Drivers and impact of the early silent invasion of SARS-CoV-2 Alpha

Abstract

SARS-CoV-2 variants of concern (VOCs) circulated cryptically before being identified as a threat, delaying interventions. Here we studied the drivers of such silent spread and its epidemic impact to inform future response planning. We focused on Alpha spread out of the UK. We integrated spatio-temporal records of international mobility, local epidemic growth and genomic surveillance into a Bayesian framework to reconstruct the first three months after Alpha emergence. We found that silent circulation lasted from days to months and decreased with the logarithm of sequencing coverage. Social restrictions in some countries likely delayed the establishment of local transmission, mitigating the negative consequences of late detection. Revisiting the initial spread of Alpha supports local mitigation at the destination in case of emerging events.

Fig. 1. Factors associated with the pattern of observed Alpha dissemination.

A Change in outbound international traffic from the UK over time, including air-travel, train, ferry and Channel Tunnel. The 73 countries contributing to GISAID during 1 Sep 2020–31 Dec 2020 are shown as an example. Traffic is rescaled to the maximum over the period. To improve readability, different months of traffic maximum are associated with a different color. B Date of first detection, i.e. collection of the first Alpha sequence submitted to GISAID, for each of the 73 countries as in (A), according to sequencing coverage and international traffic (passengers/day) averaged over 1 Sep 2020–31 Dec 2020. For each day, the sequencing coverage of a country is defined as the number of collected SARS-CoV-2 sequences on GISAID—regardless the date of submission—divided by reported cases. The dashed line provides a guide to the eye, as, under simplifying assumptions^,, we expect the date of first detection to be a function of log(sequencing coverage) + log(traveling flaw) (Supplementary Information). C Number of countries with at least one Alpha submission plotted by date of collection and date of submission. The black line shows the average rescaled sequencing coverage. In each country, the sequencing coverage was rescaled by the maximum over the period displayed in the plot to highlight the trend. Countries’ rescaled time series were then averaged. For the sake of visualization, the sequencing coverage is here smoothed over a 2 weeks sliding window. The purple line indicates the date of Alpha international alert (18 Dec 2020). The dashed black line indicates the censoring date used in the analysis (31 Dec 2020). D Distributions of delay (in days) from collection to submission for Alpha and non-Alpha sequences collected outside the UK from December 2020 to mid-January 2021 and submitted up to June 2021 (non-Alpha sequences n = 149699, Alpha sequences n = 6992). Boxplots represent the median (white bar), the quartiles and the 95% range (whiskers). The violin plot shows the Kernel estimation of the underlying distribution. Additional details are reported in Supplementary Fig. 1.

Fig. 2. Comparison between the international dissemination model and the data.

A Date of collection of the first Alpha sample submitted to GISAID and corresponding date of submission for the 24 countries submitting Alpha sequences before 31 Dec 2020. Data are shown by purple circles (collection) and green triangles (submission). Median date obtained from the model is indicated by gray circles (collection) and gray triangles (submission). The horizontal bars display the 95% prediction interval over n = 500 simulations. B Median model predicted cumulative number of countries submitting a first Alpha sequence to GISAID compared with observations. In panels A and B, the purple vertical line indicates the date of Alpha international alert (18 Dec 2020). C Alpha incidence in the UK and median model-predicted epidemic profile in the UK. Both model predictions and data are rescaled to the sum over the period considered to allow comparing the profiles of the curves. To account for testing delays model predictions are shifted right of one week. The gray colored ribbon represents the 95% credible interval.

Fig. 3. Timing of first importation and silent spread as estimated by the international dissemination model.

A Cumulative number of countries with an Alpha introduction as predicted by the model. The quantity is computed from the median predicted date of introduction in each country with 95% prediction interval obtained over n = 500 simulations. B, C Median date of first introduction when occurring before Dec 31 (vertical dashed line) for each country estimated by the model with 95% prediction interval computed over n = 500 simulations. For each country, we report the date of first Alpha detection (i.e. collection of first submitted sequence) (light pink) and the date of the first ever collected Alpha sequence (dark pink) from the data. For El Salvador, Papua New Guinea and Madagascar, no Alpha sequence had been reported before June 2021. D Duration of silent spread in days vs sequencing coverage. The distribution of the durations of silent spread is reported in Supplementary Fig. 3. Duration of silent spread is computed as the difference between the median date of first detection and the median date of first introduction as predicted by the model. We restricted the analysis to countries for which both first introduction and first detection were predicted to occur before 7 Jan 2021. Dashed line represents least-squares linear regression. P-value is computed from Wald test with t-distribution.

Fig. 4. Local spread of Alpha in six destination countries.

A Model vs. empirical Alpha infections. In the main plot, the empirical estimates of Alpha cases are computed by multiplying the Alpha frequency from virological investigations by the reported COVID-19 incidence at the same date—the date is indicated in the plot. Model estimates are obtained with the autochthonous model A (AM A in the plot). Gray lines show ratios of 100%, 50% and 25% between observed and predicted infections attributable to reporting. In the inset, the frequency of Alpha in France obtained from the autochthonous model B (AM B in the plot) is compared with the empirical data. In both panels, black error bars indicate the prediction interval over 500 stochastic simulations obtained with the median volume of Alpha introduction, output of the international dissemination model assuming a 7-day delay between case and infection. Dark colored bars account for the variability in the output of the autochthonous models accounting for the upper and the lower limit of the prediction interval of the Alpha introductions as given by the international dissemination model. Light colored bars account for variability in the delay from infection to case reporting (ranging from 4 days to 10 days). B Empirical Alpha infections vs average international traffic. C Comparison between the date of first introduction as predicted by the international dissemination model and the seeding time of the transmission chains survived until 31 Dec 2020, predicted with the autochthonous model A. Circles indicate medians and segment the 95% prediction interval. Colors indicate the effective reproduction number of the historical strain, $R_t$, computed from weekly mortality data (Methods). The star shows the date of first Alpha detection as a comparison. D Difference between the median delay of seeding predicted by the autochthonous model A and the same quantity in the reference case—i.e. when $R_t$ is the same in all countries and traveling fluxes do not change in time, plotted against the median $R_t$ during the period from first introduction to seeding.