Unusual Distribution of Mutations Associated with Serial Bottleneck Passages of Human Immunodeficiency Virus Type 1

Abstract

Repeated bottleneck passages result in fitness losses of RNA viruses. In the case of human immunodeficiency virus type 1 (HIV-1), decreases in fitness after a limited number of plaque-to-plaque transfers in MT-4 cells were very drastic. Here we report an analysis of entire genomic nucleotide sequences of four HIV-1 clones derived from the same HIV-1 isolate and their low-fitness progeny following 7 to 15 plaque-to-plaque passages. Clones accumulated 4 to 28 mutations per genome, with dominance of A → G and G → A transitions (57% of all mutations) and 49% nonsynonymous replacements. One clone—but not three sibling clones—showed an overabundance of G → A transitions, evidencing the highly stochastic nature of some types of mutational bias. The distribution of mutations along the genome was very unusual in that mutation frequencies in gag were threefold higher than in env. Particularly striking was the complete absence of replacements in the V3 loop of gp120, confirmed with partial nucleotide sequences of additional HIV-1 clones subjected to repeated bottleneck passages. The analyses revealed several amino acid replacements that have not been previously recorded among natural HIV-1 isolates and illustrate how evolution of an RNA virus genome, with regard to constant and variable regions, can be profoundly modified by alterations in population dynamics.

Retroviruses and in particular human immunodeficiency virus type 1 (HIV-1) mutate and recombine at high rates (14, 15, 39, 55, 60, 61, 70, 74). Rapid genetic variation, together with the short replication times of HIV-1 (8), generates complex and highly dynamic mutant swarms termed viral quasispecies (10, 18–22, 42, 66, 69). The mutant spectra of viral quasispecies constitute reservoirs of phenotypically relevant variants, as evidenced by the nonsyncytial-to-syncytial switch in infected individuals or the rapid selection of antibody-, cytotoxic-T-cell-, or inhibitor-resistant mutants in viral populations in vivo (2–4, 32, 33, 36, 37, 41, 46, 47, 56, 59, 65). Although most individual mutations in mutant swarms of RNA viruses may not be of immediate or even long-term selective value for the virus (63), evolution of viral quasispecies can be adaptive and may exert an influence in viral pathogenesis (26, 34, 76; reviews in 10, 28, 29). The adaptive potential of viral quasispecies is manifested by quantitatively important fitness variations as viruses evolve in constant or changing environments (for recent examples and reviews, see 10, 11, 13, 31, 48, 50–52, 75).

Large-population passages under a constant environment tend to produce fitness gains in viral populations (10, 11, 13, 25, 51, 52). In contrast, bottleneck events—experimentally realized in an extreme case by serial plaque-to-plaque transfers of virus on cell monolayers—often lead either to average fitness losses (6, 17, 24, 78) or to limitations of fitness gains (23, 51, 52). The decrease in fitness mediated by repeated bottlenecks has been interpreted as the result of an accentuation of Muller's ratchet effect (45). According to this model, asexual populations of organisms with a small population size will tend to incorporate deleterious mutations unless compensatory mechanisms such as recombination can restore mutation-free genomes (45). For RNA virus quasispecies, accumulation of deleterious mutations is expected from successive rounds of random sampling of genomes from the mutant spectrum (reviewed in 10, 11).

In retroviruses, decreases in fitness as a result of serial bottleneck passages were first documented with HIV-1 following plaque-to-plaque passages on MT-4 cells (78). In this virus, fitness losses were unexpectedly drastic when compared with the fitness losses experienced by other RNA viruses, such as bacteriophage φ6, vesicular stomatitis virus, or foot-and-mouth disease virus (FMDV) subjected to similar passage regimens (6, 17, 24). Only 4 out of 10 HIV-1 clones could produce viable progeny after 15 plaque-to-plaque transfers, and 3 of the 4 survivors displayed important decreases in fitness (78). Very little is known about the numbers and types of mutations which accompany fitness decreases of RNA viruses when they are subjected to sequential bottleneck passages. In the case of the animal picornavirus pathogen FMDV, debilitated clones showed some unusual genetic lesions (infrequent or absent in populations evolved without intervening bottlenecks; 24, 25). Such lesions included mutations that resulted in amino acid replacements at internal sites of the viral capsid and a unique elongation of five adenylate residues which resulted in an internal polyadenylate tract of variable length preceding the second functional AUG initiation codon of the FMDV genome (24, 25). No information on genetic lesions associated with fitness losses in retroviral genomes is available. Here we report complete HIV-1 genomic sequences of HIV-1 clones subjected to plaque transfers that led to severe fitness losses. The results reveal a broad spectrum of mutations associated with fitness decrease and an unexpected distribution of mutations along the HIV-1 genome.

MATERIALS AND METHODS

HIV-1 clones.

The origin and passage history of the biological clones of HIV-1 used in the present study have been previously described (78). Briefly, virus clones were isolated by plating a natural isolate of HIV-1, termed S61, on MT-4 cells. Virus populations D1, G1, I1, and K1 are from randomly chosen, individual plaques. After the first plating, viruses from individual plaques (in the range of 10² to 10⁵ PFU) were diluted in 300 μl of culture medium and plated on fresh MT-4 cells, and this process was repeated a number of times (Fig. 1). Plaques appeared 7 to 10 days after infection. Fitness of the clonal populations of HIV-1 was determined by growth-competition experiments in MT-4 cells, and populations were analyzed by the heteroduplex tracking assay as previously described (78).

FIG. 1.

Scheme of passages of HIV-1 clones subjected to plaque-to-plaque transfers in MT-4 cells. Clonal populations (HIV-1 isolated from individual plaques) are depicted as filled squares. The experimental procedures and the origins of natural HIV-1 isolate S61 and clones B1 to K1 are given in reference 78 and in Materials and Methods. HIV-1 clones are indicated by letters followed by a number which gives the total number of plaque-to-plaque transfers undergone by the clone. Infectious virus could not be rescued from viral populations B15, E13, G7, H13, and J15 as described in reference 78.

DNA extraction, PCR amplification, and nucleotide sequencing.

DNA was extracted using an Instagene purification matrix (Bio-Rad) according to the manufacturer's instructions. To determine the consensus nucleotide sequence of the entire HIV-1 genome in individual virus clones, a collection of overlapping sets of oligonucleotide primers was used. They either have been previously described (46, 47) or were designed for the present experiments (Table 1). HIV-1 DNA was amplified using nested PCR. The first amplifications (external primers) were carried out using the GeneAmp PCR kit (Perkin-Elmer) and resulted in the copying of 1,235-, 4,253-, and 6,686-bp fragments, comprising residues 1 through 1235, 546 through 4799, and 2975 through 9661, respectively; residue numbers correspond to those of the genome of HIV-1 isolate HXB2 (35). Internal amplifications yielded fragments of 500 to 1,500 residues, which were used for nucleotide sequence determination. For amplification of short regions of gag (positions 1337 to 1598) and env (positions 7071 to 7333), a single PCR amplification was carried out. Both external and internal amplifications involved 35 cycles with temperatures chosen according to the composition of the oligonucleotide primers (78). Before sequencing, the PCR mixture was digested with exonuclease I and shrimp alkaline phosphatase (Amersham Life Sciences). Nucleotide sequences were determined on the two cDNA strands, with an ABI 373 automatic sequencer. Multiple sequence alignments were obtained using the CLUSTAL W program (71).

TABLE 1.

Synthetic oligonucleotide primers employed for nucleotide sequence determination of the HIV-1 genome^a

Nucleotide sequence
GCTTCAAGTAGTGTGTGCCCGTCTG (563, S)
AGAGTCACACAACAGACGGG (582, A)
TGACTAAAAGGGTCTGAGGG (614, A)
CTCTGGTAACTAGAGATCCC (615, S)
TCTCTAGCAGTGGCGCCCGAACAGGGAC (626, S)
AGACAGGATCAGAAGAA (995, S)
GGTGATATGGCCTGATGTACCATTTGCCCCTG (1204, A)
CCAGGCCAGATGAGAGAACCAAGGG (1462, S)
TGTCCAGAATGCTGGTAGGG (1642, A)
CTCTCAGAAGCAGGAGCCG (2205, S)
GTATTTAGTAGGACCTACACCT (2475, S)
TCTTCTGTCAATGGCCATTGTTTAAC (2635, A)
GGATTAGATATCAGTACAATGTGCTT (2971, S)
CATGGATCCGATATCTAATCCCTGG (2991, A)
GCTGGTGACCTTTCCATCC (3022, A)
TAGATATCAGTACAATGTGCTTCCAC (2975, S)
TATTGCTGGTGATCCTTTCC (3026, A)
AAACATCAGAAAGAACCTCC (3207, S)
GTTCATAACCCATCCAAAGG (3249, A)
GCGGAATCTGTATGTCATTGACAGTCCAGCT (3300, A)
CATGGAGTGTATTATGACCC (3492, S)
CTTTCCCCATATTACTATGC (3704, A)
AGTTTGTCAATACCCCTCCC (3793, S)
AATCATTCAAGCACAACCAG (4061, S)
TTAGATGGAATAGATAAGGC (4233, S)
CTTGAAGCTTATCTATTCCATCTAAAAATAGT (4257, A)
GGCGAATTCACTAGCCATTGCTCTCCA (4284, A)
AGTGATTTTAACC (4299, S)
GCTTCTATATATCCACTGGC (4486, A)
AAGTATGCTGTTTCTTGCCC (4528, A)
AAGGGGGGATTGGGGGGTACAGTGCAGGG (4792, S)
TAGCCCTTCCAGTCCCCCCTTTTCTTTTA (4799, A)
CTTTCCCCTGCACTGTACCC (4825, A)
TACTAATCTAGCCTCCCCTAGTGGGATGTG (5235, A)
GCCTCTGTGGCCCTTGGTCTTCTGGGG (5595, A)
CAGAAAAGCTTGTCGACATAGCAGAATAGG (5576, S)
TTAGGCATCTCCTATGGCAGGAAGAAGCGG (5957, S)
CCCATAATAGACTGTGACCC (6347, A)
TGTGGGTTGGGGTCTGTGGG (6469, A)
ATGGGATCAAAGCCTAAAGCCATGTG (6557, S)
AGGATACCTTTGGACAGGCC (6852, A)
TCAGCACAGTACAATGTACACATGG (6949, S)
ATAAGCTTGCAGTCTAGCAGAAGAAGA (7004, S)
CAATCCTCAGGAGGGGAC (7311, S)
TGCATCTCAATTTCTGGGCTCCCCTCCTGAG (7345, A)
AGGAGTCCTCCCCTGGGTCTTAAGTA (7648, A)
AGTGCTTCCTGCTGCTCCCAAGAACCCAAG (7811, A)
TCTTGCCTGGAGCTGTTTGATGCCCCAGAC (7961, A)
TTGGAATTGGATAAGTGGGC (8205, S)
GTGAATAGAGTTAGGCAGGG (8337, S)
GAAATGACAATGGTGAGTATCCCTGCC (8376, A)
CGATTCCTTCGGGCCTGTCGGGTCCCC (8424, A)
TGTGGAACTTCTGGGACGCAGGGGGTGGG (8567, S)
CAAGGAGGAGGAGGAGGTGGGTTTTCC (8977, S)
TGGAAGGGCTAATTTGGTCCCAGA (9086, S)
CTGGGACCAAATTAGCCCTTCCAGTCC (9108, A)

^aSequences are written from 5′ to 3′; data in parentheses are the position of the 5′ nucleotide of each primer—according to the genomic nucleotide numbering of isolate HXB2 (35)—and primer orientation (S, sense [same orientation as genomic RNA]; A, antisense [complementary to genomic RNA]). Oligonucleotide primers were purchased from Isogen (Maarssen). 

The newly determined nucleotide sequences have been deposited in the EMBL sequence database with accession numbers AF256204, AF256205, AF256206, AF256207, AF256208, AF256209, AF256210, and AF256211.

RESULTS

Mutation accumulation as a result of bottleneck transfers.

HIV-1 clones underwent severe fitness losses as a result of serial plaque-to-plaque transfers in MT-4 cells (78). To determine the types and numbers of mutations accumulated during bottleneck transfers, the entire genomic nucleotide sequence of four HIV-1 clones (D1, G1, I1, and K1) and their derivatives after 15, 7, 15, and 15 plaque-to-plaque passages, respectively (termed D15, G7, I15, and K15, respectively [Fig. 1]), were obtained. The comparison of genomic nucleotide sequences of each passaged clone relative to the corresponding initial clone showed that transitions were 2.8-fold more frequent than transversions and that A → G and G → A accounted for 20% and 36%, respectively, of all mutation types; nonsynonymous mutations represented 49% of the total; an insertion of one nucleotide was present in clone I15 (Table 2). Mutation frequencies varied up to sevenfold among lineages (range, 4.4 × 10⁻⁴ to 3.1 × 10⁻³ substitutions per nucleotide [Table 2]). There was no obvious correlation between fitness decrease and mutation types or frequencies; for example, clone G7, which was debilitated to the point of not allowing a reliable determination of fitness value (78), showed a mutation frequency that was sevenfold lower than that of I15 (Table 2).

TABLE 2.

Number and types of mutations in the genome of HIV-1 clones subjected to serial plaque transfers

Clones compareda	Fitness decrease (%)b	Mutation frequencyc	No. of mutationsd
			A → C	A → G	C → A	C → G	C → T	G → A	G → T	T → A	T → C	T → G	Other	Total
D1, D15	85	6.6 × 10−4	0	2	1	0	0	3	0	0	0	0	0	6
G1, G7	ND	4.4 × 10−4	0	2	0	0	0	0	2	0	0	0	0	4
I1, I15	99	3.1 × 10−3	1	2	0	0	1	12	1	1	5	4	1	28
K1, K15	63	1.2 × 10−3	0	4	1	1	0	3	0	0	2	0	0	11
Total			1	10	2	1	1	18	3	1	7	4	1	49

^aThe entire genomic nucleotide sequence of the indicated clones was determined as described in Materials and Methods. 

^bFitness decrease refers to the comparison of D15, I15, and K15 with their corresponding initial clones D1, I1, and K1; it is expressed as percent reduction, calculated as described in reference 78. For G7 a relative fitness value could not be calculated because the virus yield was insufficient to perform competition passages (78). 

^cMutation frequency is the number of mutations found in the genome of D15, G7, I15, or K15 (when compared with the corresponding initial clones D1, G1, I1, or K1) divided by the total number of nucleotides sequenced (in this case, the entire HIV-1 genome, or 9.1 kb); therefore, values are expressed in substitutions per nucleotide. 

^dTransversions A → T and G → C were not found in the genomes of the clones analyzed. Other, nsertion of one nucleotide at genomic residue 584 (according to the numbering for the genome of HXB2 [35]) in the leader sequence. The location of mutations in the HIV-1 genome is given in Fig. 2 and is discussed in the text. 

Unusual distribution of mutations.

Mutation frequencies in env are generally higher than in gag and pol when natural HIV-1 isolates are compared (35, 44, 57, 58). In contrast, the HIV-1 clones debilitated by serial plaque transfers showed average mutation frequencies that were threefold higher for gag than env (Fig. 2). No mutations were found in tat, vpu, and rev. Most remarkable was the absence of mutations in the regions encoding variable loops of gp120, particularly the V3 loop. The asymmetric distribution of mutations was visualized by dividing the HIV-1 genome into three arbitrary regions of similar length: region 1, residues 1 through 3028; region 2, residues 3029 through 6065; and region 3, residues 6066 through 9035. Taking into account all mutations scored for clones D15, G7, I15, and K15 (as quantitated in Table 2), we found mutation frequencies for regions 1, 2, and 3 of 2.5 × 10⁻³, 9.0 × 10⁻⁴, and 6.6 × 10⁻⁴ substitutions per nucleotide, respectively. The results suggest that plaque-to-plaque transfers of HIV-1 lead to accumulation of mutations at multiple sites of the HIV-1 genome, following a pattern which is quite different from that observed in the natural evolution of HIV-1 in infected hosts.

FIG. 2.

Location of mutations found in HIV-1 clones D15, G7, I15, and K15, relative to their parental counterparts. The upper part indicates HIV-1 genes and regulatory regions based on the compilation of Korber et al. (35). The four horizontal bars in the center of the figure indicate the positions of mutations along the genome (9.1 kb) in the four clones analyzed (from top to bottom, D15, G7, I15, and K15 [described in Materials and Methods]); vertical lines within these bars indicate one, two, or three mutations, according to thickness. Mutations were found at positions 35, 171, 377, 379, 570, 584, 760, 807, 833, 988, 1128, 1161, 1188, 1351, 1467, 1545, 1578, 1596, 1810, 1863, 1875, 1937, 1961, 1966, 2145, 2174, 2329, 2668, 2804, 3068, 3114, 3129, 3945, 4458, 5239, 5270, 5342, 5422, 5684, 5686, 6588, 6655, 6670, 7962, 8095, 8890, 8900, and 8989 according to the numbering of HIV-1 isolate HXB2 (35). The gp120-coding region of env has been enlarged to depict the positions of variable loops V1 to V5; two mutations affected the V1-coding region. The two shaded rectangles correspond to genome positions 1337 to 1598 (gag [left shaded rectangle]) and 7071 to 7333 (env [right shaded rectangle]), which have been sequenced for a number of additional HIV-1 clones to confirm the asymmetric distribution of mutations (see text). The mutations found in these additional clones are not included in this scheme. The bottom part shows the three arbitrary regions into which the HIV-1 genome was divided to illustrate the bias in the distribution of mutations along the genome. Procedures used for nucleotide sequence determination are described in Materials and Methods, and the oligonucleotide primers are listed in Table 1.

Conservation of the V3 loop.

The unusual distribution of mutations found in D15, G7, I15, and K15 and the absence of mutations in the region encoding the variable V3 loop of gp120 prompted us to extend nucleotide sequence determinations to additional HIV-1 clones subjected to serial bottleneck events. These additional analyses included two clones from each of the lineages D15, G7, and K15, one clone from I15, and clones from independent lineages which were previously described (78), including three clones from F15 and one clone from H13 (Fig. 1). The analysis involved nucleotide sequences of residues 7071 to 7333 (which include the V3-coding region) and residues 1337 to 1598 (within the p24-coding region). The comparison of nucleotide sequences with those of the corresponding initial clones fully confirmed the bias in the distribution of mutations; in all, 13 replacements were found in the gag region analyzed (which represents a mutation frequency of 4.5 × 10⁻³ substitutions per nucleotide), while no replacements were found in the V3-coding region (mutation frequency, <3.5 × 10⁻⁴ substitutions per nucleotide). The statistical significance of the biased distribution of mutations was evaluated by comparing the expected versus the actual number of mutations in gag, pol, and env for regions 1, 2, and 3 into which the genome was arbitrarily divided (Fig. 2), as well as for the short gag and env stretches for the additional clones from populations D15, F15, G7, H13, I15, and K15. These results (Table 3) indicate high statistical significance of the biased mutant distribution. The degree of statistical significance of the biased distribution did not vary when either all G → A mutations or just G → A mutations found in clone I15 were excluded from the calculations (in all cases, P < 0.001 [χ² test]). Therefore, accumulation of mutations that were associated with serial bottleneck events and with fitness loss of HIV-1 affected genomic regions that are less variable during the natural evolution of the virus.

TABLE 3.

Expected versus actual number of mutations in different genomic regions of HIV-1 clones subjected to serial plaque transfers

Genomic regions compared	No. of mutations (Pf)
	Expected	Found
gag, pol, enva	8 gag, 15.7 pol, 13.2 envb	19 gag, 10 pol, 5 env (<0.001)
R1, R2, R3ac	15.8 R1, 15.8 R2, 15.8 R3b	30 R1, 11 R2, 8 R3 (<0.001)
gag (1337–1598), env (7071–7333)d	6.8 gag, 6.8 enve	13 gag, 0 env (<0.001)

^aComparisons involve mutations found in clones D15, G7, I15, and K15 relative to their corresponding parental clones D1, G1, I1, and K1 (Table 2). 

^bThe expected numbers of mutations are based on the mutation frequencies given in Table 2, assuming a random distribution of mutations along the genome. 

^cR1, R2, and R3 indicate genomic regions 1, 2, and 3, as depicted in Fig. 2. 

^dComparisons involve mutations at genomic residues 1337 through 1598 and 7071 through 7333 found in three clones from lineage F15, two clones each from lineages D15, G7, and K15, and one clone each from lineages H13 and I15, as described in the text and in Fig. 1. 

^eThe expected numbers of mutations are based on the average mutation frequencies for genomic residues 1337 through 1598 (gag) and 7071 through 7333 (env) assuming a random distribution of mutations along the two genomic stretches. 

^fP values were calculated by the χ² test. 

DISCUSSION

The results reported here describe the numbers and types of mutations associated with fitness loss of HIV-1 as a result of the operation of Muller's ratchet (Table 2; Fig. 2). For clones D15, G7, and K15, the average mutation frequency (1.3 × 10⁻⁴ substitutions per nucleotide, measured relative to the genomic nucleotide sequence of their respective parental clones D1, G1, and K1) was 12 mutations per genome (range, 10 to 15), with a predominance of transitions (75% of all mutations) over transversions. These figures are similar to those of previous determinations of the number of mutations accompanying Muller's ratchet in the FMDV genome—an average of six mutations per genome, with 77% of these transition mutations (24). However, clone I15 displayed a different pattern in that its mutation frequency corresponded to 28 mutations per genome with an overabundance of G → A transitions (43% of all mutations [Table 1]). G → A is one of the substitutions that have been associated with hypermutagenesis in HIV-1 (72, 73) and a number of other viruses (reviewed in 5, 43). In clone I15, G → A transitions were distributed rather uniformly along the genome. Several possible mechanisms have been proposed to explain biased hypermutagenesis (5, 43), including alterations in intracellular deoxynucleotide pools. The sequence context of 67% of the G → A transitions in clone I15 was GpA or GpG, which suggests a possible influence of low dCTP levels during minus cDNA synthesis in the origin of this mutation type (43). Our results with debilitated HIV-1 clones emphasize the stochastic nature of the G → A mutation bias because it was observed in only one of four clones derived from the same viral isolate, and these clones were subjected to identical treatment during the serial plaque transfers (Fig. 1). If the mechanism of nucleotide pool bias was in operation (43), it must have been triggered either by extremely subtle perturbations in the intracellular environment or by differences among the genomes of the four clones, differences that existed initially or that were generated in the course of passaging. In the latter case, the mutational biases must be subjected to indeterminations derived from the dynamics of mutant generation within the quasispecies swarms (reviewed in 10, 20).

A dominance of G → A transitions was also found in an analysis of mutations in a lacZα-based reporter gene, which was constructed to study a single cycle of HIV-1 replication (39). However, there are important differences in the mutant repertoire found following a single cycle of replication and in our clones subjected to serial plaque transfers. In the former study, G → A, C → T, and T → C mutations occurred at frequencies of 1.7 × 10⁻³, 7.1 × 10⁻⁴, and 1.3 × 10⁻⁴ substitutions per nucleotide, respectively, while in our study these same mutations occurred at frequencies of 4.9 × 10⁻⁴, 2.7 × 10⁻⁴, and 1.9 × 10⁻⁴, respectively (Table 2). After a single-cycle replication, T → G, T → A, and A → G each occurred at a frequency of 6.5 × 10⁻⁵ substitutions per nucleotide, while in our clones the values for these substitutions ranged from 2.7 × 10⁻⁴ to 2.7 × 10⁻⁵ substitutions per nucleotide. In addition, A → C, C → A, C → G, and G → T mutations found in HIV-1 clones (Table 2) were not represented among the mutations found after a single infectious cycle (39). These variations in mutation types and frequencies probably arose not only from differences in the number of replication cycles but also from the sequence context in the template being copied in a different biological environment (43).

The most unexpected finding in the analysis of low-fitness HIV-1 clones was the distribution of mutations along the genome (Fig. 2), with a statistically significant accumulation of mutations in gag and the first third of the genome, relative to env, which appears as the most conserved genomic region in all clones examined (Fig. 2; Table 3). This is in sharp contrast to results with natural HIV-1 isolates (35, 44, 57, 58) and with large-population passages of HIV-1 clones—derived also from natural isolate S61—in cell culture (64); in all of these analyses, gag and pol showed more nucleotide and amino acid sequence conservation than env. Several mechanisms could contribute to this striking difference in the distribution of mutations. In the course of plaque-to-plaque transfers, fitness gains or purifying selection is probably diminished since competition among genomes from the quasispecies is limited to the period of plaque development (10, 25). In this view, the higher variation normally seen in env relative to gag and pol would be essentially due to selection for immune evasion, for adaptation to alternative cellular receptors, increased particle stability, etc. Evidence for selection in vivo has been obtained for HIV-1 and for simian immunodeficiency virus (2, 3, 26, 34, 48, 54, 62, 77). However, even if replicative optimization of the mutant spectrum was limited as a result of bottlenecking, the deficit in mutations in env remains to be explained. There are at least two possibilities (which are not mutually exclusive): env may have a lower tolerance for mutations than gag and the 5′ third of the genome under the cell culture environment in which the passages were carried out, or HIV-1 genome replication may be more error prone in the process of copying gag than when copying env. Constraints to accept replacements in surface proteins were documented through functional and structural studies with FMDV (12 and references therein). The mutant repertoire in viral quasispecies could be strongly influenced by tolerance to nucleotide and amino acid substitutions, including silent replacements in coding regions (12, 38, 40, 67). In HIV-1, constraints in surface proteins could come about from the need always to use the cellular receptors presented by MT-4 cells in culture and to enter this same cell type monotonously in an invariant cell culture environment. Purifying selection during plaque growth, which must include many cycles of replication, may have contributed to the observed bias in the distribution of mutations along the genome.

A possible molecular basis for a difference in accuracy during copying of different genome segments of HIV-1 is not obvious. The lowest number of mutations was seen in the genomic region copied immediately after the first strand transfer in the synthesis of minus-strand DNA by reverse transcriptase (RT) (reviewed in reference 68). Since processivity of RT is limited, it could be proposed that as synthesis proceeds inside the nucleocapsid, accuracy may decrease as a result of environmental alterations (ionic composition, deoxynucleoside triphosphate pools, etc. [43]). It has been suggested that misalignment mutagenesis could be more frequent during dissociation and reinitiation of RT-catalyzed reactions (reviewed in 1). However, examination of the sequence context of each mutation suggests that the frequency of mutations which may have occurred as a result of template misalignment (1, 53) is not significantly different for regions 1, 2, and 3, into which we have divided the HIV-1 genome (52, 64, and 63%, respectively). Other effects on fidelity or, more likely, a combination of factors outlined in previous paragraphs may converge to produce the unusual distribution of mutations seen in HIV-1 clones as a result of operation of Muller's ratchet.

A total of 8 out of 20 nonsynonymous replacements found among the HIV-1 clones analyzed have not been recorded in current sequence data banks (35; Table 4). An attractive possibility is that replacements in gag could have multiple effects in RNA-protein interactions, nucleocapsid assembly, and protein stability (16, 30) and that such effects could contribute to fitness loss. It must also be considered that Vif, Vpr, and Nef are dispensable functions for HIV-1 replication in some permissive cell lines, including MT-4 (7, 49). Therefore, the accumulation of nonsynonymous replacements in vif, vpr, and nef genes (Fig. 2) could be of little consequence for plaque formation in MT-4 cells. However, an evaluation of the influence of amino acid replacements (individually or in combination) in viral fitness would require analysis of the effects of candidate mutations when introduced into infectious clones or examination of possible reversions and fixation of compensatory mutations upon fitness recovery of the debilitated HIV-1 clones. These studies are now in progress.

TABLE 4.

Amino acid replacements associated with nonsynonymous mutations during plaque-to-plaque transfers of HIV-1

Amino acid replacementa	Acceptabilityb	Frequency in databasec	HIV-1 protein and possible structural and functional significanced
R15K	5	High	p17 matrix
S67A	5	High	p17 matrix
S209A	5	High	p 24 capsid C-terminal dimerization domain
K14R	5	Not found	p7 nucleocapsid flanking first zinc finger motif
F16L	4	Not found	p7 nucleocapsid in first zinc finger motif
I14S	2	High	p6
E40G	4	Not found	RT in an α-helix in fingers subdomain
V189Ie	5	Not found	RT in palm subdomain near active site
E194K	4	Not found	RT in palm subdomain near active site
V466I	5	High	RT in a β-sheet of RNase H
V77I	5	Not found	Integrase, in catalytic D, D-35-E domain (β2)
P67T	4	High	Vif
K77R	5	High	Vif
T101N	4	Low	Vif
V128I	5	High	Vif
G43E	4	Low	Vpr
S144I	2	Not found	gp160, in V1 loop in gp120
E148G	4	Not found	gp160, in V1 loop in gp120
V580I	5	High	gp160, in gp41 ectodomain
D624G	4	Low	gp160, in gp41 ectodomain
A32T	5	Not found	Nef
R35Q	3	High	Nef
E65Stopf			Nef
E154K	5	Low	Nef

^aAmino acids are numbered for each individual protein, according to the numbering for isolate HXB2 found in the data banks (35). 

^bThe degree of acceptability of the amino acid substitution is given according to reference 27; the acceptability scale is from 0 to 6, with the latter value representing replacement by the same amino acid. 

^cThe database consulted is the one presented in reference 35, and it was again retrieved on April 18, 2000, by entering GenBank and EMBL databases. 

^dLocation of amino acid replacements and their possible significance is based on current databases (35) and overviews on HIV-1 structure-function relationships (references 7, 9, 28, and 30 and references therein). 

^eThis replacement has been correlated with resistance to nonnucleoside RT inhibitors (references 11, 28, 33, 35, and 46 and references therein). 

^fThis mutation leads to an amber termination codon in Nef. 

In conclusion, some mutations associated with the operation of Muller's ratchet in HIV-1 have not been previously reported among natural isolates evolved over more than two decades (35; Table 4). Furthermore, the mutations were distributed along the viral genome, unlike mutations in natural HIV-1 isolates, and env was the most conserved genomic region. An interesting possibility is that, in vivo, HIV-1 is not subjected to as severe bottleneck events as in experiments designed to accentuate Muller's ratchet. The observations reported here add to the complexities inherent in the relationships between occurrence of mutations and what can be eventually observed upon examination of genomic nucleotide sequences. The HIV-1 mutational pattern could be made to vary with respect to hundreds of sequences recorded in data banks simply by changing the passage regimen, without intervening, externally applied, selective forces.