Rift Valley fever in northern Senegal : a modelling approach to analyse the processes underlying virus circulation recurrence

Rift Valley fever (RVF) is endemic in northern Senegal, a Sahelian area characterized by a temporary pond network that drive both RVF mosquito population dynamics and nomadic herd movements. To investigate the mechanisms that explain RVF recurrent circulation, we modelled a realistic epidemiological system at the pond level integrating vector population dynamics, resident and nomadic ruminant herd population dynamics, and nomadic herd movements recorded in Younoufere area [1]. To calibrate the model, serological surveys were performed in 2015-2016 on both resident and nomadic herds in the same area. Mosquito population dynamics were obtained from a published model trained in the same region [2]. Model comparison techniques were used to compare five different scenarios of virus introduction by nomadic herds associated or not with vertical transmission in Aedes vexans. Our serological results confirmed a long lasting RVF endemicity in resident herds (IgG seroprevalence rate of 15.3%, n=222), and provided the first estimation of RVF IgG seroprevalence in nomadic herds in West Africa (12.4%, n=660). Multivariate analysis of serological data suggested an amplification of the transmission cycle during the rainy season with a peak of circulation at the end of that season. The best scenario of virus introduction combined yearly introductions of RVFV from 2008 to 2015 (the study period) by nomadic herds, with a proportion of viraemic individuals predicted to be larger in animals arriving during the 2nd half of the rainy season (3.4%). This result is coherent with the IgM prevalence rate (4%) found in nomadic herds sampled during the 2nd half of the rainy season. Although the existence of a vertical transmission mechanism in Aedes cannot be ruled out, our model demonstrates that nomadic movements are sufficient to account for this endemic circulation in northern Senegal. Author summary Rift Valley fever (RVF) is one of the most important vector borne disease in Africa, seriously affecting the health of domestic ruminants and humans and leading to severe economic consequences. This disease is endemic in northern Senegal, a Sahelian area characterized by a temporary pond network that drive both RVF mosquito population dynamics and nomadic herd movements. Two non-exclusive mechanisms may support this endemicity: recurrent introductions of the virus by nomadic animals, and vertical transmission of the virus (i.e. from infected female mosquito to eggs) in local Aedes populations. The authors followed up during 1 year resident and nomadic herds. They used the data thus obtained to model a realistic epidemiological system at the pond level integrating vector population dynamics, resident and nomadic ruminant herd population dynamics. They found that the best scenario explaining RVF remanence combined yearly introductions of RVFV by nomadic herds, with a proportion of viraemic predicted to be larger in animals arriving during the 2nd half of the rainy season, which is consistent with an amplification of virus circulation in the area during the rainy season. Although the existence of a vertical transmission mechanism in Aedes cannot be ruled out, their results demonstrates that nomadic movements are sufficient to account for this endemic circulation in northern Senegal.


26
Rift Valley fever (RVF) is endemic in northern Senegal, a Sahelian area characterized by a temporary 27 pond network that drive both RVF mosquito population dynamics and nomadic herd movements. To 28 investigate the mechanisms that explain RVF recurrent circulation, we modelled a realistic arriving during the 2 nd half of the rainy season (3.4%). This result is coherent with the IgM prevalence 42 rate (4%) found in nomadic herds sampled during the 2 nd half of the rainy season. Although the existence 43 of a vertical transmission mechanism in Aedes cannot be ruled out, our model demonstrates that nomadic 44 movements are sufficient to account for this endemic circulation in northern Senegal.

45
46 Author summary: more substantial evidence of these mechanisms. A last putative mechanism implies rodents [42].

120
However, and despite evidences of association between rodents and RVFV circulation, their potential 121 implication in RVFV maintenance remains controversial [43]. Because of the absence of mosquitoes 122 during the dry season in the study area, it is unlikely that rodents contribute to maintain the virus during 123 this unfavourable season [42].

124
In the Ferlo, the set composed of a temporary pond, associated vectors, sedentary ruminants living 125 around, and nomadic herds seasonally settling around, constitutes the elementary unit of the RVF 126 epidemiological system. Rainfall variation is the main driver for pond surface fluctuations which affects 127 both the vector population dynamics and nomadic migration patterns. In addition, resident and nomadic 128 herd immunity impact the ability of RVFV to circulate. A better understanding this complex system is 129 needed to improve our comprehension of RVF epidemiology in this region, to quantify the main 130 determinants of RVF transmission and emergence and help establishing better surveillance, prevention 131 and control strategies.

132
The aim of this work was to model RVFV transmission in this epidemiological system, and to use this 133 model to infer on the respective contribution of nomadic movements and VT in RVFV recurrence. We

196
In May 2016, to assess the serological status of resident herds ling in the study area, a cross sectional 197 serological survey was performed in 6 resident herds living in 3 small villages located in the close 198 vicinity of Younoufere (Kodediare, Nacara and Soringo (Fig. 1)). Animals were randomly chosen 199 among herds whose owner agreed to participate to the study. As a whole, an age-structured sample of 200 222 small ruminants were blood sampled (no sample from resident cattle could be obtained     Two between-host transmission routes were considered: vector-borne, and direct, considering that when 294 viraemic hosts abort (or calve/lamb), the abortion (or calving/lambing) products are infectious, and 295 susceptible animals are exposed to these highly infectious materials (Fig. 2). Vector-borne transmission 296 was parameterized by a scaling factor for vector population sizes ( ), and direct transmission by a 297 transmission parameter ( ). In vectors, RVFV transmission from an infected female to its eggs was 298 parameterized by a VT probability ( ).

299
Since the host and vector population sizes remained constantly high in the modelled system throughout 300 the year (>500 individuals) and because the RVFV circulation level was also expected to be high due to 301 the documented recurrence of RVFV transmission in the area, we chose to implement the model in a 302 deterministic framework, with a discrete daily time step.

303
A full description of (i) the epidemiological model, (ii) its coupling with the entomological model, and

307
The model was simulated during an 8 years duration corresponding to the maximal lifespan of hosts.

308
Entomological inputs (vector population dynamics) were obtained from the EM for the 2008-2015 309 period, using daily rainfall, temperature and humidity data collected from the study area during this

341
The summary statistics consisted of the observed numbers of seropositive resident small ruminants per 342 age class (6 classes) found when simulating the serosurvey described above (i.e. during the 2016 dry 343 season, with the age-specific numbers of tested animals given in Table 4). The recommended settings of particles in the sequential Monte-Carlo procedure; and a final size of 10,000 particles to build 347 posterior probabilities [50].

348
The scenarios A and B were compared using a model selection procedure based on random forest 349 classification methods, specifically designed to allow model comparison in an approximate Bayesian 350 computation framework [52]. Using the same method, the need for VT and/or direct transmission 351 between hosts was tested. Five models were compared (Table 3) and the best model was used for 352 parameter estimation. When two models were equivalent, the most parsimonious was selected.

Surveys in sedentary herds and in nomads 360
One hundred-sixty eight resident sheep and fifty four goats were sampled (

407
Model fit obtained after the parameter estimation of model M4 appeared satisfactory (Fig. 3): the 408 predicted age-specific distributions of IgG seroprevalence rate matched the results of the serosurvey in 409 sedentary small ruminants, and the 95% confidence intervals of age-specific predicted distributions of 410 seroprevalence rates always included the observed value (Fig. 3).

Figure 3. Predicted distribution of seroprevalence rate by age class in sedentary small ruminants 413
Model is model M4 which combines direct transmission between hosts but no VT in Aedes, and repeated   in the Ferlo, (ii) herd movements in some specific environment can allow endemicity at a regional scale 536 while circulation is epidemic at a local scale, we quantitatively demonstrate in this work that although 537 the existence of a vertical transmission mechanism in Aedes cannot be ruled out, nomadic movements 538 are sufficient to account for this endemic circulation in the Ferlo area.

539
A next outbreak will inevitably occur in the Ferlo area, and the increased national and transboundary