Abstract
In the ENSEMBLE randomized, placebo-controlled phase 3 trial (NCT04505722), estimated single-dose Ad26.COV2.S vaccine efficacy (VE) was 56% against moderate to severe-critical COVID-19. SARS-CoV-2 Spike sequences were determined from 484 vaccine and 1,067 placebo recipients who acquired COVID-19. In this set of prespecified analyses, we show that in Latin America, VE was significantly lower against Lambda vs. Reference and against Lambda vs. non-Lambda [family-wise error rate (FWER) p < 0.05]. VE differed by residue match vs. mismatch to the vaccine-insert at 16 amino acid positions; significantly decreased with physicochemical-weighted Hamming distance to the vaccine-strain sequence for Spike, receptor-binding domain, N-terminal domain, and S1 (FWER p < 0.001); and decreased (p = 0.011) with neutralization resistance level to vaccinee sera. VE against severe-critical COVID-19 was stable across most sequence features but lower against the most distant viruses.
Introduction
Most COVID-19 vaccines were designed using the Spike protein sequence from the original "Reference" strain of SARS-CoV-2. As the virus evolved into variants like Delta, Lambda, and Mu, vaccine efficacy (VE) began to wane. While clinical trials established that efficacy was dropping, a precise, quantitative understanding of which sequence features mattered most was lacking. This study, published in Nature Communications, employs a "sieve analysis" on the ENSEMBLE Phase 3 trial data to quantify how specific mutations and overall genetic "distance" from the original vaccine strain correlate with a drop in protection.
Methods
The researchers utilized data from the ENSEMBLE Phase 3 trial, which enrolled over 40,000 participants across Latin America, South Africa, and the United States. They analyzed 1,345 SARS-CoV-2 Spike amino acid sequences from 1,224 participants who developed COVID-19.
The primary analytical tool was sieve analysis, a statistical approach comparing viral sequences of breakthrough infections in vaccinated individuals against those in a placebo group. Key metrics included:
- Physicochemical-weighted Hamming distance: Measuring the "distance" between protein sequences, weighting changes based on amino acid property alterations.
- Antibody-escape scores: Using Deep Mutational Scanning (DMS) data to predict how mutations help the virus evade antibodies.
- Feature-specific proportional-hazards models: To estimate VE as a function of specific sequence traits.
Results
Variant-Specific Efficacy
The study found that VE is not a binary state but a gradient. In Latin America, where viral diversity was highest during the trial, VE varied significantly by lineage. While protection against the Reference strain and Zeta variant remained relatively high, efficacy against the Lambda variant dropped precipitously.
The "Sieve" Effect and Genetic Distance
A clear relationship emerged: as the physicochemical distance from the vaccine strain increased, VE against moderate-to-severe disease declined. The sieve analysis demonstrated that the vaccine was more effective at preventing infections from viruses genetically similar to the vaccine strain, leading to an enrichment of "distant" sequences in the breakthrough cases of the vaccine group compared to the placebo group.
Structural Insights: The Lambda Escape
The Lambda variant was identified as the "most distant" virus in the study. Structural analysis revealed a unique deletion/mutation combination (RSYLTPGD246-253N) in the N-terminal domain (NTD). This change allows the virus to add a new glycan molecule, effectively shielding a major antibody target site. Interestingly, while the S1 region evolved rapidly to escape antibodies, the S2 region (the "stalk") remained conserved, likely explaining why the vaccine remained more effective at preventing severe-critical disease than moderate disease.
Discussion
These findings provide a roadmap for predicting how future variants might impact vaccine performance. Computational "escape scores" based on laboratory antibody-binding data were highly predictive of clinical efficacy, validating these lab models as reliable proxies for real-world performance.
While the analysis was limited to the double-blind period and preceded the emergence of Omicron, the methods established—specifically the use of weighted Hamming distances and DMS escape scores—provide a robust framework for analyzing current and future variants. Future work will apply these sieve models to bivalent and variant-specific boosters to determine if they provide a broader shield against viral evolution.
Summary updated