The Fitness Cost of Mutations Associated with Human Immunodeficiency Virus Type 1 Drug Resistance Is Modulated by Mutational Interactions

Abstract

It is generally accepted that the fitness cost of resistance mutations plays a role in the persistence of transmitted drug-resistant human immunodeficiency virus type 1 and that mutations that confer a high fitness cost are less able to persist in the absence of drug pressure. Here, we show that the fitness cost of reverse transcriptase (RT) mutations can vary within a 72-fold range. We also demonstrate that the fitness cost of M184V and K70R can be decreased or enhanced by other resistance mutations such as D67N and K219Q. We conclude that the persistence of transmitted RT mutants might range widely on the basis of fitness and that the modulation of fitness cost by mutational interactions will be a critical determinant of persistence.

Antiretroviral drug resistance is an important cause of treatment failure in human immunodeficiency virus type 1-infected persons treated with reverse transcriptase (RT) and protease inhibitors. Emergence of resistance during treatment usually involves the initial selection of deleterious mutations that reduce drug susceptibility and decrease replicative capacity. Compensatory evolution through the acquisition of additional mutations generally results in partial restorations of viral fitness (19). Despite the accumulation of compensatory mutations, drug-resistant viruses tend to replicate less efficiently than wild-type viruses.

The transmission of drug-resistant mutants with diminished fitness and the evolution of these viruses in the absence of drug are typically associated with the reversion and loss of resistance mutations (3, 9, 22). Different rates of persistence and reversion of mutations have been documented in vivo and have usually been explained by the impact of mutations on viral fitness. For instance, less-fit zidovudine-resistant mutants carrying T215Y/F are generally replaced by more-fit 215D/C/S revertants within less than a year, while more-fit M41L or D67N mutants tend to revert more slowly (1-3, 5, 6, 8, 13, 17, 20, 25). However, in vivo observations do not always show the expected relationship between fitness cost and persistence, and a range of persistence for the same mutation has been noted among patients infected with multidrug-resistant viruses. For instance, M184V mutants have been found to persist between 4 and 16 months after primary infection, while persistence of K103N can range between 1 and 3 years (2, 3, 5, 17). We hypothesize that mutational interactions in multidrug-resistant viruses might modulate the fitness cost of resistance mutations and might influence the rate of reversion and persistence of mutations.

To better understand the persistence of transmitted resistance, we performed a systematic evaluation of the impact of RT mutations on fitness and assessed if such an impact could be enhanced or decreased by interactions with other resistance mutations. We evaluated the fitness cost of 11 key RT mutations (M41L, K65R, D67N, K70R, L74V, K103N, Y181C, M184V, L210W, T215Y, and K219Q) associated with resistance to nucleoside and nonnucleoside RT inhibitors. Mutations were introduced into the HXB2 genetic background by site-directed mutagenesis (QuikChange site-directed mutagenesis kit; Stratagene, La Jolla, CA) (10). The impact of mutations on fitness was assessed using a growth competition assay described previously (11, 12). Briefly, a mixture of the two competing variants was used to infect 7.5 × 10⁵ MT-4 cells at a multiplicity of infection of 0.001. After 4 to 6 days in culture, 50 μl of the supernatant was used to reinfect a fresh aliquot of 7.5 × 10⁵ MT-4 cells; this procedure was repeated for a total of 6 to 7 passages. Changes in the relative proportion of the two competing variants were monitored by sequence analysis of proviral DNA as described previously (11, 12). Passages were therefore limited to 4 to 6 days to maximize MT-4 cell viability. Fitness differences were calculated from the slope of the fitness vectors generated by plotting the changes in the relative proportion of the mutant variant overtime (14). Slopes can be negative when the mutation confers a fitness cost or positive if the mutation confers a fitness gain. The absolute value of the slope is the fitness difference between the two variants. For the calculation of fitness differences, we assumed exponential growth or decline of the two competing variants based on the relatively constant density of susceptible MT-4 cells achieved during our short passages (not shown).

The initial virus mixtures in the competitions were normalized based on equal amounts of wild-type and mutant viruses (50%-50%) or a higher proportion (75%) of the mutant virus. The use of an excess of the mutant allowed us to detect relatively unfit mutant strains for multiple time points. We conducted an initial evaluation to explore if such differences in the initial ratios could influence our fitness calculations. Figure 1 shows that when HXB2_K65R and wild-type HXB2 (HXB2_wt) or HXB2_K103N and HXB2_wt were mixed at either 50%-50% or 75%-25%, the results of the competition experiments were similar and that, in both cases, HXB2_wt outcompeted the mutant virus with similar kinetics. At the two initial ratios of 75:25 and 50:50, the calculated fitness cost for the K103N mutation was 4.7- and 7.9-fold, respectively (a 1.7-fold difference) and was 37- and 24-fold for K65R (a 1.5-fold difference). These findings suggest that in our assay system, these differences in the initial proportion of the two variants do not have a substantial effect on fitness determinations.

FIG. 1.

Competition dynamics between HXB2_wt and HXB2_K103N or HXB2_K65R. Experiments were initiated at a 50:50 (A and B) or a 75:25 (C and D) mutant-to-wild-type ratio. Panels A through D show the changes in the relative proportion of the two competing variants over time measured in a single competition experiment. Panels E and F show the average fitness vector obtained for K103N and K65R, respectively. To calculate the fitness vector, the proportion of the mutant with respect to the wild-type (WT) virus (Rn) was divided by its ratio in the initial mixture (Ro), and this value (Rn/Ro) was plotted versus the competition passage. Bars denote the standard errors of the means.

Figure 2 shows the absolute fitness differences calculated in all the competitions between HXB2_wt and single mutants. With the exception of L210W, all mutant viruses were outcompeted by HXB2_wt, with fitness differences that varied from 0.4-fold for K70R to 29-fold for K65R (a 72-fold range). The fitness vector obtained for L210W had a positive slope suggestive of a fitness gain (not shown). However, given the low difference in fitness between HXB2_wt and HXB2_L210W (0.4-fold), the possibility of assay variability cannot be excluded. Such small variability was noted for K70R, which showed a very low fitness cost (onefold) or gain (0.6-fold) in two separate experiments (not shown).

FIG. 2.

Absolute fitness differences between HXB2_wt and drug-resistant mutants. Results obtained from duplicate experiments (K70R, K103N, M184V, and K65R) or single determinations are shown. Error bars indicate the standard errors of the best-fit values obtained from the fitness vectors. All mutations, with the exception of L210W, showed a fitness cost as indicated by a negative slope in the fitness vector. The absolute fitness differences (n-fold) (95% confidence interval) were 0.4 ± 0.2 (0.4 to 1.2) for L210W, 0.4 ± 0.2 (0.1 to 1) for K70R, 0.6 ± 0.4 (0.4 to 1.5) for Y181C, 1.4 ± 0.3 (0.8 to 2) for K219Q, 2 ± 0.1 (1.6 to 2.5) for L74V, 3.7 ± 0.2 (2.6 to 4.9) for D67N, 4.5 ± 0.7 (2.8 to 6.2) for M41L, 6.2 ± 0.2 (5.7 to 6.6) for K103N, 12.9 ± 1 (10 to 15.8) for T215Y, 14 ± 0.7 (12 to 15.8) for M184V, and 29 ± 0.5 (29 to 37) for K65R.

We next investigated if the fitness cost of M184V, K70R, or T215Y/F could be affected by the presence of other resistance mutations. We performed competition experiments among isogenic multidrug-resistant viruses that differed by only one mutation. For instance, the fitness cost of M184V in a 41L/210W/215Y background was evaluated by mixing HXB2_{41L/210W/215Y} and HXB2_{41L/210W/215Y/184V}, and the fitness cost of K70R in a 67N/219Q background was evaluated by mixing HXB2_67N/219Q/70R and HXB2_67N/219Q. The use of viruses that differ by only one amino acid change was necessary to control for potential recombination and diversification of mutant genotypes. This approach allowed us to define the specific fitness cost of M184V, K70R, or T215Y/F in backgrounds that had additional resistance mutations.

When M184V was present in HXB2_{41L/210W/215Y/103N}, HXB2_{67N/70R/219Q/215F}, or HXB2_{41L/210W/215Y}, the fitness cost was high (16.1-, 14.5-, and 8.9-fold, respectively) and was similar to that seen in HXB2_wt (14-fold) (Table 1). However, such a fitness cost was only 2.3-fold in HXB2_67N/70R/219Q, suggesting that the combination of 67N/70R/219Q has a compensatory effect on the fitness cost conferred by M184V. Interestingly, the opposite effect was observed for K70R. Table 1 shows that while K70R alone had a minimal impact on fitness (0.4-fold), such an impact was increased to 6-fold in HXB2_D67N and 9.4-fold in HXB2_D67N/219Q. In contrast to M184V and K70R, the T215Y and T215F mutations had a similar impact on viral fitness in all the three RT backgrounds tested. Figure 3 (A, B, and C) shows the fitness vectors calculated for all the 11 HXB2 mutants.

TABLE 1.

Fitness cost of 184V, 70R, and 215Y/F in wild-type viruses or in viruses with other thymidine analog mutations or K103N

Mutants tested in competition assays	Fitness difference (fold) ± SEM (95% CI; r2)
M184V
HXB2wt, HXB2184V	14.0 ± 0.7 (12.2-15.8; 0.99)
HXB241L/210W/215Y/103N, HXB241L/210W/215Y/103N/184V	16.1 ± 0.6 (14.6-17.6; 0.99)
HXB267N/70R/219Q/215F, XB267N/70R/219Q/215F/184V	14.5 ± 1.2 (11.4-17.7; 0.98)
HXB241L/210W/215Y, XB241L/210W/215Y/184V	8.9 ± 0.8 (6.8-10.9; 0.93)a
HXB267N/70R/219Q, HXB267N/70R/219Q/184V	2.3 ± 0.1 (2.0-2.5; 0.97)a
K70R
HXB2wt, HXB270R	0.4 ± 0.2 (0.1-1; 0.37)
HXB267N, HXB267N/70R	6.0 ± 0.8 (4-7.9; 0.79)a
HXB267N/219Q, HXB267N/219Q/70R	9.4 ± 1.1 (6.6-12.2; 0.89)a
T215Y/F
HXB2wt, HXB2215Y	12.9 ± 1.0 (10-15.8; 0.98)
HXB2210W, HXB2210W/215Y	9.7 ± 0.8 (7.8-11.6; 0.96)
HXB267N/70R/219Q, HXB267N/70R/219Q/215F	11.7 ± 0.5 (10.6-12.8; 0.99)

^aP < 0.005 compared to HXB2_184V or HXB2_70R.

FIG. 3.

Modulation of the fitness cost of M184V, K70R, or T215Y/F by other drug resistance mutations. Panels A, B, and C show the fitness plots of different HXB2 mutants carrying M184V, K70R, or T215Y/F alone or in association with other mutations. Panel D shows the fitness vectors obtained in patient-derived viruses carrying the M184V mutation alone (isolate RD160) or in combination with K103N (isolate RD110), K70R (isolate RD583), or M41L/T215Y (isolate RD879). Results obtained from duplicate experiments (A, B, and D) or single determinations (C) are shown. Bars denote the standard errors of the means. WT, wild type.

We performed the same analysis using recombinant viruses generated from four subtype B-infected, treatment-naïve patients that had M184V alone (isolate RD160) or in combination with K70R (isolate RD583), 41L/215Y (isolate RD879), or K103N (isolate RD110). Patients RD583, RD879, and RD110 were estimated to be recently infected, while patient RD160 was chronically infected (Vironostika HIV-1; bioMerieux, Inc., Raleigh, NC) (24). Recombinant viruses were generated by cotransfection of MT-4 cells with cloned RT sequences from these patients and the RT-deleted proviral molecular clone pHIVΔ RTBstEII as described previously (10, 23). The fitness cost conferred by M184V in each background was measured by mixing individual RT clones with the corresponding site-directed mutant that lacked 184V. For instance, the fitness cost of M184V in patient RD879 was calculated by mixing RD879_{41L/215Y/M184V} with RD879_41L/215Y. Figure 3D shows that in these isolates, the M184V mutation also conferred a wide range of fitness costs. While M184V conferred a 15-fold reduction in viral fitness in isolate RD583, this value was reduced to 5.4-fold in isolate RD879 and 3.2-fold in isolate RD110 and was only 0.2-fold in isolate RD160.

Our study provides a basic understanding of the fitness cost of individual RT resistance mutations and demonstrates that mutational interactions can modulate such a fitness cost. While considerable information regarding the fitness cost of nucleoside RT inhibitor- and nonnucleoside RT inhibitor-resistant viruses has been gathered during the past few years, the use of methods that differ in their abilities to detect and quantify small fitness differences has in many cases resulted in discordant results (4, 11, 12, 15, 16, 18, 21). Our systematic analysis in the HXB2 background shows that such a fitness cost can vary within a 72-fold range, which predicts that the persistence of transmitted RT mutations might range widely solely on the basis of this parameter. These results also suggest that estimates of the rate of transmitted resistance might be biased towards mutations that confer a low fitness cost and have a higher potential to persist and be detected by standard sequence analysis.

The in vitro results on fitness cost in our RT mutants were generally consistent with the trends seen in longitudinal cohorts of primary infections showing a longer persistence of more-fit mutants such as M41L, D67N, and L210W and a more rapid reversion of mutations such as M184V and T215Y/F (2, 3, 5, 20, 25). These findings were also consistent with observations among newly diagnosed persons showing a higher frequency of less-fit mutants such as M184V among recently infected persons compared to persons with chronic infections. In contrast, mutations that confer moderate or low fitness, such as M41L and T215Y/F revertants, have been seen more frequently in chronic infections (12, 24). However, our demonstration that the fitness cost of a particular resistance mutation is not constant and may vary among multidrug-resistant isolates suggests that fitness modulations may alter the persistence of resistance mutations in transmitted isolates. While M184V mutants were generally unfit, we found that the fitness cost of this mutation was reduced in a multidrug-resistant background that had D67N, K70R, and K219Q. These data predict that M184V might persist longer in some isolates than in others and may explain why this mutation has been found in some isolates 14 to 24 months after primary infection, compared to the relatively rapid reversion (6 to 10 months) seen in most isolates (2, 3, 5). The opposite was also observed, and the low fitness cost of K70R was significantly enhanced by the presence of D67N and/or K219Q, despite the fact that both D67N and K219Q are compensatory mutations. It is possible that the compensatory effect of D67N and K219Q might be limited to increase virus replication in the presence of drugs and that such mutations have a deleterious effect in the absence of drug pressure. This is consistent with recent findings showing that compensatory mutations such as L210W improve the fitness of a M41L/T215Y mutant in the presence of zidovudine but have a fitness cost in the absence of drug (15).

The transmission of multidrug-resistant viruses and the initial establishment of a low-fit founder virus population provide an interesting model of molecular evolution. In wild-type infections, the individual genomes that compose the quasispecies have discrete fitness values that are lower that the fitness of the average population (7). Selection of drug-resistant escape mutants is generally expected to generate virus subpopulations with inferior fitness (7). In contrast, during drug-resistant infections, the fitness value of the average population is lower than the discrete values of the individual genomes generated by reversion or by compensatory changes. This scenario results in the coexistence of more-fit revertant viruses that may have multiple combinations of mutations. In this setting, the modulation of the fitness cost of mutations by mutational interactions is critical for dictating both the rate and order of reversion of individual resistance mutations. Our results demonstrate the complexity of resistance reversion in the setting of multidrug-resistant viruses and suggest that reversion may be less predictable than expected on the basis of the fitness costs of individual resistance mutations.

In summary, we show a wide range of fitness costs associated with RT resistance mutations, which predicts that the persistence of transmitted resistance may range widely. We also demonstrate that mutational interactions can modulate the fitness costs of individual RT mutations. We conclude that modulations of fitness cost will shape the rate of reversion of mutations, suggesting that the persistence of transmitted resistance may be less predictable than previously thought.

Nucleotide sequence accession numbers.

RT sequences from the clones reported in this paper have been submitted to the GenBank database under accession numbers DQ871032 to DQ871035.