Feeding Ecology and Host Preferences of Mosquitoes in two Zoological Gardens in the United Kingdom

Background: Zoological gardens contain unique congurations of exotic and endemic animals and plants that create a diverse range of developing sites and potential sources of blood meals for local mosquitoes. This may imply unusual interspecic pathogen transmission risks involving zoo animals, like avian malaria to captive penguins. Understanding mosquito ecology and host preferences is necessary to improve mosquito control and disease prevention measures in these environments. Methods: Mosquito sampling took place in Chester Zoo for three years (2017 to 2019) and for one year in Flamingo Land (2017) using different trapping methods. Blood-fed mosquitoes were identied and their bloodmeal was amplied by PCR, sequenced, and blasted for host species identication. Results: In total, 640 blood-fed mosquitoes were collected (Culex pipiens (n = 497), Culiseta annulata (n = 81), Anopheles maculipennis s.l. (n = 7), An. claviger (n = 1) and unidentiable (n = 55)). Successful identication of the host species was achieved from 159 of the 640 blood-fed mosquitoes. Mosquitoes fed on birds (n = 74), non-human mammals (n = 20) and humans (n = 71). There were mixed bloodmeals from two hosts (n = 6). The proportions of blood-fed mosquitoes varied across sampling seasons and sites within the zoos. The use of resting traps and aspiration of vegetation were more ecient techniques for capturing blood-fed mosquitoes than traps for host-seeking or gravid mosquitoes. By relating the locations of zoo animals to where fed mosquitoes were trapped, the minimum travelling distances were calculated (13.7 to 366.7 meters). Temperature, precipitation, relative humidity, proximity to zoo animal exhibits and vegetation level were found to be signicantly associated with the proportion of captured blood-fed mosquitoes by generalized linear modelling. Conclusions: Mosquito feeding behaviour in zoos is inuenced by environmental variables and host availability, which highlights the value of mosquito monitoring in complex settings to plan control strategies and potentially reduce inherent disease transmission risks for humans and threatened zoo animals. consistent, indicate that the blood-meal was mixed, with the mosquito having fed on more than one host species [33]. Mosquitoes belonging to the Culex spp. genus include, in the UK, two sympatric species that are indistinguishable morphologically, Cx. pipiens and Cx. torrentium. Females from this genus were identied using the PCR and enzyme restriction protocol developed by Hesson et al. [34]. Likewise, Cx. pipiens has two morphologically identical biotypes with different biological traits and epidemiologic roles, pipiens and molestus; the rst one of which is reported predominately as ornithophilic and the second as mammalophilic [15, 35, 36]. We found Cx. pipiens feeding on humans, we tested them using the multiplex PCR protocol proposed by Bahnck et al. [37] to differentiate their biotype. binomial family (or quasibinomial in case of overdispersion) and a logit link, with a backwards elimination process to nd minimal adequate models. Mosquito data was consolidated per week and weather data was averaged for the week before collection. The models were constructed for each sampling separately including weather and landscape variables, and for the overall mosquito collection using weather variables. Regional and local temperature data was expected to be strongly correlated so the one producing the best model t was used. Models were constructed using the MASS package of the R software [39]. activity and the dispersal after blood-feeding. Wind speed was signicant in two models but in opposite directions, so we cannot conclude its inuence in our samplings. In one model, scarce vegetation was associated with an increase in blood-fed mosquitoes capture suggesting that mosquitoes that are looking for shelter are attracted to the traps in absence of natural resting areas. Finally, a close distance to zoo animal exhibits signicantly increased the likelihood of capturing blood-fed mosquitoes in one model, possibly due to dispersal behaviour as discussed before in relation to travelling distances.


Background
In zoological gardens, mosquitoes have access to a broad range of hosts and breeding sites due to the combination of endemic and exotic species of fauna and ora that create the zoo environment [1]. Mosquitoes collected at zoos have been found feeding on zoo animals, free wild animals, and humans, and mixed blood-meals have been reported [2][3][4]. The zoo environment poses a risk of interspeci c transmission of pathogens among vertebrate groups that usually cannot happen in natural environments [1]. The relevance of mosquito ecology in zoos can be exempli ed by the outbreak of West Nile virus (WNV), transmitted by Culex spp. mosquitoes, in the Bronx Zoo/Wildlife Conservation Park in relation to the rst outbreak of WNV in America in humans [5]. The rst fully sequenced strain of WNV in North America was isolated from a amingo kept at this zoo [6]. Furthermore, it has been observed that host preference changes in Culex pipiens between humans and birds at different times in the season could be related to the infection risk with WNV for humans [7]. Mosquito-borne diseases can be also a threat for the health of zoo animals; avian malaria (caused by Plasmodium spp. haemosporidians) clearly exempli es this as it is a major cause of severe disease and death for captive penguins worldwide [8,9]. Moreover, as the parasite and its mosquito vectors have a broad distribution, this infection should be considered as a constant threat and regular surveillance of mosquitoes in zoos has been encouraged [10].
The choice or availability of hosts on which mosquitoes feed is an important factor of the ability of mosquito populations to transmit pathogens, a measure known as 'vectorial capacity'. Different mosquito species can exhibit diverse feeding patterns despite being exposed to the same vertebrates in a single location [11]. Mosquitoes can be generalists, feeding on a wide range of species or groups of vertebrates, or specialists, preferring a narrow range of host species or types [12,13]. Generalist mosquitoes can facilitate pathogen transmission among unrelated species (for example, birds and mammals) and in such cases, they are known as 'bridge vectors' [14][15][16][17]. However, the feeding of mosquitoes on hosts than cannot harbour the pathogen may lead to a dilution effect in pathogen transmission [2]. In the case of specialized mosquitoes, changes in the host community structure could force them to feed on unusual hosts [14] which could also facilitate interspeci c transmission of pathogens. For example, Aedes japonicus that usually feeds on mammals (mammalophilic) has been found to feed on birds in nature [18]. Likewise, when the mammalian-biting biotype of Culex pipiens (molestus) and its birdfeeding biotype (pipiens) hybridize, they resulting population may present mixed host preferences and therefore, could act as a bridge vector for pathogens from birds to people [15]. Therefore, both, generalist and specialist mosquitoes can function as bridge vectors under certain circumstances, and the combination of vector's host preference and host competence determine if cross-species transmission occurs and whether a dilution or ampli cation effect is observed in relation to host diversity [19].
Mosquito dispersal is another factor related to disease transmission risks. Female mosquitoes can travel long distances when seeking a blood-meal. Some species, like Aedes vexans, migrate to invade new habitats [20] and Culex spp. for example, can disperse for several kilometres [21]. Previous studies on ying distances of blood-fed mosquitoes in zoos have proved that they feed on captive animals, mostly travel short distances and are implicated in the local transmission of vector-borne diseases to zoo animals [4,22].
Here, the feeding ecology, host preferences and dispersal of mosquitoes in two zoological gardens in the UK are described to better understand the dynamics of mosquito populations and get insights into the potential for cross-species transmission of pathogens. Over two years, we identi ed bloodmeals from mosquitoes captured using traps for host-seeking mosquitoes and for gravid mosquitoes looking to oviposit; special techniques for capturing blood-engorged mosquitoes were implemented in a third sampling year in one zoological garden. As mosquitoes are highly adaptable, this information is relevant for planning mosquito control strategies in modi ed environments and to protect vulnerable hosts.

Sites and sampling methods
We performed general sampling of mosquitoes for two years (2017 and 2018) in Chester Zoo (Upton by Chester, Cheshire, UK) and for one year (2017) in Flamingo Land (Kirby Misperton, North Yorkshire, UK). In 2019, we speci cally sampled for blood-fed mosquitoes in Chester Zoo only.
The traps that we used in all samplings were the BG-Mosquitaire trap and the CDC-Gravid trap model 1712. The BG-Mosquitaire trap has a fan inside a plastic container that is connected to AC power; this trap captures host-seeking mosquitoes with its basic attractant, the BG-Sweetscent®, based on lactic acid to mimic mammalian sweat [23]. The CDC-Gravid trap model 1712 consists of an electric fan powered by a 6 V battery located inside a plastic cylinder and covered with a capture net, which is placed over a tray that contains 4 L of infusion media as attractant for egg laying mosquitoes [24]. The infusion media is prepared with tap water (40 L), hay (200 gr), brewer's yeast (2 gr) and milk powder (2 gr), and the mix is rested for at least one week before use [25].
For the 2019 season, four capture methods were used: BG-Mosquitaire trap upgraded with dry ice as source of CO 2 , CDC-Gravid trap, resting trap, and aspiration in mosquito resting areas. The BG-Mosquitaire traps operated using the basic attractant (BG-Sweetscent®) and dry ice to increase the chances of capturing host-seeking mosquitoes and the species range [26]. For this, a polyurethane box was placed next to each trap, connected to the trap with a plastic tube and sealed with silicone glue; the lid was sealed to the box with tape and a strap with a combination lock to prevent accidental opening. The CDC-Gravid traps were used as described previously. The resting trap is a 40 × 30 × 30 cm wooden box with an open side which provides shelter; there is no attractant used, mosquitoes simply go inside the box to rest, especially during daytime. We placed these traps with the open face on the underside, leaning against a structure (e.g. walls, trees, or rocks) at 45° to allow a gap for mosquito entry. We used an Improved CDC-Backpack Aspirator model 1412 [27] to aspirate the inside of the resting traps. Finally, the same aspirator was used to systematically aspirate potential resting places, such as vegetation, walls, fences, and the outside of buildings, for ve minutes in each sampling area (Fig. 1).
Sampling areas were de ned as 30 m diameter circles that contained one trap of each type, and set in public areas, staff areas or animal exhibits. Inside sampling areas, the traps were at least 10 m apart from each other to minimize interference between them, and located considering safety for people and animals, logistic implications, surrounding vegetation (avoiding coverage at least 2 m above them to reduce the possibility of leaves interfering with the trap's fan), and with protection from direct sunlight and rain when possible.

Sampling protocols
We established ten sampling areas in Chester Zoo in the 2017 season; in 2018 we discontinued three areas due to low catches, and added a new area located inside the penguin exhibit (Fig. 2). Sampling took place for 32 weeks in Chester Zoo in 2017 (May to December) and for 31 in 2018 (April to November). In Flamingo Land we established four sampling areas (Fig. 3), which were sampled for 25 weeks in 2017 (June to November). In both sites, BG-Mosquitaire traps were emptied once a week after being operational for six consecutive nights and once a week after operating for one night; the CDC-Gravid traps were operated one night per week. This resulted in a six day collections from the BG-Mosquitaire traps, followed by a one day collection from both traps every week.
In addition, speci c sampling of blood-fed mosquitoes took place in Chester Zoo for 11 weeks (April to July) in 2019 using ve sampling areas close to the penguin exhibit (Fig. 2). In this occasion, the weekly schedule consisted of one collection from the BG-Mosquitaire traps after ve days with the basic attractant plus two collections after one day each adding dry ice, two aspirations from resting traps and two from surrounding areas, and one collection from the CDC-Gravid traps after two days of operation.

Environment monitoring
The environment was monitored for landscape and weather variables. Landscape variables were categorized with three possible values and included vegetation (dense, medium, or scarce), proximity to suitable mosquito oviposition sites (close, medium, or remote), mosquito resting areas (abundant, medium or rare) and proximity to zoo animal exhibits (close, medium, or remote). The vegetation value was changed to scarce for all traps when foliage decreased in autumn. The de nition of variable values can be found in supplemental material Table S1 and the corresponding values per trap in Table S2.
Weather variables included regional temperature, relative humidity, wind speed and precipitation. Daily values were obtained using the R package "Climate", which extracts information from the OGIMET Weather Information Service (www.ogimet.com). We downloaded data from the closest inland weather stations; for Chester Zoo it is Hawarden (13 km) and for Flamingo Land, Linton On Ouse (52.2 km). We did not use the data from closer stations to Flamingo Land because they are on the coast and, therefore, in uenced by marine climate.
In 2017 and 2018 we used TinyTag© loggers, 13 of the Plus 2 TGP-4500 model and two of the Ultra 2 TGU-4500 model, (Gemini Data Loggers, UK) to record local temperature and humidity. Loggers were placed next to the BG-Mosquitaire traps, programmed to record every hour and average readings were used. The loggers were not used in 2019.

Laboratory protocols
Collected nets were transported in a cooled icebox and placed into a -20 °C freezer on arrival for approximately two hours. Mosquitoes were separated from accompanying insects and identi ed by morphology under a stereomicroscope following identi cation keys [20,28]. During the morphological identi cation, the abdomens of the blood-fed mosquitoes were cut off using entomological forceps and disposable scalpels; the materials were cleaned with 70% ethanol and DNA Away® between specimens. The abdomens were stored in individual reaction tubes at -80 °C for no more than three weeks until processing. As eld captured mosquitoes have different degrees of blood digestion that affects host identi cation [29], mosquitoes were selected for analysis if they were engorged with a red, dark red or blackish abdomen indicating blood content. This is equivalent to the stages II and III of the Sella classi cation system [30,31].
Mosquito abdomens were homogenized with 200 µl of PBS (phosphate buffered saline) per sample [32] using sterile plastic pestles or with a stainlesssteel bead and a QIAGEN© TissueLysser at frequency of 24 Hz per second for 2 minutes. For DNA extraction, we used the OMEGA Bio-Tek E.Z.N.A ® Tissue DNA kit following the manufacturer's instructions; extracts were stored at 4 °C until further processing for no more than two weeks and at -20 °C for long term storage. Afterwards, the PCR protocol and primers proposed by Alcaide et al. [33] were used to obtain a 758 bp amplicon of the Cytochrome c Oxidase Subunit I (COI) gene. Negative controls (nuclease-free water) were added every ve samples and DNA extract from liver of Black headed gull (Larus ridibundus) was used as a positive control. The ampli cation was veri ed by electrophoresis in a 1% agarose gel. Positive PCR products were sent to Macrogen Europe B. V. for Sanger sequencing using the M13 primer in the forward direction. All the PCR negative samples were tested at least twice.
Successful sequences were edited and analysed using the software BioEdit© and compared to the reported sequences in the Basic Local Alignment Search Tool (BLAST®), optimized for the highly similar sequences, and the Barcoding of Life Data System© (BOLD). The most similar sequences were aligned and compared using BioEdit© to nd the best match considering the identity and query covers and excluding wild native species absent in the area and exotic species not included in the zoo collections. With the same software, we inspected the electropherograms for double peaks in a single base position, which, if consistent, indicate that the blood-meal was mixed, with the mosquito having fed on more than one host species [33].
Mosquitoes belonging to the Culex spp. genus include, in the UK, two sympatric species that are indistinguishable morphologically, Cx. pipiens and Cx. torrentium. Females from this genus were identi ed using the PCR and enzyme restriction protocol developed by Hesson et al. [34]. Likewise, Cx. pipiens has two morphologically identical biotypes with different biological traits and epidemiologic roles, pipiens and molestus; the rst one of which is reported predominately as ornithophilic and the second as mammalophilic [15,35,36]. We found Cx. pipiens feeding on humans, so we tested them using the multiplex PCR protocol proposed by Bahnck et al. [37] to differentiate their biotype.

Flying Distances
Knowing the zoo animals on which mosquitoes have fed, we estimated the distance between the relevant animal exhibits and the location of the trap where the blood-fed mosquitoes were captured. The software Species360 ZIMS (Zoological Information Management Software) [38] was used to determine the location of the zoo animals at the time when mosquitoes were captured. The Open Street Map layer was used in the QGIS 3.2© software to delineate the polygons of the exhibits. Then, the centroid of the exhibits was estimated, and the minimum travelling distance of mosquitoes was represented as the length of the line from the centroid to the corresponding trap.

Data analysis
A gross analysis comprised the comparison of blood-fed mosquito proportions separately, by zoo and sampling year, to detect differences by sampling area, month, and capture method. Host preferences were compared by vertebrate group (birds, non-human mammals, and humans) and mosquito species using the Chi-squared test of independence exploring the test residuals to nd sources of signi cance. Fisher's exact test of independence was used when appropriate. Specimens with damaged or missing abdomens were excluded and only completely identi ed mosquitoes were included in the analyses by species.
To understand environmental in uences in blood-fed mosquito captures, we used generalized linear models (GLMs) for a proportion response using a binomial family (or quasibinomial in case of overdispersion) and a logit link, with a backwards elimination process to nd minimal adequate models.
Mosquito data was consolidated per week and weather data was averaged for the week before collection. The models were constructed for each sampling separately including weather and landscape variables, and for the overall mosquito collection using weather variables. Regional and local temperature data was expected to be strongly correlated so the one producing the best model t was used. Models were constructed using the MASS package of the R software [39].

Results
Excluding males and mosquitoes with damaged abdomens, the number of blood-fed mosquitoes captured (and percentage of all female mosquitoes caught) in Chester Zoo was 213 (3.5%) in 2017, 245 (9.7%) in 2018 and 107 (4.1%) in 2019. We caught 75 (7.2%) in Flamingo Land in 2017. In total, we caught 640 blood-fed mosquitoes (5.2%) ( Table 1). Most blood-fed mosquitoes in both sites were Culex pipiens (n = 497) and Culiseta annulata (n = 81), although we also captured one Anopheles claviger and four An. maculipennis s.l. in 2018 and three An. maculipennis s.l. in 2019. Fifty-ve mosquitoes were damaged and unable to be identi ed to species level. PCR testing con rmed that all engorged Culex spp. mosquitoes were Cx. pipiens and all of those which fed on humans were Cx. pipiens biotype pipiens. Other species captured include Aedes annulipes, Ae. vexans, Ae. detritus, An. plumbeus, An. claviger, Cs. morsitans and Cx. torrentium, although these mosquitoes were not blood-fed and captured in low numbers.
In Chester Zoo, eight species of free wild birds, two zoo birds, four zoo mammals, one cattle and 37 samples from human were identi ed in 2017 ( Table 2). The following year, we found nine species of free wild birds, two zoo birds, one chicken, one cattle, one pig, two zoo mammals and 17 humans (Table 3). In 2019, we identi ed ve species of free wild birds, two zoo birds and two humans (Table 4). In Flamingo Land (2017), one free wild bird, three species of zoo mammals, one dog and 10 humans were identi ed ( Table 5). Proportions of hosts are shown overall per sampling in Fig. 4, for Cx. pipiens in Fig. 5 and for the other species in supplementary material Fig. S1.      Blood-fed Cx. pipiens and Cs. annulata were captured in su cient numbers for a more detailed host preference analysis. Signi cant differences were found in Chester Zoo in 2017 (Fisher's exact test P < 0.001) where Cx. pipiens preferred to feed on humans and birds, and Cs. annulata preferred to feed on humans and non-human mammals. In the following year, Cx. pipiens also preferred birds and humans and Cs. annulata preferred non-human mammals (Fisher's exact test P = 0.025). In 2019, the number of samples per group was not su cient for statistical analysis as most of the Cx. pipiens were feeding on birds (n = 17). The difference in host preference was also signi cant in Flamingo Land (Fisher's exact test P = 0.021), with Cx. pipiens preferring humans and Cs. annulata, non-human mammals. Analysing differences in host preferences by month, we found a signi cant difference in Chester Zoo in 2017 (Fisher's exact test P = 0.003), Cx. pipiens preferred to feed on humans during June and on birds during July (Fig. 6). Data was insu cient for other comparisons, or differences were not signi cant.

Distribution of feeding activity
We observed signi cant differences in blood-fed mosquito catches among sampling areas in all our samplings. In Chester Zoo, during the rst year, areas A1 and A3 captured the largest numbers of mosquitoes and highest proportions of blood-feds; areas A7 and A10 also captured more blood-fed mosquitoes than expected randomly but not in relation to a large mosquito catch (X 2 = 17.556, df = 9, P = 0.041). The following year, area A10 captured more blood-fed mosquitoes, along with areas A1 and A11 (X 2 = 17.894, df = 7, P = 0.0125). In the last year, areas A12, A13 and A3 yielded high proportions of blood-fed mosquitoes and in this occasion, captures in area A1 were less than expected by chance (X 2 = 16.55, df = 4, P = 0.002) (Fig. 7a). We also found a signi cant difference in Flamingo Land related to a large catch of blood-fed mosquitoes in area A2 and smaller catches in areas A3 and A4 (X 2 = 24.868, df = 3, P < 0.0001) (Fig. 8a).

Seasonality of feeding activity
The proportion of blood fed mosquitoes trapped also changed between months. A signi cantly higher proportion of blood-fed mosquitoes in June and May and a lower proportion in July, were observed in Chester Zoo in the rst year of sampling (X 2 = 50.596, df = 6, P < 0.001). In 2018, catches larger than expected randomly were observed in July and August and smaller than expected in September and October (X 2 = 54.346, df = 6, P < 0.0001). The sampling in 2019 captured more blood-fed mosquitoes in July and May and less in June (X 2 = 7.15, df = 2, P = 0.028) (Fig. 7b). Likewise, in Flamingo Land a higher proportion of blood-fed mosquitoes was observed in July and August compared to other months (X 2 = 106.51, df = 5, P < 0.0001) (Fig. 8b).  1, 2). A comparison by species (excluding An. maculipennis s.l. due to low numbers) showed no signi cant difference overall (Kruskal-Wallis, X 2 = 0.279, df = 1, P = 0.597). One third of the mosquitoes that fed on zoo animals (n = 9) were captured within 50 m from the exhibits where they had fed on and more than half (n = 15) within 100 m. a : Mixed blood-meals, includes the distance between the exhibits of both animals and the trap. * These mosquitoes were only identi ed to the Culicinae subfamily due to damaged legs. Sampling areas are presented in Fig. 2 for Chester Zoo and Fig. 3 for Flamingo Land.

Environmental factors
Six variables showed signi cant in uence on blood-fed mosquito captures, although the direction of in uence was not consistent. Temperature had a positive in uence in blood-fed mosquito captures in four models (close to signi cance in one, P = 0.073) and a negative in uence in another, humidity had a negative in uence in three models and a positive in uence in one, precipitation was a positive in uence in three models, wind speed was positively in uencing in one model and negatively in another, and scarce vegetation and close distance to zoo animal exhibits had a positive in uence in one model.
A summary of estimates and P values can be found in supplemental material Table S3.

Discussion
We con rmed the intrinsic preference of Culex pipiens for birds, mainly free wild birds and some birds from the zoos' collections, and of Culiseta annulata for mammals. Cx. pipiens is primarily an ornithophilic species [35,40], however we observed high proportions of human hosts in all our samplings, except for 2019; similarly Heym et al. [3] and Börstler et al. [41] found mosquitoes feeding on humans but the proportion we observed was higher overall. We discount the possibility of major sample contamination as negative controls did not produce a positive result and no sequenced samples matched the positive control. These mosquitoes fed on humans from April to August, when the zoos have more visitors and temporary staff, and no human bloodmeals were found at other times, although the occurrence of birds and other mammalian blood-meals continued to be observed. Additionally, the preference for humans was signi cantly higher in Cx. pipiens than in Cs. annulata, which is unexpected as Cs. annulata has been reported as a biting nuisance for people in the UK [42]. PCR identi cation showed that all Cx. pipiens that fed on humans belong to the pipiens biotype; as this biotype is typically described as ornithophilic, abundance of visitors and staff seem to be a relevant in uence in mosquito feeding preferences in zoos.
Anopheles maculipennis s.l. prefers to feed on mammals than on birds [32,43]. However, two mosquitoes from this group fed on Humboldt penguins (Spheniscus humboldti), which to our knowledge is the rst report of this host choice. Due to the low sample size, we cannot conclude if there is a host preference or simply a tendency of capturing blood-fed mosquitoes in proximity to the animals they feed on as these mosquitoes were captured close to the penguin exhibit (< 22 m). Therefore, targeted sampling of this group is needed and should include the molecular identi cation of the species as they have different host preferences [43].
A correlation between mosquito abundance and proportion of blood-feds has been noticed before [21]. During the rst year in Chester Zoo, we found high mosquito abundance and high proportion of blood-feds in areas A1 and A3 and again for A1 in 2018. In Flamingo Land, we observed a similar situation in area A2. Nevertheless, other areas with lower general catches showed high proportions of blood-fed mosquitoes, like A7 and A10 in Chester Zoo in 2017 and areas A10 and A11 in 2018. Therefore, the mosquito abundance is not the only explanatory factor. For instance, area A3 in Chester Zoo captured a high number of blood-fed mosquitoes in 2017 possibly due to its proximity to a picnic area and a children's playground. Fewer mosquitoes were caught in this area in 2018, when the children's playground was closed for renewal over several weeks, and the off-show aviaries next to the sampling area were expanded, reducing a considerable portion of the vegetation. In Chester Zoo, area A7, which caught more blood-feds than expected in 2017, is also near a picnic garden and an area with high transit of visitors. In 2017 and 2018, area A10 also captured a high proportion of blood-feds which matched mainly wild birds. This area is inside a wetland aviary where several species are kept, and the abundance of wild passerines could be high as they are attracted to the waterfowl's food despite the netting of the exhibit. In Flamingo Land, area A2, also with a higher proportion of blood-feds, is in the boundary of the South American exhibit which contains large mammals such as capybara (Hydrochoerus hydrochaeris), on which we found two mosquitoes feeding. It appears from this result that the constant presence of suitable hosts is an attractant factor for mosquitoes.
The mosquito host feeding preferences varied each year. There is evidence that host selection by Culex spp. is in uenced primarily by the availability of preferred hosts [44]; therefore, these variations possibly depend on the mosquito abundance and host availability, both of which increase during the summer. We observed a signi cant host shift in the case of Cx. pipiens in Chester Zoo in 2017, preferring humans in June and birds in July. Tuten et al. [2] also observed a host shift for Cx. pipiens, although in this case preferring birds in the summer and birds and mammals in autumn. These changes could be related to host availability in uenced for instance, by the migration and breeding seasons of birds (nestlings are more prone to mosquito bites [3]).
Furthermore, when the preferred hosts of Culex spp. are scarce, this mosquito can shift to other hosts, including humans [12].
Some species like Cx. pipiens breed and rest close to their host's habitat [20], so it is reasonable that the chances of capturing blood-fed mosquitoes are higher closer to the location of potential hosts as we observed for zoo animals. Other authors reported maximum travelling distances between 170 to 770 m [2][3][4]22] and the maximum that we observed was within this range (367 m). However, we found mosquitoes feeding on domestic animals (cattle, pig, chicken, and dog) that we assume came from nearby farms. Thus, dispersal was probably further than our estimate.
It is possible that in some areas the landscape forms ight paths that aids mosquito's movement in a certain direction [22] and the dominant wind direction could be a relevant factor in uencing dispersal and ight direction [45]. For example, Area A1 in Chester Zoo captured blood-fed mosquitoes that feed on zoo animals located from the southwest to the northwest from this area, and the wind on the day before collection came from a similar direction in the case of four out of six mosquitoes. The use of portable weather stations could improve the study of wind and landscape in uence in mosquito dispersal. It is important to consider that using the exhibit's centroid, or any other measurement to the exhibit, assumes that the animals distribute randomly within their exhibits, when in reality, animals spend more time in certain areas, like around the feeders during the day or in the enclosure at nighttime. The estimation of animal occupancy degrees inside the exhibits should be considered for a more precise assessment of mosquito travelling distances.
Temperature affects mosquito feeding activity, reproduction, longevity, and development [46,47], which could explain the positive relationship with the capture of blood-fed mosquitoes showed in the models. High relative humidity may enhance the effect of odorant cues for host-seeking mosquitoes [12]; however, our models showed mostly a negative relation, possibly because higher humidity promotes mosquito dispersal [20] and thus decreases the chances of capture. Precipitation on the other hand, signi cantly increased the chances of capturing blood-fed mosquitoes according to three models; this is counterintuitive as mosquitoes are not expected to y under rainy conditions. However, this variable was aggregated weekly over the sampling units; thus, higher values do not imply more constant rain but more rain over a week period. Therefore, rainfall may reduce mosquito dispersal increasing the capture chances but does not prevent their feeding activity. Both, Brugman et al. [32] and Karki et al., [45] reported that the wind diminishes the capture of blood-fed mosquitoes, which can be explained by mosquitoes taking shelter under windy conditions, thus reducing the host-seeking activity and the dispersal after blood-feeding. Wind speed was signi cant in two models but in opposite directions, so we cannot conclude its in uence in our samplings.
In one model, scarce vegetation was associated with an increase in blood-fed mosquitoes capture suggesting that mosquitoes that are looking for shelter are attracted to the traps in absence of natural resting areas. Finally, a close distance to zoo animal exhibits signi cantly increased the likelihood of capturing blood-fed mosquitoes in one model, possibly due to dispersal behaviour as discussed before in relation to travelling distances.
After blood feeding, mosquitoes look for a resting place to digest the blood and produce eggs; thus blood-engorged females tend to be captured in low proportions in mosquito traps that target host seeking individuals (e.g. [26]) or gravid females looking to oviposit, which varies depending on the sampling methods and study sites (e.g. [2][3][4]22]). It is not clear if blood-fed mosquitoes were attracted by the lactic acid of host-seeking mosquito traps because they are looking for a second blood-meal, or by the fermenting media of gravid traps because they prefer to rest near potential oviposition sites.
Alternatively, they may be attracted by the dark colour and location of the traps. However, we cannot exclude the possibility that they were randomly captured as no signi cant differences were found between BG-Mosquitaire and CDC-Gravid trap types. We captured higher proportions of blood-fed mosquitoes using resting traps and aspirating resting areas, as reported by other authors [2,32,48]. In some cases, the use of resting traps has failed [2,49], thus their positioning and orientation could be a determinant and should ensure that mosquitoes get protection from direct sunlight, rain, and wind.
Our blood-meal identi cation success could have been in uenced by the time before collecting the mosquitoes and weather conditions. If the mosquitoes remained alive, they continued digesting the blood, and if dead, they started to desiccate. We observed more mosquitoes completely dehydrated, despite their evident blood-fed status, during the 2018 sampling in Chester Zoo which was drier and hotter compared to other sampling years. To improve the results, storing the samples at -80 °C or in lter paper and processing them promptly has been recommended [29,31,32], which would also minimise physical damage facilitating mosquito morphological identi cation.
The high proportion of Cx. pipiens mosquitoes feeding on humans that we observed represents not only a likely nuisance for visitors and staff at the zoos, but also a potential risk for disease transmission. Culex spp. are vectors of viruses hosted by wild birds, such as West Nile virus (WNV), Sindbis virus (SINV) and Usutu virus (USUV), all of which have been reported circulating in mainland Europe [17,50] and could pose a serious threat if they are introduced to the UK. Moreover, all mixed blood-meals that we found were from Cx. pipiens involving a bird host and the mixed blood meals that included humans were combined with blood of Eurasian magpie (Pica pica), which is a proven natural reservoir of WNV and an effective target for its surveillance [51]. In addition, it has been shown that the temporal and spatial variation in host preferences by Culex spp. could in uence the timing and severity of WNV infections, probably in relation to its seasonal shifts between ornithophilic and anthropophilic cycles [7,21]. Although the host shift in Cx. pipiens that we observed in one of our samplings occurred from humans to birds, the potential of this mosquito as a bridge vector for humans and domestic animals (i.e. horses) [49,52] should be constantly monitored, despite the lack of evidence con rming WNV's establishment in the UK [17,53], as well as for the other mentioned viruses (SINV and USUV).
The interspeci c transmission of vector borne diseases is also important for the health of the animals in the zoo collections. For instance, mosquitoes have been involved in the transmission of Eastern equine encephalitis virus to African penguins (Spheniscus demersus) in North America, USUV to great grey owls (Strix nebulosa) in Austria and WNV that has caused the death of exotic animals in roughly one hundred zoos in the United States [1]. The risk of disease transmission between local bird species and zoo animals is present in our study sites, as avian malaria has caused outbreaks in Humboldt penguin (Spheniscus humboldti) colonies in both Chester Zoo and Flamingo Land [10]. Interestingly, host preference changes could also in uence the transmission dynamics of avian malaria parasites in bird communities [44]. We found four mosquitoes feeding on Humboldt penguins, two Anopheles maculipennis s. l., one Cx. pipiens and an unknown Culicinae, which is likely to be Cx. pipiens as it had all the corresponding morphological features except for those evaluated on the legs. Anopheles spp. mosquitoes are considered as potential vectors for avian Plasmodium spp. and have been found susceptible to the parasite infection experimentally [54]. Therefore, this genus could have a relevant role in the transmission of avian malaria, although An. maculipennis s. l. has not been found infected with avian Plasmodium yet [55,56]. To clarify the host preferences of this mosquito genus and Plasmodium spp. transmission risks, the precise identi cation of its species is needed.

Conclusions
Zoological gardens provide unique opportunities for the study of mosquitoes; thus, abundance, host choice and dispersal can be explored to assess disease transmission risks. We con rmed that mosquitoes in zoos conserve their characteristic host preferences up to certain degree, feed on a wide range of hosts, and presented an important and unexpected preference for humans which varied between 10 and 63% in our samplings. This highlights the risk of zoonotic viruses transmitted between humans and birds such as WNV, USUV and SINV if ever stablished in the UK. There is an implied health risk for zoo birds as we found mosquitoes feeding on them. Of especial concern are the Javan green magpies (Cissa thalassina) which are critically endangered, and the Humboldt penguins (Spheniscus humboldti) due to their high susceptibility to avian malaria. Mosquito feeding behaviour is in uenced by different factors and it changes temporally and spatially. In our samplings, the main period of feeding activity varied with year and location, it corresponded with the overall increase in mosquito abundance and it was mainly in uenced by temperature in a positive sense. We captured a high proportion of blood-fed mosquitoes in areas with high captures of mosquitoes in general, but this was not always the case. Mosquito dispersal after a blood-meal was variable and it is likely that landscape features in uence their movements. Therefore, mosquitoes feeding behaviour is in uenced by the weather, landscape, and abundance of potential hosts. Our results highlight the complexity of mosquito ecology in zoos and the relevance of assessing interspeci c transmission risk of pathogens which need to be thoroughly understood for the e cient control of mosquito populations and reducing vectorhost contact in zoological gardens.

Ethics approval
This project was approved by the science committees of Chester Zoo and Flamingo Land, and by the University of Liverpool Veterinary Research Ethics Committee (reference VREC532a).