Abstract
The SARS-CoV-2 pandemic was characterised by continual emergence of variants. For improved future pandemic preparedness it is important to understand whether each successive variant has been more infectious and more widely spread. In this paper, we used genetic sequencing data from the COVID-19 Genomics UK Consortium in England and robust statistical models to quantify transmissibility advantage and spatial heterogeneity of successive SARS-CoV-2 variants and their sub-variants circulating between September 2020 and December 2022.
Our results show that each variant was progressively more transmissible and more heterogeneously spread. Alpha was 10-40% more transmissible than B.1.177, Delta was 40-100% more transmissible than Alpha, and Omicron was 80-120% more transmissible than the Delta variant. Progressive variants were also more spatially heterogeneous: Alpha was mostly clustered in London and Southeast England, Delta was less clustered and prevalent in both Northwest and Southeast England, while Omicron was dispersed across the country. Successive sub-variants of different clade were more transmissible and spatially spread than the previous ones (e.g. the Omicron BA.1.1 variant was 20-60% more transmissible than the previous Delta AY.4 variant). However, sub-variants of the same clade didn’t differ in transmissibility or spatial spread. Our study improves understanding of transmission intensity and spatial heterogeneity across SARS-CoV-2 variants.
Ascertaining the threat posed by emerging variants and across infectious diseases is crucial to be better prepared for the next pandemic. Our method provides a tool for analysis of variant surveillance data, translatable across pathogens and settings, that can capture an emerging variant early and as part of the pandemic preparedness strategy.
Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), the virus causing COVID-19, had continued to spread in England throughout 2020, 2021 and 2022. The spread was facilitated by the emergence of new viral variants such as the B.1.177 during the summer 2020, the Alpha and Delta variants, which dominated in late 2020 and early 2021 and the Omicron variants which dominated in 2022. By 30 December 2022, over 20.4 million confirmed cases and over 183 thousand deaths related to COVID-19 had been reported in England [1].
Different non-pharmaceutical interventions (NPIs) including three national lockdowns over 2020 and 2021, with varying levels of social restrictions, Testing-Tracing-Isolation strategies from September 2020, alongside a large scale vaccination strategy from late 2020, have been employed in England to mitigate the epidemic waves caused by these variants. In 2021, alongside the ongoing vaccination against SARS-CoV-2, there was also a gradual lifting of social restriction measures via the reopening Roadmap, starting with reopening of schools in March 2021 as a step 1, and culminating with full lifting of social restrictions from July 2021. In late 2021 and 2022, during the Omicron epidemic waves, additional vaccination strategies were employed including vaccination of adolescents, and boosting immunisation strategies for the elderly.
The emergence of new viral variants, including SARS-CoV-2 variants, is a consequence of mutations of the virus and as a result new variants have potential to be intrinsically more transmissible. Over the last few years, a number of studies have evaluated the transmissibility of different SARS-CoV-2 variants (Figure 1). Across previous studies, the quantified progressive transmissibility of the variants was across different populations, with studies in the USA [2], Japan [3][4], Norway [5][6], Spain [7], South Africa [8] and the UK (e.g. [9] [10]). Most studies estimated the transmissibility of variants using case data from population wide testing. However, a few estimated the secondary attack rate (SAR) of the variants, by calculating the number of close contacts of a case who became infected after a given time period, either using contact tracing [6] or by studying the transmission in households and using this to estimate the SAR [11] [5][3][4].
Previous estimates of relative transmissibility of variants of concern from existing published literature. Dashed lines correspond to ranges of plausible values quoted.
In the autumn 2020, when the Alpha variant emerged, a number of studies suggested it was more transmissible than the previously dominating B.1.177 variant [12][9][13][5][6][3][10][14][4][15][16], with estimated advantage ranging from 15% [3] to 130% [4] (circles in Figure 1). The Delta variant, emerging in Spring 2021 and spreading during 2021, was shown across studies to be considerably more transmissible than the Alpha variant [11][12][2][10][8][7] (rhombuses in Figure 1). The estimates in increased transmissibility ranged from an increase of 32% [7] to an increase of 167% [2]. Several studies compared the transmissibility of the Omicron variant to that of the Delta variant, [17] [18] [19], with estimates for the relative transmissibility ranging from 3.19-4.20 (triangles in Figure 1). One study, [20], contained a systematic review of studies estimating the reproduction number of the Omicron variant. These authors then compared these estimates to a value for the reproduction number of the Delta variant, calculated in a previous systematic review, [21], and found the pooled estimate for the increase in transmissibility to be 2.71 (with a 95% confidence interval of (1.86, 3.56)).
Finally, [22] estimated the reproduction number for variants Alpha, Beta, Gamma, Delta and Omicron, in different countries, using country-wide case data. This study found the highest reproduction number to be that of Omicron, at 1.90 in South Africa, but also found that the estimates for the reproduction number of each variant differed greatly from country to country.
The genetic sequencing of detected cases of SARS-CoV-2 in England by the COVID-19 Genomics UK (COG) consortium has enabled the analysis of the relative properties associated with each variant to be determined. In our previous work we used this data in early 2021, to compare the relative transmission rates of the Delta variant between May and July 2021, to the previously dominating Alpha variant [10].
Here, we extend this work to explore the transmissibility advantage of the progressive variants B.1.177, Alpha, Delta and Omicron variants, and their sub-variants, that were circulating in England between September 2020 and December 2022. We use genetic surveillance data to quantify these but also to compare the progressive spatial heterogeneity over English lower-tier local authority (LTLA) areas.
Methods
Using genetic sequencing data from the COG consortium, we developed and applied a set of hierarchical generalised additive models (HGAMs) to each of the successive dominating SARS-CoV-2 variants in England. In the next four subsections we detail the data and the statistical analyses.
Data
The publicly-available data from the COG consortium [23] corresponded to weekly aggregate counts of each genomic variant sequenced at the Wellcome Sanger Institute to monitor COVID-19 dynamics within each LTLA of England. The dates correspond to the date on which the sample was collected.
The corresponding latitude and longitude of each LTLA were obtained form [24] and matched to the LTLA code in the COG data. These were then used to inform the spatial variation in counts within the regression model. The temporal aspect of the model is informed by the number of weeks since the start of the study period (here September 5th 2020).
Statistical Models
We used the same methodology proposed in [10], adapting and extending the code from [9], to determine the relative multiplicative rate of transmission of competing variants across LTLAs in England. Hierarchical generalised additive models (HGAMs) [25] with a negative binomial response distribution were fitted with count of each variant per week and per LTLA as the response, and using a GI structure as described in [25]. This contained both a global thin plate regression spline and thin plate regression splines fitted separately to each lineage, in which the parameters and smoothness were allowed to vary between lineages. The gam.check function from the mgcv package was used to ensure appropriate number of effective degrees of freedom and knot locations had been used. Four simpler models were also fitted in each case but were discarded due to higher Akaike Information Criteria (AIC).
Quantifying progressive variants’ transmissibility
We initially studied the relative transmissibility of combined principal lineages relative to the previous where the count of cases was summed over principal sublineages of the variant, and proxied by the multiplicative effective reproduction number Rt.
Conditional on the fitted model, each LTLA was treated as a separate experimental unit, which may be correlated to a lesser or greater degree to other LTLAs. Relative multiplicative transmission rates were estimated for each LTLA using linear trends extrapolated from the fitted hierarchical models, as interactions across the determined dates of emergence. Linear slopes during the emergence period (Table 1) for each (sub)variant were calculated using the emmeans package [26] and the ratios of these slopes determined in each LTLA. This way we compared relative rates of transmission of B.1.177 vs Alpha, Alpha vs Delta and Delta vs Omicron variants i.e. each successive variant relative to the previously circulating variant over the time period in which they showed exponential growth across England. We then compared the Delta and Omicron sub-variants to the previously dominating sub-variant. Specifically, we analysed the transmissibility of sub-variants: B.1.617.2 vs the Alpha variant, the B.1.617.2 and Delta AY.4 variant, the Delta A4.4 variant and the Omicron BA1.1 variant, and the Omicron BA.1.1 and BA.2 variants; these were circulating over 2021 and 2022 period for which the COG data was available. The time period of interest for each variant and sub-variant is shown in Table 1. We note that for the comparisons using combinations of variants (i.e. Alpha vs Delta and Delta vs Omicron), we used the union of the time periods of the emerging sub-variants involved.
Unlike in [10], we fitted the model to the sub-variants pairwise, and so the values for the relative multiplicative rate of transmission depended only on the number of cases of the two sub-variants being compared. This meant that the calculation for the relative multiplicative advantage was not influenced by previous variants affecting the global smoother. Additionally, the multiplicative outcome can in this way be proxied by a multiplicative temporal reproduction number Rt which gives the secondary number of cases emerging from an infected case.
Quantifying the spatial heterogeneity in successive variants
The estimated relative transmissibilities from each of the analyses were then tested for spatial auto-correlation using Moran’s I statistic [27]. The statistical test determined evidence against the null hypothesis of no spatial auto-correlation, which would correspond to random spread across LTLAs. Higher positive test statistics tended towards significant p-values and indicated local clustering of high transmission during the period of emergence, whilst significant negative values indicated dispersion or anti-correlation.
Results
Quantifying the progressive transmissibility advantage of successive overall Alpha, Delta and Omicron variants
Our results showed that each new emerging variant was more transmissible than the previously dominant variant. Alpha was 10-40% more transmissible than B.1.177 (Figure 2a), Delta was 40-100% more transmissible than Alpha (Figure 2b), and Omicron was 80-120% more transmissible than the Delta variant (Figure 2c).
Maps showing progressive transmissibility advantage of the Alpha, Delta and Omicron variants across LTLAs in England.
Quantifying the progressive transmissibility advantage of emerging sub-variants
When comparing sub-variants, we found large variability in relative transmissibility of successive sub-variants. Successive sub-variants that belonged to a principal variant different to the one previously dominating, were always more transmissible. For example, the B.1.617.2 variant was 10-40% more transmissible than the previous Alpha variant (which didn’t have sub-variants), while the Omicron BA.1.1 variant was 20-60% more transmissible than the previously dominating Delta AY.4 variant (Figure 3c. However, when the emerging sub-variant belonged to the same principal variant (such as the Delta B.1.617.2 and AY.4 variants or the Omicron BA.1.1 and BA.2 variants), we didn’t find significant difference in their progressive transmissibility (Figure 3).
Maps showing the progressive transmissibility advantage of each sub-variant.
Spatial heterogeneity and clustering
The Moran’s I statistics (Table 2) suggested significant clustering (positive test statistics) across all variants, but no significant spatial structure within variants. For example, progressive variants were more spatially heterogeneous: Alpha was mostly clustered in London and Southeast England (Figure 2a and Table 2), Delta was less clustered and prevalent in both Northwest England and Southeast England (Figure 2b and Table 2), while Omicron was heterogeneously spread across the country and not as evidently clustered (Figure 2c and Table 2). Emerging sub-variants were more clustered if they belonged to different principal variant. For example, both the Delta 1.617.2 sub-variant, which emerged after the Alpha variant, and the Omicron BA.1.1 variant, which followed the AY.4 Delta variant, were more clustered around London and across Northeast England (Figure 3). The Delta AY.4 variant, which followed the Delta B.1.617.2 variant, was more heterogeneously spread, as was the Omicron BA.2 compared to the Omicron BA.1.1 variant.
Discussion
We use genetic surveillance data and flexible regression models to quantify the progressive advantage and spatial spread of SARS-CoV-2 variants and sub-variants that were circulating in England over periods between September 2020 to December 2022. Our results suggest that emerging variants were progressively more transmissible and more heterogeneously spread across England: Alpha was 10-40% more transmissible than B.1.177, Delta was 40-100% more transmissible than Alpha, and Omicron was 80-120% more transmissible than the Delta variant. Progressive variants were also more widely spread: Alpha was mostly clustered in London and Southeast England, Delta was less clustered and prevalent in both Northwest England and Southeast England, while Omicron was heterogeneously spread across the country. Finally, we show that successive sub-variants that belonged to the same principle variant were more transmissible and spatially spread than the previous ones (e.g. the Omicron BA.1.1 variant was 20-60% more transmissible than the previous Delta AY.4 variant). This was not the case with emerging sub-variants from the same principal variant, with our results showing that the consecutive Delta and Omicron variants having similar transmissibility and spatial spread.
The distribution of multiplicative rates is bimodal in nature. Around one third of the LTLAs observe a relative Rt close to one, suggesting that the variant does not out-compete the other circulating variants. For those LTLAs in which the new variant does become a dominant variant, there is a relatively symmetrical distribution of multiplicative rates.
Our results of progressive transmissibility of successive principal variants are in line with previous findings. As noted in our short review of previous results in the introduction, Alpha was 15-130% more transmissible than the B.1.177 variant, with our determined range well within this. Similarly, Delta was found in previous studies to be 32-167% more transmissible than the Alpha variant, and this is similar to our range of 40-100% advantage in transmissibility. A smaller number of studies have previously quantified the transmissibility advantage of Omicron vs Delta to be between 80-320%, and this range includes our predicted range of 80-120%. Overall, our results add to the existing literature. But importantly, unlike existing studies, as well as comparing principle variants, we also compared successive sub-variants. Our results that transmissibility advantage is a characteristic of sub-variants from a new principal variant, are novel. Furthermore, we also showed that successive new variants, and sub-variants of different principle variant, are more spatially spread and less clustered. Overall, we found that with each emerging variant, whilst still statistically significant, the level of clustering determined by the Moran’s I test statistic has reduced. This implies a tendency towards more random spread of the disease with genetic evolution of variants. Both of these results are useful in preparing surveillance and response when a new principal variant, or a new sub-variant emerges.
Our study has a number of strengths. Firstly, we extended and adapted established and existing methodology to longer time periods and sub-variants. We evidence the important observation that if we have groups of variants (e.g. all of Delta or all of Omicron) we have enough power to detect differences, while if we look at sub-variants it is more difficult to get significant results. Whether this is due to a weaker signal from shorter circulating times and greater numbers of sub-variants circulating, or a closer genetic similarity in those sub-variants, is less clear. Using flexible regression models, compared to other published studies, allows analysis of large data sets, including spatial smoothing and predictive ability, which can be achieved with good surveillance.
Overall, using flexible statistical models, we were able to show that successive variants of SARS-CoV-2 in England between September 2020 and December 2022 were progressively more transmissible and increasingly more widely spread across England. Importantly, genetically new principal variants or sub-variants of different clade, have a tendency to be more intrinsically transmissible and able to spread widely. This highlights the importance of continued surveillance to capture an emerging variant early, as this will allow for timely local and targeted interventions to prevent wider and larger-scale spread of emerging variants and improved pandemic preparedness.
Our study is the first to consider both temporal and spatial heterogeneity of emerging SARS-CoV-2 variants using genetic data. Our findings of spatial homogeneity/heterogeneity and clustering estimates support a tendency of new and more mutated variants to spread more rapidly and over larger space. Spatial modelling of adaptive viruses remains a significant challenge in [28, 29]. The advantage of regression-based statistical models is the ability to incorporate additional features or factors into the modelling framework to explain the variability in spatial spread and improve predictive ability of the models.
Whilst these are not formal experimental conditions, they do enable the comparison of relative transmission rates observed in the wider population as if they were an experimental study. Similarly, LTLAs in England, assuming that any underlying correlation between them has been accounted for, can also be treated as multiple experimental units. Determining the relative behaviours of the transmission rates across the population and between different variants that emerged has the potential to highlight important variation in transmission across space and between different successive variants and against the wild-type SARS-CoV-2 variant. The results support previous analyses from more directed methods for detecting relative transmissibility of emerging variants of concern.
One challenge of using happenstance data is that there is a restriction to utilise the data as they have been collected. In this instance, it was not possible to look at the course of the entire epidemic since the genetic data did not have enough power for the later Omicron variants when surveillance was reduced. Continued and sustained surveillance is therefore crucial if methods like this are going to be used as a public health benefit.
We showed there was substantial heterogeneity in transmissibility between sub-variants of different lineages, but there was a lack of power in detecting significant differences in transmissibility between sub-variants of the same dominant variant.
Additional challenges were observed in later sub-variants of Omicron as wide-scale national testing was terminated. Our results suggest that this is indeed the case for all Omicron sub-variants were more transmissible than AY.4. Figure 3d also shows the spatial variability in relative transmissibility for BA.1.1 and BA.2 relative to delta variant AY.4.
From publicly available data, we were able to replicate more labour and cost-intensive methods for estimating transmissibility. In addition our analyses are able to determine spatial variability in observed transmissibility. One of the challenges in these observed data are separating out the impacts of changes in pharmaceutical and non-pharmaceutical interventions from the inherent variation in the transmissibility of the variants. The upper bound on ranges in relative transmissibility frequently correlates well with previous more direct estimates of transmissibility, in areas where interventions are perhaps not having significant impact, however there are frequently regions where the transmissibility is coupled with interventions that impact transmission where the observed transmission may not reach its full potential.
In summary, in this paper, we show how combining robust statistical models with genetic data can be used to quantify the transmissibility and dispresion advantage of emerging pathogens, here proxied by different SARS-CoV-2 variants and sub-varaints. Hence we illustrate how this can be readily used as part of an improved pathogen surveillance and pandemic preparedness strategy.
Data Availability
Data and code to reproduce the analyses and figures presented in the paper are available at the author github without restriction.
Data Sharing
Data and code to reproduce the analyses and figures presented in the paper are available at the author’s github without restriction.
Funding
No funding was used to conduct this research.
Acknowledgments
References
