How much should we sequence? An analysis of the Swiss SARS-CoV-2 surveillance effort

Abstract

During the SARS-CoV-2 pandemic, many countries directed substantial resources toward genomic surveillance to detect and track viral variants. There is a debate over how much sequencing effort is necessary in national surveillance programs for SARS-CoV-2 and future pandemic threats. We aimed to investigate the effect of reduced sequencing on surveillance outcomes in a large genomic data set from Switzerland, comprising more than 143k sequences. We employed a uniform downsampling strategy using 100 iterations each to investigate the effects of fewer available sequences on the surveillance outcomes: (i) first detection of variants of concern (VOCs), (ii) speed of introduction of VOCs, (iii) diversity of lineages, (iv) first cluster detection of VOCs, (v) density of active clusters, and (vi) geographic spread of clusters. The impact of downsampling on VOC detection is disparate for the three VOC lineages, but many outcomes including introduction and cluster detection could be recapitulated even with only 35% of the original sequencing effort. The effect on the observed speed of introduction and first detection of clusters was more sensitive to reduced sequencing effort for some VOCs, in particular Omicron and Delta, respectively. A genomic surveillance program needs a balance between societal benefits and costs. While the overall national dynamics of the pandemic could be recapitulated by a reduced sequencing effort, the effect is strongly lineage-dependent-something that is unknown at the time of sequencing-and comes at the cost of accuracy, in particular for tracking the emergence of potential VOCs.IMPORTANCESwitzerland had one of the most comprehensive genomic surveillance systems during the COVID-19 pandemic. Such programs need to strike a balance between societal benefits and program costs. Our study aims to answer the question: How would surveillance outcomes have changed had we sequenced less? We find that some outcomes but also certain viral lineages are more affected than others by sequencing less. However, sequencing to around a third of the original effort still captured many important outcomes for the variants of concern such as their first detection but affected more strongly other measures like the detection of first transmission clusters for some lineages. Our work highlights the importance of setting predefined targets for a national genomic surveillance program based on which sequencing effort should be determined. Additionally, the use of a centralized surveillance platform facilitates aggregating data on a national level for rapid public health responses as well as post-analyses.

Keywords: SARS-CoV-2; genomic surveillance; modeling; molecular epidemiology; platform; public health; surveillance studies; whole genome sequencing.

Fig 1

(A) Number of positive COVID-19 cases in Switzerland per week. (B) Percentage of cases that were sequenced per week. (C) Number of sequenced cases per week colored by their Pango lineage assignment. This represents the total number of sequences submitted to SPSP (n = 143,260). The black arrows denote the first detection of the VOCs Alpha, Delta, and Omicron. The green wave corresponds to the Alpha wave, the blue and purple colors to the Delta wave, and the pink colors to Omicron.

Fig 2

(A) Delay in detecting the first sequence of the VOC upon downsampling, as compared to using the full data set (“all”). Delays are shown for 100 iterations at each downsampling size. (B) Delay for detecting the first cluster of a VOC after detecting its first introduction, shown for the full data set (“all”) and upon downsampling. Delays are shown for 100 iterations at each downsampling size. Boxplots display the median and interquartile range, with minimum and maximum values shown with whiskers. Color code: delays for Alpha in green, Delta in blue, and Omicron in pink.

Fig 3

(A) The speed of introduction (i.e., the slope of the linear model of the growth phase of each VOC) upon downsampling in relation to the complete data set (indicated by the red line). (B) Number of days required to reach 50% prevalence upon downsampling compared to using the full data set (“all”).

Fig 4

(A) Normalized density of clusters per day for all available sequences. (B) Normalized density of clusters per day for a sample of 50k sequences. The absolute number of active clusters detected each day has been normalized by the number of sequences obtained in a time window of ±15 days. The colors indicate the VOC (Alpha in green, Delta in blue, and Omicron in pink). Black arrows indicate the date at which the first sequence of each variant was isolated; colored arrows highlight the date at which the first cluster of each VOC was detected when considering all 143k samples; colored dashed arrows show the date at which the first cluster of each VOC would be detected if considering only 50k samples. Note: for Omicron and this particular sample of 50k sequences, the date of the first cluster detection was identical as for the complete data set.

Fig 5

Effect of downsampling on LDVM. (A) Absolute number of LDVM detected at each downsampling size. (B) Percentage of clusters that were found to be LDVM at each downsampling size.