Predictors of durable immune responses six months after the last vaccination in preventive HIV vaccine trials

Abstract

Background: The evaluation of durable immune responses is important in HIV vaccine research and development. The efficiency of such evaluation could be increased by incorporating predictors of the responses in the statistical analysis. In this paper, we investigated whether and how baseline demographic variables and immune responses measured two weeks after vaccination predicted durable immune responses measured six months later.

Methods: We included data from seven preventive HIV vaccine regimens evaluated in three clinical trials: a Phase 1 study of four DNA, NYVAC and/or AIDSVAX vaccine regimens (HVTN096), a Phase 2 study of two DNA and/or MVA vaccine regimens (HVTN205), and a Phase 3 study of a single ALVAC/AIDSVAX regimen (RV144). Regularized random forests and linear regression models were used to identify and evaluate predictors of the positivity and magnitude of durable immune responses.

Results: We analyzed 201 vaccine recipients with data from 10 to 127 immune response biomarkers, and 3-5 demographic variables. The best prediction of participants' durable response positivity based on two-week responses rendered up to close-to-perfect accuracy; the best prediction of participants' durable response magnitude rendered correlation coefficients between the observed and predicted responses ranging up to 0.91. Though prediction performances differed among biomarkers, durable immune responses were best predicted by the two-week response level of the same biomarker. Adding demographic information and two-week response levels of different biomarkers provided little or no improvement in the predictions.

Conclusions: For some biomarkers and for the vaccines we studied, two-week post-vaccination responses can well predict durable responses six months later. Therefore, if immune response durability is only assessed in a sub-sample of vaccine recipients, statistical analyses of durable responses will have increased efficiency by incorporating two-week response data. Further research is needed to generalize the findings to other vaccine regimens and biomarkers. Clinicaltrials.gov identifiers: NCT01799954, NCT00820846, NCT00223080.

Keywords: Binding antibody multiplex array; Immunogenicity; Intracellular cytokine staining; Regularised random forest; Statistical power.

Conflict of interest statement

Conflict of interest

No conflicts of interest.

Copyright © 2017 Elsevier Ltd. All rights reserved.

Figures

Fig. 1

Individual immune response trajectories from wk26 to wk52 across the 7 vaccine regimens in HVTN096, HVTN205 and RV144. Net IgG BAMA responses to three different HIV antigens (gp120, gp41, and V1V2) (Panel A) and CD4+ ICS immune responses to two different HIV antigens (Env, Gag) (Panel B) were assessed using serum samples (Panel A) or peripheral blood mononuclear cells (Panel B) from vaccine recipients in HVTN096, HVTN205, and RV144 with n = 15–18/group, 32–46/group, and 38–40, respectively. Note that RV144 BAMA assays used a 1:40 dilution and responses were baseline-subtracted, whereas HVTN096 and HVTN204 BAMA assays used a 1:50 dilution and responses were not baseline-subtracted. Red and blue dots indicate positive and negative responses, respectively. Responses for biomarkers with a wk52 response rate 1N/N/NA/NA, 2NA/NA/NA/NA, 3D/D/NA/NA, 4DA/DA/NA/NA, 5DDMM, 6MMM. N, NYVAC (Viral Vector – Pox); NA, NYVAC/AIDSVAX (Viral Vector – Pox/Protein); D, DNA (DNA-based); DA, DNA/AIDSVAX (DNA/Protein); DDMM, DNA/DNA/modified vaccinia virus Ankara (MVA)/MVA; MMM, MVA/MVA/MVA. More details on the vaccine regimens can be found in Table 1. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 2

Spearman correlation coefficients between observed wk26 vs. wk52 responses of the same biomarker or of the observed wk26 response of one biomarker vs. the observed wk52 response of another biomarker, as measured by Cellular and Antibody assays. Spearman correlation coefficients were determined for observed immune responses of the same biomarker at wk26 vs. wk52 (Panel A) or for the observed wk26 response of one biomarker vs. the observed wk52 response of another biomarker (Panel B) as measured by Cellular (ELISpot, ICS, and Luminex) and Antibody (binding antibody, avidity, and neutralizing antibody) assays. Correlation coefficients were determined for responses in vaccine recipients in HVTN096, HVTN205 and RV144 who received all planned vaccinations and who were HIV-uninfected at the time of immune response measurement. ELISpot, enzyme-linked immunospot; ICS, intracellular cytokine staining.

Fig. 3

Scatter plots of RRF-predicted vs. observed antibody responses in RV144 comparing the ability of the two RRF models to predict wk52 responses. Serum IgA antibody responses of RV144 vaccine recipients to gp120 (Panels A and B; 1A244gp120gDpos293Tmon and 2GNE8rgp120Qpool) and serum IgG antibody responses to V1V2 (Panel C; 3BCaseA2V1V2mut3) antigens were determined at wk26 and wk52. RRF models were used to predict the wk52 responses using either the wk26 response of the same biomarker (left panels, “Same WK26 Response”) or using the wk26 response of the same biomarker in addition to the wk26 responses of all other measured biomarkers (right panels, “Same + Other WK26 Responses”). ρ indicates the Spearman correlation coefficient between the observed and predicted wk52 response magnitudes. P-value indicates the significance of ρ being different from zero. A Lowess smoother line using locally-weighted polynomial regression was included in each panel. Ig, immunoglobulin; RRF, regularized random forest.