Abstract
Many countries have attempted to control COVID-19 through the implementation of non-pharmaceutical interventions. However, it remains unclear how different control strategies have impacted SARS-CoV-2 virus transmission dynamics at the local level. Using complete SARS-CoV-2 genomes, we inferred the relative frequencies of virus importation and exportation, as well as virus transmission chain dynamics in Nordic countries - Denmark, Finland, Iceland, Norway and Sweden - during the first months of the pandemic. Our analyses revealed that Sweden experienced more numerous transmission chains, which tended to have more cases, and were of longer duration, a set of features that increased with time. Together with Denmark, Sweden was also a net exporter of SARS-CoV-2. Hence, Sweden effectively constituted an epidemiological and evolutionary ‘refugia’ that enabled the virus to maintain active transmission and spread to other geographic localities. This analysis highlights the utility of genomic surveillance where active transmission chain monitoring is a key metric.
Introduction
Since its initial description in December 2019, SARS-CoV-2, the causative agent of COVID191,2, has rapidly led to an unprecedented global health crisis. The pandemic has caused tens of millions of infections and over 1.3 million deaths worldwide and continues to accelerate, imposing a significant impact on health care systems, societies and the global economy. Countries are continuously struggling with how to effectively counteract the pandemic, balancing the protection of health with social and economic considerations. In the absence of therapeutics and available therapies and vaccines, efforts have centred on so-called ‘non-pharmaceutical strategies’, particularly initial short-term large-scale restrictions to population movement (e.g. ‘lock-downs’), increased testing and various levels of social distancing. Analysis of the local epidemiological consequences of different control strategies provides key information on the most effective approaches to reduce the rate of virus transmission within communities.
The Nordic countries, defined here as Denmark, Finland, Norway, Iceland and Sweden, provide a powerful example of geographically, politically and socially related countries that differ markedly in how control strategies for COVID-19 were implemented during the first six months of the pandemic. In particular, Sweden took a less restrictive approach where no general population movement restrictions was enforced, schools for children below the age of 16 years old remained open, no quarantine was imposed for infected households, and businesses continued almost as usual3. In contrast, Norway and Denmark enforced a more invasive population movement restriction that included enforced home office for workers in the public sector, home-bound schooling, targeted private sector close downs, as well as closed international borders for non-residents. Iceland, a relatively small homogenous island population (of ∼350,000) never initiated a population movement restriction as Norway and Denmark, but rather focused on large-scale testing and contact tracing to limit virus spread within the community. In relation to population size, Sweden has had a higher number of COVID19-related cases and deaths, than all other Nordic countries4,5, with a total of around 1450/58 cases/deaths per 100,000 people in Sweden, compared to around 951/13 in Denmark, 428/5 in Norway, 321/7 in Finland, and 3 in Iceland, as of 30th of October 20206, and predominantly occurring between April and May 2020.
Although the relative ‘success’ of COVID-19 control measures are normally gauged in the number of cases and deaths at the country level, it is also the case that intervention and mitigation strategies may lead to marked differences in transmission dynamics among populations, which may in turn impact the evolution of the virus. Using a comparative analysis of genome sequence data we addressed whether the different approaches to COVID-19 control employed by the Nordic countries resulted in differences in virus transmission dynamics and the relative frequencies of virus importation/exportation during the first seven months of the pandemic. Accordingly, we compiled a representative data set of SARS-CoV-2 virus genomes and performed a phylo-epidemiological study to identify any differences in transmission chain dynamics between these countries.
Results
We conducted a set of analyses on transmission chains of sampled genomes, also known as “transmission lineages”7, which are defined as monophyletic groups of at least two genomes sampled from a Nordic country. The estimated time of emergence of sampled transmission chains provides information about the onset of community transmission, although such an association is sensitive to sampling bias, particularly the time-scale of country-specific sequencing efforts. Hence, we estimated that the first sampled transmission chains emerged from around January in Sweden (15th January, 95% confidence interval, CI: 30th December – 16th February) to late February in Norway (25th February, 95% CI: 5th February – 7th March) (Table 1). In Norway the average date of emergence of all transmission chains was around early June (2nd June, 95% CI: 24th May – 13th June), over a month later than in other countries (Table 1). This likely reflects that there are fewer genome sequences from between February and May – that is, during the first phase of the pandemic – than between June and October in this country (Supplementary Figure S1). The average time of emergence of transmission chains occurred approximately two weeks and one month, respectively, prior to the increase in COVID-19 cases and induced deaths (Figure 1, Panels F and G). This was most notable in Sweden which, in relation to the other Nordic countries, experienced a steady increase in the number of daily deaths beginning in mid-March. Correspondingly, lower numbers of daily deaths were recorded in Denmark, Finland, Norway and particularly Iceland (Figure 1, Panels F and G).
Summary statistics computed for SARS-CoV-2 data from each Nordic country. The 95% confidence interval (95% CI) is shown for statistics computed via non-parametric bootstrapping. The 95% highest posterior density (95% HPD) is shown for estimates from the Bayesian stochastic mapping analysis.
Transmission chain summaries (panels A–E). White circles mark chain duration (left y-axis; in days) against to their origin date (x-axis). Black circles represent last sample collected from each chain, with the right y-axis denoting their size, and the horizontal grey lines to their left span their duration. The trend lines represent an ordinary regression line for chain duration as a function of time, with respective p-values. New daily laboratory-confirmed cases and deaths over time are shown in the same x-axis in panels F and G, respectively. Note that Sweden reported case numbers weekly, causing their saltatory trajectory.
Our phylogenetic analyses revealed that SARS-CoV-2 was imported to all Nordic countries between mid-January to mid-February 2020, with detectable community transmission from the end of February as inferred by the presence of transmission chains (Figure 1). Sustained community transmission continued for all Nordic countries beyond April, with the exception of Iceland which was characterised by a relatively short one-month period of community transmission, and Norway, which had a low number of cases combined with a low sequencing proportion of 0.023 (the second lowest in our study after Sweden; Table 1).
The number of transmission chains observed was 46 (95% CI: 41.0–92.0) for Denmark, 21 (95% CI: 20.0–40.8) for Finland, 45 (95% CI: 34.0–62.0) for Iceland, and 26 (95% CI: 22.0– 30.8) for Norway. Notably, the highest number of transmission chains - 98 (95% CI: 72.2– 105.0) - was observed in Sweden. Similarly, the mean duration of such transmission chains (number of days from detection to last sampled case) was higher in Sweden than other localities: 26.3 (95% CI: 20.0–36.0) for Sweden, 14.0 (95% CI: 10.1–15.8) for Denmark, 13.5 (95% CI: 9.9–19.6) for Finland, 6.8 (95% CI: 5.0–7.2) for Iceland, and 18.1 (95% CI: 10.9– 23.7) for Norway (Table 1). In Denmark, Finland and Iceland both the duration and the number of transmission chains decreased with time. Conversely, Sweden was characterised by an increase in the duration, size and number of transmission chains over time (Figure 1, Panels A–E). No particular trend was seen in Norway, characterised by an essentially flat trend line. However, only the transmission chain duration trends for Finland and Iceland were statistically significant.
The largest transmission chain was found in Denmark, with 94 cases (95% CI: 22.7–95.2), followed by Iceland with 82 (95% CI: 26.0–82.0), Norway with 57 (95% CI: 52.0–57.0), Finland with 35 (95% CI: 8.0–36.0), and Sweden with 33 (95% CI: 14.1–34.0). Although the largest transmission chain in Sweden was smaller than that for other countries, this country had the highest Ht-index (i.e. the greatest number Ht of transmission chains, each with at least Ht cases) (Table 1). In particular, the Ht-index for Sweden was 9 (95% CI: 8–9), whereas those for other countries were 8 for Denmark (95% CI: 4–9), 7 for Finland (95% CI: 4–7), 7 for Iceland (95% CI: 4–8), and 6 for Norway (95% CI: 4–7), suggesting that Sweden had the least-interrupted community transmission.
To investigate the importation and exportation dynamics of SARS-CoV-2 we employed a phylogeographic approach that estimates putative number of migration events between the Nordic countries and a genomic sample of the global diversity8. As expected, genomes from all of the Nordic countries were broadly distributed throughout the phylogenetic tree, implying multiple migration events (Figure 2, Panel A). For Denmark and Sweden we inferred that most migration events were exportations of the virus, whereas Norway, Iceland and Finland were characterised by virus importations (Figure 2, Panel B). The ratios of exportation to importation events for each country were: 2.1 for Sweden (95% highest posterior density, HPD: 1.5–2.5), 5.3 for Denmark (95% HPD 4.3–6.9), 0.5 for Iceland (95% HPD: 0.4–0.6), 0.01 for Norway (95% HPD: 0.01–0.06). We estimated <1 exportation event for Finland, with a large number of importations, such that the exportation to importation ratio is effectively 0. The total inferred number of exportation events was larger for Denmark (335, 95% HPD: 304.0–377.3) than for Sweden (187, 95% HPD: 154.7–211.0). However, because the sequencing proportion in Denmark was 7-fold higher than in Sweden (Table 1), it is likely that the number of exportation events from Sweden has been underestimated. Most inferred exportation events from Sweden were into Finland, whereas those from Denmark were into non-Nordic countries, followed by Sweden and Iceland (Figure 2, Panel B).
Phylogenetic tree of SARS-CoV-2 from Nordic countries as well as global genetic diversity. The tips are coloured according each Nordic country (panel A). Circular migration diagram of migration events between the five Nordic countries and other locations. The size of the coloured arrows denotes the posterior mean number of inferred migration events from the Bayesian phylogeographic analysis. The black arrows represent the migration direction (Panel B).
Discussion
Up to the end of October 2020, Sweden had experienced a greater number of COVID-19-related cases and deaths in relation to population size compared to all other Nordic countries6. Our genomic analysis shows that whilst SARS-CoV-2 was imported to all Nordic countries during the same time-period in early 2020, Sweden experienced more transmission chains, which lasted longer, and tended to be more numerous. Moreover, transmission chains in Sweden beginning later tended to be greater in size and duration, suggesting increasing spread in the population. These trends emerged despite Sweden having the lowest virus genome sampling proportion, suggesting that community transmission was extensive and well-established relative to the other Nordic countries. In addition, both Denmark and Sweden were characterised by having acted as exporters, of SARS-CoV-2 to its Nordic neighbours. Conversely, Iceland, Norway and in particular Finland showed lower levels of virus exportation. At the same time, our analyses suggest that Finland was the main receiver of the exportations from Sweden, suggesting a unidirectional mode of exportation between the two countries.
During the first six months of the pandemic Sweden never enforced strict community mobility limitations. Instead, social distancing, limiting sizes of social events and moving to distance education for students above 16 years were employed as the main mitigation efforts. Conversely, Denmark, Finland and Norway enforced a relatively strict population movement restrictions of their communities, closure of borders and government run facilities4. Iceland, in a similar manner to South Korea, utilised large-scale testing and contact tracing combined with social distancing and voluntary home-based quarantine without the need to regulate population movement9 (see also Supplementary table S1). Thus, while regulating population movement appears to provide an efficient mode to reduce community transmission10,11, non-population movement-targeted mitigation efforts are also viable options under certain circumstances. Norway experienced relatively low numbers of transmission chains during the first six months of the pandemic, with the bulk of the transmission chains sampled here emerging in mid-April and towards the end of July, although this might in part reflect a delayed start to genome sequencing such results should be interpreted with caution (Figure 1, Panel D). Indeed, we note that the sequencing intensity of Norwegian cases is lower than some of the other Nordic countries with similar case burdens (Table 1). When sample size per country increases with time, the effect of such potential sampling biases are expected to decrease (Supplementary figure S2).
To what extent then did Sweden’s mitigation strategy affect the epidemiological situation both within the country and the Nordic region as a whole? Sweden received a lot of attention due to the number of deaths reported. Sweden has been shown to have had one of the longest durations of excess deaths and highest case numbers between February and May 20205 and it has been suggested that a greater level of self-isolation likely would have reduced the number of deaths in Sweden12. However, although Sweden exhibited a relatively high number of cases compared to the other Nordic countries, and our results are consistent with a relatively high community transmission for Sweden during the first six months, such an increase in community transmission does not necessarily imply a causal relationship with excess deaths. Whilst we employ a robust genomic-based modelling approach to study transmission chain changes, equating or directly relating such changes with the number of deaths may lead to incorrect conclusions as we have not taken into account any patient-related epidemiological parameters such as gender or age. Furthermore, it has been argued that the higher number of deaths in Sweden may also have been influenced by how intensive care unit admission criteria were applied, particularly for the elderly population in care homes, during the early phase of the pandemic12. Thus, official mitigation strategies and efforts alone are unlikely to explain epidemiological differences between countries, which were also impacted by the compliance and practical enforcement from the general public, businesses, and health care providers.
A critical issue, although one that has received little attention, is that if transmission chains are allowed to remain active, they also provide increased opportunity for the virus to evolve and adapt to local populations, potentially acquiring mutations that in some way enhance virus fitness. Although mutations routinely appear in the SARS-CoV-2 genome, including viruses sampled from Sweden13, their functional relevance is currently unknown. In addition to providing increased mutational potential, sustained transmission chains also provide epidemiological ‘refugia’ for the virus to be transmitted to other localities. This is in line with our results showing that Sweden had a greater frequency of exportation events among neighbouring Nordic countries. Thus, interrupting and ultimately stopping transmission chains is not only important to minimise virus spread within populations, but also to reduce the chances for the virus to accumulate beneficial mutations.
Overall, our study highlights the utility of continuous genomic surveillance and retrospective studies to compare and understand differences in pandemic responses with respect to transmission dynamics. In particular, the data presented suggests that transmission chain monitoring may prove to be a useful metric in comparing outbreak mitigation outcome.
Materials and Methods
Data set construction
On the 23rd of June 2020, more than 40,000 complete SARS-CoV-2 high coverage genomes were downloaded from the GISAID platform (www.gisaid.org/). This alignment was updated on October 6th 2020 with additional genomes from the Nordic countries. Initially, the genomes were binned according to country of isolation and all available genomes from the Nordic countries were analysed. We calculated sequencing proportion by dividing the number of sequences per country by the number of cumulative confirmed cases at the most recent sequence collection date (see Table 1).
For all non-Nordic countries, the number of genomes per country was subsampled via cd-hit v.4.8.1 (https://github.com/weizhongli/cdhit) to include 99% of the total diversity per country, apart from USA for which 97% clustering was applied. All genome sequences were aligned with Mafft v.7.455 (https://mafft.cbrc.jp/alignment/software/) and the ‘auto’ option, whereafter poorly aligned regions were removed using TrimAl (https://github.com/scapella/trimal) employing the ‘automated1’ algorithm. The GISAID accession numbers of all sequences included in the final sequence alignment, comprising 3,211 SARS-CoV-2 genome sequences, are available as Supplementary Table S2.
Estimating a time-scaled phylogenetic tree
We estimated maximum likelihood phylogenetic trees using IQ-TREE v2.0.614 employing the GTR+ Γ model of nucleotide substitution. A root-to-tip regression of genetic divergence against sampling time was performed on the resultant maximum likelihood tree using TempEst v.1.5.315. Due to the quantity of sequences and because it did not display obvious temporal signal, we obtained a time-scaled phylogenetic tree using LSDv0.316 with a fixed evolutionary rate of 1×10-3 subs/site/year as estimated previously17, and with the sampling dates as additional calibration information.
Detection of virus transmission chains
We defined transmission chains as monophyletic groups of genomes (at least 2) all collected from one of the Nordic countries, which is analogous to transmission lineages as defined previously7. We computed key statistics from each transmission chain; the duration (i.e. the length of time in days from the first to the last collected genome), the size (i.e. the number of genomes), and the time of origin. For each country we also calculated an Ht-index of transmission chains, where Ht is the number of transmission chains with at least Ht cases (analogous to the H-index citation metric). Larger values reflect increase in the number and size of lineages. For example, of Sweden’s lineages, 9 had at least 9 cases. To assess the impact of phylogenetic uncertainty we obtained 1000 time-trees from standard nonparametric bootstrap replicates and computed the key transmission chain statistics in each case. The 95% quantile of the statistics from the bootstrap replicates represent a confidence interval (CI) due to phylogenetic uncertainty.
Inferring virus migration dynamics between Nordic countries
To infer the frequency of importation or exportation events of SARS-CoV-2 between the Nordic countries and from the rest of the world, we employed a Bayesian stochastic mapping approach, also known as a discrete phylogeographic approach, as implemented in BEAST v1.10.418–20 using guidelines from Dellicour et al. for very large genomic data sets8. We fixed the time-tree described above for the analysis. This method is broadly similar to that used to infer geographic movement of the virus in Belgium17.
The stochastic mapping generates a posterior distribution of ‘type-changes’ between locations along the branches. Our locations consisted of the five Nordic countries plus an ‘other’ category for genomes from any other location, which represent the global diversity of the virus as described above. We ran a Markov chain Monte Carlo of length 107 steps, recording every 1000th step. Sufficient sampling from the posterior was assessed using Tracer with all effective sample size values above 200 as estimated in Tracer v.1.721. We inferred the posterior number of migration events between the six possible states and the amount of time spent at each state, known as Markov rewards.
Data Availability
Accession numbers to the sequence data used in the study (full data available at the GISAID platform) are available in supplementary table 2.
Supplementary materials for this manuscript including the following
Number of SARS-CoV-2 genomes sampled with time for all Nordic countries and global samples (i.e. ‘other’) included in the final alignment.
Markov rewards for each Nordic country and for the global diversity in the data (labelled as “other”). The Markov rewards correspond to the portion of time that all lineages spend at either of the six states in the stochastic mapping. The y-axis is the Markov rewards as a proportion of total tree length. Note that countries with most exportation events, Sweden and Denmark are also occupy the largest portion of the tree.
Supplementary Table S1. Summary table of mitigation response take by the Nordic countries until 2020-10-21 according to the European Centre for Disease Control.
Supplementary Table S2. Table including all GIASID accession numbers for all sequences in the final multiple sequence alignment of SARS-CoV-2.
Acknowledgements
We thank all people who have contributed SARS-CoV-2 genome data to the GISAID platform. Details on specific contributions are available via www.gisaid.org/. JHOP is funded by the Swedish research council FORMAS (grant no: 2015-710). ECH is funded by an ARC Australian Laureate Fellowship (FL170100022). SD and LF are funded by an ARC Discovery Early Career Award awarded to SD (DE190100805).
Footnotes
Competing interest statement All authors have read the manuscript and have no conflict of interest relating to the manuscript. The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.
References
