Microclimate and the vertical stratification of potential bridge vectors of mosquito‑borne viruses captured by nets and ovitraps in a central Amazonian forest bordering Manaus, Brazil

In the Americas, some mosquito-borne viruses such as Zika, chikungunya, and dengue circulate among humans in urban transmission cycles, while others, including yellow fever and Mayaro, circulate among monkeys in sylvatic cycles. The intersection of humans and wildlife at forest edges creates risk for zoonotic virus exchange. We built a scaffold tower at the edge of a treefall gap in rainforest bordering Manaus, Brazil, to identify vectors that may bridge transmission between humans and monkeys. We vertically sampled diurnally active, anthropophilic mosquitoes using handheld nets at 0, 5, and 9 m and container-breeding mosquitoes in ovitraps at 0, 5, 10, and 15 m. Haemagogus janthinomys and Psorophora amazonica were present in high relative abundance in nets at each height sampled, while anthropophilic species were uncommon in ovitraps. Hg. janthinomys was more abundant at elevated heights than at ground level, while Ps. amazonica abundance was not significantly stratified across heights. The presence of each species increased with increasing 7-day rainfall lagged at 1 week, and at 1 and 4 weeks prior to collection, respectively. In addition, Hg. janthinomys was most frequently collected at 29.9 °C, irrespective of height. These data provide insight into the potential role of each species as bridge vectors.

www.nature.com/scientificreports/ the automated INMET meteorological station (Code: Manaus-A101, OMM: 81730) located within Manaus (3.103682° S, 60.015461° W), were used to calculate cumulative 7-day rainfall at lags of 1, 2, 3, and 4 weeks prior to each collection to test associations between past rainfall and the occurrence of adult mosquitoes.
Ovitrap collections. The stratification and trap preference of ovipositing mosquitoes was investigated by collecting mosquito eggs in three types of water-holding container (Figs. 2b, c, Supplementary Data S2). These were standard black plastic ovitraps (Geronimo Strategic, El Paso, Texas), Brazil nut tree (Bertholletia excelsa) fruit pods, and 30 cm sections of bamboo with a lid and 2 cm diameter lateral hole. One of each container type was placed at each of four heights (0, 5, 9, and 15 m) in a factorial blocked design. The containers at 0, 5, and 9 m were placed on the tower, while those at 15 m were suspended on a rope hanging from a branch above the tower. Each container was lined with seed germination paper (Nasco, Fort Atkinson, Wisconsin) as an oviposition substrate, along with several decaying leaves and water from a nearby stream. Containers were left in the field for seven days before the water and germination paper were collected and replaced. Ten collections were made between 17 January and 9 April 2020. Eggs were reared to adults at ambient temperatures (26 °C ± 1) and relative humidity (70-80%) in an insectary at FMT-HVD. Germination papers and water from each container were placed in rearing trays and topped-up with distilled water if necessary. If eggs did not hatch after three days, papers were dried for three days and then resubmerged, up to a maximum four times. Larvae were fed with fish food (Oscar Gold Ornamental Fish Food, Hai Feng, Taiwan) as required. Pupae were separated and placed in cups of distilled water in cages containing cotton wool soaked with 10% glucose solution. Adults were kept alive for 24 h after eclosion to allow male genitalia to rotate before being stored at -20 °C until identified. Any larvae remaining in trays one month after rearing were discarded.

Mosquito identifications.
Mosquitoes were placed on a chill table (BioQuip, Rancho Dominguez, California, USA) and identified to species using a stereomicroscope and taxonomic keys as previously described 5,15 . Genus and species names follow the Walter Reed Biosystematics nomenclature 36 . Samples were stored at −80 °C for future arbovirus screening.

Statistical analyses.
Where possible, analyses were performed that enabled comparisons with our previous work using BG-Sentinel traps 15 . Two approaches were used to investigate changes in the community composition of diurnally active, anthropophilic mosquitoes at each height. The first was the Morisita overlap index 37 calculated using the PAST software package 38 . This is a commonly used measure of dispersion which provides a value between 0 (indicating no similarity between sample units) and 1 (complete similarity). The second approach was a principal components analysis of the relative frequency of each species followed by hierarchical clustering of communities. Contingency tables and Pearson's chi-squared tests were used to compare the relative abundance of genera and species collected using nets and BG-Sentinel traps; rare genera were grouped and 1 count was added to each cell in the analysis to avoid zero values. Kruskal-Wallis tests followed by posthoc Wilcoxon Each Pair tests were used to test the effect of height on the abundance of target species, and lag to first approach. Linear regressions were also used to test the effect of mosquito abundance, height, and mean daytime temperature on lag to first approach for Hg. janthinomys. Nominal logistic regressions were used to test the effect of date, height, mean daytime temperature, and 7-day cumulative rainfall at lags of 1, 2, 3, and 4 weeks on the occurrence (presence or absence) of target species. Nominal regressions were also used to investigate relationships between temperature, which was inversely correlated with relative humidity, and occurrence of these species.
For ovitrap collections, the Morisita overlap index was used to compare mosquito community composition between container types and among heights. Target species in ovitraps were insufficiently abundant for analysis of stratification in this study. All analyses were performed using JMP 14 39 unless stated.
Ethics and permits. We note that our method of collection differed from a human landing collection in that we measured mosquito approach without permitting mosquito landing. No skin was deliberately exposed to attract mosquitoes, and mosquitoes were netted prior to landing on a collector. All collectors are listed among the coauthors and were fully aware of the nature of the research. They wore trousers and either a long-sleeved shirt or repellent to minimize the risk of being bitten and had been vaccinated against yellow fever. Mosquito collections at the Ducke reserve were approved by local environmental authorities (SISBIO license 57003-6) and the study did not involve endangered or protected species.

Results
Variation in microclimate. Microclimate at the tower varied across the daily sampling period with temperatures highest and relative humidity lowest around midday and the early afternoon hours, although the time of peak temperature and nadir humidity varied by height (Fig. 3a, b). Mean temperature was highest at ground level at 11:30 (30.0 °C) when it was on average 0.2 °C hotter than at 9 m, whereas at 5 m and 9 m, it was highest at 13:30 (29.7 °C and 30.3 °C, respectively). The inverse was true for mean relative humidity, which was lowest at ground level at 11:30 (83.8%) and lowest at 5 m and 9 m at 13:30 (80.1% and 76.4%, respectively). Both variables showed substantial overlap in means and standard errors across the sampled heights during the morning hours, before diverging in the afternoon. For comparison, we extracted microclimate data from the corresponding sampling period in the BG-Sentinel trap study 15 , which revealed clear differences in temperature and humidity at each height sampled (Fig. 3c, d). BG-Sentinel traps were often hung beneath the forest canopy where it was considerably cooler and more humid than at the treefall gap, particularly at ground level. www.nature.com/scientificreports/ Community composition of diurnally active, anthropophilic mosquitoes. A total of 2146 adult mosquitoes representing seven genera and 34 species were collected using nets (Fig. 4a), of which 99.8% (2142/2146) were female and 99.7% (2140/2146) were identified to species level. Mosquito abundance was similar at ground level and 9 m but was slightly lower at 5 m, while species richness was higher at ground level (28 species), than at 5 m (18 species) and 9 m (22 species). Psorophora was the most abundant genus (1231 mosquitoes, 57.4% of the total catch), followed by Haemagogus (32.3%), and Sabethes (6.6%). The genera Limatus (1.4%), Culex (1.2%), Wyeomyia (1.0%), and Onirion (< 0.1%), formed just 3.7% of the total catch. In agreement with our BG-Sentinel trap collections from the same area ( Fig. 4b), Ps. amazonica (82.4% of identified Psorophora) and Hg. janthinomys (97.3% of identified Haemagogus) dominated collections within their respective genera, while Ps. albigenu (16.8% of identified Psorophora) was also relatively abundant. Sabethes was the most diverse genus with 11 species collected, and Sa. chloropterus (32.4% of identified Sabethes) was the most abundant of these. The Morisita overlap index for specimens identified to species level (n = 2140 mosquitoes) showed substantial overlap in community composition between the sampled heights. However, elevated (5 m and 9 m) communities were more similar to each other (0.966) than they were to the ground level community (0.945 and 0.896 for 5 m and 9 m comparisons with ground level, respectively). These results were consistent with clustering based on principal components generated using the summed numbers for each species per height (Fig. 4c) and agreed with the results of our BG-Sentinel trap collections (Fig. 4d).
Comparing nets with BG-sentinel traps. When comparing handheld net collections in this study with those made using BG-Sentinel traps in the same area of the reserve 15 , there was a significant difference in relative genus abundance, pooled across height (Pearson's chi-squared test, DF = 8, χ 2 = 423.9, P < 0.0001) (Fig. 4). The relative abundance of Haemagogus was higher in nets than in traps, while for Psorophora, the reverse was true. The relative abundance of Sabethes was similar using both methods. When comparing the methods for individual species stratified by height, there was a significant difference in the relative abundance of Hg. janthinomys (DF = 2, χ 2 = 22.7, P < 0.0001) which was collected in higher numbers at ground level compared with elevated heights in nets than in traps. There was also a significant difference in the relative abundance of Ps. amazonica (DF = 2, χ 2 = 102.1, P < 0.0001), which was collected in lower numbers at ground level compared with elevated heights in nets than in traps. Small sample sizes precluded analysis of Sabethes mosquitoes at species level. Species abundance by height. As expected, Hg. janthinomys showed significant stratification in abundance across the sampled heights when analyzed using data summarized by date and height (Kruskal-Wallis test, www.nature.com/scientificreports/ DF = 2, χ 2 = 10.05, P = 0.007). Post-hoc Wilcoxon Each Pair tests revealed this species to be more abundant at 9 m than at 0 m (P = 0.01), but there was no difference in abundance between 0 and 5 m or 5 m and 9 m (P > 0.09 for both comparisons). In contrast, there was no significant stratification in abundance of Ps. amazonica or Ps. albigenu across sampled heights (P > 0.12 for both comparisons). Data were further summarized by morning (10:00-11:59) and afternoon (12:00-14:59) due to the differences in stratification of microclimate during these periods (Fig. 3a, b). This showed that Hg. janthinomys abundance was not significantly stratified in the morning (Kruskal-Wallis test, DF = 2, χ 2 = 4.7, P = 0.1) when microclimate generally overlapped across heights, but it was in the afternoon as microclimate became more distinct (Kruskal-Wallis test, DF = 2, χ 2 = 11.4, P = 0.003). Again, post-hoc Wilcoxon Each Pair tests revealed Hg. janthinomys to be more abundant at 9 m than 0 m (P = 0.001) in the afternoon, whereas comparisons between 0 and 5 m (P = 0.085) and 5 m and 9 m (P = 0.051) were not significant. Neither Ps. amazonica or Ps. albigenu showed significant stratification in the morning or afternoon sampling periods (P > 0.08 for all comparisons). Sabethes species were not included in analyses of abundance due to small sample sizes.
Lag to first approach. Lag to first approach was defined as the number of minutes that elapsed from the start of collections until the first mosquito of a given species was caught. If no mosquito of that species arrived during a particular collection, lag was set to 300 min (the full duration of collection). Lag to first approach differed significantly between heights for Hg. janthinomys (Kruskal-Wallis, DF = 2, χ 2 = 6.4, P = 0.04), which arrived A linear regression showed that, across all heights, lag to first approach decreased significantly as Hg. janthinomys abundance increased (DF = 1, F = 52.1, P < 0.001), and the same pattern held when data from each height was analyzed separately (P < 0.005 for each analysis). As Hg. janthinomys abundance was higher at 9 m, this suggests that the difference among heights in lag to first approach is attributable to differences in abundance. When only considering collections in which least one Hg. janthinomys was caught, a multiple linear regression showed that lag to first approach did not differ by height but did differ by abundance (DF = 1, F = 16.7, P = 0.0001) and mean daytime temperature (DF = 1, F = 4.2, P = 0.04), with faster approaches occurring at higher abundance and higher temperature. Lag to first approach did not differ among heights for either Ps. amazonica or Ps. albigenu. Again, small sample sizes precluded analysis of Sabethes at species level. The mean time (+ / − 1 S.E.) to first approach for each species at each height is listed in Supplementary Table S4.
Species occurrence. Based on data summarized by date and height of collection, nominal logistic regressions revealed significant positive associations between several of the predictor variables and the occurrence of target species. Seven-day cumulative rainfall at a lag of 1 week (DF = 1, χ 2 = 4.7, P = 0.03) was a significant positive predictor of Hg. janthinomys occurrence (Fig. 5), while height also had a marginal effect but was not significant in this analysis (DF = 2, χ 2 = 5.1, P = 0.08). Rainfall at a lag of 1 week was associated with an increase in Haemagogus occurrence in our earlier study 15 . Date of collection (DF = 1, χ 2 = 27.8, P < 0.0001), rainfall at a lag of one week (DF = 1, χ 2 = 8.9, P = 0.003), and rainfall at a lag of four weeks (DF = 1, χ 2 = 4.6, P = 0.03) were all significant predictors of Ps. amazonica occurrence (Fig. 5). Mean daytime temperature followed a similar www.nature.com/scientificreports/ but non-significant trend (DF = 1, χ 2 = 3.3, P = 0.07) and height had no significant effect on the occurrence of this species. The association between Ps. amazonica occurrence and rainfall at lags of one and four weeks corresponded with our previous results 15 . Rainfall at a lag of one week (DF = 1, χ 2 = 9.3, P = 0.002) was a significant predictor of Ps. albigenu occurrence, while height (DF = 2, χ 2 = 5.0, P = 0.08) had a marginal but non-significant effect on this species which we had not previously analyzed 15 . Height was a significant predictor of Sa. chloropterus occurrence (DF = 2, χ 2 = 14.5, P = 0.0007), but there was no effect of 7-day cumulative rainfall (Fig. 5). We had previously shown that Sa. chloropterus occurrence significantly increased with rainfall at a lag of 3 weeks 15 .

Relationships between temperature and occurrence. BG-Sentinel trap collections had previously
shown that Haemagogus and Sabethes species mosquitoes increased in occurrence with increasing temperature and decreasing relative humidity when analyzed at genus level 15 . In the current study, we found that these mosquitoes were mainly active between 26 °C and 33.9 °C when classified in ≈4 °C increments (Fig. 6), which were chosen to provide a balance between the range and sample size of each temperature class. Both Ps. amazonica and Ps. albigenu were more tolerant of lower temperatures (22-25.9 °C), while Ps. amazonica was the most abundant of the species collected at temperatures higher than 33.9 °C. When analyzing relationships between species occurrence and microclimate using data summarized at 15-min time intervals by date and height, a nominal logistic regression revealed a significant interaction in the impact of height and temperature on the occurrence of Hg. janthinomys (DF = 1, χ 2 = 4.4, P = 0.036). Subsequent nominal regressions analyzing the effect of temperature stratified by height showed a significant positive effect of temperature on occurrence at 0 m (DF = 1, χ 2 = 13.9, P = 0.0002) and 5 m (DF = 1, χ 2 = 17.0, P < 0.0001), but not at 9 m (DF = 1, χ 2 = 2.0, P = 0.157). Interestingly, the mean temperature for the intervals in which Hg. janthinomys was collected was 29.9 °C at all three heights. Nominal logistic regressions showed no effect of temperature on the occurrence of Ps. amazonica (DF = 1, χ 2 = 0.3, P = 0.58) but did reveal an effect on the occurrence of Ps. albigenu (DF = 1, χ 2 = 7.3, P = 0.007), which preferred cooler than average conditions. There was no effect of temperature on the occurrence of either Sa. chloropterus (DF = 1, χ 2 = 2.0, P = 0.15) or Sabethes when analyzed at genus level (DF = 1, χ 2 = 0.3, P = 0.59), although sample sizes were small (n = 44 and n = 132, respectively).

Community composition of mosquitoes in artificial and natural containers.
A total of 2070 adult mosquitoes (50.9% female) representing seven genera and 11 species were reared from ovitraps, of which 99.9% www.nature.com/scientificreports/ were identified to species level (Fig. 7). In contrast to net collections, Culex was the most abundant genus (84.7% of the total catch), followed by Limatus (13.7%), Sabethes (1.3%), Wyeomyia (0.2%), Haemagogus (0.05%), Orthopodomyia (0.05%), and Toxorhynchites (0.05%). Collections in standard and fruit pod ovitraps were dominated by Culex urichii which comprised 75.2% of the catch in these containers, while Limatus flavisetosus (11.5%) was also relatively abundant. Mosquitoes reared from bamboo represented just 6.1% of the overall catch but included Sa. quasicyaneus (22 mosquitoes), Sa. batesi (5), and Hg. janthinomys (1), which were absent from other container types. Mosquito communities in standard and fruit pod ovitraps overlapped substantially when analyzed using the Morisita index at species level (Supplementary Table S4) but overlap decreased when comparing these traps with bamboo. When comparing heights for the combined standard and fruit pod ovitraps, there was substantial overlap at 0, 5, and 9 m, but this decreased slightly when comparing each height with 15 m. When comparing heights for individual container types (Supplementary Table S5), the standard ovitrap followed a similar pattern, but the fruit pod showed substantial overlap among all heights sampled. Bamboo communities were quite dissimilar when comparing 0 m with elevated heights, and those at 15 m were distinct from all other heights sampled, but this was based on a small number of specimens.

Discussion
The intersection of humans, mosquitoes, and monkeys in forests bordering urban environments creates risk for spillover and spillback of mosquito-borne viruses between urban and sylvatic cycles. We constructed a tower at the edge of a treefall gap to investigate relationships between microclimate and the vertical stratification of potential mosquito bridge vectors. Our results showed little difference in mean temperature and relative humidity across heights during the morning hours, but in the afternoon, it was hotter and less humid at 9 m than at ground level. This contrasted with the clear stratification of microclimate throughout the corresponding sampling period in our BG-Sentinel trap study 15 . Our findings at the tower mostly agreed with a study by Shuttleworth investigating the vertical stratification of microclimate at the Ducke reserve 40 . This showed that differences in temperature and humidity between heights gradually increased throughout the day and decreased at night, although temperature divergence began earlier in the morning than in our study. Others have also reported steep vertical gradients in temperature and humidity in the afternoon relative to the morning 41,42 . We observed higher mean temperatures at ground level than at 9 m during the morning collection hours and reasons for this may be explained by a breakdown in the vertical stratification of microclimate at forest edges, which has been shown to occur in temperate rainforest 43 . www.nature.com/scientificreports/ We compared diurnally active, anthropophilic mosquito communities sampled with nets in this study with our previous BG-Sentinel trap collections 15 . In the current study, Ps. amazonica, Hg. janthinomys, and Ps. albigenu were the most abundant species, while Sa. chloropterus was poorly represented. In agreement with our earlier work 15 , the Morisita index showed substantial overlap in species composition between the sampled heights but indicated that elevated mosquito communities were more like each other than they were the ground level community. This was also reflected in the results of the principal components analysis. Other studies [44][45][46] have shown similar changes in community composition with increasing height above the forest floor and found that species richness and mosquito abundance are generally higher at ground level than at the heights sampled in this study. We also found species richness to be greatest at 0 m, although overall abundance was similar between 0 and 9 m. The latter finding may be explained by our sampling methods which limited collections to one mosquito per minute, potentially underestimating abundance at ground level. This limitation was applicable to Ps. amazonica which was present in high relative abundance at ground level compared to elevated heights in BG-Sentinel traps but not in nets.
Our analyses of stratification showed that lag to first approach was closely negatively correlated with the abundance of our target species. The stratification in abundance of Hg. janthinomys sampled using nets was less prominent than in the BG-Sentinel trap study (Fig. 4). However, it differed between the morning and afternoon sampling periods, corresponding with differences in the stratification of microclimate ( Fig. 3a and b). This species was only more abundant at 9 m than at ground level after midday, when temperatures were higher at elevated heights. Whereas several studies have indicated that Hg. janthinomys stratification shifts seasonally with changes in microclimate 18,28 , there is little known about diurnal shifts in this regard. Unfortunately, we were unable to investigate seasonal changes in stratification or community composition since our collections were truncated by the COVID-19 pandemic. We found no evidence of stratification in the abundance of Ps. amazonica or Ps. albigenu which were frequently caught at each height. While Psorophora species are known to be predominantly ground dwelling, they are often collected in the forest canopy 46,47 . In our study, the high relative abundance of both Ps. amazonica and Hg. janthinomys at each height sampled indicates a clear potential route for arbovirus transmission between the ground and forest canopy.
In contrast with our previous study 15 , the occurrence of Hg. janthinomys was not significantly stratified by height. There was also no effect of height on the occurrence of Ps. amazonica and Ps. albigenu. However, height did influence the occurrence of Sa. chloropterus in keeping with its acrodendrophilic behavior 15,47 . Several species showed significant associations with rainfall lag, an important predictor of mosquito emergence and vector-borne disease 48,49 . On this occasion, Hg. janthinomys occurrence increased with increasing rainfall at a lag of 1 week, an association that was previously reported for Haemagogus mosquitoes at genus level 15 . The occurrence of Ps. amazonica increased at lags of 1 and 4 weeks, concurrent with our previous results 15 . These findings support our earlier work and highlight reliable traits for modeling the emergence of both species. In addition, the occurrence of Ps. albigenu was associated with rainfall at a lag of 1 week which reflects the need for Psorophora species to develop quickly in transient breeding sites 50 . However, the association between Sa. chloropterus occurrence and rainfall at a lag of 3 weeks that we previously reported 15 was not detected on this occasion, despite the sample size being slightly larger than in the earlier study. Since Sa. chloropterus develops slowly in large tree holes with small openings that minimize water loss 51,52 , its emergence may be less dependent on rainfall in the weeks immediately prior to collection than other species.
In agreement with our BG-Sentinel trap study 15 , we found that Haemagogus and Sabethes species were mainly active between ≈26 °C and 34 °C (Fig. 6). In addition, Ps. amazonica was more tolerant of more extreme temperatures while Ps. albigenu also tolerated cooler conditions. We previously suggested that relationships exist between the occurrence of Haemagogus and Sabethes mosquitoes and microclimate 15 but had insufficient data to draw inference at species level. In this study, we showed that the mean temperature at time intervals in which Hg. janthinomys was collected was 29.9 °C at each height. This was greater than the mean daily temperature at ground level (29.0 °C) and shows that this species prefers warmer conditions. We also found that Hg. janthinomys occurrence increased with increasing temperature at 0 m and 5 m, but not at 9 m, where microclimate was more likely to be within its preferred range. Bates 18 remarked that this species mainly occupied the lower canopy during his extensive investigations in Colombia, while Bates and Roca-Garcia 53 showed that it was more tolerant than other local mosquitoes to laboratory temperatures of 30 °C. These findings agree with our evidence that Hg. janthinomys abundance was not significantly stratified in the morning when temperatures were similar at each height. They may also explain the differences in relative abundance of Hg. janthinomys collected at ground level in nets and in traps (Fig. 4). Whereas net collections were often made in direct sunlight, BG-Sentinel traps were mainly hung beneath the forest canopy where microclimate was cooler and more humid than in the current study 15 . This was especially true at ground level where the mean temperature was never higher than 27.2 °C and Hg. janthinomys was rarely caught. In addition to microclimate influencing stratification, a recent study of the sylvan malaria vector, Anopheles cruzii, suggests that mosquitoes present at ground level may have a different genetic structure to those found in the forest canopy, and that anthropogenic habitat fragmentation may drive the vertical dispersal of particular genotypes 54 .
In addition to collecting diurnally active, anthropophilic host-seeking mosquitoes, the tower also enabled us to conveniently sample certain container-breeding species. The composition of mosquitoes collected with ovitraps was strikingly different to those sampled with nets and BG-Sentinel traps. Despite rearing over 2000 adults from the three container types, the abundance of target species was too low to investigate vertical stratification. The predominant mosquito in standard and fruit pod ovitraps was Cx. urichii, a diurnal species 55 recently found in high abundance near forest edges at a rural settlement north of Manaus 56 . However, there is little evidence to suggest that Cx. urichii is attracted to humans 55,57 . The bamboo community was quite distinct from the other ovitraps and was the only container type to yield Sabethes mosquitoes. While bamboo ovitraps are useful for collecting certain Haemagogus and Sabethes species 25 www.nature.com/scientificreports/ Hg. janthinomys and Sa. chloropterus sampled with this method is not always commensurate with the abundance of adult populations. We only reared a single Hg. janthinomys specimen from bamboo despite this species being common at our study site, although our methods differed slightly from others that have proven more successful for collecting this species 59,60 .
Our study primarily focused on the population of diurnally active, anthropophilic mosquitoes most likely to effect spillover and spillback in rainforest bordering Manaus. While we identified several potential routes of transmission for mosquito-borne viruses between the ground and forest canopy, there is scope to further investigate crepuscular and nocturnal species not detected during the current study.

Data availability
All data generated or analyzed during this study are included in this published article (and its Supplementary  Information files).