Selection of HIV vaccine candidates for concurrent testing in an efficacy trial

Abstract

Phase IIb or III HIV-1 vaccine efficacy trials are generally large and operationally challenging. To mitigate this challenge, the HIV Vaccine Trials Network is designing a Phase IIb efficacy trial accommodating the evaluation of multiple vaccine regimens concurrently. As this efficacy trial would evaluate a limited number of vaccine regimens, there is a need to develop a framework for optimizing the strategic selection of regimens from the large number of vaccine candidates tested in Phase I/IIa trials. In this paper we describe the approaches for the selection process, including the choice of immune response endpoints and the statistical criteria and algorithms. We illustrate the selection approaches using data from HIV-1 vaccine trials.

Figure 1

Down-selection scheme

Figure 2

Lists of immune classes for the second step of down-selection. (a) Inclusive set of immune classes, which are putatively part of a CoP based on one of the following conditions: i) a known CoP for one or more licensed vaccines; ii) evidence as a CoP based on CoR analysis of the RV144 trial and assumptions for assessing CoRs as CoPs; or iii) evidence as a CoP in non-human primates challenge trials for HIV-1 vaccines. (b) Subset of significant inverse CoRs of HIV-1 infection in RV144 for Step 2 [weighting by strength of correlation]

Figure 2

Lists of immune classes for the second step of down-selection. (a) Inclusive set of immune classes, which are putatively part of a CoP based on one of the following conditions: i) a known CoP for one or more licensed vaccines; ii) evidence as a CoP based on CoR analysis of the RV144 trial and assumptions for assessing CoRs as CoPs; or iii) evidence as a CoP in non-human primates challenge trials for HIV-1 vaccines. (b) Subset of significant inverse CoRs of HIV-1 infection in RV144 for Step 2 [weighting by strength of correlation]

Figure 3

Demonstration of the superiority and non-redundancy criteria. (a) Suppose six regimens with mean immune endpoint scores enter the down-selection. The pairwise relationship is shown in (b), where → indicates that one regimen is superior to the other with respect to their immune profiles, with the arrow pointing to the inferior regimen; and ↔ indicates equivalence between two regimens with respect to their immune profiles. Panels (c)–(g) display some possible outcomes based on the down-selection. The non-redundancy criterion is satisfied if there is not a → or ↔ among the selected set. Here the non-redundancy criterion is satisfied in (f) and (g) but not in (c)–(e). Among the two panels where the non-redundancy criterion is satisfied, the superiority criterion is also satisfied in (g) but not in (f). For the latter, regimen B is selected but there is a regimen A superior to B in the original set before the down-selection.

Figure 4

Results of the data example. (a) Assay-specific mean immune endpoint scores for five HIV-1 vaccine regimens. The eight endpoint scores are: IgG binding antibody responses to six different gp120 antigens measured using the binding antibody multiplex assay (BAMA) (log-transformed blank-subtracted MFI readout at a 1:50 dilution), mean NAb response (log-transformed ID50 titer) to six HIV-1 isolates measured using the TZM-bL assay and the A3R5 assay, and CD4+ T-cell response measured using the intracellular cytokine staining (ICS) assay (log-transformed % of T cells expressing IFN-γ and /or IL-2). The scores have been scaled by the standard deviations of the corresponding endpoint measures in RV144.T. (b) Principal component (PC) biplot by regimen. The x-axis is the value from the first principal component and the y-axis is the second principal component, where each axis label includes the percentage of variation in the total set of readouts captured by the principal component. Points on the plot represent the values of the principal components of each observation. Points that are close together correspond to observations that have similar values on the components displayed in the plot. The top axis is the first principal component loadings and the right axis is the second principal component loadings, where loadings are the weights by which each original immunogenicity endpoint score should be multiplied to get the value of the first or second principal component. An arrow (vector) is drawn for each immunogenicity endpoint from the origin to the point defined by its first two principal component loadings. Vectors that point in the same direction correspond to endpoints that have similar response profiles based on the first two PCs. The observations whose points project furthest in the direction in which the vector points are the observations that have the most weight of whatever the endpoint measures. Those points that project at the other end have the least weight of whatever the endpoint measures. The angle between two arrows conveys information about the correlation of the two endpoint scores, with a zero degree angle denoting perfect correlation and a 90 degree angle denoting no correlation. (c) Heatmap of the mean immune endpoint scores for the vaccine regimens. The number of clusters is estimated to be 4 based on the GAP statistics proposed in Tibshirani et al. (2001) [29]. The value for each endpoint score has been centered by the average of mean scores for this particular endpoint across regimens. Hierarchical clustering analysis is performed based on the complete linkage method; the distance matrix is generated as weighted Manhattan distance of mean endpoint scores. The blue vertical line cuts the hierarchical tree into four clusters. (d) Process of down-selection by RFS and CR+RFS.

Figure 4

Results of the data example. (a) Assay-specific mean immune endpoint scores for five HIV-1 vaccine regimens. The eight endpoint scores are: IgG binding antibody responses to six different gp120 antigens measured using the binding antibody multiplex assay (BAMA) (log-transformed blank-subtracted MFI readout at a 1:50 dilution), mean NAb response (log-transformed ID50 titer) to six HIV-1 isolates measured using the TZM-bL assay and the A3R5 assay, and CD4+ T-cell response measured using the intracellular cytokine staining (ICS) assay (log-transformed % of T cells expressing IFN-γ and /or IL-2). The scores have been scaled by the standard deviations of the corresponding endpoint measures in RV144.T. (b) Principal component (PC) biplot by regimen. The x-axis is the value from the first principal component and the y-axis is the second principal component, where each axis label includes the percentage of variation in the total set of readouts captured by the principal component. Points on the plot represent the values of the principal components of each observation. Points that are close together correspond to observations that have similar values on the components displayed in the plot. The top axis is the first principal component loadings and the right axis is the second principal component loadings, where loadings are the weights by which each original immunogenicity endpoint score should be multiplied to get the value of the first or second principal component. An arrow (vector) is drawn for each immunogenicity endpoint from the origin to the point defined by its first two principal component loadings. Vectors that point in the same direction correspond to endpoints that have similar response profiles based on the first two PCs. The observations whose points project furthest in the direction in which the vector points are the observations that have the most weight of whatever the endpoint measures. Those points that project at the other end have the least weight of whatever the endpoint measures. The angle between two arrows conveys information about the correlation of the two endpoint scores, with a zero degree angle denoting perfect correlation and a 90 degree angle denoting no correlation. (c) Heatmap of the mean immune endpoint scores for the vaccine regimens. The number of clusters is estimated to be 4 based on the GAP statistics proposed in Tibshirani et al. (2001) [29]. The value for each endpoint score has been centered by the average of mean scores for this particular endpoint across regimens. Hierarchical clustering analysis is performed based on the complete linkage method; the distance matrix is generated as weighted Manhattan distance of mean endpoint scores. The blue vertical line cuts the hierarchical tree into four clusters. (d) Process of down-selection by RFS and CR+RFS.