SARS-CoV-2 transmission dynamics in Belarus revealed by genomic and incidence data analysis

Since the emergence of COVID-19, a series of non-pharmaceutical interventions (NPIs) has been implemented by governments and public health authorities world- wide to control and curb the ongoing pandemic spread. From that perspective, Belarus is one of a few countries with a relatively modern healthcare system, where much narrower NPIs have been put in place. Given the uniqueness of this Belarusian experience, the understanding its COVID-19 epidemiological dynamics is essential not only for the local assessment, but also for a better insight into the impact of different NPI strategies globally. In this work, we integrate genomic epidemiology and surveillance methods to investigate the emergence and spread of SARS-CoV-2 in the country. The observed Belarusian SARS-CoV-2 genetic diversity originated from at least eighteen separate introductions, at least five of which resulted in on- going domestic transmissions. The introduction sources represent a wide variety of regions, although the proportion of regional virus introductions and exports from/to geographical neighbors appears to be higher than for other European countries. Phylodynamic analysis indicates a moderate reduction in the effective reproductive numberRe after the introduction of limited NPIs, with the reduction magnitude generally being lower than for countries with large-scale NPIs. On the other hand, the estimate of the Belarusian Re at the early epidemic stage is comparable with this number for the neighboring ex-USSR country of Ukraine, where much broader NPIs have been implemented. The actual number of cases by the end of May, 2020 was predicted to be 2-9 times higher than the detected number of cases.

COVID-19, SARS-CoV-2, Belarus, genomic epidemiology, phylodynam- ics, effective reproduction number ◦ Corresponding author: [email protected] * The authors contributed equally. 1 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprintthis version posted April 19, 2021. ; https://doi.org/10.1101/2021.04.13.21255404doi: medRxiv preprint NOTE: This preprint reports new research that has not been certified by peer review and should not be used to guide clinical practice. 1 Introduction1 The Republic of Belarus is a country in Eastern Europe with a population of approxi-2 mately 9.5 million. In comparison to other ex-USSR and Central European countries, it is3 characterized by weaker socio-economic and political ties with the neighboring European4 Union [20, 52] and lower outward population mobility [9]. At the same time, Belarus has5 a relatively modern healthcare system [45], and the country’s Human Development Index6 (HDI) is categorized as “very high” (vhHDI) [13].7 The COVID-19 epidemic has reached Belarus later than most of Western European8 countries and approximately at the same time as its neighbors. The first confirmed9 imported case reported on February 28, 2020 was a person who arrived from Iran [5, 3].10 Since then, there was a steady increase in the number of officially reported laboratory-11 confirmed cases that has surpassed 300, 000 on March 13, 2021.12 The major feature of the COVID-19 pandemic in Belarus is the significantly narrower13 scope of non-pharmaceutical interventions (NPIs) in comparison to other vhHDI countries14 [17]. The implemented NPIs included a mandatory 14-day self-isolation for individuals15 who were arriving from abroad or were identified as close contacts of individuals with16 confirmed COVID-19; some social distancing measures such as the increase in frequency17 of public transportation operations to reduce crowding; remote teaching and delaying the18 class starting times at schools and higher education institutions [6]. No large-scale quar-19 antines, lockdowns or other strict social distancing measures have ever been administered.20 The other widely practiced measures such as mask regimen and border closures were not21 mandated until November and December of 2020, respectively.22 Given the uniqueness of the Belarusian experience, understanding of COVID-19 epi-23 demiological dynamics in this country is essential not only for assessment of its past and24 current public health situation, but also for a better insight into the impact of different25 NPI strategies around the globe. However, the development of such understanding has26 been impeded by the limited amount of available data. Until the last quarter of 2020 the27 only available data have been the officially reported country-level counts that included28 daily incidence, numbers of conducted diagnostic tests, and COVID-related mortality.29 Such statistics are prone to biases and underreporting [31, 53]. While these drawbacks30 are well-known and common for all countries, they have a potential to be exacerbated in31 Belarus due to limited testing capacities provided by a handful of national-level labora-32 tories [6].33 In the meantime, whole-genome sequencing (WGS) data analyzed using genomic34 epidemiology methods provides a complementary and independent source of informa-35 tion. WGS SARS-CoV-2 data have already been used to study transmission histo-36 ries and epidemiological dynamics in a variety of countries and administrative regions37 [35, 27, 15, 47, 37, 23, 28, 24, 22]. For Belarus sufficiently representative genomic dataset38 has become available only in the late 2020, when the limited sequencing data produced39 outside of the country on behalf of the World Health Organization (WHO) were extended40 by the locally produced sequences.41 In this paper, we combined WGS genomic data and epidemiological data to carry42 out the first study of SARS-CoV-2 transmission dynamics in Belarus. In the absence43 of significant amounts of reliable epidemiological statistics, the integrated genomic and44 incidence analysis allowed to fill the information gap and provide a plausible picture of45 the emergence and spread of SARS-CoV-2 in the country. The obtained results also gave46 2 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprintthis version posted April 19, 2021. ; https://doi.org/10.1101/2021.04.13.21255404doi: medRxiv preprint insight into the effect of limited NPIs during the first epidemic wave.47 2 Methods48 2.1 Data49 The SARS-CoV-2 genomic data for analysis were downloaded from GISAID [48] on March50 15, 2021. The Belarusian dataset consists of 41 full-length genomes sampled between51 March 2020 and February 2021. One sequence was obtained from a citizen of Azerbaijan52 who was tested in Belarus to be allowed to return to his home country. This sequence is53 marked as Azerbaijanian in GISAID, but is considered as Belarusian here. The Ukrainian54 dataset that was analyzed for comparison purposes consists of 116 sequences. Daily55 numbers of new cases and conducted tests were collected from the official Telegram channel56 of the Ministry of Health of the Republic of Belarus [10]57 2.2 Global phylogenetic analysis58 For the phylogeny reconstruction, we utilized the SARS-CoV-2-specific phylogenetic in-59 ference pipeline implemented in Nextstrain [29]. The sequences from Belarus were an-60 alyzed together with 12, 064 background sequences from the global SARS-CoV-2 popu-61 lation. To obtain a representative sample with those background sequences, a country-62 specific Nextstrain context subsampling was used [29]. The sequences were aligned using63 MAFFT [34], and a maximum likelihood (ML) phylogenetic tree was constructed using64 IQ-TREE [39] under Hasegawa-Kishino-Yano (HKY)+Γ nucleotide substitution model65 with a gamma-distributed site rate variation [30].66 In the resulting time-labelled tree, ancestral geolocation traits have been inferred using67 so-called “mugration model” [46]. In this model, countries of origin of the tree nodes are68 considered as discrete traits, and the virus spread between countries is considered as a69 general time reversible process. We augmented this model by incorporating the human70 mobility statistics provided by European Commission Knowledge Center on Migration71 and Demography (KCMD) [44] via KCMD Dynamic Data Hub [9]. Even though global72 travel has been affected by COVID-19-related restrictions, this statistics are still assumed73 to representatively reflect the relative density of human mobility between countries even74 in quarantine settings. Specifically, the transition rates between traits were assumed to75 be proportional to the normalized average numbers of inter-country trips. The resulting76 transition rate matrix has been used to estimate the maximum joint likelihood traits77 of internal nodes using the dynamic programming algorithm [40]. This trait inference78 algorithm has been implemented in Matlab (v. R2019b).79 Belarusian clades were defined as those having the most recent common ancestors80 (MRCA) with “Belarus” trait, and intra-Belarusian lineages were inferred as the maximal81 subtrees inside these clades. Upon examination of the Belarusian clades, we joined two82 clusters that have the same estimated source trait and the MRCA at the tree distance of 483 from both of them. Finally, global lineages of sequences were determined using Pangolin84 SARS-CoV-2 Lineage Assigner [11].85 3 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprintthis version posted April 19, 2021. ; https://doi.org/10.1101/2021.04.13.21255404doi: medRxiv preprint 2.3 Intra-country phylodynamic analysis86 In this work we largely followed a general analytic pipeline adopted in other similar87 country-level studies (see e.g. [27, 36, 35, 47]), with several modification tailored for the88 specifics of the analyzed data. At first, temporal signal was evaluated by constructing89 an ML phylogeny under HKY+Γ nucleotide substitution model and by regressing root-90 to-tip genetic divergence against sampling dates using TempEst (v.1.5.3) [43]. Next,91 BEAST (v.2.6.3) [18] was used to fit the Coalescent Bayesian Skyline model to the full92 set of Belarusian sequences. As before, HKY+Γ nucleotide substitution model was used93 together with a strict molecular clock. The clock rate was assumed to follow a gamma94 (Γ) distribution with the mean equal to 8 × 10−4 mutations/site/year and the standard95 deviation of 5× 10−4 [16, 27], where the distribution density was parametrized using the96 corresponding shape and rate parameters. Four segments were assumed for the effective97 population size that roughly corresponded to growth and decline periods of the first98 and second COVID-19 epidemic waves. The model parameters were sampled from the99 corresponding posterior distribution using Markov Chain Monte Carlo (MCMC) method100 with 3× 107 iterations, sampling every 3 × 103 iterations and the initial 10% “burn-101 in” iterations. The MCMC sampling quality was assessed using Tracer (v.1.7.1) [41]102 and accepted if all parameters had effective sampling sizes (ESS) higher than 200. The103 obtained maximum clade credibility (MCC) tree was annotated using Tree Annotator104 (v.1.8.4) [33]. The reliability of intra-Belarusian clusters detected by the ML phylogenetic105 inference was re-confirmed by verifying their correspondence to monophyletic clades in106 the MCC tree. For each cluster, time to the most recent common ancestor (TMRCA)107 was estimated.108 The effective reproduction number Re and the sampling proportion have been es-109 timated for the two best-sampled Belarusian transmission lineages with a total of 19110 genomes (Supplemental Table S3) using Birth Death Skyline Serial (BDSKY) model [49]111 implemented in BEAST. The analyzed lineages were likely co-circulating over the same112 susceptible population (see Results). Thus, we used a linked model where both lineages113 evolve and are being sampled independently but share the substitution model parameters,114 the molecular clock rate and the effective reproduction number drawn from the same re-115 spective priors. Given the relative sparsity of available genomic data, this approach allows116 to use larger and more representative combined sample for the analysis. The same settings117 as above have been used for the substitution model, molecular clock and MCMC. Since118 BDSKY model is parameter-rich, we equipped it with the informative priors on several119 parameters. Specifically, the sampling proportions were assumed to have a Beta(α,β )120 distribution prior with parameters α = 1 andβ = 9.99·105, thus reflecting the sparsity of121 Belarusian sequence sample (the proportion of sequenced cases from the total number of122 cases is assumed to vary between 10 −6 and 10−3). The prior for the origin of each cluster123 was assumed to be normally distributed with the mean equal to the time estimated using124 the Coalescent Bayesian Skyline. For the rate of becoming non-infectious, we assumed an125 infectious period of 10 days [27, 32, 35, 38]. Finally, we considered the models with one126 and two changes of the effective reproduction number Re and the sampling proportion.127 The times of the parameters change were fixed to July 1, 2020 for the first model and128 May 1 and July 1, 2020 for the second model. The list of model parameters is reported129 in the Supplemental Table S1.130 4 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprintthis version posted April 19, 2021. ; https://doi.org/10.1101/2021.04.13.21255404doi: medRxiv preprint 2.4 Inference of case counts131 Here we used two complementary approaches. In the first approach, trees and BDSKY132 parameters sampled by BEAST were used to reconstruct cumulative case count trajecto-133 ries using the particle filter algorithm implemented in EpiInf (v.7.3.0) [51]. In the second134 approach, we utilized the method of [53]. It quantifies the case counts underestimation135 from the numbers of confirmed cases and conducted tests up to a specified date in a semi-136 Bayesian way under the assumptions that the observed data are subject to sampling,137 reporting and diagnosis biases. The model [53] has been used with the default settings.138 The case count trajectories were inferred by taking 10 4 samples from model-defined prior139 distributions of testing probabilities for individuals with different severity of symptoms.140 3 Results141 SARS-CoV-2 genomic diversity Despite the sparse sampling, the observed Belaru-142 sian SARS-CoV-2 sequences belong to 11 genomic lineages (by the nomenclature of [42],143 Fig.1A). In particular, the genome that was sampled on February 23, 2021 belongs to144 B.1.1.7 lineage that emerged in the UK in November 2020 and had been rapidly spread-145 ing toward fixation [21]. The root-to-tip regression analysis demonstrated moderately146 strong temporal signal (R 2 = 0.56, p< 10−6, Fig.1B).147 SARS-CoV-2 transmission history We identified 18 distinct intra-Belarusian clades148 that most likely correspond to separate introductions of SARS-CoV-2 into the country.149 The inference of between-country importations of SARS-CoV-2 is usually complicated,150 since during the global pandemic close genomic variants can be observed in multiple151 geographic locations. Therefore, the results of such inference should be treated with152 caution. With that in mind, we note that the inferred transmission history agreed with the153 travel records for those cases when they were available. In particular, the first confirmed154 SARS-CoV-2 case was the individual who arrived from Iran [3], and the phylogenetics155 reaffirmed that. The agreement also held for the second detected case brought by the156 travelled from Italy [1]. The first introduction produced at least one secondary case as157 indicated by the tree; however, both lineages were not sampled after March, 2020 (Fig.2B).158 This can be attributed to the timely isolation of those individuals and their first order159 contacts [2]. In general, SARS-CoV-2 importations into the country could be attributed160 to a mixture of regional and global transmissions. As illustrated in Fig.1B the most161 frequent alleged virus introduction sources were the neighboring countries of Russia (5162 introductions) and Poland (3 introductions).163 Five SARS-CoV-2 introductions (28%) are associated with clusters of two or more164 sequences, and thus are hypothesized to establish intra-country transmission lineages.165 Three largest transmission lineages are paraphyletic and may indicate virus re-export166 from Belarus to other countries. Even though some alleged export cases could be sampling167 artefacts, those of them involving large lineages are more reliable. Such cases include two168 SARS-CoV-2 introductions to the neighboring country of Latvia in June, 2020 (95% CI:169 May 31, 2020 - June 26, 2020) and in October, 2020 (95% CI: October 1, 2020 - October 22,170 2020) that established substantial Latvian transmission lineages (Supplemental Fig.S2)171 Effective reproduction number and the effect of NPIs. Most observed clades orig-172 inated between March and July of 2020, and the majority of their times to MRCA fall173 5 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprintthis version posted April 19, 2021. ; https://doi.org/10.1101/2021.04.13.21255404doi: medRxiv preprint Lineage Counts 0 2 4 6 8 10 12 B.1 B.1.1.164 B.1.36B1.1 B.1.374 B.4 B.1.1.7B.1.243B.1.1.294B.1.1.317B.1.1.422 A Introductions to Belarus 0 1 2 3 4 5 RussiaPoland ItalyUAEUSAIran Montenegro Turkey UK B Temporal Signal Root to Tip Divergence ● ●● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ●● Regression Fit Sampled Sequences R2 = 0.62 p − value < 10−6 2020−02−272020−03−312020−05−032020−06−052020−07−082020−08−102020−09−122020−10−152020−11−172020−12−202021−01−222021−02−24 0.0002 0.0004 0.0006 0.0008 0.001 0.0012 C Log−Lineages Through Time Log−Lineage Log−Median 95% Log−Bands 2020−01−062020−02−132020−03−222020−04−292020−06−062020−07−142020−08−212020−09−282020−11−052020−12−132021−01−202021−02−27 0.69 1.3 1.9 2.51 3.11 3.71 D Figure 1: Lineages summaries: A) Abundances of genomicthe c lineages. B) Estimated

sources to Belarus. C) Temporal signal. D) Lineages-through-time (on logarithmic scale). into April (Fig. 1D, Fig.2B and Supplementary Table S4). The majority of branching174 events also belonged to the same time period. It implies that, despite sequencing being175 performed mostly in late 2020 – early 2021, the phylodynamic analysis of currently avail-176 able Belarusian genomes allows us to reliably assess only the first epidemic wave prior177 to July 2020. Another reason to choose July, 1 for the endpoint of our phylodynamics178 analysis is the dynamics of the daily percentage of positive tests. The WHO criterion for179 influenza-like-illnesses (ILI) assumes that epidemic is “under control” if the percentage of180 positive tests is below 5% for at least two weeks [14]. According to the officially reported181 data, Belarus reached this state with respect to the first COVID-19 wave by the end of182 June, 2020 (Fig.4D), even though the reported incidence peaked several weeks earlier.183 Best-sampled transmission clusters are well-mixed and have representatives from at184 least two Belarusian administrative regions (Fig.2B). This fact and the relative homo-185 geneity of the Belarusian demographical characteristics suggest that the corresponding186 viral lineages co-circulated over the same susceptible population. Thus, we estimated the187 effective reproduction number Re for these lineages using a linked BDSKY model. The188 model with three segments shows a moderate decline of the median ˆRe from ˆRe = 1.95189 (95% highest posterior density (HPD) interval: (1.03; 2.99)) in March-April to ˆRe = 1.59190 (95% HPD interval: (0 .82; 2.39)) in May-June (Fig. 3B). The obtained HPD intervals,191 however, are rather wide due to the relatively small genome sample size. Thus, we also192 6 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprintthis version posted April 19, 2021. ; https://doi.org/10.1101/2021.04.13.21255404doi: medRxiv preprint 0.2 Figure 2: The annotated maximum clade credibility tree. A) Administrative regions of Belarus. B) The tree with the leaves color-coded by sampling regions (using the colors from panel A) and with the branches coded by the corresponding clusters. C) Cluster sources marked on the world map by their ids that correspond to panel B. Maps for figures were downloaded from Vemaps.com estimated the median ˆRe for the entire period of March-June, which turned out to be193 ˆRe = 1.70 (95% HPD interval: (1.45; 1.96)). The Kolmogorov-Smirnov test was used for194 the formal comparison of prior and posterior distribution samples for ˆRe and resulted in195 p< 10−10 for all of them.196 In addition, we matched the estimate of the effective reproduction number ˆRe for197 Belarus against that for Ukraine - the neighboring ex-USSR non-EU country with sim-198 ilar demographics. The major difference in COVID-19 epidemics between Belarus and199 Ukraine is the scope of NPIs, with Ukraine implementing much stricter lockdown and200 physical distancing policies [17]. The same Birth-Death Skyline Serial model was applied201 to two best-sampled Ukrainian clusters with the total of 28 sequences defined as in [25]202 (Supplemental Table S3). The median Ukrainian ˆRe over the same tine period was es-203 timated to be ˆRe = 1.64 (95% HPD interval: (1 .49; 1.81)). This assessment agrees with204 the previous estimation based on Exponential Coalescent model [25] and appeared to be205 comparable to ˆRe estimates for Belarus.206 Cumulative incidence and case counts. Cumulative case count trajectories for Be-207 7 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprintthis version posted April 19, 2021. ; https://doi.org/10.1101/2021.04.13.21255404doi: medRxiv preprint Prior and Posterior Distributions of Re Density 0 1 2 3 4 5 6 0.0 0.5 1.0 1.5 2.0 2.5 3.0 Prior Posterior (March−June) A Prior and Posterior Distributions of Re Density 0 1 2 3 4 5 6 0.0 0.2 0.4 0.6 0.8 Prior Posterior (March−April) Posterior (May−June) B BDSKY Cummulative Trajectories Cummulative Cases Trajectories Quantiles 2020−03−052020−03−122020−03−192020−03−262020−04−022020−04−092020−04−162020−04−232020−04−302020−05−072020−05−142020−05−212020−05−282020−06−042020−06−112020−06−182020−06−25 100 101 102 103 104 105 106 C Figure 3: BDSKY model estimations. A), B) prior (green) and posteriors (blue and orange) distributions of the effective reproduction number estimate ˆRe for Belarus during the first COVID-19 wave. C) The cumulative case count trajectories on log 10 scale. Solid blue and dashed lines represents a median and 95% confidence intervals, respectively. larus implied by the BDSKY model are reported on Fig.3C. The cumulative number of208 cases by July 1, 2020 falls into the 95% prediction interval: (364; 17066). It should be209 kept in mind that these estimates apply only to two transmission lineages out of possibly210 many more.211 The results of the complementary analysis based on the number of officially reported212 cases D(t) and conducted tests T (t) over time are presented in Fig.4. The model [53]213 was designed for the initial phase of the epidemics with the exponential growth of D(t).214 Therefore, we calibrated and used it to estimate the cumulative number of infections215 C(t) for the time interval from April 1 (the first date when the number of conducted216 tests was available) to May 16, 2020 (officially reported peak of the first wave) with a217 15-day increments. The obtained results suggested a substantial underestimation of the218 cumulative number of cases through the study period (Fig.4B). In particular, on t∗ =219 May 16, 2020 the model predicted C(t∗) = 118 , 521 cases (95% PI: (54 , 057; 249, 000))220 while the reported number was D(t∗) = 28, 681. Hence, 76% of infections occurred by221 that date were supposedly undetected (95% PI: (47%; 88%)). The model-inferred case222 detection rate D(t)/C(t) increases over time as more tests are being conducted (Fig.4C).223 4 Discussion224 In this paper we presented the first detailed study of COVID-19 epidemic in Belarus using225 the officially reported incidence data, testing data and genomic data collected between226 March, 2020 and February, 2021. The reported results significantly expand our under-227 standing of COVID-19 dynamics and effects of limited NPIs in Belarus, and reflect several228 key epidemiological issues that it shares with other countries around the globe.229 First, the analysis revealed the diverse history of transmissions of SARS-CoV-2 into,230 from and inside the country. It identified 18 introductions within 13 genomic lineages, but231 this estimate is most likely a lower bound on the real number of introductions, since only a232 very small fraction of all SARS-CoV-2 genomic diversity has been sampled. In contrast to233 most Western European and North American countries [27, 15, 47, 37, 23, 28, 24, 38, 22],234 the larger portion of estimated transmission links was with geographic neighbors. It235 8 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprintthis version posted April 19, 2021. ; https://doi.org/10.1101/2021.04.13.21255404doi: medRxiv preprint Total and Positive Cumulative Tests Counts (in Thousands) Total Tests Positive Tests 2020−04−01 2020−04−16 2020−05−01 2020−05−16 0 67 134 201 268 335 A Cumulative Predicted Cases and Positive Tests Counts (in Thousands) Predicted Cases Positive Tests 2020−04−01 2020−04−16 2020−05−01 2020−05−16 0 50 101 151 201 252 B Cumulative Proportions of Detected Cases and Positive Tests Proportion Detected Proportion 2020−04−01 2020−04−16 2020−05−01 2020−05−16 0.005 0.109 0.213 0.316 0.42 0.524 C Daily Proportions of Positive Tests Proportion Daily Tests Threshold (0.05) 2020−04−152020−04−302020−05−152020−05−302020−06−142020−06−292020−07−142020−07−29 0 0.02 0.04 0.06 0.08 0.1 0.12 D Figure 4: The summary of counts data analysis. A) Input data: officially reported cumula- tive numbers of casesD(t) (orange) and conducted tests T (t) (black). B) The cumulative numbers of officially reported cases (D (t), orange) and the counts that were inferred by the model (C(t), blue). C) Model-based case detection rate D(t)/C(t) (green). D) The daily proportion of positive tests (light blue) together with the suggested WHO threshold of 0.05 (red). In panels B and C, solid blue and green lines represent median estimates across 104 model runs, while dashed lines depict 2.5th and 97.5th percentiles. is not entirely surprising, given the comparatively lower outward mobility of Belarusian236 population. It is also worth mentioning that much stricter travel restrictions implemented237 by the Belarus’ neighbors failed to stop the flow of SARS-CoV-2 across the borders in238 both directions. Furthermore, approximately half of estimated introductions did not239 appear directly across the border, which emphasize that Belarus, like most countries in240 the world, is a part of global interconnected environment and as such, affects and is241 affected by epidemiological developments in other countries.242 Second, the estimation of the effective reproduction number Re allowed the prelimi-243 nary assessment of the effect of limited NPIs implemented in the country during the first244 epidemic wave. These estimates should be interpreted only in comparison with similar es-245 timates for other countries. The analysis suggests a moderate but statistically significant246 decrease ofRe after the NPIs were put in action (Fig. 3B). The magnitude of decrease,247 however, is lower in comparison to the countries with broader and stricter NPIs (Table248 S2). Furthermore, the estimated median effective reproduction number ˆRe = 1.70 (CI:249 (1.45; 1.96)) over the entire analyzed period for Belarus is comparable with the estimates250 9 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprintthis version posted April 19, 2021. ; https://doi.org/10.1101/2021.04.13.21255404doi: medRxiv preprint ofRe in developed countries before the introduction of strict NPIs [47, 37, 19, 38]. For251 example, for Victoria, Australia this value is 1 .63 (CI: (1.45; 1.8)) [47]. On the other252 hand, the estimate of Re for Belarus is also close to the estimate of Re over the same253 time period for neighboring Ukraine, where the scope of implemented NPIs has been sig-254 nificantly broader. In our opinion, the latter fact is not entirely surprising and is more255 reflective of the reported extensive violations of lockdown and distancing measures in256 Ukraine and limited ability of authorities to control the epidemics [4, 7]. Similar estimate257 Re = 1.76 (0.91; 2.71) has been also reported for Russia [35] which borders both Belarus258 and Ukraine. This comparison of three ex-USSR countries suggests that regional demo-259 graphic and social specifics could be important factors for COVID-19 epidemiology along260 with NPIs. Study of such factors should be the subject of further investigation.261 Third, the true number of infections by the end of May, 2020 is most likely ∼ 4 (CI262 :(2; 9)) times higher than the detected number of cases, which is expected for respiratory263 diseases in general and for COVID-19 in particular [53, 50]. For example, according264 to observed seroprevalence of SARS-CoV-2 antibodies, in the USA the total number of265 COVID-19 infections in March-May, 2020 was probably between 6 to 24 times the number266 of reported cases [31].267 It is important to highlight that the presented study has several limitations. The first268 of them is the scarcity of currently available genomic data, especially in comparison with269 most of other European countries. Our approach strives to compensates for it by utilizing270 informative priors and linked models for phylogenetic and phylodynamics inference. BD-271 SKY models are also sensitive enough and suitable for inference even for smaller genomic272 datasets. For example, the numbers of sequences and/or density of branching events in273 this study is similar to those in other studies [49, 38, 26], where meaningful estimates274 ofRe have been produced for several epidemics, including SARS-CoV-2. Nevertheless,275 the inference precision could have been higher, if more SARS-CoV-2 genomes have been276 available. For Belarus, however, significant expansion of the available genomic dataset in277 the near future is unlikely, and in our opinion the lack of other studies justifies the need278 to fill the knowledge gap and to report the results based on the existing data. We also279 hope that this study will serve as a trigger for further SARS-CoV-2 genomic epidemiology280 studies in Belarus and will encourage funding increase and the corresponding development281 and expansion of sequencing facilities for molecular surveillance.282 The second limitation is that phylogeographic inference of introduction sources can be283 sensitive to sampling bias and can be affected by relatively slow accumulation of mutations284 in SARS-CoV-2 genomes [35, 28, 38]. In particular, even though no transmission links285 with Ukraine has been detected, it is likely that such links will emerge when more data286 from both countries will become available. Thus, SARS-CoV-2 phylogeography analysis287 should always be treated with a grain of salt, even though transmission history presented288 in this study is consistent enough and agrees with the travel records for those cases289 when they are available. The source inference for Belarus during the early pandemic can290 actually be more accurate than for some other regions, since Belarusian lineages were291 established after most of their source lineages were already sufficiently diversified. The292 incorporation of the global travel statistics into the “mugration model” of [46] may also293 have contributed towards the increase in transmission inference accuracy. Finally, for294 the case of Belarus, even if new data refine estimation of sources of some lineages, the295 obtained results are likely reflecting a true trend towards the higher prevalence of regional296 and neighbor-to-neighbor virus importations.297 10 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprintthis version posted April 19, 2021. ; https://doi.org/10.1101/2021.04.13.21255404doi: medRxiv preprint The third limitation is the sparsity of the SARS-CoV-2 incidence and testing data. In298 contrast to other countries [8, 12], Belarusian COVID-19 statistics are currently reported299 only for the entire country rather than for specific regions. The reported numbers of300 tests are not dichotomized into first-time tests and retests, those conducted by state301 or commercial laboratories, PCR and antibody tests. Furthermore, the sampling for302 testing is likely incomplete and biased towards individuals with COVID-19 symptoms303 and their close contacts and, for instance, persons who were tested upon arrival or prior304 to departure from the country. These issues may result in underestimation of the true305 number of cases, even though we are employing a method that is supposed to take them306 into account. For example, if a significant number of recovered individuals were tested at307 least twice, then the adjusted proportion of positive tests among those who are getting308 tested the first time will be higher and, consequently, the estimates of the number of309 cases will also increase. Furthermore, the aforementioned issues impede the development310 of stochastic agent-based models that otherwise can be used for high-precision analysis311 and forecasting. If (or when) more precise data will become available, it can be used to312 improve the precision and accuracy of our estimates.313 In conclusion, this study demonstrates the power of SARS-CoV-2 surveillance using314 combined genomic and epidemiological data. For such resource-constrained countries as315 Belarus, it is vitally important to develop sequencing facilities, detailed statistics and316 analytical resources to the level already established in other countries. These facilities317 and resources should become integral parts of the national mechanism to respond to318 emergence, re-emergence and spread of SARS-CoV-2 and other pathogens.319 5 Data availability320 The sequences used in this study are available at GISAID [48]. The Matlab scripts,321 Nextstrain configuration files, BEAST 2 XML files used to perform the described analy-322 ses and the acknowledgements table with sequence accession numbers and names of re-323 searchers and laboratories who produced the sequences are available athttps://github.com/324 compbel/COVID-Belarus.325 6 Acknowledgements326 PS was supported by the National Institutes of Health grant 1R01EB025022 and by the327 National Science Foundation grant 2047828.328 7 Contributions329 AN performed a phylogenetic analysis, analyzed genomic data and wrote the paper. AEA330 performed a phylogenetic analysis and analyzed genomic data. EG, KB, LV and AK pre-331 pared and handled genomic and associated epidemiological data, carried out the primary332 sequence processing. OG analyzed genomic data and wrote the paper. AK supervised333 the incidence data analysis, processed and analyzed incidence data, wrote the paper. PS334 designed and supervised the study, designed and implemented bioinformatics algorithms,335 performed a phylogenetic analysis, analyzed genomic data and wrote the paper.336 11 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprintthis version posted April 19, 2021. ; https://doi.org/10.1101/2021.04.13.21255404doi: medRxiv preprint References337 [1] Ministry of Health of Republic of Belarus: Belarusian woman from Vitebsk has338 been tested positive for coronavirus (in Russian). http://minzdrav.gov.by/ru/339 sobytiya/minzdrav-respubliki-belarus-informiruet/, 2020. Accessed: 2021-340 03-31.341 [2] Ministry of Health of Republic of Belarus: Six new coronavirus cases have been342 confirmed in Belarus (in Russian). https://minzdrav.gov.by/ru/sobytiya/shest-343 sluchaev-koronavirusa-podtverzhdeno-v-belarusi/ , 2020. Accessed: 2021-03-344 31.345 [3] Ministry of Health of Republic of Belarus: The first imported case of the coronavirus346 has been registered in Belarus (in Belarusian). http://minzdrav.gov.by/sobytiya/347 v-belarusi-zaregistrirovan-zavoznoy-sluchay-koronavirusa/ , 2020. Ac-348 cessed: 2021-03-31.349 [4] UN report: Impact of COVID-19 on human rights in Ukraine . https://350 www.ohchr.org/Documents/Countries/UA/Ukraine COVID-19 HR impact EN.pdf,351 2020. Accessed: 2021-03-31.352 [5] WHO Coronavirus disease 2019 (COVID-19). Situation report 40. http://353 hdr.undp.org/en, 2020. Accessed: 2021-03-31.354 [6] WHO COVID-19 Technical mission of experts to the Republic of Belarus: 8–11 April355 2020. Executive summary. https://www.euro.who.int/en/countries/belarus/356 publications/covid-19-technical-mission-of-experts-to-the-republic-357 of-belarus-811-april-2020.-executive-summary, 2020. Accessed: 2021-03-31.358 [7] Atlantic Council: Ukraine’s local authorities and the Covid-19 pandemic.359 https://www.atlanticcouncil.org/blogs/ukrainealert/ukraines-local-360 authorities-and-the-covid-19-pandemic/, 2021. Accessed: 2021-04-02.361 [8] COVID-19 Map - Johns Hopkins Coronavirus Resource Center. https://362 coronavirus.jhu.edu/map.html, 2021. Accessed: 2021-03-31.363 [9] KCMD Dynamic Data Hub. https://bluehub.jrc.ec.europa.eu/migration/app/,364 2021. Accessed: 2021-03-31.365 [10] Ministry of Public Health of the Republic of Belarus - Official Telegram Channel (in366 Russian). https://t.me/s/minzdravbelarus, 2021. Accessed: 2021-03-31.367 [11] pangolin.cog-uk.io. https://pangolin.cog-uk.io/, 2021. Accessed: 2021-03-31.368 [12] The National Health Service of Ukraine (NHSU) - COVID-19 Dashboard. https:369 //nszu.gov.ua/e-data/dashboard/covid19, 2021. Accessed: 2021-03-31.370 [13] United Nations Human Development Reports. http://hdr.undp.org/en, 2021. Ac-371 cessed: 2021-03-31.372 12 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprintthis version posted April 19, 2021. ; https://doi.org/10.1101/2021.04.13.21255404doi: medRxiv preprint [14] WHO. Public health criteria to adjust public health and social measures in the con-373 text of COVID-19. https://www.who.int/docs/default-source/coronaviruse/374 situation-reports/20201012-weekly-epi-update-9.pdf, 2021. Accessed: 2021-375 03-31.376 [15] Claudia Alteri, Valeria Cento, Antonio Piralla, Valentino Costabile, Monica Tallarita,377 Luna Colagrossi, Silvia Renica, Federica Giardina, Federica Novazzi, Stefano Gaiarsa,378 et al. Genomic epidemiology of sars-cov-2 reveals multiple lineages and early spread379 of sars-cov-2 infections in lombardy, italy. Nature communications, 12(1):1–13, 2021.380 [16] Kristian G Andersen, Andrew Rambaut, W Ian Lipkin, Edward C Holmes, and381 Robert F Garry. The proximal origin of sars-cov-2. Nature medicine, 26(4):450–452,382 2020.383 [17] Anders ˚Aslund. Responses to the covid-19 crisis in russia, ukraine, and belarus.384 Eurasian Geography and Economics, pages 1–14, 2020.385 [18] Remco Bouckaert, Joseph Heled, Denise K¨ uhnert, Tim Vaughan, Chieh-Hsi Wu,386 Dong Xie, Marc A Suchard, Andrew Rambaut, and Alexei J Drummond. Beast387 2: a software platform for bayesian evolutionary analysis. PLoS Comput Biol,388 10(4):e1003537, 2014.389 [19] Gonch´ e Danesh, Baptiste Elie, Yannis Michalakis, Mircea T Sofonea, Antonin Bal,390 Sylvie Behillil, Gr´ egory Destras, David Boutolleau, Sonia Burrel, Anne-Genevi` eve391 Marcelin, et al. Early phylodynamics analysis of the covid-19 epidemic in france.392 medRxiv, 2020.393 [20] Alena A Dastanka. Multilateralism in foreign policy of belarus: European and eu-394 arasian dimension. Regional Formation and Development Studies , 16(2):16–23, 2015.395 [21] Nicholas G Davies, Sam Abbott, Rosanna C Barnard, Christopher I Jarvis, Adam J396 Kucharski, James D Munday, Carl AB Pearson, Timothy W Russell, Damien C Tully,397 Alex D Washburne, et al. Estimated transmissibility and impact of sars-cov-2 lineage398 b. 1.1. 7 in england. Science, 2021.399 [22] Xianding Deng, Wei Gu, Scot Federman, Louis Du Plessis, Oliver G Pybus, Nuno R400 Faria, Candace Wang, Guixia Yu, Brian Bushnell, Chao-Yang Pan, et al. Genomic401 surveillance reveals multiple introductions of sars-cov-2 into northern california. Sci-402 ence, 369(6503):582–587, 2020.403 [23] Louis du Plessis, John T McCrone, Alexander E Zarebski, Verity Hill, Christopher404 Ruis, Bernardo Gutierrez, Jayna Raghwani, Jordan Ashworth, Rachel Colquhoun,405 Thomas R Connor, et al. Establishment and lineage dynamics of the sars-cov-2406 epidemic in the uk. Science, 371(6530):708–712, 2021.407 [24] Joseph R Fauver, Mary E Petrone, Emma B Hodcroft, Kayoko Shioda, Hanna Y408 Ehrlich, Alexander G Watts, Chantal BF Vogels, Anderson F Brito, Tara Alpert,409 Anthony Muyombwe, et al. Coast-to-coast spread of sars-cov-2 during the early410 epidemic in the united states. Cell, 2020.411 13 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprintthis version posted April 19, 2021. ; https://doi.org/10.1101/2021.04.13.21255404doi: medRxiv preprint [25] Yuriy Gankin, Vladimir Koniukhovskii, Alina Nemira, Gerardo Chowell, Thomas A412 Weppelmann, Pavel Skums, and Alexander Kirpich. Investigating the first wave of413 the covid-19 pandemic in ukraine using epidemiological and genomic sequencing data.414 medRxiv, 2021.415 [26] Lily Geidelberg, Olivia Boyd, David Jorgensen, Igor Siveroni, Fabr´ ıcia F Nascimento,416 Robert Johnson, Manon Ragonnet-Cronin, Han Fu, Haowei Wang, Xiaoyue Xi, et al.417 Genomic epidemiology of a densely sampled covid-19 outbreak in china. Virus evo-418 lution, 7(1):veaa102, 2021.419 [27] Jemma L Geoghegan, Xiaoyun Ren, Matthew Storey, James Hadfield, Lauren Jelley,420 Sarah Jefferies, Jill Sherwood, Shevaun Paine, Sue Huang, Jordan Douglas, et al.421 Genomic epidemiology reveals transmission patterns and dynamics of sars-cov-2 in422 aotearoa new zealand. Nature communications, 11(1):1–7, 2020.423 [28] Ana S Gonzalez-Reiche, Matthew M Hernandez, Mitchell J Sullivan, Brianne Ciferri,424 Hala Alshammary, Ajay Obla, Shelcie Fabre, Giulio Kleiner, Jose Polanco, Zenab425 Khan, et al. Introductions and early spread of sars-cov-2 in the new york city area.426 Science, 369(6501):297–301, 2020.427 [29] James Hadfield, Colin Megill, Sidney M Bell, John Huddleston, Barney Potter, Charl-428 ton Callender, Pavel Sagulenko, Trevor Bedford, and Richard A Neher. Nextstrain:429 real-time tracking of pathogen evolution. Bioinformatics, 34(23):4121–4123, 2018.430 [30] Masami Hasegawa, Hirohisa Kishino, and Taka-aki Yano. Dating of the human-ape431 splitting by a molecular clock of mitochondrial dna. Journal of molecular evolution,432 22(2):160–174, 1985.433 [31] Fiona P Havers, Carrie Reed, Travis Lim, Joel M Montgomery, John D Klena, Aron J434 Hall, Alicia M Fry, Deborah L Cannon, Cheng-Feng Chiang, Aridth Gibbons, et al.435 Seroprevalence of antibodies to sars-cov-2 in 10 sites in the united states, march436 23-may 12, 2020. JAMA internal medicine, 180(12):1576–1586, 2020.437 [32] Xi He, Eric HY Lau, Peng Wu, Xilong Deng, Jian Wang, Xinxin Hao, Yiu Chung438 Lau, Jessica Y Wong, Yujuan Guan, Xinghua Tan, et al. Temporal dynamics in viral439 shedding and transmissibility of covid-19. Nature medicine, 26(5):672–675, 2020.440 [33] Joseph Heled and Remco R Bouckaert. Looking for trees in the forest: summary tree441 from posterior samples. BMC evolutionary biology, 13(1):1–11, 2013.442 [34] Kazutaka Katoh and Daron M Standley. Mafft multiple sequence alignment soft-443 ware version 7: improvements in performance and usability. Molecular biology and444 evolution, 30(4):772–780, 2013.445 [35] Andrey B Komissarov, Ksenia R Safina, Sofya K Garushyants, Artem V Fadeev,446 Mariia V Sergeeva, Anna A Ivanova, Daria M Danilenko, Dmitry Lioznov, Olga V447 Shneider, Nikita Shvyrev, et al. Genomic epidemiology of the early stages of the448 sars-cov-2 outbreak in russia. Nature communications, 12(1):1–13, 2021.449 [36] Alessia Lai, Annalisa Bergna, Carla Acciarri, Massimo Galli, and Gianguglielmo450 Zehender. Early phylogenetic estimate of the effective reproduction number of sars-451 cov-2. Journal of medical virology , 92(6):675–679, 2020.452 14 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprintthis version posted April 19, 2021. ; https://doi.org/10.1101/2021.04.13.21255404doi: medRxiv preprint [37] Danielle Miller, Michael A Martin, Noam Harel, Omer Tirosh, Talia Kustin, Moran453 Meir, Nadav Sorek, Shiraz Gefen-Halevi, Sharon Amit, Olesya Vorontsov, et al. Full454 genome viral sequences inform patterns of sars-cov-2 spread into and within israel.455 Nature communications, 11(1):1–10, 2020.456 [38] Sarah A Nadeau, Timothy G Vaughan, J´ er´ emie Scir´ e, Jana S Huisman, and Tanja457 Stadler. The origin and early spread of sars-cov-2 in europe. Proceedings of the458 National Academy of Sciences, 118(9), 2021.459 [39] Lam-Tung Nguyen, Heiko A Schmidt, Arndt Von Haeseler, and Bui Quang Minh.460 Iq-tree: a fast and effective stochastic algorithm for estimating maximum-likelihood461 phylogenies. Molecular biology and evolution , 32(1):268–274, 2015.462 [40] Tal Pupko, Itsik Pe, Ron Shamir, and Dan Graur. A fast algorithm for joint re-463 construction of ancestral amino acid sequences. Molecular biology and evolution ,464 17(6):890–896, 2000.465 [41] Andrew Rambaut, Alexei J Drummond, Dong Xie, Guy Baele, and Marc A Suchard.466 Posterior summarization in bayesian phylogenetics using tracer 1.7. Systematic biol-467 ogy, 67(5):901, 2018.468 [42] Andrew Rambaut, Edward C Holmes, ´Aine O’Toole, Verity Hill, John T McCrone,469 Christopher Ruis, Louis du Plessis, and Oliver G Pybus. A dynamic nomenclature470 proposal for sars-cov-2 lineages to assist genomic epidemiology. Nature microbiology,471 5(11):1403–1407, 2020.472 [43] Andrew Rambaut, Tommy T Lam, Luiz Max Carvalho, and Oliver G Pybus. Ex-473 ploring the temporal structure of heterochronous sequences using tempest (formerly474 path-o-gen). Virus evolution, 2(1):vew007, 2016.475 [44] Ettore Recchi, Emanuel Deutschmann, and Michele Vespe. Estimating transnational476 human mobility on a global scale. Robert Schuman Centre for Advanced Studies477 Research Paper No. RSCAS, 30, 2019.478 [45] Erica Richardson, Irina Malakhova, Irina Novik, Andrei Famenka, and WHO. Be-479 larus: health system review . World Health Organization. Regional Office for Europe,480 2013.481 [46] Pavel Sagulenko, Vadim Puller, and Richard A Neher. Treetime: Maximum-482 likelihood phylodynamic analysis. Virus evolution, 4(1):vex042, 2018.483 [47] Torsten Seemann, Courtney R Lane, Norelle L Sherry, Sebastian Duchene, An-484 ders Gon¸ calves da Silva, Leon Caly, Michelle Sait, Susan A Ballard, Kristy Horan,485 Mark B Schultz, et al. Tracking the covid-19 pandemic in australia using genomics.486 Nature communications, 11(1):1–9, 2020.487 [48] Yuelong Shu and John McCauley. Gisaid: Global initiative on sharing all influenza488 data–from vision to reality. Eurosurveillance, 22(13):30494, 2017.489 15 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprintthis version posted April 19, 2021. ; https://doi.org/10.1101/2021.04.13.21255404doi: medRxiv preprint [49] Tanja Stadler, Denise K¨ uhnert, Sebastian Bonhoeffer, and Alexei J Drummond.490 Birth–death skyline plot reveals temporal changes of epidemic spread in hiv and hep-491 atitis c virus (hcv). Proceedings of the National Academy of Sciences, 110(1):228–233,492 2013.493 [50] Silvia Stringhini, Ania Wisniak, Giovanni Piumatti, Andrew S Azman, Stephen A494 Lauer, H´ el` ene Baysson, David De Ridder, Dusan Petrovic, Stephanie Schrempft,495 Kailing Marcus, et al. Seroprevalence of anti-sars-cov-2 igg antibodies in geneva,496 switzerland (serocov-pop): a population-based study. The Lancet, 396(10247):313–497 319, 2020.498 [51] Timothy G Vaughan, Gabriel E Leventhal, David A Rasmussen, Alexei J Drummond,499 David Welch, and Tanja Stadler. Estimating epidemic incidence and prevalence from500 genomic data. Molecular biology and evolution , 36(8):1804–1816, 2019.501 [52] Stephen White, Ian McAllister, and Valentina Feklyunina. Belarus, ukraine and502 russia: East or west? The British journal of politics and international relations ,503 12(3):344–367, 2010.504 [53] Sean L Wu, Andrew N Mertens, Yoshika S Crider, Anna Nguyen, Nolan N505 Pokpongkiat, Stephanie Djajadi, Anmol Seth, Michelle S Hsiang, John M Colford,506 Art Reingold, et al. Substantial underestimation of sars-cov-2 infection in the united507 states. Nature communications, 11(1):1–10, 2020.508 16 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprintthis version posted April 19, 2021. ; https://doi.org/10.1101/2021.04.13.21255404doi: medRxiv preprint