1. Introduction
Emerging or re-emerging mosquito-borne viruses (moboviruses) are of growing concern in Europe [
1]. Several moboviruses circulate on the European continent [
2]. Thereby, West Nile virus (WNV, genus
Flavivirus, family
Flaviviridae) is of particular importance. This zoonotic virus belongs to the Japanese encephalitis serocomplex and is one the most widespread moboviruses in the world [
3,
4,
5]. Enzootic transmission takes place between birds as amplifying hosts and mosquitoes as vectors. WNV can cause high mortalities in birds, but spillover events also have significant public health consequences, e.g., headache, rash and even neurological complication [
6,
7]. Various outbreaks of WNV infections have been reported in southern and southeastern Europe, resulting in several thousand human cases with dozens of fatal outcomes [
8,
9]. Thereby, Romania is a hotspot for WNV circulation [
6,
8,
10,
11,
12,
13,
14]. Over the last two decades, the country has experienced at least three large outbreaks of WNV (1996, 2010, 2018) with a mortality rate of up to 20%. Genetic and phylogenetic analyses grouped the WNV strains into eight distinct evolutionary lineages, from which the most spread worldwide and associated with disease and outbreaks belong to lineages 1 and 2 [
15,
16]. The virus is transmitted and maintained in the natural cycle by mosquitoes (mostly of the
Culex genus) as vectors with birds as the main amplifying hosts, while humans and horses are considered incidental or dead-end hosts. Nowadays, West Nile virus exhibits a worldwide distribution throughout Africa, the Middle East, Europe, western Russia, southwestern Asia, and Australia [
15]. Starting early 1990s, the frequency, severity and geographic range of human WNV outbreaks increased with the appearance of new viral strains in Romania, Russia, Israel, and Greece [
17,
18]. In the western hemisphere, West Nile virus spread from its 1999 appearance in New York City throughout the Pacific Coast and Argentina in 2005 [
19,
20,
21]. Nowadays, the severity, magnitude and geographic location of the WNV outbreaks differs greatly, being instrumented by the local ecological conditions and increased anomalies of seasonal temperature. The Danube Delta Biosphere Reserve (DDBR) is the second largest wetland in Europe. This complex of ecosystems is predominantly located in Romania, with small parts also located in the Ukraine. The heterogeneous area of the DDBR has a high biodiversity with an important function as a major hub for bird migration [
22,
23]. Thus, the area has a high risk for the introduction of zoonotic pathogens. Introduced moboviruses find an abundant and diverse mosquito fauna [
9,
23]. Most of the DDBR is only accessible by boat. This makes comprehensive studies on mosquito fauna and associated viruses a difficult task, e.g., showcased by a recent pilot study, detecting two new mosquito species for Romania (
Aedes hungaricus and
Anopheles algeriensis) [
9].
The circulation of WNV in the DDBR was reported before 2014 [
12,
24]. WNV dynamics in the Danube Delta are positively correlated with temperature and negatively correlated with rainfall. However, there is still a lack of knowledge driving the risk of WNV transmission under near-natural conditions as present in the DDBR. Therefore, in this study, classical virus screening of mosquitoes was combined with a
xenosurveillance approach. Testing for WNV-specific antibodies in sentinel horses or chicken is a common monitoring tool in Europe [
25]. However, such a surveillance system is difficult to implement under the remote conditions of a wetland system and the results might not reflect the natural transmission cycle. In addition, comprehensive sampling from wild animals needs a lot of effort. In this study, mosquitoes were used as
“biological syringes”, i.e., blood-fed specimens were screened for WNV-specific antibodies and viral RNA. Experimental studies by Leigthon et al. [
26] demonstrated the potential of mosquitoes for sero-epidemiological studies. This was further supported by a field-study in Thailand detecting antibodies against dengue virus and Japanese encephalitis virus in blood-fed mosquitoes in two different mosquito species [
27]. However, a broad application of this method for different wild mosquito species, which feed on different vertebrate hosts, was missing.
Host-feeding patterns of blood-sucking arthropods shape the transmission cycle of vector-borne pathogens, offering direct insights into the interaction between vectors and hosts. However, there is still a lack of knowledge about the host spectrum of mosquitoes in Europe [
28]. Previous studies predominantly investigated certain combinations of vector species and pathogens, e.g.,
Culex spp./WNV [
29,
30,
31,
32],
Culex spp./avian malaria [
33],
Anopheles spp./malaria [
33,
34] and
Aedes albopictus as an invasive vector species for a variety of pathogens [
35]. Only few European studies analyzed the host-feeding patterns of a wide range of species [
36,
37,
38], which is required to better understand pathogen circulation.
Thus, in order to get comprehensive insight into the ecology of WNV in the DDBR, the mosquito fauna was studied in a longitudinal surveillance program over three years. Molecular assays were applied to (i) screen for WNV infections in mosquitoes, analyze the evolutionary mechanism of the virus and its dispersal patterns in Europe, in particular in Romania and the DDBR, (ii) detect WNV-specific antibodies in the blood meals from horses, dogs, humans and birds and (iii) identify potential vector species by analyzing the host-feeding patterns of the blood-fed mosquitoes.
2. Materials and Methods
Mosquitoes were collected within a longitudinal arbovirus surveillance program between 2014 and 2016 at two sampling sites in a rural/urban environment (Letea, Sulina) and two near-natural sampling sites (Dunărea Veche and Lake Roșuleț) in the DDBR. Each year, on average, every tenth day between April and September, three to four (2014, 2015) or one (2016) carbon dioxide-baited Heavy Duty Encephalitis Vector Survey trap(s) (Bioquip Products Inc., CA, USA) were installed at each site. A detailed description of the collection sites can be found in Török et al. [
9]. The DDBR Authority issued research permits (9/25.04.2014, 10692/ARBDD/25.04.2014; 7717/ARBDD/28.04.2016, 11/28.04.2016). The collected specimens were transported on dry ice, stored in the freezer and identified by morphology on chill tables using a stereomicroscope (Olympus SZX12, Tokyo, Japan) [
39]. Blood-fed mosquitoes were separated from unfed specimens. Furthermore, morphologically identified
Culex pipiens specimens were typed to species level (
Cx. pipiens pipiens f.
pipiens,
Cx. pipiens pipiens f.
molestus or
Cx. torrentium) using a molecular assay [
40].
For the WNV screening, mosquito pools between 1 and 250 specimens were pooled per sampling site and sampling date. Mosquitoes were put in 2 mL safe-lock tubes (Eppendorf, Hamburg, Germany) or 50 mL centrifuge tubes (Sarstedt, Nümbrecht, Germany) with zirconia beads (2 mm, BioSpec Products, Bartlesville, OK, USA) and 0.5 or 3 mL chilled high-glucose (4.5g/L) Dulbecco’s modified Eagle’s medium (DMEM) (Sigma-Aldrich, St. Louis, MO, USA). Mosquitoes were homogenized in a TissueLyser or TissueLyser II (Qiagen, Hilden, Germany) for 2 min at 30–50 Hz. The suspension was clarified by centrifugation for 1 min at 8000 rpm and 4 °C. RNA was extracted with a KingFisher Flex 96 Deep-Well Magnetic Particle Processor using the MagMAX CORE Nucleic Acid Purification Kit (ThermoFisher Scientific, Waltham, MA, USA). Samples were tested with pan-flavivirus RT-PCR modified from Chao et al. [
41] as described in detail by Becker et al. [
42]. WNV-positive mosquito pools were subjected to Sanger sequencing (LGC Genomics, Berlin, Germany) for complete genome sequencing [
43].
The blood-fed specimens were individually placed into 2 mL safe-lock tubes. Homogenization and extraction were conducted using the same protocol as described above. Thereby, 30 μL supernatant from each of ten specimens was pooled for WNV screening. Detection of WNV-RNA was conducted with the RealStar WNV RT-PCR Kit 1.0 (altona Diagnostics, Hamburg, Germany).
For the host identification, the supernatant of individual blood-fed specimens was heat-inactivated at 99 °C for 1 min in a Peqlab thermocycler (VWR International GmbH, Darmstadt, Germany) for the reduction of possible inhibitors. The PCR assay used the Phusion Blood Direct Master Mix (Thermo Fisher Scientific, MA, USA), 5 μL of the homogenate was used in a total of 30 μL reaction volume for PCR amplification of the cytochrome b gene [
44,
45]. Amplification was conducted by incubation for 5 min at 98 °C, followed by 40 cycles of 1 s at 98 °C, 5 s at 57 °C and 30 s at 72 °C, ending with incubation for 1 min at 72 °C. If the reaction with the first primer set yielded no result, the PCR reaction was repeated using another pair of vertebrate-specific primers targeting the 16S rRNA gene fragment [
46]. The same applied to potential mixed blood meals, as indicated by double peaks in the sequence electropherograms at different positions, resulting in unreadable chromatograms. For this PCR, amplification was conducted by incubation for 5 min at 98 °C, followed by 40 cycles of 1 s at 98 °C, 5 s at 50 °C and 30 s at 72 °C, concluded by incubation for 1 min at 72 °C. The amplicons were sequenced (LGC Genomics, Berlin, Germany) and analyzed with Geneious v9.1.7 (Biomatters, Auckland, New Zealand). Sequences were compared to available sequences from GenBank database (
https://blast.ncbi.nlm.nih.gov/). Host species were identified if the percentage identity was 95% or higher. The statistical computer program R [
47] was used for all data analyses. Data manipulation and visualization was conducted with functions from the packages plyr [
48], dplyr [
49], magrrittr [
50] and ggplot2 [
51]. Spearman’s rank correlation was used to analyze the statistical relationship between the number of analyzed specimens per mosquito species and the number of detected host species. For each mosquito species, higher order taxa (e.g., Anatidae, Bovidae, Chiroptera) were only considered for the calculations of host species, if no corresponding taxa of lower ranks were detected. The frequencies of detected birds, non-human mammals or humans between the six most abundant mosquito species and between the four sampling sites were compared with Chi-square tests with Bonferroni corrected p-values for multiple pairwise comparisons.
Horse-, human-, dog- and bird-derived blood meals were tested for WNV-specific IgG/IgY, using an indirect immunofluorescence (IIF) assay as described previously [
52]. Host species were selected, which are important amplifying hosts (bird), known to become critically ill from WNV infections (human, horse) or were previously identified to be suitable sentinel species for WNV (dog, horse) [
6,
7,
53,
54]. In brief, Vero cells infected with WNV NY99 were seeded on microscope slides with 12 reaction wells (Marienfeld, Lauda-Königshofen, Germany). Slides were treated with Acetone (99%), 15 µL of each sample (single mosquito homogenized in 500 µL) was transferred into one reaction well. Cells were washed with PBS and stained with Alexa Fluor
® 488-conjugated Alpaca Secondary Antibodies (Jackson ImmunoResearch, West Grove PA, USA 1:200 in 1% Evans blue solution), namely goat anti-human IgG, goat anti-horse IgG, rabbit anti-chicken IgY and rabbit anti-dog IgG antibodies, depending on the identified blood-meal source. In order to test for cross-reactivity with heterologous flaviviruses potentially circulating in the sampling area, the WNV IgG positive samples were also tested for Usutu virus- (USUV) and tick-borne encephalitis virus- (TBEV) specific IgG using the same assay with the respective virus.
Genomes obtained for WNV strains from Danube Delta were compared with all complete and partial publicly available NS5 gene sequences from Europe and Africa. Phylogenetic trees were inferred using the Bayesian Markov chain Monte Carlo (MCMC) approach available in BEAST v1.10 [
55]. Analyses were performed under the best fit nucleotide substitution model identified as the GTR +Γ for complete genome and TN93+Γ for partial NS5 datasets using jModelTest 2 [
56] and a prior MCMC was chosen by testing all models and determining Bayes factors (log
10 BF). We employed TempEst for an interactive regression approach to explore the association between genetic divergence through time and sampling dates [
57]. In order to assess the spatial temporal dynamics of WNV, the time to most recent common ancestor (tMRCA), and the effective population dynamics of WNV, we employed a relaxed uncorrelated log normal (UCLN) molecular clock, a flexible demographic model (coalescent Gaussian Markov Random field Bayesian Skyride model, GMRF) as the best demographic scenario detected. In all cases, each of the MCMC chain lengths was run for 10
8 generations (with 10% burn-in) and subsampled every 10
4 iterations to achieve convergence. The Bayesian maximum clade credibility (MCC) trees were visualized using FigTree v1.4.1 (
http://tree.bio.ed.ac.uk/software/figtree/). To test the hypothesis that WNV is periodically imported from Africa into Europe, a phylogeographic analysis was conducted using a discrete model attributing state characters represented by the detection locality of each strain and the Bayesian stochastic search variable (BSSV) algorithm implemented in BEAST v1.10 [
55].
4. Discussion
In this study, we elucidated the possible origin, pattern of spatial-temporal dynamics, and eco-epidemiological factors of WNV in the ecosystem of DDBR. Our phylogeographic analysis identified at least two distinct introduction events of WNV lineage 2 to Romania. It circulates under a number of different virus variants (EEC1 and WEC1) with South Africa/Senegal and Greece as a possible hub for the progenitor of WNV strains involved in the outbreaks in Romania. The presence of a geographically distinct WNV clade (WEC1) is likely due to very recent introduction, adaptation to the local ecological conditions and some geographic barriers such as climate, vegetation, and vector species. Furthermore, the long-term circulation (EEC1) and adaptation of the virus to the host populations and its enzootic maintenance lead to spread into new geographic regions and local virus variants (in situ evolution).
Although the overlap between the phylogenetic and geographical clustering of the Romanian and Russian members of the Eastern European clade of WNV lineage 2 was expected, it is interesting to note that the clade also contains an Italian strain. This suggests a new, independent introduction of the EEC1 in the south-central part of the continent. Similarly to Eastern Europe, the Italian Peninsula is crossed by major Afro-European bird migration routes. To date, the dispersion pattern of WNV into temperate Eurasia can be best explained by bird migration [
59,
60,
61,
62], with short-distance migratory species as potential mode of WNV spread within Europe [
62]. Interestingly, we found evidence of adaptive evolution in the WNV from Danube Delta also in non-structural genes, which likely indicates that the host immune selection pressure does not cause increases in viral fitness [
63]. Mutations observed at amino acid positions T108I (C), S196P and R361K (E), I1192V (NS2A) and G2932R (NS5) have been found to be involved in the formation of EEC1. Although the impact of these mutations mostly from the nonstructural genes is unclear (likely occurred due to introduction of WNV in this country), similar changes modulated the host antiviral response by inhibition of interferon signaling [
64]. The residue alternations R851K (NS1), I1462M (NS2B), R1516K (NS3), T2296A (NS4), N2305S (NS4) and R2719K (NS5) are specific for African variants. Similar patterns of convergent evolution have been described for WNV and suggest that a limited number of residue changes are permitted due to functional constraints [
65].
This study successfully used a xenosurveillance approach to monitor the presence of WNV-specific antibodies in different host species. As demonstrated previously [
66], mosquito-based surveillance allows non-invasive blood-sampling from free-roaming vertebrate hosts (e.g., feral horse) and from species which are rare or have a cryptic behavior (e.g., raccoon dog (
Nyctereutes procyonoides), European mink (
Mustela lutreola), Eurasian otter (
Lutra lutra) or Golden jackal (
Canis aureus)). This study supports previous studies, which identified horses and dogs as suitable sentinel species for WNV [
53,
54]. WNV seroprevalence in dogs (6.3%) and horses (3.1%) was similar to conventional sampling of the species in different areas with WNV activity [
67,
68,
69]. Nevertheless, the seroprevalence in horses was markedly lower than in southeastern Romania (15.1%) [
70]. WNV-specific antibodies were found in blood meals from horses and dogs in all four sampling sites, but not in mosquito blood-meals from human or bird. Thus, this indicate widespread, continuous WNV circulation, but probably only on a low level. In addition, due to potential cross-reactivity of the applied serological assay, we cannot exclude the possibility that one of the samples was also positive for USUV, a virus with a similar transmission cycle to WNV.
Feral horses and free-ranging cattle were the two most commonly detected host species, accounting for more than 50% of the detected hosts. These animals have their origins in pre-1990 state-owned collective farms and private homesteads from where they were released in more recent years. There is no official census published, but it is estimated that 4000 horses and a few other thousand cattle roam and reproduce freely in the DDBR. The high abundance in combination with the relatively huge body size [
71] might explain why both host species are facilitated so often. However, there were differences between the sampling sites. Non-human mammals dominated the detected hosts for the two sampling sites located in the interface between anthropic and natural landscapes (Sulina and Letea), i.e., homesteads in direct proximity of livestock. In contrast, the other two near-natural sampling sites (Lake Roșuleț and Dunărea Veche) were both located deep inside the Danube Delta and only insignificantly anthropogenically influenced. However, humans are commonly present in a fishing cabin and an agricultural holding. In the absence of high abundances of cattle and horse, mosquitoes might rather select other available hosts, e.g., birds and humans. Thus, host availability probably is a decisive factor for the host-selection of mosquitoes, influencing the risk of local pathogen transmission.
At the same time, this study highlights the importance of
Cx. pipiens s.l./
torrentium as a WNV vector in Europe [
72]. The only two WNV positive pools belonged to this taxon. Previous studies described the species complex as predominantly ornithophilic [
30,
31,
32,
60,
73]. However, this study again demonstrates its catholic host-feeding pattern. As the other five most abundance taxa,
Cx. pipiens s.l./
torrentium predominantly fed on non-human mammals and humans but had the highest proportion of birds, i.e., a more than nine times higher proportion, making the species a potential bridge vector. This is in line with studies from Africa [
74], Middle East [
75], Europe [
37], and North America [
76]. Although several other collected mosquito species (e.g.,
An. hyrcanus) have been found positive for WNV-RNA in Romania [
11,
13,
24], members of the
Culex pipiens complex are considered the main vectors for WNV in both urban and rural/natural transmission cycles [
13].
Active WNV circulation in the DDBR is strongly implicated by WNV infections of unfed mosquito specimens and serological evidence of WNV-specific antibodies in the hosts. However, this study collected only few specimens of the most competent vector
Cx. pipiens s.l./
torrentium. In previous studies in the DDBR [
24,
77], >95% of mosquitoes captured with bird-baited traps belonged to this species complex, while the most abundant species in our study (
Cq. richiardii) was absent. The usage of a single type of trap in only a few sampling sites across the delta’s heterogenous landscape is likely to have contributed to a biased sampling outcome [
78,
79]. In addition, mobovirus transmission is generally restricted to small foci in non-epidemic years [
80]. Therefore, the presence of WNV can be underestimated depending on the number, type and location of traps. Further studies are needed to identify and further understand the driving factors of landscape and time.