. 4. Relative fitness of motif B variants. A, sequencing chromatograms showing changes in the relative frequencies of wild type and mutant genotypes during serial passage. Viral RNA in culture supernatants was amplified by RT-PCR, and the products were analyzed by automated DNA sequencing. The wild type nucleotide sequence for each randomized codon is shown in parentheses. Green, black, red, and blue peaks correspond to A, G, T, and C, respectively. Data are shown in order of increasing variant fitness (top to bottom). Data for Lys 154 are from a pool randomized with primer 154BNN as described under "Experimental Procedures." B, fitness vectors for motif B variants. The slope of each vector indicates the rate of change in the proportion of mutant HIV-1 genotypes relative to wild type during serial passage of a single random virus pool (Fig. 2 right). Each pool contains a mixture of wild-type HIV-1 and mutant viruses with substitutions at the indicated RT residue (see Fig. 3B). The relative mutant frequencies in each population (except Gln 161 ) were determined from chromatogram peak heights (see   Table I and "Experimental Procedures" for details. B, motif B variants identified in the random pools following a single passage in culture. The subscripts indicate the number of clones containing wild type (top line) or variant sequences (lower lines) and are the combined results of all mutant pools (see "Experimental Procedures" for pool descriptions). The black letters indicate mutants shown to be replication-competent based on their persistence in subsequent passages and/or infectivity in the single cycle assay (Table I and data not shown). Mutants in gray letters were detected only in passage 1 supernatants and were not assayed for replication competence. However, most of these mutants are probably infectious, because they were present at frequencies similar to other replication-competent mutants in the passage 1 pools. K154Q was absent in passage 1 but detected in passage 2 (1 of 32 clones sequenced). random mutant pools during the course of serial passage (Fig.  4B). Similar rates of wild type outgrowth were observed in independent serial transfers of Lys 154 mutant pools, illustrating the reproducible nature of this assay. Sequencing of individual clones from each passage of the Lys 154 , Pro 157 , and Gln 161 mutant pools confirmed a steady increase in the proportion of wild type virus and the persistence of variants at low levels in these populations through passage 4 (Fig. 5, A-C).
Although there was a general trend toward reduced fitness, several mutant pools exhibited slow or undetectable outgrowth of wild type virus (Fig. 4). Only a gradual increase in the relative proportion of wild type virus was apparent in the Lys 154 and Gln 161 pools, whereas no wild type outgrowth was detected in the pool randomized at position Val 148 (vector slope not statistically different from zero; p ϭ 0.42, F test). Thus, several amino acid substitutions at Val 148 were neutral under these culture conditions. Surprisingly, a single alanine variant emerged from the Ser 156 random pool that increased in frequency relative to wild type during the course of serial passage. The positive slope of the resulting fitness vector indicates that the S156A mutation improved the relative replication rate of HIV-1 in HeLa-P4 cell culture (slope ϭ 0.05 Ϯ 0.005; p Ͻ 0.001, F test).
To determine whether sampling errors or the presence of wild type virus affected variant selection in our protocol, we performed a second serial passage experiment using a Lys 154 mutant pool that contained no detectable wild type sequences (Ͻ1%). After 13 passages, wild type virus remained undetectable (0 of 27 clones sequenced) (Fig. 5C, bottom), indicating that Lys 154 variants replicate in HeLa-P4 cells as a relatively stable mutant virus population. Sequencing of individual clones from passages 1-4 showed that the spectrum of replication-competent mutants in this second experiment (Fig. 5C, bottom) was similar to that observed in the original experiment (Fig. 5C,  top). Thus, sampling errors did not contribute significantly to the outcome of these serial passage experiments, and the presence of wild type virus did not influence the diversity of emergent genotypes. Mutagenesis strategies that stringently exclude wild type sequences are required to perform similar analyses at other amino acid positions.
Motif B Variants Exhibit Unique Phenotypes-To characterize the phenotypic properties of variants selected from the virus pools, we first examined the relative infectivities of individual mutants selected in passage 1 ( Fig. 3B and Table I). We tested 23 different mutant clones, each containing a single amino acid replacement in motif B. These were constructed by site-directed mutagenesis or by subcloning PCR products that contained only the desired RT mutation (confirmed by DNA sequencing). Among the 23 mutant clones tested, only one (L149E) failed to produce infectious virus (data not shown). Approximately half of the viable mutants ( Percentages of wild type virus in the input populations are indicated. RT-PCR products amplified from culture supernatants were cloned into a plasmid vector, and Ն25 individual clones from each passage were sequenced (exact number in parentheses). Amino acids are grouped and colored according to biochemical properties: yellow/brown bars, nonpolar aliphatic; gray, aromatic; green, polar uncharged; red, acidic; blue, basic. Data for Pro 157 represent the combined sequencing results from two independently passaged pools, one randomized with primer 157NDN and one randomized with primer 157DNN. Data for Gln 161 were obtained from two independent pools produced using primers 161NBN and DNN. Data for Lys 154 were obtained from three independent pools produced using primers 154BNN, BNY, and NNY. See "Experimental Procedures" for primer details. The percentage of wild type (WT) and mutant genotypes in each pool prior to serial passage ("input") was determined by sequencing Ͼ40 individual clones from the fulllength plasmid libraries and random virus pools. retained at least 50% of wild type infectivity, and several of these were not statistically different from wild type (Table I,  column 2, and data not shown). Only two variants (V148R and Q151V) exhibited substantial replication defects (Ͻ5% of wild type infectivity). Thus, the majority of mutants observed in the passage 1 pools retained relatively high replication capacity and did not require second-site mutations for this viability.
To further evaluate the phenotypes of the selected variants, we determined their susceptibilities to two nucleoside analog inhibitors of HIV-1 RT that function as chain terminators of DNA synthesis (45): AZT and 3TC (Table I, columns 3

and 4).
In agreement with previous studies (46,47), the Q151M mutation in RT conferred low level resistance to AZT, and the addition of four other amino acid substitutions to the Q151M background (A62V, V75I, F77L, and F116Y) increased the level of resistance Ͼ100-fold. Other substitutions at Gln 151 and mutations at Val 148 , Lys 154 , Gly 155 , and Gln 161 had the opposite effect, resulting in 4 -50-fold hypersensitivity to AZT. We also observed substitutions that affected virus susceptibility to 3TC. Replacements at Gln 151 , Pro 157 , and Gln 161 increased virus resistance to 3TC 5-10-fold, whereas G155N conferred a slight (3-fold) hypersensitivity to the drug.
In total, mutations at 7 of the 11 motif B sites analyzed resulted in a significant change in EC 50 for AZT or 3TC. Three of the Gln 151 substitutions (Ala, Val, and Ile) conferred both AZT hypersensitivity and 3TC resistance. Taken together, these data indicate that many residues in RT motif B influence nucleoside analog susceptibility. DISCUSSION Motifs A, B, and C form a core structure conserved in all RTs and RNA-dependent viral polymerases (Fig. 1A) (2)(3)(4)(5)(6). Although the amino acid sequences of these motifs are conserved within a particular viral family, alignments of sequences from evolutionarily distant RNA viruses suggest that only a few residues in each motif are absolutely indispensable for polymerase function and viral replication (Fig. 1B). To test this idea, we performed random mutagenesis of motif B in HIV-1 RT and identified the range of replacements that preserve viral replication in culture (Fig. 2). Replication-competent viruses selected from the random pools contained amino acid substitu-tions throughout motif B, including residues highly conserved in retroviral and retroelement RTs (Fig. 3). Serial passages of mixtures of wild type and mutant viruses revealed high fitness variants (Fig. 4), and a number of mutants exhibited singlecycle infectivities comparable to that of the wild type strain (Table I and data not shown). In addition, amino acid replacements at several motif B residues altered the susceptibility of HIV-1 to the nucleoside analog inhibitors AZT and 3TC (Table I).
Mutations in motif B are rarely observed in HIV-1 sequences obtained from infected patients. Among the nearly 10,000 RT sequences in the Stanford HIV RT Sequence Database (available on the World Wide Web at hivdb.stanford.edu/hiv) (16), only 75 contain mutations at the motif B residues examined in this study (excluding the drug-resistant Q151M variant). This Ͻ1% diversity occurs despite continuous mutation pressure due to error-prone HIV-1 replication (48,49). Thus, motif B is subject to strong negative selection in vivo (23,24). This purifying selection, however, masks the underlying complexity of HIV-1 populations. Our screen of random mutations in vitro reveals the mutational flexibility of RT prior to purifying selection and permits the characterization of replication-competent variants not readily detected in vivo. In addition, this strategy of random mutagenesis coupled to virus selection in culture identifies mutations in RT that are biologically active (Fig. 3), phenotypically unique ( Fig. 4 and Table I), and thus well suited for biochemical studies of RT structure/function.
Our screen of random motif B mutations revealed a diverse array of amino acid substitutions that preserve HIV-1 replication (Fig. 3B). Among the ϳ150 different mutations screened, 53 supported virus replication in culture, and only one randomized position (Gly 152 ) failed to yield replication-competent variants. The immutability of Gly 152 is consistent with its absolute evolutionary conservation (Fig. 1B) and with published mutational analyses of purified HIV-1 RT (9 -11) and other RNA virus polymerases (12)(13)(14)(15), indicating that interactions between this residue and the templating nucleotide are essential for polymerase function (Fig. 1A) (4,5). Our data, together with random mutagenesis studies of the "fingers" domain in RT and motif A in other DNA polymerases (19 -22), indicate that mu- tational flexibility of conserved structural elements is a general property of polymerases from diverse organisms. Similar to the in vivo situation, we observed that amino acid substitutions in motif B were generally deleterious to viral fitness (Fig. 4). We note, however, that most of the mutants retained 50% or more of wild type infectivity (Table I and data  not shown), and randomization only slightly impacted the relative fitness of several mutant pools (Fig. 4). The Leu 149 , Gln 151 , Lys 154 , Gly 155 , Pro 157 , and Gln 161 pools exhibited rates of variant loss relative to wild type virus of Ͻ4% per replication cycle (assuming a wild type replication rate of 1-2 days Ϫ1 ) (42). These data suggest that motif B variants would also replicate in vivo, based on previous studies of patient-derived isolates (42,50). Perhaps most striking were the observations that several Val 148 substitutions appear neutral and that the S156A mutant replicates faster than wild type virus in culture (0.5-1% S156A outgrowth per replication cycle). There are other examples of HIV-1 mutants replicating better than wild type in culture, although the majority of these involve multiple substitutions in RT or protease (51)(52)(53)(54). We conclude that a number of RT motif B mutations are only slightly deleterious for HIV-1 replication and that some are nearly neutral or even adaptive.
Mutations throughout motif B conferred significant changes in the susceptibility of HIV-1 to the nucleoside analog inhibitors AZT and 3TC (Table I). Generally, these involved hypersensitivity to AZT or resistance to 3TC; mutants Q151A and Q151V exhibited both phenotypes. Our observation that Gln 151 mutations result in AZT hypersensitivity was unexpected, since a methionine substitution at this site confers weak AZT resistance in clinical isolates, and the combination of four additional mutations with Q151M results in high level resistance to AZT and several other nucleoside analogs (46,47) (Table I). Analyses of purified RTs show that the Q151M and Q151N mutations confer resistance to nucleoside analogs as well as mild to moderate increases in the fidelity of nucleotide incorporation in cell-free assays (11,55,56). Substitutions of alanine or asparagine at the equivalent position of murine leukemia virus RT (Gln 190 ) also confer increased fidelity but render the enzyme hypersensitive to dideoxynucleoside triphosphates (57). Thus, mutations at this position have similar effects on HIV-1 and murine leukemia virus RT fidelity but opposite effects on nucleoside analog susceptibility, suggesting that differences in local protein environment modulate the effects of Gln 151 substitutions. Our data also show that Gln 151 mutations can either increase or decrease drug sensitivity, depending on the nature of the amino acid substitution and the structure of the drug (Table I) and that some Gln 151 variants are hypersensitive to a number of nucleoside inhibitors. 2 Studies of purified mutant RTs are required to identify biochemical mechanisms that contribute to these viral phenotypes (45,58).
In patients receiving antiretroviral therapy, the development of nucleoside analog resistance is commonly associated with specific mutations in motifs A-D or the "fingers" region of RT (4,59). The discovery of a significant number of replicationcompetent mutants with altered drug sensitivity in our relatively small survey of RT variants (encompassing only 11 amino acid residues) suggests that the range of resistance mutations observed clinically represents a fraction of the total number of mutations with the potential to confer drug resistance. By extension, it is likely that other HIV-1 proteins exhibit similar mutational flexibility and that a large repertoire of viable phenotypic variants continuously arise in replicating HIV-1 populations in vivo.
The mutational flexibility of RT has important implications for the evolution of HIV-1 variants. Although variation at conserved RT positions is rarely observed in isolates from infected patients (see above), the persistence of variants during serial passage in culture (Figs. 4 and 5) suggests that motif B and other RT mutants are present at low frequencies in the complex mixtures of variants that comprise HIV-1 populations in vivo (44). Components of these mutant "swarms" or "quasispecies" can emerge as the dominant member of a population as a result of changing environmental demands (24,44,60). This is clearly evident in the development of resistance to antiviral therapy, where positive selection often results in the outgrowth of rare, drug-resistant variants that preexist in drug-naive patients (42,50,(61)(62)(63)(64). There is accumulating evidence that random genetic drift also contributes to HIV-1 RT variation (64), and our observation that S156A mutants replicate faster than wild type virus in culture (whereas the reverse must be true in vivo) demonstrates the importance of slightly deleterious mutations when negative selection barriers are lifted or changed (23). The fixation of new RT variants in HIV-1 populations by all of these mechanisms (positive selection, genetic drift, and relaxed negative selection) hinges on the ability of the virus to tolerate RT mutations without severe fitness loss. The plasticity of RT observed in our study suggests that substitutions at the majority of amino acid sites are only slightly deleterious to virus replication, underscoring the tremendous evolutionary potential of HIV-1 RT.
In summary, we show that a conserved region of HIV-1 RT tolerates a number of mutations that preserve virus replication capacity. Several of the variants selected in culture retain sufficient infectivity to support viral replication in vivo (42,50). A number of the selected mutants also exhibit significant hypersensitivity to AZT and/or resistance to 3TC, indicating that many sites in motif B contribute to nucleoside analog susceptibility. We conclude that, in addition to HIV-1 population size, diversity, and turnover (60), the plasticity of RT is an important contributor to escape from selective pressures in infected individuals.