Estimating the risk of arbovirus transmission in Southern Europe using vector competence data

Arboviral diseases such as chikungunya, dengue, and Zika viruses have been threatening the European countries since the introduction in 1979 of the major vector Aedes albopictus. In 2017, more than three hundred of CHIKV autochthonous cases were reported in Italy, highlighting the urgent need for a risk assessment of arboviral diseases in European countries. In this study, the vector competence for three major arboviruses were analyzed in eight Ae. albopictus populations from Europe. Here we show that Southern European Ae. albopictus were susceptible to CHIKV, DENV-1 and ZIKV with the highest vector competence for CHIKV. Based on vector competence data and vector distribution, a prediction risk map for CHIKV was generated stressing the fear of CHIKV and to a lesser extent, of other arboviruses for Europe, calling us for new public health strategies.


Results
Southern european Ae. albopictus are highly susceptible to chikungunya and to a lesser extent, to dengue, and Zika viruses. To analyze the vector competence of European mosquito populations for CHIKV, DENV, and ZIKV, F1-F4 mosquito populations (Table 1) were used for each virus challenge experiment. The number of viral particles was estimated in the body, head, and saliva at 7 and 14 days post-infection (dpi) for CHIKV and 7, 14, and 21 dpi for DENV-1 and ZIKV (Table 2).
Among the three viruses, CHIKV ( Fig. 1a-c) provided the highest indexes of vector competence (infection, dissemination and transmission) than DENV-1 ( Fig. 1d-f) and ZIKV ( Fig. 1g-i). For CHIKV, while populations could show different infection rates (IR) at 7 dpi (Fisher's exact test: p < 10 −4 ), IR reached values higher than 79% at 14 dpi in all populations (Fig. 1a). Once the midgut is infected, mosquitoes should allow an active viral dissemination inside the mosquito general cavity; while only AAF (Faliro, Greece), AAE (Tenero, Switzerland), and AAC (Canton, China) populations showed dissemination efficiency (DE) higher than 70% at 7 dpi, all populations reached high and comparable DEs at 14 dpi ( > 62%; Fisher's exact test: p = 0.391; Fig. 1b). Active viral dissemination could lead to viral transmission with virus detected in mosquito saliva; except AAE, all populations showed transmission efficiency (TE) lower than 33% at 7 dpi and 37.5% at 14 dpi (Fig. 1c). Collectively, all populations showed the same profiles at 14 dpi, IR > 80%, DE > 62%, and TE between 8% and 33% suggesting a slightly higher viral blocking at the salivary glands than at the midgut level.

Discussion
Vector competence data is a reliable predictor for emergence of arboviral diseases in European countries. Ae. albopictus populations were highly competent to CHIKV and to a lesser extent, to DENV-1 and ZIKV. Here we show how vector competence data combined with vector distribution, can provide an accurate risk map for CHIKV transmission, which matched with the occurrence of human local cases in sampled sites of Rome, Anzio and Guardavalle in Italy 23 .    www.nature.com/scientificreports www.nature.com/scientificreports/ Aedes albopictus has been first introduced in Europe in 1979 in Albania 24 and again, in Italy in 1990 25 . The species is present in 20 European countries 26 . Several studies using different genetic markers showed collectively a limited genetic differentiation among geographically distant populations reinforcing the recent invasion of the species mostly associated with human activities in mediating Ae. albopictus dispersal 22,27,28 .
Since 2007, this mosquito has been responsible in Europe for local CHIKV cases 11,13,23,29-31 and DENV cases 8,10 stressing that the species is a competent vector to both arboviruses. The vector competence assessing the ability of a mosquito to transmit a pathogen is a measure which can be modulated by genetic, epigenetic and also environmental factors 32 . Firstly, the outcome of infection depends on the specific combination between vector and pathogen genotypes described under genotype-by-genotype (G x G) interactions 33 . Notably, different mosquito generations in insectaries were used in this study; their vector competence for viruses could be affected due to adaptation to laboratory conditions. However, we initiated the mosquito population from a large size of field-collected mosquitoes to avoid inbreeding. This allowed to keep the population genetic diversity, and limit changes in mosquito fitness performance for 13 generations 34 . Thus, the results of mosquito vector competence for arboviruses were expected to be reliable. Using experimental viral challenges, it has been shown that Ae. albopictus from Southeast France was highly efficient to transmit CHIKV with virus detected in mosquito saliva from day 3 post-infection 35 , as was also Ae. albopictus from Italy 36 . Interestingly, CHIKV strains belonged exclusively to the ECSA genotype 11,13,23,29,31 . Nine years after the emergence of CHIKV in the Indian Ocean region, CHIKV was detected in October 2013 in Saint-Martin Island in the Caribbean 37 which against all expectations, belonged to the Asian genotype 38 . Despite several hundreds of CHIKV cases imported to continental Europe from the Americas, no autochthonous transmission of the Asian CHIKV was reported. A previous study showed that low temperatures limit the transmission of the Asian genotype 39 and not of the ECSA genotype providing further evidence that environmental factors such as the temperature can intervene in modulating the G x G interactions through genotype-by-genotype-by-environment (G x G x E) interactions 40 . www.nature.com/scientificreports www.nature.com/scientificreports/ To a lesser extent, local cases of DENV were also reported in Europe; in 2010, autochthonous transmission was related in Southern France 10 and Croatia 8 . Since then, several transmission episodes were periodically detected in Europe 41 . It has been demonstrated that Ae. albopictus in France were able to transmit DENV-1 from day 9 post infectious blood meal 35 . Mosquito populations tested in this study showed moderate infection, and low dissemination and transmission efficiencies. As for CHIKV, it has been shown that the environmental temperature can increase transmission of DENV 42,43 . The risk of dengue outbreaks is still a threat to Europe and recalls the past when dengue caused ~1 million cases in Athens, Greece, in 1927-1928 44 .
Zika has caused an outbreak impressive by its magnitude and rapid spread 45 . After its first detection in 2015 in Brazil, several million cases were reported in the Caribbean, and the Americas. Unexpectedly, severe symptoms have been described including neurological disorders and microcephaly in newborns leading to a global drive to limit this new health threat 46 . With the increasing number of imported ZIKV cases reported in Europe, local transmission of ZIKV was expected. The first European Zika autochthonous cases were reported in Hyères, France 47 , although the transmission pathway is still uncertain; the abundantly distributed vector, Ae. albopictus, in Southern Europe has increased the risk of mosquito-mediated Zika transmission. However, using experimental infections, it has been demonstrated that Ae. albopictus in Europe were poorly susceptible to ZIKV infection (Asian genotype) requiring at least 14 days to be excreted in mosquito saliva after an infectious blood meal [48][49][50][51] . Compared to CHIKV and DENV-1 mentioned above, ZIKV is the less-transmitted virus by Ae. albopictus indicating that the risk of autochthonous transmission of ZIKV in Europe is still minimal 52 .
Based on the vector competence analysis, European Ae. albopictus showed the highest susceptibilities to CHIKV. Therefore CHIKV dissemination and transmission of each population were analyzed by determining the theoretical thresholds for virus to escape from each anatomical barrier, midgut and salivary glands. According to the dissemination model, three groups could be roughly divided depending on their theoretical thresholds of dissemination, whereas the transmission model showed a similar theoretical threshold of transmission among all European populations. These results suggest that the significance of the midgut as barrier to viral dissemination depends on the mosquito population. However, once the populations ensured an efficient dissemination of CHIKV, transmission is quite similar as they are sharing the same potential to transmit. Interestingly, the non-European control, Canton population has distinct dissemination and transmission features than the other European populations; it showed the highest potential to disseminate and the lowest probability to transmit CHIKV, stressing again the genetic basis of vector competence depending on pairings vector and pathogen genotypes 33 .
In 2017, CHIKV autochthonous outbreaks have caused hundreds of infections in Italy 16 , raising the need for a risk prediction map. We elaborated an Ae. albopictus-driven prediction of CHIKV transmission risk. Based on previously predicted probability of occurrence of Ae. albopictus 4 , an European Ae. albopictus vector competence data-driven map of CHIKV transmission risk was generated. Although the predicted Ae. albopictus probabilities in the sampling localities were not particularly high except in Faliro region (Greece), the predicted risk of CHIKV transmission in European regions was significant: several regions share the same, and even higher risk of CHIKV transmission than in Anzio, Guardavalle, and Rome regions, where autochthonous CHIKV cases were reported in 2017 16 . Of note is that the spatial model used in this study relies on assumptions that may not fully reflect the diversity of interactions between Ae. albopictus and CHIKV. First, dissemination and transmission models were informed through feeding assays using fixed virus titers in blood meals offered to mosquitoes bred under controlled conditions in the lab, in contrast with wild mosquitoes feeding from humans with varying viral loads. In particular, the contribution of varying meteorological conditions on viral pathogenicity, dissemination or transmission efficiencies cannot be explored and is likely to modulate the ability to transmit back to a human host 53,54 . Likewise, the estimated risks are based on the population-specific median viral titers recorded in lab experiments. Second, the probability for a mosquito to acquire an arbovirus requires an estimate of its prevalence in humans; in absence of such data, the computed probabilities represent higher boundaries of the risk to transmit CHIKV to the human population. The use of data from Kraemer et al. 4 furthermore assumes constant probability of encountering Ae. albopictus throughout the year, ignoring seasonal variations in vector abundance. Third, the inverse-distance weighted spatial model would benefit from data taken at more scattered locations. However, the current experimental design, with most sampling sites being doubled within a few kilometers, allowed for robust results in a leave-one-out sensitivity analysis.
To conclude, environmental and biological factors are shaping the global distribution of mosquito vectors, and consequently, changing the epidemiology of associated diseases. As a newly emerging arboviral disease in Europe, CHIKV has become a new threat to European public health. Several information is crucial for the control of arboviral diseases; here, the vector competence of European Ae. albopictus for CHIKV, DENV, and ZIKV were analyzed. This study provides complete information on viral dissemination and transmission at different periods of viral incubation; the dissemination and transmission models help in understanding the virus propagation in mosquitoes, and reveal differences in vector competence among populations highlighting the risk of CHIKV outbreaks in Europe associated with Ae. albopictus. Adding data on vector competence to existing information on mosquito distribution holds particular promise for addressing epidemiological risks of CHIKV transmission at local, national and European scales. www.nature.com/scientificreports www.nature.com/scientificreports/ Animal Care and Use Committee (IACUC) at the Institut Pasteur. All infection experiments were conducted under biosafety level 3 conditions. This study did not involve endangered or protected species.

Materials and
Mosquito collections and rearing. Mosquito population samples were collected using ovitraps placed in different localities and countries in Southern Europe (Table 1) Seven-day-old female adults were fed on a blood meal containing 1.4 mL of washed rabbit red blood cells and 0.7 mL of viral suspension ( Table 1). The blood meal was supplemented with ATP as a phagostimulant at a final concentration of 1 mM. Mosquitoes were exposed to the blood using a Hemotek ® membrane feeding system.
Virus titers of blood meals were at 10 7 ffu/mL for CHIKV and DENV, and 10 7 pfu/ml for ZIKV. Engorged mosquitoes were transferred into boxes and fed ad libitum with 10% sucrose solution. Mosquitoes were maintained under a photoperiod of 12:12, at 28 °C until analysis. preparation of samples. Mosquito saliva was collected using the forced salivation technique 35  Assessment of vector competence. Three indexes were used to describe the vector competence of each combination virus -mosquito population. The infection rate (IR), dissemination efficiency (DE) and transmission efficiency (TE) illustrate the migratory route of the virus after ingestion by the mosquito. IR measures the proportion of mosquitoes with the midgut infected after the infectious blood meal resulting from a successful entry of the virus in the midgut epithelial cells followed by active replication. DE corresponds to the proportion of mosquitoes able to disseminate the virus from the midgut into the mosquito general cavity where the hemolymph contributes to the dissemination and infection of mosquito internal tissues and organs; the detection of the virus in mosquito heads means that the virus has disseminated from the midgut. Lastly, TE refers to the proportion of mosquitoes with virus having infected the salivary glands after penetration into acinar cells, replication and release of produced virus in the salivary conduct during mosquito blood feeding.
Modeling of vector competence. The virus efficiency to disseminate from the midgut to other tissues and then to salivary glands were modeled using a two-step logistic regression by restricting the mosquito population to those with a detectable viral load in the compartment of origin (i.e. midgut or head). We compared a model using viral load and mosquito population as explanatory covariates to a null model, not accounting for mosquito population, using a likelihood-ratio test. The performance of the resulting classifier was evaluated using the area under the curve (AUC) of the receiver operating characteristic curve (ROC). The probabilities of a successful dissemination and transmission were then derived from these models through inverse logit transformation, yielding population-specific probabilities of dissemination as a function of viral load. In the text, P50 and P75 values refer to 50% and 75% probability of dissemination or transmission, along with their corresponding viral load.
Kriging has often been used to model spatial processes, including in epidemiology 57 . Here, we used a simpler inverse distance weighting spatial model to calculate continuous variations of population-specific dissemination efficiencies. This model was applied to the vertices of the polygon defined by the geocoded sampling locations, with predicted probabilities taken at the median of observed viral titer in each compartment. Let d(i) and t(i) denote modeled probabilities of dissemination and transmission at sampled location i, with i corresponding to all European locations from Table 1 www.nature.com/scientificreports www.nature.com/scientificreports/ (2015) 4 , yielding an overview of mosquitos ability to get infected from human blood and transmit back in the general population. The resulting surface can be written as: The p exponent was fixed at 2 and distances computed as geodesic.
Statistical analysis. Statistical tests were conducted using the STATA software (StataCorp LP, Texas, USA) or R v3.5.2. Proportions were compared using Fisher's exact. P-values above 0.05 were considered non-significant. The maps in this manuscript were generated using the R software (packages raster v3.0.7 58 and gstat v2.0.3 59,60 ), based on data published by Kraemer et al. 4 .

Data availability
The data that support the findings of this study are available from the corresponding authors upon reasonable request.

Code availability
All analyses conducted in this manuscript and associated code will be made available upon request to the authors.