Integration of genomic sequencing into the response to the Ebola virus outbreak in Nord Kivu, Democratic Republic of the Congo

Abstract

On 1 August 2018, the Democratic Republic of the Congo (DRC) declared its tenth Ebola virus disease (EVD) outbreak. To aid the epidemiologic response, the Institut National de Recherche Biomédicale (INRB) implemented an end-to-end genomic surveillance system, including sequencing, bioinformatic analysis and dissemination of genomic epidemiologic results to frontline public health workers. We report 744 new genomes sampled between 27 July 2018 and 27 April 2020 generated by this surveillance effort. Together with previously available sequence data (n = 48 genomes), these data represent almost 24% of all laboratory-confirmed Ebola virus (EBOV) infections in DRC in the period analyzed. We inferred spatiotemporal transmission dynamics from the genomic data as new sequences were generated, and disseminated the results to support epidemiologic response efforts. Here we provide an overview of how this genomic surveillance system functioned, present a full phylodynamic analysis of 792 Ebola genomes from the Nord Kivu outbreak and discuss how the genomic surveillance data informed response efforts and public health decision making.

Extended Data Figure 1:. Frequent lineage migration between health zones sustained the outbreak.

Here, the overall phylogeny (see Figure 2 in the main text) is separated to show patterns of introduction and circulation within individual health zones for all lineages in the tree. Lineages are grouped by the health zone in which they circulated. Introductions are shown as circles at the beginning of each lineage. The color of the introduction circle indicates the donor health zone, and the x-axis position indicates the inferred timing of the introduction. While some lineages circulated in a health zone for long periods of time, most were short lived before moving into another health zone, as indicated by the relatively short branch lengths of many lineages. Visualization produced using BALTIC (github.com/evogytis/baltic/).

Extended Data Figure 2:. Patterns of transmission between health zones.

(A, B) The number of introductions of EVD into a health zone positively correlates with the number of exportations out of a health zone (r²=0.48, p<0.001), with most movement events occurring into and out of the same 5 health zones (Mabalako, Kalunguta, Katwa, Beni, and Mandima). State reconstructions that are less than 80% certain are excluded. (C) Heatmap showing the frequency of lineage migration between all pairs of affected health zones. A migration event is counted only if the phylogeographic reconstruction for both the source and the sink health zones is at least 80% certain. (D) The duration of time that a lineage circulated within a health zone is weakly correlated with the number of introduction events that a lineage seeded into other health zones (r²=0.21, p<0.003).

Extended Data Figure 3:. Inferred transmission dynamics are robust to sampling.

(A) Kernel density estimates for the same metrics presented in Figure 3. This analysis used a dataset subsampled to include 3 genomes per health zone per month (total n = 323 genomes). (B) Kernel density estimates for the same metrics presented in Figure 3. This analysis used a dataset subsampled to include 5 genomes per health zone per month (total n = 433 genomes). Inferences from the subsampled datasets recapitulate the findings shown in Figure 3, suggesting that phylogeographic inferences are robust to sampling frame.

Extended Data Figure 4:. Genomic characterization of transmission after unsafe burial of a pastor.

The horizontal axis represents nucleotide substitutions relative to the EBOV genome sequence from the pastor (KAT5915, orange). Three other samples had identical genome sequences to KAT5915. One case was from Oicha (light brown), one case was from Ariwara (neon yellow), and another was from Beni (green). Additional cases diverged by only one nucleotide were detected in Beni (green), Butembo (orange), and Kalunguta (purple).

Extended Data Figure 5:. Secondary transmission associated with infection of a motorcycle taxi driver.

The horizontal axis represents nucleotide substitutions relative to the EBOV genome sequence from the infected motorcycle taxi driver (MAN12309). Twenty other samples had identical genome sequences, as indicated in the figure by their position at 0 nucleotides diverged. Distance along the y-axis has no meaning, and only serves to separate samples for visualization. Additional sequenced cases in Mabalako were more genetically diverged from MAN12309, indicating additional propagated transmission following this event.

Figure 1:. Progress of genomic surveillance over the course of the outbreak.

(A) Total numbers of sequenced (orange) and unsequenced (grey) laboratory-confirmed cases of EVD as reported in WHO situation reports. (B) Correlation between the number of laboratory-confirmed cases reported in a health zone and the number of sequenced cases from a health zone. (C) Time lags between sample collection and release of phylogenetic analyses. In this figure each row represents a sample. The x position of a colored dot represents the date when a specific action occurred, and the color represents what the action was. Thus each row shows the amount of time that passed between different events for a single sequenced sample. Vertical lines represent events that occurred for a large proportion of samples. Dashed black lines represent when the WHO declared the outbreak start and end. (D) Kernel density estimates of lag times between sample collection and sequencing (orange), between sequencing and private release of the data (teal), and between sequencing and public release of the data (purple), prior to September 2019. (E) Kernel density estimates of lag times between sample collection and sequencing (orange), between sequencing and private release of the data (teal), and between sequencing and public release of the data (purple), after switching to privately-released Nextstrain Narrative situation reports in September 2019.

Figure 2:. Broad scale spatiotemporal dynamics of EVD in Nord Kivu.

(A) Temporally-resolved phylogenetic tree of 792 EBOV genomes colored by reporting health zone. The health zone of internal nodes is inferred via a discrete model and reduced confidence is conveyed by transitioning colors to gray. (B) Geographical spread of samples over four disjoint time intervals which span the entire outbreak. Figure adapted from Nextstrain visualizations. Note that three health zones, Manguredjipa (2 samples), Rwampara (4 samples) and Mwenga (4 samples), are located outside of the map as shown here.

Figure 3:. Transmission dynamics within and between health zones.

(A) Kernel density estimate of the inferred distance in kilometers between a source and a sink health zone, for 188 high confidence events where a viral lineage moved between two health zones; 50% of movement events occur between health zones that are less than 49km apart, and 95% of movement events occur between health zones less than 200km apart. (B) Kernel density estimate of the number of times a lineage was introduced into a different health zone. 50% of lineages seed less than 5 introduction events, and 95% of lineages seed less than 25 introduction events. (C) Kernel density estimate of the number of times EBOV was introduced into each health zone; 50% of health zones experienced less than 3 introduction events and 95% of health zones experienced less than 8 introduction events. (D) Kernel density estimate of the duration of time a lineage circulated within a single health zone; 50% of lineages circulated within a single health zone for less than 10 weeks, and 95% of lineages circulated within a single health zone for less than 40 weeks.

Figure 4:. Initial genomic evidence for an infection relapse event.

(A) Root-to-tip plot showing genetic divergence of all 792 genomes as a function of their sampling date. The regression line indicates the average substitution rate across this outbreak (1.17×10⁻³ substitutions per site per year, as annotated). (B) Temporally resolved phylogenetic tree showing the patient’s June sample (MAN4194), and December sample (MAN12309). (C) Phylogenetic tree showing nucleotide divergence from the root of this clade. The June infection (MAN4194) and December infection (MAN12309) are diverged by only 2 substitutions, T5587C and A6867G.