Genomics, social media and mobile phone data enable mapping of SARS-CoV-2 lineages to inform health policy in Bangladesh

The study addresses how SARS-CoV-2 was introduced and disseminated within Bangladesh and how integrating viral genomics with population mobility data can inform national health policy. In the context of widespread global genomic surveillance and the necessity of such efforts in LMICs, Bangladesh implemented movement restrictions and non-pharmaceutical interventions early in the pandemic. However, assessing their impact was challenging due to constraints in testing and epidemiological surveillance. With a large, mobile population and substantial rural demographics, Bangladesh required high-resolution understanding of lineage introductions, community transmission, and the role of human mobility—particularly mass migration events—in shaping epidemic dynamics. The purpose was to characterize lineage emergence, timing of introductions, spatial spread, and the impact of mobility patterns on transmission to guide targeted interventions.

The paper situates its work within global genomic epidemiology efforts enabled by GISAID, which have supported vaccine development and tracking of variants of concern (VOCs). Prior studies have mapped introductions and lineage dynamics in high-income countries (e.g., UK, US, New Zealand) and underscored differences in transmission dynamics in low- and middle-income settings. Research on the effects of non-pharmaceutical interventions (NPIs) has largely focused on HICs, with limited evidence on how measures like stay-at-home orders may have different consequences in LMICs due to population mobility and labor patterns. Additionally, studies have linked host genetic factors (e.g., Neanderthal-derived haplotypes) to COVID-19 severity, with high carrier frequencies reported in Bangladesh, reinforcing the need for robust surveillance. Work using mobility data (Facebook, telecom CDRs) has shown associations between human movement and infectious disease spread, motivating integration of mobility with genomic data in this setting.

Design: Observational genomic epidemiology integrated with digital mobility data. Sampling and sequencing (2020): Sequenced 67 SARS-CoV-2-positive nasopharyngeal swab samples collected by IEDCR between 5 March and 5 July 2020; combined with 324 publicly available Bangladeshi genomes for a total of 391 Bangladesh sequences in 2020 analyses. Global context included 68,870 GISAID sequences (as of July 2020) to contextualize introductions. Sampling and sequencing (2021): Sequenced an additional 85 samples collected between 11 November 2020 and 15 April 2021 to investigate VOC emergence (Alpha/B.1.1.7 and Beta/B.1.351) and ongoing transmission. Laboratory methods: RNA re-extraction (QIAamp Viral RNA Mini Kit); RT–PCR confirmation using WHO-recommended E and N gene assays; samples with Cq<31 selected. cDNA synthesis with SuperScript II or LunaScript RT SuperMix; second-strand synthesis with Q5 High-Fidelity DNA Polymerase. Libraries prepared with NEBNext Ultra II kits and sequenced (MiSeq and nanopore MinION flow cells per noted protocols: initial set per Novel SARS-CoV-2 sequencing protocol v2; later per 2019-nCoV sequencing protocol v3.0). Consensus genomes generated by mapping to Wuhan reference (GenBank MN908947.3) using pipeline tools (ARTIC-cg/biome guidelines for SNP calling and consensus generation; software versions as cited). Phylogenetics and phylodynamics: Maximum-likelihood phylogeny of 391 Bangladesh genomes within a global tree (68,870 sequences). Quality filtering removed sequences with >12 ambiguous sites. Temporal signal assessed with TempEst v1.5 (R=0.2471), supporting molecular clock analyses. Time-scaled phylogenies inferred using BEAST v1.10.4 with an exponential coalescent prior, HKY+I substitution model, strict clock; MCMC sampling (100,000 steps, sampling every 1,000; 10% burn-in); convergence assessed with Tracer v1.7.1; maximum clade credibility trees visualized. Nextstrain builds (as of 20 April 2021; 1,489 representative global sequences, 166 Bangladesh sequences highlighted) used to contextualize 2021 lineages and VOCs. Mobility data: Facebook Data for Good—aggregated, anonymized daily location data (six-hour windows) from users providing location services via Facebook/Instagram; district-level daily aggregates from 2 March 2020; baseline averages computed over the 45 days preceding 22 March stratified by day-of-week and time-of-day; daily percent change relative to baseline computed per district. Telecom CDRs—aggregated trips derived from changes in subscribers’ assigned cell towers day-to-day, provided by three operators (Grameenphone, Banglalink, Robi Axiata); data aggregated daily at the Upazila level; origin-destination trip counts between all Upazila pairs from 27 April 2020 onward; coverage ~100 million subscribers. Long-distance trips defined as >50 km; monthly distributions analyzed, highlighting Eid (late July 2020). Statistical notes: No a priori sample size calculations; no exclusions unless C>3; descriptive statistics used for mobility (percent change, CI for long-distance trip proportions).

- Introductions and timing: Bayesian time-scaled phylogenetics indicate SARS-CoV-2 was introduced into Bangladesh from abroad around mid-February 2020, preceding the first reported cases on 8 March 2020. - Dominant early lineages: By late March–April 2020, three lineages dominated—B.1, B.1.1.25 (also written as B.1.125 in parts of the text), and B.1.36. During end of March, they comprised approximately 19% (B.1), 55% (B.1.1.25/B.1.125), and 8% (B.1.36) of sequenced samples. - Origins: B.1.1 and B.1.1.25 were derived from lineages established outside Bangladesh, with evidence for importations from Europe and the United States; B.1.36 likely originated from a single importation linked to a traveller from Saudi Arabia (tested positive on 22 March in Chattogram). - Spatial spread: B.1.1 and B.1.1.25 were widely dispersed across divisions; B.1.36 predominated in southern Bangladesh, with 64% of its isolates in Chattogram. - Lineage A: Not detected until April 2020 and did not expand in Bangladesh through 2020–2021. - Mobility and dissemination: Facebook data show ~14.2% of users left Dhaka between 23–26 March 2020, evidencing mass migration at the start of the national General Holiday. Telecom data (~100 million subscribers) show substantial long-distance travel continuing through summer 2020; during Eid (July), 71.1% (95% CI 70.1–71.2) of trips were >50 km versus 57.9% (95% CI 57.9–58.0) in August. - Link between mobility and lineage expansion: MRCA dates for the three dominant lineages immediately precede the late-March migration; rapid, countrywide dissemination of B.1, B.1.1.25, and B.1.36 followed the mass movement from Dhaka. - VOCs in 2021: Multiple introductions of Alpha (B.1.1.7) and Beta (B.1.351) occurred after December 2020; a sustained community transmission chain of B.1.351 was established by February 2021, with Beta accounting for 47% of randomly sampled sequences and predominating in Dhaka. - Policy relevance: Findings informed government actions including quarantine of VOC cases/contacts and restrictions on intercity movement to limit spread beyond Dhaka.

Integrating genomic and mobility data revealed that while multiple international introductions seeded SARS-CoV-2 in Bangladesh until late March 2020, subsequent mass migration out of Dhaka coincided with clonal expansion and rapid nationwide dissemination of three dominant lineages. This pattern underscores how LMIC-specific social and economic contexts—such as large populations of transient urban workers—can alter the impact of NPIs, with stay-at-home orders inadvertently triggering mass movements that amplify spread. The findings directly address the research question by showing that mobility was a key driver of lineage dynamics and spatial spread, and that real-time genomic surveillance can detect introductions and expansions, including VOCs like Beta. The study highlights the necessity of tailoring interventions (e.g., targeted travel restrictions, focused quarantine, and enhanced surveillance) to national mobility patterns, especially around holidays like Eid. Given reduced vaccine efficacy against Beta and Delta variants reported elsewhere, timely detection and containment of VOCs are critical to mitigate transmission and severe outcomes in Bangladesh.

This work demonstrates the value of combining pathogen genomics with large-scale digital mobility data to reconstruct introductions, track lineage dynamics, and inform public health policy in Bangladesh. The study documents early international importations, the pivotal role of late-March 2020 mass migration in disseminating B.1, B.1.1.25, and B.1.36 nationwide, and the subsequent emergence and dominance of Beta (B.1.351) in early 2021, particularly in Dhaka. These insights supported rapid policy responses, including quarantines and intercity travel restrictions, to limit spread beyond urban centers. Future work should expand continuous genomic surveillance across all divisions, integrate additional mobility and behavioral datasets, assess vaccine effectiveness against circulating VOCs, and develop predictive frameworks to anticipate migration-driven transmission surges around policy changes and holidays.

- Sampling and representativeness: Although hundreds of genomes were analyzed, sequencing coverage relative to total cases was limited and may under-sample certain regions or time periods. The initial 2020 dataset combined 67 newly sequenced genomes with publicly available sequences, which can introduce biases related to where and when samples were collected. - Mobility data biases: Facebook data represent a subset of the population with enabled location services; telecom data, while broad (~100 million subscribers), are aggregated and infer trips from tower changes, which may misclassify short movements or undercount multi-leg trips. - Phylodynamic signal: Temporal signal was modest (TempEst R=0.2471), which can limit precision in MRCA dating and rate estimates. - Methodological constraints: No a priori statistical power calculations; descriptive analyses predominate. Potential inconsistencies or errors in some reported lineage percentages may reflect data heterogeneity or reporting artifacts. - Causal inference: While associations between migration and lineage expansion are strong and temporally consistent, causal attribution is observational and may be influenced by unmeasured confounders (e.g., testing access, local interventions).