Epidemiological and molecular forensics of cholera recurrence in Haiti

Cholera has affected Haiti with damping waves of outbreaks since October 2010. However, mechanisms behind disease persistence during lull periods remain poorly understood. By mid 2014, cholera transmission seemed to only persist in the northern part of Haiti. Meanwhile, cholera appeared nearly extinct in the capital, Port-au-Prince, where it eventually exploded in September 2014. This study aimed to determine whether this outbreak was caused by local undetected cases or by re-importation of the disease from the north. Applying an integrated approach between November 2013 and November 2014, we assessed the temporal and spatial dynamics of cholera using routine surveillance data and performed population genetics analyses of 178 Vibrio cholerae O1 clinical isolates. The results suggest that the northern part of the country exhibited a persisting metapopulation pattern with roaming oligoclonal outbreaks that could not be effectively controlled. Conversely, undetected and unaddressed autochthonous low-grade transmission persisted in the Port-au-Prince area, which may have been the source of the acute outbreak in late-2014. Cholera genotyping is a simple but powerful tool to adapt control strategies based on epidemic specificities. In Haiti, these data have already yielded significant progress in cholera surveillance, which is a key component of the strategy to eventually eliminate cholera.

S1 Appendix: List of analyzed VNTRs S1  We also identified and characterized rainy seasons (S2 Table). The 2014 rainy season was among the shortest, and was the less wet of the four rainy seasons since 2011. It also was slightly warmer than previous ones. Departments). The overall characteristics of the three zones were summarized in S3 Table. S3

S4 Appendix: Comparison of epidemic period characteristics
The characteristics of the identified epidemic periods were summarized and compared using Kruskal-Wallis rank sum test for count data indicators (rainfall, cases, stool samples cultures and genotypes) and using Fisher's exact test for proportions (suspected cases with confirmation culture, and culture positivity ratio). Indicators and p-values are presented in S4 Table. S4

S5 Appendix: Comparison between FST and other genetic differentiation indices
To ensure reliability of our population genetics results, we compared FST estimated by the Weir and Cockerham q 3 , with three other common differentiation indices: the indexes D (Host) 4 , GST (Nei) 5 , GST (Hedrick) 6 .
The pairwise comparisons between FST and other indices point out strong linear relationships (S5 Figure). This confirms our results regardless of the index selected.
We initially chose FST because this index is widely used in the literature and facilitates informative comparisons, and because debates on possible biases in certain situations neither take away the interest of this marker in assessing population structure, nor seem to us to be conclusive 4,7-14 .

(Panel B) distribution by department and (Panel C) distribution by trimester.
Overall, the differentiation index FST between the 5 trimesters exhibited a similar temporal structure to that between the 3 periods (S6 Table). Indeed, no significant genetic differentiation was observed between trimesters T1 and T2, which nearly encompassed period P1, and between T3 and T4, which nearly encompassed period P2. Conversely, populations T1 and T2 appeared genetically different from T3 and T4. Besides, T3 and T4 were significantly different from T5, which nearly encompassed period P3 (S6 Table).
Spatially, the differentiation index FST between the 10 departments exhibited a similar structure to that between the three zones (S6 Table). DSO population, which includes PaP zone, exhibited a strong genetic differentiation with DSA, DSC, DSN and DSNE, which nearly encompassed North zone.
FST between DSO and departments covering the South zone (DSSE, DSNI, DSS, DSGA), were not significant, partly because of the small number of isolates (S6 Table).
Combining the five trimesters and the 10 departments, no significant genetic differentiation was observed between the five populations from the Ouest department (DSO1, DSO2, DSO3, DSO4 and DSO5) (S6 Table). This matches with the low FST indexes between PaP1, PaP2 and PaP3. Conversely,      Table). Distribution of these clusters among the V. cholerae O1 populations defined by the epidemic periods and the three zones is summarized in S4 Table. Cluster 1 grouped 116 of the 178 isolates. It mostly included isolates from the North zone throughout the three periods of the study and was the main cluster in these three populations. Cluster 2 grouped only nine isolates from the North1, North2 and South2 populations.
Cluster 3 grouped 53 isolates, including 27 isolates from the PaP zone throughout the three periods, and was the main cluster in these three populations. Cluster 3 also included isolates from the North1 and South1 populations. This confirms the marked genetic differentiation between the North and PaP zones. S8 Appendix: Multiple linear regression analysis of genetic, spatial and temporal distances between isolates

S7
The structuration of the entire population was further assessed via multiple linear regression, which analyzed the relationship between the pairwise genetic distances of all 178 isolates and the associated spatial and temporal distances: The Genetic distance corresponded to the number of locus variants between each pair of isolates. The temporal distance corresponded to the lag between the sampling of isolates (number of days /365).
The Euclidean distance was used to estimate the spatial distance from their coordinates. Analyses were computed using R version 3.2.1 for Mac 16 , and the car package The genetic distance between isolates appeared significantly correlated with their spatial distance (pvalue <0.0001) and their temporal distance (p-value <0.0001) (S8 Table). This confirmed the marked genetic differentiation of V. cholerae O1 isolates in time and space between November 2013 and November 2014 in Haiti. Another regression model analyzed the distances between the 16 isolates sampled in the PaP zone during the P3 period (PaP3 population), and the 157 isolates previously sampled across the country.

S8
For this subset of isolates, the genetic distance was significantly associated with the spatial distance (p-value <0.0001), but not with the temporal distance (p-value = 0.3) (S8 Table). This result suggests that the V. cholerae O1 isolates from the outbreak in Port-au-Prince in late 2014 likely originated from a genetically stable autochthonous population rather than imported strains from distant areas of the country.

S9 Appendix: Bayesian clustering for spatial population genetics
In order to justify the choice of our populations (North, Pap and South) and assess its effects on our results, we completed our analysis with a Bayesian clustering algorithm for spatial population genetics implemented in the TESS program 18