Fitness Landscape of Human Immunodeficiency Virus Type 1 Protease Quasispecies

Abstract

Here we show, at a high resolution (1%), the human immunodeficiency virus type 1 (HIV-1) protease gene quasispecies landscape from three infected naïve individuals. A huge range of genetic configurations was found (67%, 71%, and 80% of the nucleotide clones from the three individuals, respectively, were different), and these configurations created a dense net that linked different parts of the viral population. Similarly, a vast diversity of different protease activities was also found. Importantly, 65% of the analyzed enzymes had detectable protease activity, and 11% of the minority individual variants showed similar or better fitness than the master (most abundant) enzyme, suggesting that the viral complexity in this genomic region does not exclusively depend on the enzyme's catalytic efficiency. Several high-fitness minority variants had only one substitution compared to the master sequence, supporting the possibility that the rugged HIV-1 protease quasispecies fitness landscape may be formed by a continuous network that can be traversed by single mutational steps without passing through defective or less-adapted proteins.

The human immunodeficiency virus type 1 (HIV-1) protease is an aspartic protease consisting of two identical 99-amino-acid monomers. The viral protease is the enzyme required for processing Gag and Gag-Pol polyproteins to yield mature infectious virions. Several protease inhibitors (PIs) have been developed to block this step of the viral life cycle and have been proven to decrease the plasma viral load of infected individuals. Numerous studies have described HIV-1 protease variability and polymorphisms found in naïve or PI-treated infected individuals (6, 10, 24, 30, 32, 48, 57, 63). Those reports have also shown that the HIV-1 protease is an enzyme that can accept a great number of amino acid changes without losing its enzyme activity (33). The observed HIV-1 protease heterogeneity originates from the high rate of incorrect nucleotide substitutions during HIV reverse transcription (10⁻⁴ to 10⁻⁵ mutations per nucleotide and per replication cycle) (35), rapid viral turnover (10⁹ to 10¹⁰ virions per day) (23, 60), large numbers of infected cells (10⁷ to 10⁸ infected cells) (9), and a high level of recombination (7, 27). The high mutation rate associated with HIV-1 replication, like other RNA viruses, results in the generation of swarms of mutants known as viral quasispecies (15, 16, 18). Since the behavior of any particular variant may be influenced by the entire viral population (17), it has been suggested that the quasispecies, and not individual viral genomes, are the target of selection and random drift (15). Therefore, the study of HIV-1 quasispecies can be relevant in order to understand viral evolution in the presence of selective pressures exerted by the host immune system and antiretroviral therapy. Recent work with poliovirus has shown that viruses carrying a high-fidelity polymerase replicate at wild-type (WT) levels but generate less genomic diversity and are unable to adapt to adverse growth conditions. In infected animals, this reduced viral diversity leads to an attenuated pathogenic phenotype (59). The genetic structure of HIV-1 quasispecies has been well characterized for different viral genomic regions. The highest degree of diversity within an infected individual (8%) has been observed in the variable regions of the viral envelope (54, 62). A lower level of intraindividual diversity, 1%, has been reported for the region encoding the protease (7, 12, 25, 32). To date, only one study, in which HIV-1 Tat quasispecies were described, has linked variant genotypes with their phenotypes (39). Thus, there is a lack of information regarding the phenotypic range represented in a single gene of an RNA virus or complete genomic quasispecies. For example, are quasispecies phenotype landscapes rugged or smooth?

Here, we analyzed the genotype and enzymatic activities of three HIV-1 protease quasispecies at high resolution. We determined the fitness of each variant present in the quasispecies in order to establish the relationships between genotype, phenotype, and fitness and constructed a phylogenetic-fitness landscape map for each quasispecies.

MATERIALS AND METHODS

Individuals.

Three HIV-1-infected individuals, M, N, and O, with no previous PI therapy were chosen for this study. Individual samples M and N had similar viral loads and CD4⁺ and CD8⁺ T-cell counts, while sample O was from an individual in an advanced stage of the disease, with CD4⁺ T-cell counts below 200 cells/μl and high viral loads (Table 1).

TABLE 1.

Clinical and virological characteristics of the three studied HIV-1 infected individuals

Parametera	Value for individual
	M	N	O
Viral load (RNA copies/ml)	17,094	15,662	297,000
Proviral load (DNA copies/106 PBMCs ± SD)	82 ± 45	223 ± 115	481 ± 247
CD4+ (cells/μl)	471	607	10
CD8+ (cells/μl)	897	975	550
Hn (%)	67	71	80
Sn	0.8	0.8	0.9
Ha (%)	20	42	42
Sa	0.4	0.6	0.5
Dg (% ± SD)	1.3 ± 0.3	1.8 ± 0.3	1.6 ± 0.3
Da (% ± SD)	1.4 ± 0.6	2.8 ± 0.8	1.2 ± 0.2
dS (proportion ± SD)	0.03 ± 0.022	0.03 ± 0.018	0.05 ± 0.028
dN (proportion ± SD)	0.01 ± 0.005	0.01 ± 0.009	0.01 ± 0.005
Ratio (dS/dN)	5.9	2.93	8.32

^aHn, nucleotide heterogeneity; Sa, amino acid-normalized Shannon entropy; Dg, genetic distance; Da, amino acid distance.

Recovery and analysis of DNA sequences.

Genomic peripheral blood mononuclear cell (PBMC) DNA extraction and purification were performed as described previously (26). Next, proviral load quantification was determined by endpoint limiting dilution and expressed as DNA copies/10⁶ PBMC ± standard deviation (Table 1) as described previously (19, 26). The HIV-1 protease gene was amplified by PCR from proviral PBMC DNA. PCR amplification and phage lambda cloning were performed as previously described (5, 36). Briefly, for the first PCR, protease oligonucleotides 5′prot1 (5′-AGGCTAATTTTTTAGGGAAGATCTGGCCTTCC-3′[HXB2 residues 2078 to 2108]) and 3′prot1 (5′-GCAAATACTGGAGTATTGTATGGATTTTCAGG-3′[HXB2 residues 2703 to 2734]) were used. At least 10 HIV-1 proviral DNA copies were used as a starting template in the first PCR amplification. A second amplification (nested PCR) was carried out with oligonucleotides HIVproL (5′-GGGGAATTCTAAGGCCAGGGAATTTTCTT-3′[HXB2 residues 2118 to 2136; underlining indicates an EcoRI restriction site]) and Xho8R (5′-GGGAGGGGCTCGAGTCAAAGGCCATCCATTCCT-3′[HXB2 residues 2591 to 2604; underlining indicates an XhoI restriction site, and a stop codon is shown in boldface type]). For both amplifications, the PCR mixture contained 10 pmol of each oligonucleotide, 200 μM deoxyribonucleoside triphosphates, 2 mM MgSO₄, 1× high-fidelity PCR buffer (Invitrogen), and 0.5 U Platinum Taq DNA polymerase (Invitrogen) in a total reaction volume of 25 μl. Cycling parameters were one cycle of denaturation at 95°C for 2 min and then 35 cycles of denaturation at 95°C for 30 s, annealing at 55°C for 30 s, and extension at 68°C for 1 min. This was followed by a 7-min incubation at 68°C. A 5-μl aliquot was again amplified in a 50-μl reaction mixture. The PCR products were digested with EcoRI and XhoI, ligated into a Uni-ZAP XR Vector kit (Stratagene), and packaged with a Uni-ZAP XR Gigapack cloning kit (Stratagene). To ensure that multiple proviral HIV-1 protease templates were present in each analyzed quasispecies, for each sample, 40 different PCR amplifications were performed and pooled before cloning. For each sample, a minimum of 100 individual lambda clones were obtained and analyzed. Twenty-five picograms of an HIV-1 HXB2 plasmid control was amplified and cloned as described above, and 14 subclones were sequenced to estimate diversity due to PCR Taq errors. The different proteases were sequenced with the flanking oligonucleotides T3 (5′-AATTAACCCTCACTAAAGGG-3′) and T7 (5′-TCGAGGTCGACGGTATC-3′) using the ABI PRISM dRhodamine terminator cycle sequencing kit (Applied Biosystems). Sequence alignment and editing were performed with Sequencer version 4.1 (GeneCodes) software. For the phylogenetic analysis, the PAUP* 4.0 (58) software package was used with a GTR + I + γ model of evolution, where GTR means general time reversible and I indicates the proportion of invariable sites. Bootstrap resampling (1,000 replicates) was applied to the neighbor-joining tree to assign approximate confidence limits to individual branches. The final graphical output was created with the TREEVIEW software program (44). Nucleotide heterogeneity data were obtained from the distance matrix generated with the PAUP*4.0 software package used in the phylogenetic analysis. The amino acid distances, with the Poisson correction, were calculated with the MEGA 2 software package (31). The nuclear-normalized Shannon entropy (S) (Sn) was calculated as follows: Sn = −Σ_i (p_iln p_i)/ln N, where N is the total number of sequences analyzed and p_i is the frequency of each sequence in the viral quasispecies. S varies from 0 (no complexity) to 1 (maximum complexity) (62). To determine possible selective pressures, the proportion of synonymous substitutions per potential synonymous sites (dS) and the proportion of nonsynonymous substitutions per potential nonsynonymous sites (dN) were calculated with the SNAP software program (http://www.hiv.lanl.gov/content/hiv-db/SNAP) using the Nei-Gojobori model of evolution (40), incorporating a statistical model developed previously by Ota and Nei (43). To estimate codon-specific selection pressures (ratio of nonsynonymous substitutions [dN] to synonymous substitutions [dS] significantly greater than 1), we used a maximum likelihood method implemented in the CODEML software program from the PAML v 3.14 software package (64). To assess evidence for positive selection, neutral models (M0, M1, and M7) were compared to selection models (M2, M3, and M8). Models of codon evolution were compared using a likelihood ratio test: M0 versus M3, M1 versus M2, and M7 versus M8. Thus, a single codon subjected to positive selection can be identified by a Bayesian method implemented in the same software (Table 2).

TABLE 2.

Positive selection in the HIV-1 protease coding region^a

Subject and models	−2(lnλ)	dfb	P value	Positions
M				37H and 71A
M0 vs M3 (k = 3)	131,9288	4	0.0005
M1 vs M2	45,9322	2	0.0005
M7 vs M8	49,0892	2	0.0005
N				41R and 65D
M0 vs M3 (k = 3)	247,972	4	0.0005
M1 vs M2	79,047	2	0.0005
M7 vs M8	78,8692	2	0.0005
O
M0 vs M3 (k = 3)	19,1164	4	0.0005
M1 vs M2	0,1538	2	0.0005
M7 vs M8	4,4492	2	0.0005

^aTo estimate codon-specific selection pressures of a dN/dS ratio significantly greater than 1, we used a maximum likelihood method implemented in the CODEML software program from the PAML v 3.14 software package (64).

^bdf, degrees of freedom for each model comparison. Only the positive selected codons (P > 0.95) under model M8 are listed.

Determination of protease enzymatic activities.

The catalytic efficiencies of the different HIV-1 proteases were calculated using a bacteriophage lambda-based genetic screening, as previously described (5, 20, 36, 37, 45, 46, 55). Briefly, Escherichia coli JM109 cells containing plasmid p2X-cI-HIV were transformed with plasmid pcI-HIV-cro. The resulting cells were grown in the presence of 0.2% maltose, harvested by centrifugation, and suspended to 2.0 optical density at 600 nm (OD₆₀₀) units/ml in 10 mM MgSO₄. Cells (200 μl) were infected with 5 × 10⁷ PFU of phages containing the different HIV-1 proteases. After 15 min at 37°C, the cells were washed with 1 ml of 10 mM MgSO₄, harvested by centrifugation, and suspended in 1 ml of LB medium containing 12.5 μg of tetracycline, 0.2% maltose, 10 mM MgSO₄, and 0.1 mM IPTG (isopropyl-β-d-thiogalactopyranoside). The cell cultures were incubated at 37°C for 3 h and harvested by centrifugation. An additional cycle of selective growth was gained by suspending the infected cells with a fresh aliquot (200 μl) of JM109 pcI-HIV-cro cells. After two selective growth cycles, the titer of the resulting phage was determined by coplating the cultures with 200 μl of E. coli XL-1 Blue cells (OD₆₀₀ of 2.0/ml in 10 mM MgSO₄) on LB plates using 3 ml top agar containing 12.5 μg of tetracycline per ml, 0.2% maltose, and 0.1 mM IPTG. After incubation at 37°C for 6 h, the plaques were counted to score lytic growth.

To ensure that the HIV-1 proteolytic cleavage sites 1 (MA/CA; amino acids 129 to 136 of the HIV-1 HXB2 strain Gag-Pol polyprotein) of the three viral quasispecies analyzed in this study were analogous to the one inserted in the cI protein, four individual clones containing the MA/CA coding region were sequenced for each sample. PCR amplification was performed with Platinum Taq high-fidelity DNA polymerase (Invitrogen), as described above. The oligonucleotides used were MaCaFout (5′-TAGCAACCCTCTATTGTGT-3′ [HXB2 residues 1034 to 1052]) and MaCaRout (5′-AATGCTGAAAACATGGGTAT-3′ [HXB2 residues 1294 to 1313]). PCR products were cloned using the p-GEM-T Easy vector system (Promega) and sequenced with oligonucleotides T7 and SP6 (5′-ATTTAGGTGACACTATAGAA-3′). The amino acid sequences of all clones sequenced were identical to the HIV-1 MA/CA sequence introduced in the cI lambda repressor.

Phylogeny fitness landscape map.

For each quasispecies, the amino acid phylogenetic relationships and the enzymatic activities of the different HIV-1 protease variants were plotted together with the median-joining method implemented in NETWORK v 4.1.0.9 software (1).

Structural analysis.

To facilitate the visualization of HIV-1 protease quasispecies variation, all amino acids were mapped onto a three-dimensional (3D) representation of the enzyme and color coded according to their mutation frequency rates using the PyMol software package (13), and the X-ray coordinates deposited in the Protein Data Bank (http://www.ncbi.nlm.nih.gov) under accession number 1AJX(8). Positively selected amino acids are also shown for each quasispecies.

Western blot.

Four proteases belonging to the O quasispecies and having different catalytic efficiencies (O239, O255, O228, and O256) were chosen to correlate the lambda genetic screening with the degradation of the cI repressor. Phages with the different proteases were excised in a pBluescript SK plasmid, which was then introduced into E. coli JM109 cells containing p2X-cI-HIV (200 μl at 2.0 OD₆₀₀ units/ml). These cells were grown for 3 h at 37°C in LB medium containing 0.1 mM IPTG. pAlter EX-2, pBluescript SK, a hepatitis C virus (HCV) NS3 protease, and the HIV-1 HXB2 protease (in the presence or absence of IPTG) were also tested. All samples were suspended in 2× Laemmli buffer and normalized to 1 OD₆₀₀ unit/ml after cell lysis. Samples were fractionated on a 4 to 12% sodium dodecyl sulfate-NU-PAGE polyacrylamide gel (Invitrogen) for 2 h, transferred onto a nitrocellulose membrane, and blocked (10% skim milk) overnight at 4°C. A polyclonal cI antiserum (Stratagene), diluted 1/1,000, was used to detect the degradation of the repressor with a horseradish chemiluminescent system (ECL-Plus; Amersham).

Single-cycle infectivity assay.

The same four proteases, O239, O255, O228, and O256, analyzed previously by Western blotting were cloned into the luciferase reporter HIV-1 infectious clone pNL4-3-Luc-E⁻R⁻ (NIH AIDS Research and Reference Reagent Program) (11). An HIV-1 NL4.3 protease D30N mutant was also included in this analysis. Recombinant pNL4-3-Luc-E⁻R⁻ plasmids were used to transfect 293T cells in the presence of CalPhos (BD). The relative replication capacity of the virus was determined by measuring the amount of p24 antigen produced 72 h after transfection. The replication capacity is expressed as the percentage of p24 antigen produced by the vectors containing individually derived protease sequences compared to the p24 antigen from the vector containing the HIV-1NL4-3 protease reference sequence (100%). Replication capacity measurements were normalized for differences in transfection efficiencies by monitoring the luciferase activity generated in transfected cells. Three replicates were performed for each sample.

Nucleotide sequence accession numbers.

The sequences reported in this paper have been deposited in the GenBank database (accession no. DQ193605 to DQ193912).

RESULTS

Quasispecies genetic structure.

Lambda phage individual clones carrying the HIV-1 protease-encoding region were obtained at a single time point from three HIV-1 PI-naïve infected individuals (Table 1). At least 100 phage clones from each individual sample were isolated and sequenced (102, 100, and 100 clones for the M, N, and O samples, respectively). Neighbor-joining phylogenetic reconstruction and bootstrap analysis of all protease sequences revealed that sequences from each sample produced a monophyletic group (Fig. 1). Intrasample distance (D), Shannon entropy (S), and heterogeneity (H) were calculated for both nucleotide and amino acid sequences (Table 1). At the nucleotide level, only slight differences were found among the three studied samples. Intrasample genetic distances were 1.3% ± 0.3%, 1.8% ± 0.3%, and 1.6% ± 0.3% (mean ± standard error) for the M, N, and O samples, respectively. Similarly, no differences were found when the Sn was calculated. High proportions of different nucleotide variant clones, 67%, 71%, and 80% for M, N, and O, respectively, were also identified within the three quasispecies analyzed, illustrating the ample number of HIV-1 protease genetic configurations that exist in infected individuals. To determine the contribution of Taq polymerase errors to the level of genetic diversity, a nested PCR was initiated with an HIV-1 protease plasmid, and 14 subcloned lambda phage clones were sequenced. Only two clones were mutated, one with a synonymous T-to-C substitution and another one with a nonsynonymous A-to-G substitution. The estimated Taq error rate was 0.068 × 10⁻⁴ misincorporations per base pair per PCR cycle, which was within the expected range for a high-fidelity Taq polymerase (2). This result suggested that few PCR errors were generated under our experimental conditions. Nevertheless, we cannot dismiss the possibility that some of the substitutions analyzed here were generated during PCR amplification. Interestingly, a different scenario was found when the amino acid sequences were compared. The values from the N sample were comparable to those from the O sample but not to the ones from the M quasispecies, which showed a lower amino acid distance, amino acid-normalized Shannon entropy, and amino acid heterogeneity (Ha) (Table 1). Overall, lower amino acid diversity was found within the M sample compared to the N and O samples, emphasizing the different pressure constraints between viral quasispecies. The proportion of synonymous substitutions per potential synonymous site (dS) and the proportion of nonsynonymous substitutions per potential nonsynonymous site (dN) were calculated to search for positive selection within the analyzed HIV-1 protease quasispecies. Although the dS/dN ratios were greater than 1 in the three quasispecies, different values were detected for each quasispecies: 5.9, 2.9, and 8.3, respectively. This suggests, again, that different selective constraints may have been acting on the different quasispecies. A second approach for assessing positive selective pressures was to determine codon-specific selection with the PAML 3.14 software package (64). We identified four positively selected codons, two in the M quasispecies (codons 37 and 71) and two in the N quasispecies (codons 41 and 65) (Table 2). Two of the four identified codons were in protease regions in which selective pressures have been previously detected. Substitutions at residues 37 and 41 have been found in viruses from infected individuals with the HLA alleles A*6802 or B*44 and B*44, respectively. Despite the high amino acid diversity found within the O sample, no positively selected codons were detected in this quasispecies.

FIG. 1.

Neighbor-joining phylogram of proviral HIV-1 protease sequences from samples M, N, and O. Phylogenetic reconstruction was generated using a GTR + I + γ model implemented in the PAUP* 4.0 beta 8 software package. Bootstrap analysis (1,000 repetitions) was performed to determine the reliability of the sample grouping (numbers at branch nodes). Only those at the main branches and greater than 800 are shown. The HIV-1 HXB2 strain (prototype clade B protease sequence) was used as the outgroup.

Amino acid alignments showed a different population structure for each quasispecies (Fig. 2). Among the 21 different variants found in the M quasispecies, five major forms, with frequencies of 35%, 25%, 8%, 8%, and 6%, were identified. Similarly, of the 42 different variants included in the N quasispecies, four major forms, with frequencies of 29%, 14%, 4%, and 3%, were found. In contrast, within the O quasispecies, 42 different variants were also identified, but only two major forms, with frequencies 58% and 2%, were found; all of the remaining 40 sequences were unique. Overall, the N quasispecies displayed the highest percentage of clones bearing more amino acid changes than the most frequent variant (Fig. 3). A common characteristic of the three studied quasispecies was the high proportion of unique variants. This diversity allows the quick adaptation of viral quasispecies to environmental changes. For instance, primary (V82A) or secondary (L10I, G16E, K20R, M36VT, D60E, I62V, L63P, A71T, V77I, and N83D) mutations conferring resistance to PIs, as previously described by others (21, 28, 32), were found in the three quasispecies (Fig. 2) (21, 28, 32).

FIG. 2.

Amino acid sequence alignments of the three HIV-1 quasispecies. (A) Quasispecies M. (B) Quasispecies N. (C) Quasispecies O. Amino acid changes relative to the master sequence are indicated. The amino acid sequences are annotated by a capital letter for each sample. The number (percentage) of occurrences within each sample of identical amino acid sequences is given on the right. Dots indicate amino acid sequence identity. A circled letter indicates a primary substitution associated with resistance to protease inhibitors. Underlined letters indicate secondary substitutions associated with resistance to protease inhibitors.

FIG. 3.

Percentage of clones with amino acids changes relative to the master sequence. For each individual quasispecies, the percentage of clones with one, two, three, four, five, or six mutations relative to the most frequent variant is shown.

For each HIV-1 protease quasispecies, a 3D image representing the crystallographic protease structure was built. The master sequence was used as a backbone, and all amino acids substitutions detected in the variants, compared with the master sequence, were highlighted (Fig. 4). Most of the variability appeared at the protease surface, leaving critical structures such as the active site or the flap region almost without mutations. When the 3D representations of the N and O quasispecies were compared, different mutational patterns were identified in the two populations, although both had identical amino acid diversity (Ha = 42%). Most of the variant proteases detected in the N viral quasispecies were generated by combinations of a few mutations, while the variability detected in the O quasispecies came from individual mutational events. Consequently, a higher number of mutated residues was found in the O quasispecies (42) than in the N quasispecies (24) (Fig. 2 and 4).

FIG. 4.

Amino acid mutational pattern of the different quasispecies, M, N, and O, on the molecular surface of the HIV-1 protease. Molecular surfaces of the HIV-1 protease crystal structure (Protein Data Bank accession no. 1AJX) in which patches have been colored according to the number of mutations found in the corresponding amino acid as observed in the three quasispecies are shown. The backbone (blue) represents the master sequence of each quasispecies. Green indicates specific codons under positive selection pressure. Of note, within the four green codons under positive selection, more than 10 amino acid mutational events were also detected.

Quasispecies fitness landscape.

To determine the enzymatic activities of the different identified proteases (21, 42, and 42 for the M, N, and O quasispecies, respectively), a bacteriophage lambda-based genetic screen was used. This genetic screen is based on the phage lambda regulatory circuit in which the viral repressor cI is specifically cleaved to initiate the lysogenic-to-lytic switch (55). The introduction of an HIV-1 protease in a WT phage will cleave a mutant cI repressor containing a specific HIV-1 protease cleavage site, allowing the phage to go into the lytic replication cycle. As we have previously demonstrated, and as also shown in Fig. 5, the cI repressor cleavage is directly proportional to the protease catalytic efficiency (5, 20, 36, 37, 46, 47). The enzymatic activities of the different master and variant proteases of the three quasispecies analyzed in this study were related to the activity of the HXB2 protease (100%) (Fig. 6). Sequences with deletions or stop codons or that were hypermutated were not analyzed. The enzymatic activities of the three master sequences were 103% ± 8%, 45% ± 0.7%, and 76.5% ± 4.5% for the M, N, and O quasispecies, respectively (mean ± standard deviation) (Fig. 6). The second most represented sequence form of the M (M10) and N (N128) quasispecies had catalytic efficiencies of 71% ± 24% and 36.8% ± 2%, respectively. Within the three viral populations, there were proteases with reduced or undetectable enzymatic activity (less than 1% of the activity of the HXB2 strain protease). The percentage of defective protease clones was 19%, 28%, and 58% for the M, N, and O quasispecies, respectively. Therefore, 65% of all analyzed proteases displayed detectable enzymatic activity (Fig. 6). Interestingly, the O quasispecies, which had the highest proportion of defective protease clones, was obtained from an infected individual with a very low CD4⁺ T-cell count (10 cells/μl) and high viral load (297,000 copies/ml). Moreover, this quasispecies had the highest number of individual mutational events, the highest number of mutated amino acid residues (42), and the highest dS/dN ratio (8.3), suggesting an absence of positive selective pressures on this protease population. Also important is the finding that 4%, 24%, and 4% of the protease variants from the M, N, and O quasispecies, respectively, showed a similar or higher catalytic efficiency compared to the master protease. Remarkably, 18% of the variants within the N quasispecies displayed higher (more than 10%) catalytic efficiency than the master protease. This result strongly suggests that in some cases, the viral complexity in this genomic region does not depend exclusively on the enzyme's catalytic efficiency. Importantly, most of the high-fitness minority variants had only one substitution compared to the master sequence (Fig. 2 and 6).

FIG. 5.

Lambda cI-HIV-1 repressor cleavage is directly proportional to the HIV-1 protease catalytic efficiency. The catalytic efficiencies of four different proteases (O255, O239, O256, and O228) from quasispecies O were tested and analyzed by Western blotting. All bacterial samples expressed the lambda cI-HIV-1 repressor except for the last sample (lane 9), in which a bacterium expressing an empty pAlter Ex-2 plasmid was used. Control protease with catalytic residue mutation D25N (O255), an HCV protease, and an empty pBluescript SK plasmid without protease were also included in this experiment (lanes 1, 7, and 8, respectively). Expression of the protease was induced with IPTG for 3 h. The optical density of the cultures after 3 h (in the presence of IPTG) was measured to ensure that equivalent amounts of total cell protein were blotted. No differences were observed when the optical densities of the different cultures were compared, suggesting that the expression of the HIV-1 protease did not affect the growth of the bacteria. The Western blot proved that the cI-HIV-1 repressor was not cleaved by the O255 protease, the HCV protease, or the pBluescript SK empty plasmid. Nevertheless, all the other O or HXB2 proteases cut the cI-HIV-1 repressor. cI-HIV-1 degradation was proportional to the catalytic efficiency of the expressed protease. The HXB2 protease was tested with or without IPTG (lanes 5 and 6, respectively).

FIG. 6.

Comparative growth of phages containing different HIV-1 protease single variants. (A) Quasispecies M. (B) Quasispecies N. (C) Quasispecies O. The growth of phages encoding a single protease variant (black and white bars) was compared to the growth of WT HIV-1 HXB2 protease (100%) (gray bar). The black bar indicates the master sequence, and the dashed bare (if any) denotes the second most represented sequence. The growth of a phage encoding an inverted HXB2 protease was included as a negative control (invHXB2).

In order to confirm that the enzyme activities determined with the genetic screen were affecting the ex vivo replication capacities (fitness) of the viruses to a similar extent, in the HIV-1 infectious clone pNL4-3luc⁺ R⁻E⁻ A, we introduced four different protease variants from the O quasispecies, namely, O255, O239, O228, and O256. These proteases, which included the O master sequence, were chosen because they displayed different catalytic efficiencies in the genetic screen. In addition, a mutated pNL4-3luc⁺ R⁻E⁻ A protease (D30N), which drastically reduces viral fitness (38), was included in the analysis. Although some differences in the percentages of enzyme activity and fitness reduction were found between these two assays, after one cycle of viral replication, the observed HIV-1 growth was proportional to the observed enzymatic activity of the corresponding protease (Fig. 7 and 8). This result suggested that the protease genetic screen system used here can be seen as a complement to the classical biochemical approach for monitoring protease proteolytic activity.

FIG. 7.

Replication capacity (fitness) of infectious HIV-1 carrying different protease variants. Four proteases, O239, O255, O228, and O256, which had different enzymatic activities (Fig. 5 and 6), were introduced into the luciferase reporter HIV-1 infectious clone pNL4-3-Luc-E⁻R⁻. An HIV-1 NL4-3 protease D30N mutant was also included in this experiment. The replication capacities of the different proteases are represented as a percentage relative to the WT HIV-1 NL4-3 strain (100%). The relative replication capacity of the virus was determined by measuring the amount of p24 antigen. The O master sequence is indicated by a black bar. Three replicates were performed for each sample.

FIG. 8.

Proportionality between HIV-1 replication capacity and catalytic efficiency. The catalytic efficiencies of three protease variants (white bars), O255, O228, and O256, are compared to the replication capacities of HIV-1 infectious clones (black bars) carrying these three protease variants. The catalytic efficiencies and the replication capacities of the three proteases are represented as percentages relative to that of the O239 variant (100%).

Finally, for each of the three HIV-1 protease quasispecies, a phylogeny fitness landscape map was constructed by correlating the amino acid phylogenetic relationship and the catalytic efficiencies of the different variants. This correlation was performed by using the median-joining network method (1). Three different landscapes were obtained, one for each protease population. The M quasispecies, which had the lowest amino acid diversity (Table 1 and Fig. 2), formed a landscape in which the ancestor (network center) sequence coincided with the master sequence and with one of the adaptive peaks (Fig. 9A).Nevertheless, other adaptive peaks formed by minority variants were detected around the central peak originated by the master sequence. A more complex scenario was found within the N quasispecies landscape. The higher amino acid diversity of this protease population generated a very complex phylogenetic network (Fig. 9B) in which the master sequence did not occupy the most adaptive peak, since there were outsider minority variants not directly related to the master sequence that formed several adaptive peaks. In this quasispecies, the master sequence was not the fittest type but rather represented a center of gravity in the population landscape that did not coincide with the sequence representing maximum fitness. The O quasispecies protease formed the simplest landscape (Fig. 9C). The master sequence was located in the center of the network forming an adaptive peak and surrounded by closely related minority variants, most of them occupying low-fitness adaptive peaks. Every viral protease quasispecies analyzed here formed distinctive individual landscapes, suggesting that there are variable evolutionary patterns and different selective pressures within each quasispecies. Nevertheless, several adaptive peaks were identified in the three populations (Fig. 6 and 9), indicating that the three HIV-1 protease quasispecies landscapes illustrated here are rugged.

FIG. 9.

Phylogeny fitness landscape map. The median-joining network shows the amino acid phylogenetic relationships and the catalytic efficiencies of the different protease variants. (A) Quasispecies M. (B) Quasispecies N. (C) Quasispecies O. This phylogenetic reconstruction was performed by using the median-joining network method (1). Circles represent the different protease single clones. The circle size is proportional to the observed clone frequency in the quasispecies. Circle color represents the protease catalytic efficiency (fitness) relative to the WT HIV-1 HXB2 protease.

DISCUSSION

HIV-1 replication generates a myriad of variants that constitute a complex interrelated structure, termed a quasispecies, that must be effectively defeated by the host immune system and by the antiretroviral treatment. Considering that natural selection acts on RNA viruses at the population level rather than on individual variants (15, 59), an exhaustive analysis of the quasispecies variants will be necessary to understand and to solve the fatal interaction of HIV-1 with its host. Several studies have analyzed the HIV-1 genetic structure and the evolution of different viral genes in detail; nevertheless, data regarding the phenotypic structure of HIV-1 and other RNA virus quasispecies are scarce. Here, we describe for the first time the fitness landscape of an HIV-1 protein quasispecies. Overall, the three quasispecies analyzed displayed a number of notable features. First, within each quasispecies, there was a large number of fitness optima or peaks. Because fitness optima were frequent (65% of all analyzed proteases displayed a detectable enzymatic activity, and 11% showed similar or better fitness than the master, the most abundant enzyme), the HIV-1 protease quasispecies complexity does not depend exclusively on the enzyme's catalytic efficiency. Moreover, in one quasispecies population (Fig. 9B), the master sequence was not the fittest sequence but rather represented a middle or low adaptive peak. Selective pressures in another region of the HIV-1 genome might favor a less-fit protease variant (14, 25, 29, 41). Second, the fitness landscape was rugged. Several single substitutions were lethal and led the master sequence to drop down the peak. However, at other positions, single substitutions sent the master sequence to a new local optimum or peak, suggesting that the master sequence may walk through the quasispecies fitness landscape by single mutational steps without being trapped at suboptimal alleles (34, 56, 61). Third, although the three analyzed quasispecies shared some traits, such as the presence of several fitness optima, every protease quasispecies formed distinctive individual fitness landscapes. This is particularly interesting because the three quasispecies had similar nucleotide diversities (Fig. 1), suggesting that different selective constraints may have been acting on different quasispecies. Even quasispecies N and O, which had similar amino acid diversities (Table 1), formed very different fitness landscapes. Quasispecies N seemed to be composed of variants formed by different combinations of a reduced number of substitutions that generated a high proportion (>70%) of fit proteases. In contrast, within the O quasispecies, there were many different amino acid changes all over the viral protease, with the majority of these variants (>60%) being defective enzymes (Fig. 2 and 6). This result, together with the higher dS/dN ratio found in the O quasispecies, strongly suggests an absence of positive selective forces shaping this protease quasispecies. Therefore, random genetic drift and selection pressures within the protease coding region or in other viral genomic regions could all be acting together to generate the protease quasispecies landscape (4, 50, 51).

A striking feature found within the quasispecies analyzed here is the presence of several low-fitness protease variants in every population. Since a low-fitness protease strongly affects the fitness of the virus (Fig. 7) (38, 42), it is intriguing that these low-fitness variants were not eliminated during successive rounds of replication. Moreover, the phylogenetic-fitness landscape map shows that some of these low-fitness variants were not generated by single mutations from the master sequence or fitter variants but rather that they had a more complex relationship with the master or more abundant enzymes (Fig. 9), suggesting that their presence in the quasispecies was not recent. The newly described phenomenon of memory in viral quasispecies, including memory subpopulations of HIV-1 in vivo, supports the adaptive role of genome subpopulations (3, 52, 53). As exemplified by the presence of minority subpopulations containing PI resistance substitutions (Fig. 2), low-fitness variants may serve as a molecular reservoir capable of reacting swiftly to a selective pressure. Thus, the viral population would preserve less-fit variants to ensure the success of the whole quasispecies.

Site-directed mutagenesis and random mutagenesis have been used to perform structure-function studies on many proteins, including the HIV-1 protease and other HIV-1 proteins (22, 33). This work highlights the usefulness of analyzing, at high resolution, RNA-virus protein quasispecies to discover the relative importance of specific residues to protein structure and function through the numbers and types of tolerated mutations. The catalytic efficiencies of the different variant proteases not only give a precise scheme of the selection forces that act in the evolution of this protein but also highlight essential positions in the enzyme. A previous study determined that a threshold of HIV-1 protease activity exists between 4-fold and 50-fold reduction, below which processing is insufficient to yield infectious particles (49). Our data are in agreement with this threshold, and mutations in the active sites D25N (O255) and T26A (O201), the flap regions G49E (M86 and O104) and G51R (N41 and O203), or the autocleavage sites P1S (N9 and O229) and I3V (O250) rendered proteases with no or extremely reduced enzymatic activity (Fig. 6).

Overall, our results demonstrate that the rugged HIV-1 protease quasispecies landscape must be able to respond to environmental changes that may threaten the virus' survival. Genetic-fitness landscape maps may provide clues to adaptive mechanisms and their relationship to genetic drift and selective pressures exerted by the host immune system or antiretroviral therapy.