The development of a mathematical modelling framework to translate TB vaccine responses between species and predict the most immunogenic dose in humans using animal data

Abstract

Background: Preclinical animal experiments measuring vaccine immunogenicity and safety are essential, not only to establish if the vaccine should progress further, but to generate information on how the vaccine should be administered in humans. Animal models that represent human vaccine responses well are vital to translate information about vaccine dose to clinical phases. Vaccine dose is a key aspect in creating an effective vaccine. However, if the wrong dose is chosen, vaccine candidates may be mistakenly discarded and considerable resources wasted. Current methods of finding optimal vaccine dose are mostly empirically based, which may be leading to sub-optimal doses progressing into later clinical trials. A current example of this is in the tuberculosis (TB) vaccine developmental pipeline, where a series of adjuvanted subunit vaccines, the H-series, have progressed through to later stages of clinical development with a high dose that has been shown to less immunogenic than lower doses. In drug development, mathematical model-based methods are routinely used alongside empirical evaluations, to inform dose-finding. I hypothesised that vaccine development may benefit from the application of similar quantitative methods. As such, I launched the new field of vaccine immunostimulation/immunodynamic (IS/ID) mathematical modelling. My aims for this thesis were 1) to establish differences in Bacillus Calmette–Guerin (BCG) Interferon-Gamma (IFN-γ) response by human subpopulation, then develop a IS/ID model to represent these response dynamics and identify the most representative macaque subpopulation for human BCG responses. Aim 2) was to predict human H-series vaccine IFN-γ response using IS/ID model calibrated to mouse multi-dose IFN-γ data and allometric scaling. Methods: For aim 1, longitudinal data on IFN-γ emitting CD4+ T cells following vaccination BCG were available in humans and macaques. Human (sub)population covariates were: baseline BCG vaccination status, time since BCG vaccination, gender and monocyte/lymphocyte cell count ratio. The macaque (sub)population covariate was colony of origin. I developed a two-compartmental mathematical model describing the post-BCG IFN-γ immune response dynamics. The model was calibrated to the human and macaque data using 4 Nonlinear Mixed Effects Modelling (NLMEM) to establish if there were differences in IFN-γ dynamics for both species subpopulations. I then established which macaque subpopulation best described human data. For aim 2, longitudinal data on IFN-γ emitting CD4+ T cells following two vaccinations with five doses of novel TB vaccine H56+IC31 in mice were generated. I then assessed the shape of the dose response curve at early and late time points. I calibrated the T cell model to the mouse data and established the change in key model parameters across dose. Using the change in model parameters across dose found in the mice, I predicted the immune response dynamics in humans for different doses and which dose was most immunogenic. Results: In aim 1, I found that BCG status in humans (baseline BCG-naive or baseline BCG-vaccinated) was associated with differences in the peak and end IFN-γ response after vaccination with BCG. When the mathematical model was calibrated to the BCG data for both macaques and humans, significant differences (p<0.05) in key model parameters were found after stratification by macaque colony and human baseline-BCG status. Indonesian cynomolgus macaques had the closest immune response dynamics to the baseline BCG-naive humans. In aim 2, a peaked curve was the best description of the mouse H56+IC31 dose response curve for early and late time points. Calibrating a revaccination model to the data and mapping changes in the estimated mouse model parameters across dose group to the estimated human model parameters, I found at day 224 (a latest time point), the model-predicted median number of human IFN-γ secreting CD4+ T cells were the highest for the dose group in the range 1-10μg H56/H1+500 nmol IC31. This suggests a dose of 1-10μg may be the most immunogenic in humans. Discussion: Finding the most predictive animal model and optimal vaccine dose is essential for efficiently accelerating the development of new, effective, TB vaccines. I demonstrated that mathematical modelling was a useful tool to quantify BCG immune response dynamics in macaques and humans. I established which macaque subpopulation should be used to represent a human BCG (or potentially new TB vaccine) induced IFN-γ response in future clinical trials. Using IFN-γ as marker of vaccine immunogenicity, mathematical modelling predictions using preclinical data suggested that doses in current novel TB vaccines clinical 5 trials on healthy BCG-vaccinated participants should be between 1-10μg H56/H1+500 nmol IC31, a result which has been recently corroborated in an empirical H56+IC31 dose-ranging trial. This project has demonstrated the potential utility of mathematical modelling in vaccine development. I believe future work on IS/ID modelling should include data on more complex immune response networks and different animal and human subpopulations. This future work is entirely feasible and would establish IS/ID modelling as a legitimate tool to accelerate vaccine development.