SARS-CoV-2 variants associated with vaccine breakthrough in the Delaware Valley through summer 2021

Abstract

The severe acute respiratory coronavirus-2 (SARS-CoV-2) is the cause of the global outbreak of COVID-19. Evidence suggests that the virus is evolving to allow efficient spread through the human population, including vaccinated individuals. Here we report a study of viral variants from surveillance of the Delaware Valley, including the city of Philadelphia, and variants infecting vaccinated subjects. We sequenced and analyzed complete viral genomes from 2621 surveillance samples from March 2020 to September 2021 and compared them to genome sequences from 159 vaccine breakthroughs. In the early spring of 2020, all detected variants were of the B.1 and closely related lineages. A mixture of lineages followed, notably including B.1.243 followed by B.1.1.7 (alpha), with other lineages present at lower levels. Later isolations were dominated by B.1.617.2 (delta) and other delta lineages; delta was the exclusive variant present by the last time sampled. To investigate whether any variants appeared preferentially in vaccine breakthroughs, we devised a model based on Bayesian autoregressive moving average logistic multinomial regression to allow rigorous comparison. This revealed that B.1.617.2 (delta) showed three-fold enrichment in vaccine breakthrough cases (odds ratio of 3; 95% credible interval 0.89-11). Viral point substitutions could also be associated with vaccine breakthroughs, notably the N501Y substitution found in the alpha, beta and gamma variants (odds ratio 2.04; 95% credible interval of 1.25-3.18). This study thus provides a detailed picture of viral evolution in the Delaware Valley and a geographically matched analysis of vaccine breakthroughs; it also introduces a rigorous statistical approach to interrogating enrichment of viral variants.

Keywords: COVID-19; Philadelphia; SARS-CoV-2; coronavirus; genome sequencing.

Figure 1.

Longitudinal data from the COVID-19 pandemic in the city of Philadelphia. The y-axis shows the daily test positivity rate (light grey) as a percent of the highest value (26.57% positivity on 4/13/2020), the hospitalization rate (dark grey) as a percent of the highest value (87 hospitalizations per day on 4/22/2020), the vaccination rate in adults 18 years old and older (black). The estimated percent of surveillance samples classified as alpha (green) or delta (red) variant was estimated from the sequence data presented in this paper. Other data is from the city of Philadelphia “Testing Data: Programs and Initiatives.”

Figure 2.

Comparison of viral genome sequence data from surveillance samples (A, B) to spike target gene failures (C) and vaccine breakthrough samples (D). A) Longitudinal stacked bar graph depicting the SARS-CoV-2 variants present in surveillance samples from the Delaware Valley, shown as the proportion of genomes classified as each variant lineage within each week. The numbers of genomes sampled each week are shown above the graph. Variants are colored according to the key at the bottom of the figure. B) Markings are the same as in A), but showing the proportions of variants estimated from the count data in A) using Bayesian autoregressive moving average multinomial logistic regression. C) Markings as in A), but showing counts of spike target gene failures samples. D) Markings as in A), but showing the counts of vaccine breakthrough samples.

Figure 3.

Frequencies of individual variants estimated using Bayesian autoregressive moving average multinomial logistic regression. Time is shown along the x-axis and estimated proportions of the surveillance population along the y-axis. The grey bars indicate raw proportions from the count data shaded by the number of observations observed in a given week (darker indicating more samples) while the colored lines indicate the proportion estimated by the Bayesian model. The light colored envelopes around each line show the 95% credible intervals for the proportion. Only lineages achieving an estimated proportion of >10% in any given week are shown.

Figure 4.

Estimated posterior probability densities for the enrichment of variants among spike gene target failures produced by Bayesian autoregressive moving average multinomial logistic regression. The x-axis shows the fold enrichment/depletion in the odds of a variant (labeled above each plot) appearing in the spike gene target failure set relative to the proportions estimated in the surveillance population and the y-axis shows the posterior probability. No enrichment (fold-change of 1) is indicated by the dashed vertical line. Increased density to the right indicates greater likelihood of inclusion among spike gene target failures, increased density to the left indicates decreased likelihood. The four lineages with highest posterior probability of enrichment are shown on the top and four lineages with high posterior probability of depletion are shown on the bottom. Color coding indicates the variant queried with colors as in earlier figures.

Figure 5.

Posterior probability densities for enrichment of variants among vaccine breakthrough samples compared to the surveillance population as estimated by Bayesian autoregressive moving average multinomial logistic regression. Markings as in Figure 4.

Figure 6.

Assessment of enrichment of specific base substitutions and deletions in vaccine breakthrough samples. A) Longitudinal frequencies of individual mutations estimated using Bayesian autoregressive moving average logistic regression models. Red indicates mutations commonly found in delta lineages. Green indicates mutations commonly found in alpha lineages. Orange indicates mutations shared by almost all lineages in the study. Black indicates mutations found in other subsets of lineages. B) Estimated posterior probability densities summarizing the fold enrichment/depletion in odds of a mutation appearing in the vaccine breakthrough samples over its proportions estimated from surveillance samples. The mutations estimated as most enriched and most depleted are shown along with other mutations of interest. Markings as in Figures 4 and 5.