HIV transmission. Selection bias at the heterosexual HIV-1 transmission bottleneck

Abstract

Heterosexual transmission of HIV-1 typically results in one genetic variant establishing systemic infection. We compared, for 137 linked transmission pairs, the amino acid sequences encoded by non-envelope genes of viruses in both partners and demonstrate a selection bias for transmission of residues that are predicted to confer increased in vivo fitness on viruses in the newly infected, immunologically naïve recipient. Although tempered by transmission risk factors, such as donor viral load, genital inflammation, and recipient gender, this selection bias provides an overall transmission advantage for viral quasispecies that are dominated by viruses with high in vivo fitness. Thus, preventative or therapeutic approaches that even marginally reduce viral fitness may lower the overall transmission rates and offer long-term benefits even upon successful transmission.

Fig. 1. HIV-1 viruses with amino acid residues matching the consensus of the study population are preferentially transmitted

For each linked transmission couple, the proportion of sites that were transmitted was defined to be the proportion of sites in which the variant observed in the recipient matched that observed in the donor. Sites with a mixture in the recipient were excluded. (A) Donor variants that matched cohort consensus were more likely to be transmitted among all non-mixture sites in the donor, while (B) a consensus residue that was observed in mixture with one other variant was more likely to be transmitted than was the other variant. A nucleotide mixture was called if more than one base resulted in a >25% Sanger peak height. An amino acid mixture was called if the nucleotide mixture resulted in a mixture of amino acids. Dashed gray line represents the expected frequency of transmission of consensus. Consensus was defined to be any amino acid observed in at least 50% of chronically infected individuals in the Zambian cohort.

Fig. 2. Viral fitness modulates selection bias in heterosexual HIV-1 transmission

The odds that the donor’s amino acid will be transmitted to the recipient is a function of the relative frequency of the amino acid in the quasispecies as well as the fitness of that amino acid, as estimated here by several independent metrics. Each plot shows the empirical transmission probability (odds on a log₁₀ scale) of a variant as a function of one or more parameters. Empirical transmission probabilities (solid colored lines) are estimated counting the proportion transmitted within a continuous sliding window of width 1 log-odds with respect to the feature represented on the abscissa. All log-odds values are smoothed by adding a pseudo-count. Grey lines represent a quadratic fit to the sliding window averages; shaded areas represent 95% confidence intervals estimated using the percentile-t method on 1000 multilevel bootstraps. P-values are taken from Table 1 (A) or Table 2 (D–F) and represent the p-value from a multilevel logistic regression model in which all features are treated as continuous variables, as described in Materials and Methods. (A) The log-odds of transmission is linearly related to the relative in vivo frequency of the variant in the donor quasispecies, with a near 1-to-1 mapping for variants that match cohort consensus. In contrast, polymorphisms are uniformly less likely to be transmitted (N=8314 observations over 5 couples). (B) Among N=3,115 donor sites containing two-amino acid mixtures from 137 couples, the probability of transmission is also strongly predicted by the relative cohort frequency of the amino acid. Transmission probability is with respect to a randomly chosen member of the mixture; the abscissa represents the relative frequency of that amino acid in the cohort compared to the other amino acid in the mixture. (C–F) Among N=228,362 non-mixture donor sites from 137 couples, the odds of transmission is predicted by: (C) the frequency of the amino acid in the cohort; (D) the relative impact of the variant on the stability of the protein structure (low impact implies high fitness); (E) the number of covarying sites statistically linked with the variant; and (F) whether the variant is consistent with immune escape from one of the donor’s HLA alleles (only polymorphic sites are shown). See methods for feature definitions.

Fig. 3. Transmission risk factors reduce selection bias in heterosexual HIV-1 transmission

The empirical log-odds of transmission is plotted as a function of the frequency of each variant in the cohort, as defined in Figure 2. Donor viral load (VL) near the time of transmission, sex of the recipient, and presentation of genital ulcers or inflammation (GUI) in male recipient partners each affect the selection bias. (A–C) Individuals are segregated by donor VL levels used in previous studies of transmission risk (27, 28). High donor VL reduces transmission selection bias in (B) female-to-male, but not (C) male-to-female, transmission. (D–F) Male recipients appear to have increased selection bias compared to female recipients, an effect that is mitigated by the presence of GUI (D, E) or high donor VL (F). 95% confidence intervals (shaded area) and quadratic polynomial fit (solid gray lines) were estimated as in Figure 2. P-values are estimated from a non-parametric, block-bootstrap method that tests the null hypothesis that the normalized area under two curves are identical (see methods for details).

Fig. 4. Transmission of low-fitness viruses changes reversion dynamics and predicts lower early set-point viral load in the recipient

(A) The proportion of donor non-consensus, polymorphisms that remain polymorphic, is plotted as a function of days after the first available recipient sample (N=6,220 polymorphisms from 81 couples). The ordinate at time 0 represents the fraction of donor polymorphisms that were transmitted. Female recipients permit transmission of more polymorphisms than males, but these revert at a faster rate. Hazard ratio and p-value was taken from a multivariable Cox-proportional hazard model (see Table S2). See Figure S4 for estimates of instantaneous reversion rates as a function of time since the estimated date of infection. (B) The number of transmitted polymorphisms at sites that were polymorphic in the linked donor negatively correlates with early set-point VL in the recipient (N=81), corroborating the fitness cost imposed by many of these variants.

Fig. 5. Sequence-derived transmission index predicts transmission

The transmission index of a sequence was calculated as the mean of the expected log-odds of transmission for each site in the sequence, as estimated by logistic regression model that included a second-order polynomial of cohort frequency, the number of covarying sites, and offsets and cohort frequency interactions for each protein domain. Transmission indices were computed out-of-sample using leave-one-out cross validation. (A–B) The transmission index for individual donor Gag amplicons compared to the transmission index for the linked founder Gag sequence. Models trained on Gag alone. (A) Transmission index for each couple. Black bar represents median transmission index of donor sequences. Four sequences with transmission index < 6.7 were excluded as outliers. (B) The transmission index of each founder virus, median-centered against the transmission indices for linked donor sequences (p-value from two-tailed Wilcoxon signed-rank test). (C) The overall transmission index of Gag, Pol and Nef is significantly different between donor and potential source partners in risk-matched discordant couples (p-value from two-tailed Mann-Whitney rank-sum test).