A total of 707 individual mosquito recordings (including the manual extractions) from 114 assays were included in the analysis (Table Individual flights lasting longer than 10 minutes were recorded in 46 mosquitoes, while flights exceeding 30 minutes were recorded in 12 mosquitoes. Total flight ranged 1 – 16,107 sec with a mean total of 1257.2 sec.
Table 1. Mosquito samples by season, gonotrophic state and species, for which flight data was collected and analyzed
Season
|
Gonotrophic stage
|
An. coluzzii
|
An. arabiensis
|
An. gambiae
|
Total
|
Dec-Feb
|
Gravid
|
62
|
0
|
2
|
64
|
|
Unfed
|
2
|
0
|
0
|
2
|
Mar-Apr
|
Gravid
|
31
|
0
|
0
|
31
|
|
Unfed
|
0
|
0
|
0
|
0
|
Jul
|
Gravid
|
75
|
0
|
3
|
78
|
|
Unfed
|
0
|
0
|
0
|
0
|
Aug-Sep
|
Gravid
|
125
|
10
|
62
|
197
|
|
Unfed
|
30
|
1
|
8
|
39
|
Oct-Nov
|
Gravid
|
114
|
95
|
76
|
285
|
|
Unfed
|
5
|
5
|
1
|
11
|
Total
|
|
444
|
111
|
152
|
707
|
Nightly flight activity and identification of putative Long-Distance Flyers (LDMs)
To determine if flight activity was concentrated in certain parts of the night, we examined three indices, namely, 1) number of hourly flight bouts, 2) longest flight bout, and 3) total hourly flight (across species, season and gonotrophic state) (Fig. 3). To consider possible differences in nightly activity between appetitive and strong flyers, we evaluated both the median and 90th percentile of each flight aptitude index.
Figure 3. Nocturnal flight activity. Hourly flight bouts and total hourly flight across all 707 mosquitoes, (across species, season, and gonotrophic state). Hourly flight activity (flight aptitude) through the night showing the median (left column) and the 90th %il (right column), representing trends of most mosquitoes and higher flight activity mosquitoes respectively. The 95% confidence interval of each hour (based on bootstrapping) not shown in full to emphasize the nightly trend.
Overall, there were no significant peaks of activity identified in the hourly flight data. Variation, as measured by hourly 95% confidence intervals, was also found to be non-indicative of clear modality. Although there was a mild modality, suggesting elevated total flight and flight bouts between 11:00 and 02:00h (but not in longest flight), this modality was not found to be statistically supported by the 95% CI which overlapped widely. We concluded that the flight activity was spread homogenously throughout the assay time and used the full length/duration (21:00-05:00h) to measure flight aptitude.
Identification of putative migrants
Considering all recorded mosquito flights (n=707 mosquitoes), the asymmetric distributions of each flight aptitude index revealed a long right tail. This observed distribution corresponds to existing literature, with frequency distributions of laboratory-measured bouts of flight [61,62] showing a majority of individuals making short flights and a only few making long ones (Fig. 4).
Figure 4. Flight aptitude index distributions: longest flight (a), total flight (b), and flight bouts (c). Flights were divided into two classes (x-axis; MAD units): LFAs; below 3.5 (left of vertical red line), and HFAs; above 3.5 (right of vertical red line) based on guidelines for outlier detection (see Methods). The data are based on 707 wild female mosquitoes representing three species, both unfed and gravid females. Y-axis denotes the frequency in percent of the sample.
Following previous flight behavior studies [31–34,37,63], the mosquitoes at the far-right tail of the distribution were suspected to represent long-distance flyers, or HFA individuals, in our study. Subsequently, we evaluated differences in the proportion of HFAs among species, seasons, and gonotrophic state.
For the most part, flight aptitude indices were significantly correlated with each other; however, correlation coefficients (Spearman) were negative (r = -0.43) between the longest flight bout and the number of flight bouts and moderately positive (r = 0.32 – 0.52) between total flight and other indices (Fig. S3). The relatively low correlation coefficients indicate that each flight index conveys unique information—the negative correlation suggests that flight bouts describes “restlessness” unlike “flight persistence” that is captured by the longest flight bout and especially total flight. Therefore, we compared all three indices to determine which were more important/informative predictors of migrants.
Variation of flight aptitude by season
Seasonal variation in flight aptitude was tested on gravid An. coluzzii, as this was the only species (and gonotrophic state) found across seasons. Significant differences were most pronounced when looking at longest flight bout (P<0.002, overall Monte-Carlo Exact test), but were also detected when examining total flight (P<0.005, overall Monte-Carlo Exact test) (Fig. 3a-b, respectively). With regards to the longest flight bout index, the highest fraction of HFA was discovered in the late wet season (Oct-Nov; 39.5%), followed by the mid-wet season (Aug-Sep; 29%), early wet season (Jul; 19%), early dry season (Dec-Feb; 22%), and finally, the lowest fraction being in the late dry season (Apr; 10%, P<0.005, 2-tailed test, Fig. 3a). A similar trend, albeit with smaller differences, was detected when looking at total flight, with only a single significant difference between late wet season and late dry season (P<0.04, Fig. 3b). Flight bouts did not follow this pattern and showed no significant difference in the overall test (Fig. 3c).
Figure 5. Variation of flight aptitude indices by season; longest flight (a), total flight (b), and flight bouts (c). Variation in An. coluzzii flight aptitude between seasons; The x-axis depicts the different parts of the year (‘seasons’); Dec-Feb and Apr represent the dry season. Jul, Aug-Sep and Oct-Nov represent the wet season. Y-axis values are percent frequencies of the HFA populations, with n above each bar. Seasonal flight aptitude comparison was carried out on gravid An. coluzzii females, the only species which had samples across seasons.
Variation of FA between species
Variations in flight aptitude between species were tested on gravid females between Oct-Nov when all species were represented. Considering total flight, An. coluzzii exhibited a significantly higher fraction of HFAs (25%) than An. arabiensis (8%), with An. gambiae displaying intermediate values (17%) (Overall Monte Carlo Exact Test χ2=10.0, P<0.015) (Fig. 6b). Contrasting tests between species showed a significant difference between An. coluzzii and An. arabiensis (Wald χ2=8.7, P<0.004). Although not statistically significant, similar trends were revealed in both longest flight and flight bout indices (Fig. 6a and c).
Figure 6. Variation of flight aptitude between species (gravid females in Oct-Nov, when sample sizes were sufficient); longest flight (a), total flight (b), and flight bouts (c). Values are percent frequencies of the flyer populations, with n above each bar. Overall test is a contingency table exact test using Monte Carlo with 10,000 replicates (P-values pertain to 2-sided tests). Specific comparisons are shown were tested using contrasts in logistic regression if overall test was significant.
Variation in flight aptitude between gonotrophic stages
The effect of gonotrophic stages on flight aptitude was tested after pooling the species, as well as with stratification by species (CMH test) in August-September, when the number of unfed females was suitable for such a test. These tests revealed that a significantly higher rate of HFAs among gravid females (11.2% vs. 0%, P<0.013 Overall Monte Carlo Exact Test) was detected in flight bouts (pooled; P<0.024, 1-tailed Fisher Exact test, and when stratified (CMH = P<0.049, 2-tailed test) (Fig. 7c). However, while there were no significant differences detected when looking at the longest- and total flight indices, a consistent trend of higher HFA among gravid females was observed (Fig. 7a and b, respectively).
Figure 7. Variation in flight aptitude between gonotrophic stages; longest flight (a), total flight (b), and flight bouts (c). Overall test is a contingency table exact test using Monte Carlo simulations with 10,000 replicates. This Gonotrophic state comparison was done when on pooled species in Aug-Sep, when sample size was suitable for comparison. Prior CMH test showed that the effect was significant across species (not shown) and no heterogeneity between species was detected.
Wing morphology and flight aptitude
Among the three species, significant differences in flight activity were found in An. coluzzii with regard to longest flight duration (Fig. 8a and d) and total flight (Fig. 8b and e), showing both longer (Fig. 8a and b) and wider wings (Fig. 8d and e) in HFAs (red) (Overall ANOVA; P<0.02).
Figure 8. Box-whisker plot of wing length (top), and width (bottom) across species (x-axis, abbreviated species names: arab.=An. arabiensis, coluz.=An. coluzzii., gamb.=An. gambiae s.s.) in longest flight (panels a and d), total flight (panels b and e) and flight bouts (panels c and f) for LFAs (blue) and HFAs (red). Mean marked as ○ (for LFAs) or + (for HFAs). Horizontal line within box represents the median. Box bottom and top are 25th and 75th percentile, respectively, whiskers extend up to 1.5 time the inter-quartile range and outliers (‘o’ or ‘+’) represent observations that extend beyond the whiskers. Significantly larger mean and median wing dimensions in HFAs vs. LFAs indicated by asterisks on the left and right of the arrow (showing direction of increase; LFA or HFA), respectively. One tailed significance levels of P<0.05 and P<0.01 measured by ANOVA (left of arrow) or Median score tests (right of arrow); shown as ‘*’ and ‘**’ respectively.
Allometry of wings to detect wing shape variation within HFA’s
In gravid An. coluzzii during the wet season, total-flight HFAs had wider wings when compared to LFAs, after adjusting for wing length (P<0.035, 1-tail test) (Fig. 9). We found no significant interaction between wing length and HFAs, indicating that the shape effect was monotonic with wing length. A similar trend was observed in An. arabiensis, but no significant difference in intercepts was detected, possibly due to smaller sample size.
Figure 9. Wing allometry in HFAs (red) and LFAs (blue) for total flight across species (the lines shown were computed based on separate linear regression models for flyer type of each species). In gravid An. coluzzii during the wet season, total-flight HFAs had wider wings than LFAs after adjusting for wing length (P<0.035, 1-tail test).
Consistent with previous studies [64,65], An. coluzzii also displayed longer wings in the early dry season (December-February), when compared to other months in the year (Fig. S4). This increase in length was isometric—in other words, it was accompanied by a proportional increase in wing width during this time (i.e. no change in wing shape).