Genomic sequencing of SARS-CoV-2 in Rwanda reveals the importance of incoming travelers on lineage diversity

COVID-19 transmission rates are often linked to locally circulating strains of SARS-CoV-2. Here we describe 203 SARS-CoV-2 whole genome sequences analyzed from strains circulating in Rwanda from May 2020 to February 2021. In particular, we report a shift in variant distribution towards the emerging sub-lineage A.23.1 that is currently dominating. Furthermore, we report the detection of the first Rwandan cases of the B.1.1.7 and B.1.351 variants of concern among incoming travelers tested at Kigali International Airport. To assess the importance of viral introductions from neighboring countries and local transmission, we exploit available individual travel history metadata to inform spatio-temporal phylogeographic inference, enabling us to take into account infections from unsampled locations. We uncover an important role of neighboring countries in seeding introductions into Rwanda, including those from which no genomic sequences were available. Our results highlight the importance of systematic genomic surveillance and regional collaborations for a durable response towards combating COVID-19.

emergence and increase in infections of lineage A.23.1 in Rwanda as in Uganda, albeit with a delay of roughly six weeks. As more and more samples were being sequenced as of the second half of December 2020, most of the infections in Rwanda could be attributed to A.23.1, with an increasing proportion of B.1.351 since the start of 2021. It hence seems fair to assume that the epidemic situation in Rwanda and Uganda are linked (to some extent), and that A.23.1 was the driving factor behind the increase in total case counts in both countries. This is clearly not the case in Kenya, where the epidemic situation is markedly different, and the proportion of A.23.1 infections was only on the rise at the end of our study period but had not (yet) attained dominance.
Our phylogeographic analysis (see Figure 5 in the main text) estimates the origin of A.23.1 to have been in Uganda, corresponding with the observation that such genomes have been detected earlier in Uganda compared to Rwanda. The key difference (between Rwanda and Uganda) lies with the lineages being present before the increase in infections with A.23.1.
Rwanda was heavily impacted by infections with lineage B.1.380 -a phenomenon not shared by Uganda and Kenya, who saw virtually no such infections -whereas Uganda saw mostly infections with a wide range of lineages -but mostly B.1 and A.23 -and Kenya with lineages B.1 and B.1.1. Our phylogeographic analysis (see Figure 6 in the main text) estimates the origin of B.1.380 to have been in Kenya, with such genomes being reported during a relatively short time before those from Rwanda and during the first part of the B.1.380 epidemic in Rwanda. This possibly points to a short B.1.380 epidemic in Kenya, although this is difficult to conclude due to the low total genome count. It is thus clear that each of these countries reacted differently -in terms of the relative proportion of infections -to the appearance of these lineages, possibly related to the lineages already circulating in those countries at that time.
Quantifying the relative contribution of introductions and local transmission is challenging in the case of limited genomic surveillance and we here refrain from comparing the estimated number of Markov jumps in our study to any size assessment of local transmission clusters in Rwanda. We have discussed in the main text that the BaTS analyses we performed indicated that local transmission played a more important role in driving the Rwandan epidemic compared to introductions, and this for both A.23.1 and B.1.380 lineages. This can also be seen Kenya (Githinji et al., 2020) have also refrained from drawing strong conclusions, which requires a highly representative genome sequencing effort not only in the country under study but also the main countries with which viral exchange takes place. An example can be found in a large-scale analysis in the United Kingdom (Du Plessis et al., 2021), where more quantitative analyses can in fact be used for this purpose.

A note on exploiting individual travel histories through phylogeographic reconstruction
The importance of using metadata associated with genomes in the form of individual travel histories has been shown in the work of Lemey et al. (2020) by an explicit (and time-consuming) comparison of performing Bayesian phylogeographic inference with and without these travel data, on two different SARS-CoV-2 data sets. Rather than repeating this type of comparison, we here compare these travel records to the estimated Markov jumps between countries, for both key lineages analysed in our manuscript. Supplementary Figures S9 and S10 show this comparison for lineages A.23.1 and B.1.380, respectively. These figures show that the travel history-aware phylogeographic reconstruction (Lemey et al., 2020) does not merely report the individual travel histories but shows clear differences -in both directions -between the estimated number of Markov jumps (with Bayes factor >3) and the number of recorded individual travel histories between pairs of countries. In general, it is to be expected that the total number of transitions (Markov jumps) between countries along the entire estimated phylogeny will be higher than the recorded travel cases, as typically only a fraction of the genomic samples are accompanied by travel information. However, and as shown in Supplementary Figure S9, in some cases the inferred number of transitions between countries will be lower than the number of travel cases between those countries, for example when samples that correspond to known travel cases cluster together (e.g. in a hypothetical case of an infected family of four crossing a country border, which will result in four individual travel histories but only in a single Markov jump, on the assumption that they have a single / identical source of infection).

Assessing sampling bias in continuous phylogeographic analysis
The continuous phylogeographic analysis discussed in the main text, and shown in Supplementary Figure S11, focuses on a within-Rwanda analysis of viral spread. The sequencing effort described in this study yielded genomes that for a large part originated from Kigali, as described in the main text. While the precise sampling location within the country is not directly relevant for discrete phylogeographic analysis, as performed in the main text, recent work does indicate that continuous phylogeographic reconstructions can be affected by this type of sampling bias, i.e. the lack of sampling from certain areas (Kalkauskas et al., 2021). However, given that international travel has been shown to be an important predictor for the spread of SARS-CoV-2 (Lemey et al., 2020) and that population density is highest in the province of Kigali, it seems reasonable to focus a fair share of genome sequencing efforts to this particular region of the country, i.e. in the region that also harbours one its main points of entry. As a result, the sampling bias will likely not lead to a shift in inferred location of origin as would be the case when sampling from only the eastern or western-side of the country for example (Kalkauskas et al., 2021).
When it comes to reconstructing the dispersal within Rwanda, the relatively higher sampling effort in Kigali indeed impacts the continuous phylogeographic reconstruction in the sense that such an analysis likely fails to highlight local circulation of lineages outside Kigali as a consequence of under-sampling in those regions. While the heterogeneous sampling effort (or sampling bias) indeed prevents us from interpreting this reconstruction as a realistic overview of the overall dispersal history of SARS-CoV-2 lineages in Rwanda, it still aims to infer the dispersal history of those lineages that were sampled in our study. Although a particular sampling will always affect the reconstructed dispersal history of viral lineages, continuous phylogeographic inference will still provide movement data that can inform on the dispersal dynamics of the virus (Dellicour et al., 2019). However, it stands to reason that investigating the local dispersal dynamic of viral lineages outside Kigali would certainly add much interesting information to this type of analysis.
In order to assess the impact of the sampling bias in our data set, we provide a sensitivity test by redoing our continuous phylogeographic analysis on ten subsampled data sets with reduced heterogeneous sampling, by downsampling the more intensively sampled areas. To this end, we have randomly selected a maximum of two sequences per administrative "sector" area in each of ten replicates continuous phylogeographic analyses. We show the results in Supplementary Figure S12, which illustrates that the phylogeographic reconstructions in these ten replicates are coherent with the phylogeographic dispersal pattern we inferred from the original data set.

Supplementary Tables
Supplementary Table S1. Non-pharmaceutical interventions by announcement date in Rwanda with key changes described. Sources are official government communiques, provided beneath the table.

Date
Details Source 2020-03-22 · Unnecessary movements and visits outside the home are not permitted except for essential services such as healthcare, food shopping, or banking, and for the personnel performing such services · Hand hygiene, social distancing, mask wearing · Schools and places of worship are closed · All borders are closed, except for goods and cargo, as well as returning Rwandan citizens and legal residents, who will be subject to mandatory 14-day quarantine at designated locations · Travel between cities and districts of the country is not permitted, except for medical reasons or essential services; transport of food and essential goods will continue · Shops and markets will remain closed, except those selling food, medicine, hygiene and cleaning products, fuel, and other essential items · All bars are closed 1 2020-05-01 · Movements are prohibited from 8pm to 5am except with permission · Mass screening and testing · Public and private businesses will resume with essential staff while other employees continue working from home · Markets will open for essential vendors not exceeding 50% of registered traders · Manufacturing and construction sectors will open with essential workers · Hotels and restaurants will operate but close at 7pm · Individual sporting activity in open spaces is permitted, however sports facilities shall remain closed · Public and private transport will resume within the same province · Funeral gatherings should not exceed 30 people 2 2020-07-30 · Kigali International airport (KIA) reopen and tourism to resume · Hotels resume activities · Land borders will remain closed, except for goods and cargo 3 2020-08-14 · Passengers arriving at KIA must present a negative COVID-19 PCR test taken within 120 hours prior to departure · Places of worship shall operate upon compliance with COVID-19 preventive measures 4 2020-08-27 · Authorized public gatherings, including conferences and weddings will resume, in adherence with health guidelines, including negative COVID-19 test and participants not exceeding 30% of the venue's capacity 5 2020-09-11 · School will resume with a gradual opening in the coming weeks · Public transport between Kigali and other provinces will resume · COVID-19 test not required for: o Weddings with fewer than 30 guests o Meetings and conferences not exceeding 30% of capacity · Movements are prohibited from 10pm to 5am 6 2020-10-13 · Offices of public and private institutions will operate at 50% capacity · Places of worship will increase to 50% of venue capacity · Church wedding ceremonies and funeral gatherings to not exceed 75 persons 7 2020-11-12 · Gyms and swimming pools to resume activities · Live performances and cultural shows will resume activities 8 2020-12-15 · From 15-21/12/2020, movements are prohibited from 9pm to 4am · From 22-12/2020 -4/01/2021, movements are prohibited from 8pm to 4am · All social gatherings including wedding ceremonies and celebrations of all kinds are prohibited both in public and private settings · Offices of public and private institutions will operate at 30% capacity · All gyms and swimming pools are closed 9 2021-01-05 · All business establishments, including restaurants, shops, markets, and malls will close operations daily by 6pm · Movements are prohibited from 8pm to 4am · Public and private transports are prohibited to and from Kigali · Domestic and international tourists may travel across districts but must possess a negative COVID-

Supplementary Figures
Supplementary Figure S1. Maps showing case and sequence counts for each region and for each month in the data set. Note that the first set of maps show all data until the end of June 2020, and the last map has data from the 1st of January to the 10th of February. Kigali, the capital city of Rwanda, and where most cases and sequences are from, is indicated on each map.

Supplementary Figure S2. Timeline showing the number of cases with direct travel history.
Light green are those cases who entered Rwanda by land, and dark green are those who entered by air. Note that the air borders were closed from 2020-03-22 to 2020-07-30 (see Supplementary  Table S2). Of the 54 countries that 424 air passengers originated in, 60 were from Tanzania (14%), 54 from Kenya (13%), 38 from Uganda (9%), 30 from Burundi (7%), 27 from Nigeria (6%), 23 from India (5%), 19 from Dubai (4%), 14 from Cameroon (3%), 13 from the USA (3%) and 11 from the DRC (3%). The remaining countries had 10 passengers or fewer, and we provide a complete overview of the incoming travellers who entered Rwanda by air in a Supplementary File.   Regarding the A.23.1 and B.1.380 lineages in this study, we note that the introduction of B.1.380

Supplementary Figure S5. Markov jump trajectory plots for three selected Rwandan infected individuals with travel history (returning) from Italy (A), Morocco (B) and the
in Rwanda was followed by its rise to dominance by the start of July, 2020. By the start of  replicates of our continuous phylogeographic reconstructions based on subsets of sequences obtained by randomly selecting a maximum of two sequences per administrative "sector" area.
We refer to the legend of Supplementary Figure S11 for additional details on the mapping of this continuous phylogeographic inference.