Conceived and designed the experiments: UDK BA. Performed the experiments: UDK. Analyzed the data: UDK BA. Contributed reagents/materials/analysis tools: UDK. Wrote the paper: UDK BA.
The authors have declared that no competing interests exist.
HIV-1 escape from the cytotoxic T-lymphocyte (CTL) response leads to a weakening of viral control and is likely to be detrimental to the patient. To date, the impact of escape on viral load and CD4+ T cell count has not been quantified, primarily because of sparse longitudinal data and the difficulty of separating cause and effect in cross-sectional studies. We use two independent methods to quantify the impact of HIV-1 escape from CTLs in chronic infection: mathematical modelling of escape and statistical analysis of a cross-sectional cohort. Mathematical modelling revealed a modest increase in log viral load of 0.051 copies ml−1 per escape event. Analysis of the cross-sectional cohort revealed a significant positive association between viral load and the number of “escape events”, after correcting for length of infection and rate of replication. We estimate that a single CTL escape event leads to a viral load increase of 0.11 log copies ml−1 (95% confidence interval: 0.040–0.18), consistent with the predictions from the mathematical modelling. Overall, the number of escape events could only account for approximately 6% of the viral load variation in the cohort. Our findings indicate that although the loss of the CTL response for a single epitope results in a highly statistically significant increase in viral load, the biological impact is modest. We suggest that this small increase in viral load is explained by the small growth advantage of the variant relative to the wildtype virus. Escape from CTLs had a measurable, but unexpectedly low, impact on viral load in chronic infection.
HIV, like many viruses, has evolved multiple strategies to evade immune surveillance. One of these strategies is the evolution of escape mutations which reduce the ability of the immune response to kill HIV-infected cells. But does HIV escape matter? Some believe that the accumulation of escape mutations leads to AIDS; many more believe escape is likely to be highly detrimental to human health. Yet, to date, it has not been possible to measure the impact of escape. We developed two independent methods to quantify the impact of escape on HIV viral load. Both methods showed that escape does lead to a detectable increase in viral load, but is unlikely to have a major impact on patient health as the increase is small. Indeed, only 6% of between-individual variation in viral load could be attributed to HIV escape. This work suggests that the current research focus on escape in chronic HIV infection might be out of proportion to its importance with other factors playing a more significant role in determining viral load.
The CTL response is thought to play a role in firstly reducing
Longitudinal studies following viral escape in HIV-1 infected individuals have shown no conclusive results
Hence, because of limited longitudinal data and difficulties in interpreting cross-sectional data, the consequences of HIV-1 escape from the CTL response are still unclear. The aim of this work was to quantify the impact of HIV escape on viral load in chronic infection.
We used an extension of the Perelson/De Boer model of HIV-1 dynamics
The variant was assumed to have reduced replicative ability compared to the wildtype (i.e. although the variant is fitter in the presence of a CTL response, it is less fit in the absence). The system was modelled under the “worst case” scenario that the CTL response is unable to target the escaped epitope again and so does not regain control of viraemia. In all 100,000 runs of the model, covering a wide range of biologically plausible parameters, there was an increase in viral load upon escape. The increase in viral load upon escape was significantly correlated with the outgrowth rate of the variant (Spearman's rank correlation: p<0.0001; rho = 0.59). For outgrowth rates representative of what is observed in chronic HIV infection (i.e. the variant replaces the wildtype at rate 0.01–0.04 per day, corresponding to variant outgrowth in 230–900 days
Results from the first 1,000 runs from the mathematical model are plotted. The increase in viral load after an escape event is significantly positively correlated with the outgrowth rate (k) of the escape variant (Spearman's rank correlation: p<0.0001, rho = 0.59). The median predicted increase in viral load for observed values of the outgrowth rate (k, shaded area) is approximately 0.051 log (95% confidence interval: 0.050–0.052).
A second scenario, in which the infection and virion production rates of the variant were set independently of the wildtype such that the variant was permitted to be fitter than the wildtype at the time of escape, was also considered. In this model, the median increase in viral load was 0.050 log copies ml−1 (95% confidence interval: 0.044–0.056). Even if the outgrowth rate of the escape variant was an order of magnitude higher (e.g. 0.4 per day, corresponding to the variant replacing the wildtype in 23 days), the median predicted increase in viral load was still only 0.27 log. As before, there was a significant positive correlation between the outgrowth rate of the escape variant and the resulting increase in viral load (Spearman's rank correlation: p<0.0001; rho = 0.59). Interestingly, for this model there was a fraction (12%) of cases in which escape resulted in a decrease in viral load.
To test the robustness of the prediction that escape leads to a small increase in viral load, additional mathematical models were simulated which tested different assumptions and explicitly modelled the CTL response as a separate dynamic population (see
Following Moore
In order to correct for the background level of mutation, we performed multivariate regression. This allowed us to test whether the number of escape events (NEE) is a significant independent predictor of viral load after a possible relationship between viral load and the background level of mutation (quantified by NSE, the number of synonymous changes in epitopes per individual) had been taken into account.
The analysis showed that the number of escape events is a significant independent predictor of log viral load when the number of synonymous changes is taken into account (
Number of escape events is a significant predictor of log viral load after the background level of mutation is taken into account (multiple linear regression: p = 0.0021, gradient = 0.11, r2 = 0.060, N = 157). The width of each bar is proportional to the number of points. Alternate representations of this data are shown in
In order to investigate the relationship between escape from protective alleles and viral load, we performed three analyses. First, we removed individuals with protective alleles from the multiple linear regression (see
To check for robustness, we performed a bootstrap analysis by sampling 10,000 times with replacement from the Full cohort. We found that 92% of the bootstrap runs resulted in a significant positive association between log viral load and the number of escape events, the median two-tailed p-value of all runs was 0.0019 (95% confidence interval: 0.0018–0.0021;
By calculating the gradient of the line of best fit through the data, after correcting for the background level of mutation, we find that each escape event causes a median increase in viral load of 0.11 log copies ml−1 (95% confidence interval: 0.040–0.18), or an absolute median increase of 8,500 copies ml−1. Overall, the number of escape events could only explain about 6% of the variation in viral load in the cohort.
Next we investigated whether escape events in any particular gene were associated with the increase in viral load, so we repeated the multi-linear analysis on each gene independently (
Data used in
Next, we investigated which gene (or genes) was driving the effect in the 92% of bootstrap runs that gave a significant association between the total number of escape events in all genes and viral load (see previous section “Relationship between number of escape events and viral load”). This analysis showed that escape events in most genes were detrimental for the host, but that escape events in Pol epitopes were most frequently (81%) behind the increase in viral load (
Finally, we investigated individual gene effects in the Extended cohort (N = 347) in order to increase power. We could not study this cohort in the previous analysis as we did not have sequence data for all genes for all individuals; however this is unnecessary for the purposes of individual gene analysis. As the Full cohort is a subset of the Extended cohort, the results must not be considered independent. Analysis of this extended cohort strengthened and extended our previous results. Escape events in
Subjects were sampled with replacement 10,000 times from the Extended cohort (N = 347). The shaded area on the left-hand side of each graph indicates the proportion of runs with a statistically significant increase in log viral load (one-tailed p≤0.025), and the shaded area on the right indicates a significant decrease in log viral load (one-tailed p≥0.975). In each panel the median change in log viral load associated with an escape event in that gene is presented.
To date, it has not been possible to quantify the impact of HIV-1 escape on viral load. Longitudinal datasets are too sparse to allow firm conclusions and studies of cross-sectional datasets have been impossible due to confounding factors. Furthermore, previous studies
Firstly, we simulated the emergence of HIV-1 escape variants using an extension of the Perelson/De Boer model of HIV-1 dynamics
Secondly, we analysed a cross-sectional cohort of 157 HIV-1 C-clade infected individuals. We used a novel method to remove confounding factors which permits the analysis of data-rich cross-sectional cohorts. By treating the number of synonymous changes accrued by an individual as a clock that keeps count of the number of “mutation events”, we corrected for differences in individual mutation rates, regardless of whether the increased mutation frequency resulted from a higher viral set point, higher viral replication rate or longer length of infection (and hence lower CD4+ count). We show a statistically significant positive association between the number of mutated epitopes and viral load. We found a small increase in viral load of 0.11 log per escape event. To put this into context, the median standard deviation of random fluctuations in log viral load over time in a chronically HIV B-clade infected cohort is 0.32 (see
The small increase in viral load per escape event that we observed in the cohort could be explained by two (non-exclusive) hypotheses. Firstly, the impact of escape on viral load is transient due to the flexibility of the CTL response which adapts to recognise either the escape variant or a previously sub-dominant epitope, resulting in renewed and long-term suppression of viral load. Secondly, the impact of escape is small because the variant has a slow outgrowth rate (i.e. a small net fitness advantage compared with the wildtype), so the accompanying increase in viral load is not large. Under the first hypothesis, escape initially causes a large increase in viral load but then the CTL response adapts and partially regains control of viraemia, either by targeting previously unrecognised or sub-dominant epitopes
Additionally, the modelling showed that the increase in viral load per escape event was significantly positively correlated with the outgrowth rate of the variant: that is, the faster the variant outgrows the wildtype, the larger the resulting increase in viral load. This is consistent with the observation that protective HLA class I alleles associated with slow progression to AIDS tend to present epitopes where escape variants have a slower outgrowth rate
A similar argument could explain our finding that escape events in
A number of studies have reported a relationship between protective HLA class I alleles and low levels of viral escape, suggesting that escape is an important determinant of progression to AIDS
We find that escape only determines approximately 6% of variation in viral load. How can we reconcile the low impact of escape events on viral load with disease progression, given the strong association between HLA and disease progression? As viral load is subject to high levels of random fluctuations within an individual over time, one possible explanation is that viral load is not always strongly associated with progression
We predicted mathematically and also showed by analysis of an HIV-1 infected cohort that HIV escape from the CTL response is associated with a small increase in viral load. Although the finding that a typical escape event has a modest impact on viral load is surprising, it is consistent with firstly, the failure to find a clear association between escape and viral load in longitudinal studies
This study shows that although the impact of HIV-1 escape from CTLs is highly statistically significant, the effect in clinical terms is mild: an increase in viral load of 0.11 log copies ml−1 will have few consequences for patient health
The Perelson/De Boer model of HIV-1 dynamics
Parameter | Range | Description (units) | Reference |
λ | 5–30 | Influx of new uninfected CD4+ cells (c d−1) | |
d | 0.013–0.078 | Natural death rate of uninfected CD4+ cells (d−1) | |
β | 0.001–0.01 | Infection rate of wildtype virus (v−1 d−1) | |
β′ | 0.001–0.01 | Infection rate of variant virus (v−1 d−1) | |
b | 0.5–1.0 | Death rate of infected CD4+ cells from all other sources (d−1) | |
c | 0.02–0.2 | Death rate of CD4+ cells by a single CTL clone (d−1) | |
h | 20–200 | Burst rate of CD4+ cells infected with wildtype (v c−1 d−1) | |
h′ | 20–200 | Burst rate of CD4+ cells infected with variant (v c−1 d−1) | |
u | 3–300 | Clearance rate of free virus (d−1) |
Note that parameter b is the rate of death of infected CD4+ cells attributable to CTL-independent death and all CTL responses other than the single CTL response which is escaped (parameter c). Units c: cells mm−3; v: virions mm−3; d: day.
The outgrowth rate (net growth advantage) of an escape variant is the rate at which the variant outgrows the wildtype virus. It is defined as the growth rate of the variant minus the growth rate of the wildtype
This list, reproduced in
Viral RNA sequence, HLA class I genotype and viral load were taken from a treatment naïve HIV-1 C-clade infected cohort attending a prenatal clinic in Durban, as described in
Following Moore
For each gene in the HIV-1 genome, all viral amino acid sequences in the cohort were aligned using TCoffee
The number of epitopes with a coding change (NEE) was calculated for each patient, either across all genes, or by separating the epitopes by gene and calculating a NEE score for each gene. HLA-matched epitopes, plus flanking regions of 3 amino acids either side of the epitope, were taken from the K37 HLA-epitope list and compared to the epitope in the cohort consensus amino acid sequence. Any deviation from the cohort consensus was marked as a putative escape mutation and the number of putative escape events (NEE) calculated. The escape event could have happened at any time point prior to sampling (i.e. during the acute or chronic phases, or was already present in the transmitted infecting virus). The number of escape events per epitope is capped at one, i.e. if there are two HLA-associated polymorphisms in the same epitope then this is counted as a single, and not multiple, escape event.
We use NSE to quantify the background level of mutations in an individual. NSE is the equivalent of NEE for synonymous nucleotide changes. The method for obtaining a score for NSE, i.e. epitopes with a silent coding change, is similar to the method for obtaining NEE, save that the individual's viral nucleotide sequence is compared with the cohort consensus nucleotide sequence and only non-coding changes enumerated.
To calculate the impact of HLA polymorphisms on viral load after correcting for the “background” number of mutations, we performed multiple linear regression with log viral load as the dependent variable and NEE and NSE as predictors:
Patients were selected from the Full cohort at random with replacement 10,000 times. For each bootstrap iteration in which total escape events over all genes was a significant predictor of log viral load, escape events were then split by gene to determine escape events in which gene were significant predictors of log viral load. When multiple genes were found to be significant predictors, we checked if they were independent predictors or correlated with escape events in other genes. Sets of independently significant genes contributing less than 1% of the total runs are grouped together as “Minor effects” and excluded from further analysis. Multiple linear regression was used to determine significance, and the number of synonymous changes was always included as a co-predictor. See
To estimate the fraction of viral load variable that can be explained by the HLA class I genotype, we calculated the “Explained Fraction” (EF) metric
Comparison of viral loads from the mathematical model before and after escape. Each line is the log viral load for a single run from the Attenuated Model, from time zero (before escape) to the end of the model run (after escape). A small amount of random noise was added to the x-axis to increase clarity. Note: 400 of the 10,000 runs were randomly chosen to represent this graph, as plotting all 10,000 runs resulted in an incomprehensible figure. The log viral load is significantly lower after escape (paired t-test: p<0.0001), but the size of the decrease is small (mean of log difference is 0.09, 95% confidence interval: 0.086–0.10).
(0.06 MB TIF)
Comparison of corrected and uncorrected viral load with different measures of sequence variation (c.f.
(0.24 MB TIF)
Bootstrap analysis of the Full cohort, sampled with replacement 10,000 times. Panel A shows the distribution of the one-tailed p-value of NEE as an independent predictor of log viral load (multiple linear regression: median two-tailed p-value was 0.0019, 95% confidence interval 0.0018–0.0021; the shaded area shows the proportion of runs where NEE was significant at a one-tailed p-value of ≤0.025). Panel B shows the distribution of the p-value of NSE as an independent predictor of log viral load (multiple linear regression: median two-tailed p-value was 0.030, 95% confidence interval 0.028–0.031; shaded area shows the proportion of runs where NSE was significant at a one-tailed p-value of ≤0.025). Panel C shows the distribution of the difference in viral load of each of the runs (median log difference: 0.1097; 95% confidence interval 0.1091–0.1104).
(0.26 MB TIF)
Distribution of one-tailed p-values from multiple linear regression of each gene in the K37 epitope list. Subjects were sampled with replacement 10,000 times from the Full cohort (N = 157). The shaded area on the left-hand side of each graph indicates the proportion of runs with a statistically significant increase in log viral load (one-tailed p≤0.025), and the shaded area on the right indicates a significant decrease in log viral load (one-tailed p≥0.975). In each panel the median change in viral load associated with an escape event in that gene is presented. See
(0.35 MB TIF)
Descriptions of the mathematical models. In addition to these models which assumed that the variant virus would always be attenuated compared to the wildtype, an ‘equal fitness’ version was run. Here, the variant was assumed to have the same fitness as the wildtype in the absence of CTL. For these models the infection (β) and virion production (h) rates were the same for wildtype and variant virus. Legend. 5D: five population; 6D: six population; MA: mass action; MM: Michaelis-Menten.
(0.03 MB DOC)
Stratification of the Durban cohort by frequent or protective alleles. A frequently occurring allele was defined as occurring in more than 6% of the cohort. An allele was defined as ‘protective’ if the median viral load of all individuals possessing that allele was lower than the median viral load of the cohort, see
(0.03 MB DOC)
Summary of multiple linear regression on the Full cohort on log viral load against the number of escaped epitopes (NEE), stratified by gene. Results which are statistically significant (two-tailed p<0.05) are shown in bold font.
(0.03 MB DOC)
Breakdown by gene the percentage of times escape events predicted log viral load. Percentage of the 10,000 bootstrap runs on the Full cohort where escape events (NEE) in a single gene were a statistically significant predictor of log viral load, independent of the number of synonymous changes (NSE). Also, see
(0.03 MB DOC)
Percentage of the 10,000 bootstrap runs on the Full cohort where escape events in a gene were a significant independent predictor of log viral load, stratified by gene and corrected for the number of synonymous changes (NSE). Where there was a statistically significant association between NEE in all genes and log viral load, the genes driving this association were identified using multiple linear regression, or if none had p-values under 0.05, the gene with the lowest p-value in the regression was chosen. N.B. the totals do not sum to 100% as escape events in multiple genes can be significant independent predictors of log viral load.
(0.03 MB DOC)
K37 HLA-epitope list. This list is taken from Kiepiela
(0.07 MB DOC)
Supplemental information for alternative mathematical models and additional methods and results.
(0.24 MB DOC)
We thank the Los Alamos HIV Database for making the viral sequences and HLA-epitope associations publicly available, and Charles RM Bangham and Marjet Elemans for helpful comments during the preparation of the manuscript. We also thank Jacques Fellay for a very helpful discussion on whole genome associations in HIV.