Fig 1.
Statistical analysis plan with participant selection.
A) Measurement data selection of Swiss HIV Cohort Study (SHCS) participants with human leukocyte antigen (HLA) alleles and next-generation sequencing (NGS) data for analysis Ιa & Ιb, HIV load (VL) data for analysis ΙΙ, and multiple sequences for analysis ΙΙΙ. B) HIV amino acid variants are tested in logistic regressions for their association with HLA alleles and their interaction effect between HLA alleles and average pairwise diversity (APD) (analysis Ιb). Pairs with significant interaction in analysis Ιb were further tested for their effect on VL (analysis ΙΙ), retested in a longitudinal setting (analysis ΙΙΙ), and assessed by computational HLA-epitope binding predictions. C) Pairs were selected if power>0.8 and Fisher’s exact test p-value<0.2 (FDR-corrected). Statistical analyses performed: [Ιa] Presence of HIV variants as function of HLA alleles. [Ιb] Presence of HIV variants as function of HLA alleles, APD, and interaction between HLA and APD. [ΙΙ] VL levels as function of HLA/HIV-variant pairs (identified as significantly associated in [Ιb]); and [ΙΙΙ] longitudinal survival analysis of HLA/HIV-variant pairs identified in [Ιb]. Only samples from ART-naïve participants were used for the grey-shaded analyses ([Ιb], [ΙΙ] and [ΙΙΙ]).
Table 1.
Characteristics of SHCS participants included.
The number of participants used for cross-sectional (Ι and ΙΙ) and longitudinal analysis (ΙΙΙ) and their characteristics are shown for each analysis, restricted to subtype B.
Fig 2.
Associations between HIV amino acid variants, HLA alleles, and average pairwise diversity interactions.
A) Shown are Gag242N:B*57:01 and 98 HLA/HIV-variant pairs with significant average pairwise diversity (APD)-HLA interactions (analysis Ιβ). Odds ratios (OR), sorted by effect size and HIV protein, are derived from Fisher’s exact tests, and colored by HLA genes (class Ι gray, class ΙΙ red). Pairs of interest are highlighted (bold label, triangle symbol). B) Multivariable logistic regressions of eight HLA/HIV-variant pairs of interest are shown. The effect of HLA allele, APD, and their interaction on the occurrence of the viral variant is presented as odds ratios with 95% confidence intervals (OR [CI]). Sample sizes are indicated (n). C) Distribution of the pairs Pol432R:A*03:01 (blue) and Rev57E:B*40:01 (red) within the study population and the likelihood of observing the variant in the presence (dark color) or absence (light color) of the HLA allele over APD is shown. Labels indicate the sample sizes for each group. Dashed lines represent the fitted values of linear regressions, while plain lines show data with standard error. D) Effect of APD on presence of viral variant (estimated OR from analysis Ιβ) in presence/absence of HLA allele. Coloration according to panel B. Positive interaction effects in yellow quadrant (= expected HLA escape mutation). *All estimates exceeding the third quartile+1.5 were adjusted to this threshold. Standardized APD used in panel B and D.
Fig 3.
Viral load association with human leukocyte antigen and HIV variant interaction.
A) Multivariate linear regressions of eight HLA/HIV-variant pairs of interest (same as in Fig 1B), depicted as their association and interaction effect on viral load (VL). B, C) VL distribution over four groups for pair Pol432R:A*03:01 (B) and Rev57E:B*40:01 (C). The absence of human leukocyte antigen (HLA) or HIV variant is coded as ‘0’ and the presence as ‘1’. Each participant is represented by one point (mean log10 VL), and violin plots show distribution, and boxplots show median and IQR of the distribution. Coloration according to panel A. D, E) Multiple linear regression estimates of HIV amino acid (AA) variant on VL (D) or HLA allele on VL (E) and HLA/HIV-variant interaction on VL plotted for Gag242N:B*57:01 and all 98 HLA/HIV-variant pairs with significant average pairwise diversity (APD)-HLA interactions (analysis Ιβ). The 12 HLA/HIV-variant pairs with a significant interaction effect on viral load (p<0.05) are represented by whole symbols and all non-significant (n.s.) by empty symbols. Shapes indicate HIV protein. Coloration according to panel A. Upper-left section in E (yellow) indicates expected HLA escape mutations.
Fig 4.
Cumulative hazard of developing an amino acid variant over time concerning the presence of HLA alleles.
A) Log odds ratios (OR) of interaction effects between APD and HLA on the viral variant versus the Log Hazard Ratio of the respective pair. Significant HLA/HIV-variant interactions (p<0.05) are represented by whole and non-significant (n.s.) by empty symbols. HLA/HIV-variant pairs of interest (same as in Fig 1B) are highlighted. B, C) Cumulative Hazards of acquiring HIV variant Pol432R (B) or Rev57E (C) in presence (dark blue (B) or dark red (C)) of respective HLA allele versus its absence (light blue (B) or light red (C)), are indicated by the respective lines, confidence intervals are illustrated by the shaded area. Censored events are displayed as vertical lines. P-values are shown as log-rank. “Number at risk” indicates the number of participants with outstanding events after the specified time in years.
Fig 5.
HLA-HIV peptide binding prediction based on mass spectrometry eluted ligands (EL) rank (%).
A) Log odds ratio (OR) of the Average Pairwise Diversity (APD)-HLA interaction versus the Log Ratio of EL Rank of mutation/EL Rank of consensus. Upper-right section (yellow) indicates expected major histocompatibility complex (MHC) escape mutations. Labels describe HLA/HIV-variant pairs of interest (same as in Fig 1B) and the epitope position of mutation (ranging from 1 to 9). B, C) Peptides are 9-mer HIV sequences, including the mutation (dark color; Pol432R/ RT277R (B) or Rev57E (C)) or consensus (light color; Pol432K/ RT277K (B) or Rev57G (C)) at different positions (ranging from 1 to 9). The rank was categorized into strong binder (≤0.5%), weak binder (≤2.0%), and non-binder (>2.0%). Non-binding positions are omitted here. All binding prediction computations are derived from NetMHCpan-4.1.