Joint epitope selection and spacer design for string-of-beads vaccines

Abstract

Motivation: Conceptually, epitope-based vaccine design poses two distinct problems: (i) selecting the best epitopes to elicit the strongest possible immune response and (ii) arranging and linking them through short spacer sequences to string-of-beads vaccines, so that their recovery likelihood during antigen processing is maximized. Current state-of-the-art approaches solve this design problem sequentially. Consequently, such approaches are unable to capture the inter-dependencies between the two design steps, usually emphasizing theoretical immunogenicity over correct vaccine processing, thus resulting in vaccines with less effective immunogenicity in vivo.

Results: In this work, we present a computational approach based on linear programming, called JessEV, that solves both design steps simultaneously, allowing to weigh the selection of a set of epitopes that have great immunogenic potential against their assembly into a string-of-beads construct that provides a high chance of recovery. We conducted Monte Carlo cleavage simulations to show that a fixed set of epitopes often cannot be assembled adequately, whereas selecting epitopes to accommodate proper cleavage requirements substantially improves their recovery probability and thus the effective immunogenicity, pathogen and population coverage of the resulting vaccines by at least 2-fold.

Availability and implementation: The software and the data analyzed are available at https://github.com/SchubertLab/JessEV.

Supplementary information: Supplementary data are available at Bioinformatics online.

Fig. 1.

Conceptual steps in EV design with the proposed framework. (a) Epitopes are extracted from a given set of antigens, and properties such as immunogenicity, coverage and conservation are computed. (b) We formulate an MILP that creates a string-of-beads vaccine by simultaneously selecting which epitopes to include and how to assemble them into the final construct. This formulation maximizes the immunogenicity of the selected epitopes subject to constraints related to patient and pathogen coverage of the resulting vaccine, as well as cleavage probability of specific residues. To connect the selected epitopes, spacers are designed to provide a high chance of cleavage at the termini of the epitopes, and epitopes that have too high a cleavage probability in their interior are discarded. The vaccine will be subject to proteolytic digestion, which has strong effects on its efficacy. To quantify these effects, (c) we perform repeated stochastic simulations of proteasomal cleavage and estimate the probability that each epitope is correctly recovered from the string-of-beads construct. (d) Based on the recovered epitopes, the vaccine is evaluated in terms of the average immunogenicity of the recovered epitopes, as well as coverage and conservation with respect to the original antigens and/or the target population. We show that approaching selection and assembly together increases the number of epitopes correctly recovered from the vaccine making the vaccine itself more effective

Fig. 2.

Comparison of cleavage scores between JessEV and a sequential approach. (a) The cleavage scores of residues at the termini, inside the epitopes and inside the spacers for 30 vaccines designed on random subsets of 5000 epitopes. We are able to enforce a strict separation with a clear gap between the scores of residues inside the epitopes and at the termini. (b) How many cleavage events, as predicted by NetChop (Nielsen et al., 2005), happened at the termini, inside the epitopes and inside the spacers in the same bootstraps used in (a). The marked differences are statistically significant ($***<0.001$). (c) The cleavage scores for each residue of a string-of-beads vaccine designed on the complete set of epitopes with a sequential approach and (d) with JessEV. The spacers are highlighted in green, and the gray vertical lines represent cleavage frequencies as computed by Monte Carlo simulations with a prior of 0.15, with darker shades being more likely. The title reports both theoretical and effective immunogenicity. Thanks to higher minimum cleavage at the termini and lower maximum cleavage inside the epitopes and spacers, the effective immunogenicity of our vaccine is about twice that of the sequential approach, even though the individual epitopes are less immunogenic

Fig. 3.

Evaluation of string-of-beads designed with JessEV and a sequential approach across different prior cleavage probabilities. (a), (b), (c) and (d) Mean, 25th and 75th percentile of the Monte Carlo simulations for effective immunogenicity, recovered epitopes, pathogen coverage and HLA coverage, respectively. JessEV was better under all metrics across all choices of prior cleavage probabilities. Note that both vaccines optimized for immunogenicity in (a) and (b), and for coverage in (c) and (d), which means that different constraints were used to produce them. (e) and (f) The probability of worsening and expected improvement of effective immunogenicity, effective pathogen coverage and effective HLA coverage. Both were estimated through 5000 bootstrap of the outcomes of the 1000 Monte Carlo simulations. String-of-bead vaccines produced by JessEV were very frequently not worse than the sequential approach, and on average between three to five times better across a realistic range of prior probabilities. At cleavage probabilities larger than 0.7, no epitopes were ever recovered for the sequential approach; hence, the expected improvement approached infinity

Fig. 4.

The effects of cleavage constraints on the immunogenicity objective and effective immunogenicity. (a) For each prior probability, we show the effective immunogenicity relative to the best obtained for that prior (y-axis) for different parameter settings ranked by effective immunogenicity (x-axis). There is a range of prior probabilities, from 0.1 to 0.3, where four or five different parameter settings were within 5% of the largest effective immunogenicity. (b) Effective immunogenicity (y-axis) as a function of the inner epitope cleavage (x-axis) for different cleavage at the termini (lighter lines for larger constraints) and nine different prior cleavage probabilities (in each sub-figure). For prior cleavage probabilities in a reasonable range, the best effective immunogenicity was obtained with an inner epitope cleavage around zero, while lower settings worked best for high priors and larger ones for low priors. (c) The effect of prior cleavage probability (annotated close to each data point) on parameters (x- and y-axes) that resulted in the largest effective immunogenicity (blue) or effective pathogen coverage (orange). Only transitions are displayed, meaning that several prior probabilities between, e.g. 0.15 and 0.5 (not included) had the same optimal settings for the effective immunogenicity. As the prior cleavage probability increased, constraints on cleavage at the termini could be relaxed, while the score inside the epitopes must be kept lower. Optimizing for effective coverage required larger possible cleavage likelihoods inside the epitopes, but similar cleavage likelihoods at the termini. (d) Immunogenicity objective for different cleavage constraints, with light background for infeasible settings. Enforcing low cleavage likelihoods inside the epitopes greatly reduced the immunogenicity objective, as many epitopes are not eligible due to higher cleavage likelihoods in the residues of their second half, which cannot be reduced through the preceding spacer under our cleavage model