Introduction

An adult female mosquito’s age is a key factor in its ability to blood feed, lay eggs and transmit pathogens. Anthropophilic species like Anopheles gambiae typically obtain their first bloodmeal 3–5 days post-eclosion1. The extrinsic incubation period (EIP), defines the time it takes for a vector-borne pathogen ingested in a competent mosquito’s blood meal to infect or penetrate the mosquito’s tissues, undergo development (in the case of Plasmodium parasites or filarial worms) and disseminate to tissues from where it can be transmitted to the next vertebrate host (e.g. salivary glands or thoracic flight muscles). For Plasmodium falciparum in African Anopheles spp., the EIP is approximately between 9 and 16 days depending on temperature and species2,3, while the typical lifespan of an Anopheles mosquito in the wild is roughly 15–20 days3,4,5. Thus, very few mosquitoes in populations ever become infectious and large sample sizes are required to reliably estimate transmission intensity. However, much effort, time and costs can be expended to capture and test for infectious mosquitoes from these large collections. As such, the ability to rapidly and accurately age-grade captured mosquitoes from a population could allow for simple and less resource-intensive estimates of transmission risk and help better evaluate the efficacy of vector control interventions that often occur before periods of high pathogen transmission.

Original mosquito age-grading techniques were qualitative, such as classifying wing degradation, assessing the presence of mites on mosquitoes, or meconium in midguts6,7,8. Eventually evaluating parity by assessing tracheole skeins on dissected ovaries became one gold standard age grading technique, which was further refined and quantified by counting ovariole dilatations from dissected ovaries9,10,11. However, these techniques can be imprecise, they only provide physiological age rather than calendar age, and it can be difficult and time-consuming to process many mosquitoes per field collection with the dissection techniques required. New age-grading techniques have been investigated, including biochemical, gene and protein profiling, and spectroscopy methods12,13,14,15,16,17,18,19,20,21,22,23,24,25. While these methods are reinvigorating mosquito age-grading processes, there remain challenges towards practical implementation in the field. These challenges include the need for expensive equipment which can be difficult to use in a field setting, or shipment of specimens for analysis by this equipment outside of the field site, as well as destructive testing of the samples. Furthermore, these techniques may require recalibration of age models at each field site and/or timepoint the samples are collected, and they can involve advanced technical and statistical expertise to run samples and analyze the data produced21,26.

We previously developed a simple, low-cost, high-throughput, and nondestructive method of age-grading lab-reared Anopheles gambiae populations by calculating the pixel intensity (PI) of wing photos, which analyzes the intensity (darkness) of the pixels of a mosquito wing photo in greyscale27. This method only requires a dissecting microsope with an integrated camera and a computer program, it takes ~ 30 s to process each mosquito, and it can quantitatively assess wing scale loss over time due to flying because a mosquito’s wing scales are often darker than the underlying membranous wing. In the study, we demonstrated this method’s ability to distinguish the age structure of two lab mosquito populations reared in different mesocosms, whereby one was regularly treated with blood meals that contained the mosquitocidal drug ivermectin (IVM), and the other control mesocosm was only provided regular blood meals. However, it was unclear if this method could be extended to field mosquitoes. Here, we have slightly refined our techniques for PI analysis of wings, applied our lab-reared An. gambiae age-model to estimate ages of wild caught An. gambiae s.l. captured during a clinical trial conducted in Burkina Faso called RIMDAMAL II (which tested the effect of ivermectin mass drug administrations (MDA) on malaria incidence)28, and demonstrate how this technique can infer the efficacy of vector control interventions that occurred during the trial.

Results

Total pixel intensity vs. mean pixel intensity

Our previously published PI analyses calculated the total PI of each mosquito’s wing photos (both left- and right-wing) and used these two values to calculate the average of the total wing PI per mosquito27. However, this analysis yielded variable results from some mosquitoes caught and processed during the early sampling periods of RIMDAMAL II trial, which we determined were due to some differences in file sizes, and perhaps photo resolution and/or relative photo brightness compared to pictures taken during rest of the trial that were standardized thereafter. Independently analyzing the mean PI of each wing photo, rather than the total PI, normalized the PI output regardless of file size and other parameters. A mean PI per mosquito was then calculated by averaging the mean PI from each wing, which allowed for more normalized comparisons across groups of mosquitoes processed during the trial. Overall, the prior published analysis of total PI gave ranges from 2.4 × 108−6.6 × 108 while our updated analysis of mean PI gave ranges from 127.85 to 133.08.

PI differences among mosquito wings

Next, we wanted to understand how the variance between the two wings per mosquito may be associated with the mean PI, which may potentially influence the predicted ages of a mosquito population. We calculated the standard deviation (SD) of the mean PI between the two wings per mosquito and then the coefficient of variation (CoV), and plotted the CoV against the mean PI. Among the lab mosquitoes aged in a mesocosm, the mean CoV was 0.13 (Range: 0.00-0.76), and among the wild mosquitoes, the mean CoV was 0.08 (Range: 0.0004-3.34). Comparisons of CoV from both mosquito populations relative to mean PI show weak correlation where little of the variation among CoVs is predictable by mean PI (Fig. 1), suggesting that as a mosquito ages and loses more scales through more flying, there is only a small likelihood of having one wing that has significantly fewer scales than the other. Finally, analysis of each pair of wings suggests individual An. gambiae s.l. do not lose scales more often from one wing compared to the other. 51.69% (95% 49.42–53.96%) of left wings had a higher PI while 48.31% (95% CI: 46.04–50.58%) right wings had a higher PI (Chi-square test for proportions P = 0.075) (Supplemental Fig. 2).

Fig. 1
figure 1

The standard deviation of the mean pixel intensity values for each set of a mosquito’s wings is modestly correlated with increasing mean pixel intensity (PI) or mosquito age. Each dot represents data from a set of a mosquito’s wings. Top panels: laboratory-reared An. gambiae aged in a mesocosm (n = 149 in both A and B). Bottom panels: An. gambiae s.l. captured during the RIMDAMAL II trial (C; n = 1,861), (D; n = 549). Regression line slopes in panels A, C, and D are significantly different from zero (panel A, P = 0.008, r2 = 0.045; panel C, P < 0.0001, r2 = 0.1474; panel D P < 0.0001, r2 = 0.054).

Full size image

Individual and binned pixel intensity of mosquito populations to examine inferred age structure

We then plotted the proportions of wild mosquitoes from the RIMDAMAL II trial from which we took wing pictures, both individually and in bins of 0.3 PI units, to observe the overall inferred age structure (Fig. 2). The plots shows distributions that mimic wild mosquitoes age-graded by the gold standard Polovodova method5 as well as newer techniques such as mid-infrared spectroscopy (MIRS)23, suggesting a population distribution dominated by nulliparous and young mosquitoes in the lowest PI ranges, and relatively few older females in the long top- or right-skewed tail of the individual or bin distributions, respectively (Fig. 2). Overall, most mosquitoes captured during the trial had PIs of ≤ 128.73 (Range: 127.81-133.43) with a top- or right-skewed distribution tail extending to a PI of 133.43. When the population was split based on mosquitoes captured in each arm of the trial, the top- or right-skewed tail of the distribution was more prominent among the population from placebo clusters but not intervention (ivermectin-treated) clusters (Fig. 3). Furthermore, when separating the data into the trial years, the top- or right-skewed tail of the distribution was evident among the population captured in 2019 but absent in 2020 (Fig. 4). The PI bin distributions among the populations captured in 2020 among clusters from both arms were grouped in only 3 of the lowest PI bins (Fig. 4).

Fig. 2
figure 2

Pixel intensities (PI) from wild Anopheles gambiae s.l. captured in the RIMDAMAL II trial reflects typical population age distributions from wild mosquitoes. (A) The distribution of individual mosquito’s mean wing PIs. Each dot represents the mean PI from a set of a mosquito’s wings. (B) Bins of mosquitoes’ mean wing PIs by 0.3 PI units/bin.

Full size image
Fig. 3
figure 3

Pixel intensities (PI) of wild Anopheles gambiae s.l. captured in the RIMDAMAL II trial, grouped by arm. (A) The distribution of individual mosquitoes’ mean wing PIs from the Placebo and Intervention arms. Each dot represents the mean PI from a set of a mosquito’s wings. (B) Bins of mosquitoes’ mean wing PIs, from mosquitoes captured from clusters of the Placebo (B) or Intervention (C) arms.

Full size image
Fig. 4
figure 4

Pixel intensities (PI) of wild Anopheles gambiae s.l. captured in the RIMDAMAL II trial, grouped by trial year. (A) The distribution of individual mosquitoes’ mean wing PIs grouped by the year they were captured during the rainy season intervention periods (late July to early November of 2019 and 2020). Each dot represents the mean PI from a set of a mosquito’s wings. (B) Bins of mosquitoes’ mean wing PIs, from mosquitoes captured during trial years 2019 (B) and 2020 (C).

Full size image

To observe how the changes in PI associates with the timings of the two mosquito control interventions that occurred during the trial, the median PIs between mosquito populations collected one and three weeks after each MDA performed in the trial were graphed (Fig. 5). Ivermectin or placebo was distributed monthly in the respective trial arms throughout both trial years, and distribution of new dual-chemistry Interceptor G2 (IG2) insecticide treated nets (ITNs; containing chlorfenapyr + alpha-cypermethrin) occurred in all clusters of both arms in early November 2019. This ITN distribution was performed near the end of the first rainy season, approximately 2–4 weeks after the 4th MDA between sampling weeks 13 and 15 when mosquito capture numbers were very low (Fig. 5)28. Thus, most of the IG2 nets were not deployed by the populace until 2020 and were used by households from both arms throughout that year.

Fig. 5
figure 5

Median pixel intensities (PIs) from wild Anopheles gambiae s.l. captured in the RIMDAMAL II trial by sampling period and relative to the timing of the mass drug administrations. Median PI of mosquitoes captured from the Placebo (blue: n = 1,129) and Intervention (red: n = 799) arms. Whiskers depict 95% CIs. Mosquito sampling weeks over the 2-year intervention period (late July-early November in 2019 and 2020) are enumerated consecutively on the X-axis, and mass drug administrations (MDA) of either placebo or ivermectin tables are marked with vertical or dotted lines and occurred over a 3-day periods. The period of IG2 ITN distributions in 2019 to all villages in the health district is marked on the X-axis.

Full size image

A lognormal generalized linear mixed model was generated using individual wing PIs from the captured mosquitoes and a log link function. Random (grouping) effect terms of individual mosquito ID, village ID, and household ID were included in the final model, along with a four-way interaction term of year, study arm, weeks after MDA, and day of the year. Results were exponentiated for interpretation. This model revealed a significant reduction in median PI of mosquitoes captured in 2020 compared to 2019 (exp(β): 0.996 [95% CI: 0.995–0.996], p < 0.001), suggesting a significant effect of the new ITNs on the median PI, lowering it below 128.5 among mosquito populations of both arms beginning with the 4th MDA and lasting to the end of the trial. Similarly, the model revealed a seasonal trend of decreasing median wing PI across the collection season overall (exp(β) of the scaled “day of year” parameter: 0.998 [95% CI: 0.997–0.998], p < 0.001). This trend was nullified in 2020 collections (exp(β): 1.002 [95% CI: 1.002–1.002], p < 0.001). The model did not attribute significant differences in median wing PI to differences between mosquitoes captured from the different arms overall (p = 0.55), or only in 2020 (p = 0.83). However, it did indicate that, in 2019, median wing PI was higher three weeks after MDA in placebo clusters relative to the one week-post-MDA timepoint (exp(β): 1.001 [95% CI: 1.001–1.002], p < 0.001), with an opposite trend observed among mosquitoes collected in IVM clusters (exp(β): 0.998 [0.997–0.999], p < 0.001). This trend was strongest earlier in the 2019 collection season compared to the later season overall (p = 0.005) and in IVM clusters (p < 0.001). In 2020, the model revealed a small increase in median PI among mosquitoes captured in IVM clusters 3 weeks-post-MDA compared to 1 week-post-MDA (exp(β): 1.002 [95% CI: 1.001–1.003], p = 0.003). This trend was strongest (i.e., the model revealed the largest differences) early in the season, trending towards a relative decrease in IVM median PI 3 weeks-post-MDA relative to 1 week-post-MDA in the later 2020 collections (p < 0.001). No significant overall trends were observed between post-MDA timepoints in 2020 mosquitoes in general (p = 0.12) or as the season progressed (p = 0.72).

Adapting and applying an age model to wild An. gambiae s.l.

To predict age for the wild mosquitoes, we sought to use the colonized mosquitoes aged in laboratory mesocosms to develop a new age model and use it to interpolate and extrapolate the ages of the wild mosquito populations from RIMDAMAL II (Fig. 6)27. The range of mean PI scores from these lab mosquitoes of known age was notably constrained (range: 128.30–129.45) compared to the range obtained from wild mosquitoes (range: 127.85–133.08), but the agreement in the magnitude of PI measured between the lab and field populations was encouraging. The lowest mean PI score from the newly eclosed adults (128.35) was 0.5 PI units higher than the lowest mean PI score obtained from wild mosquitoes (127.85), suggesting wings from the wild mosquitoes tended to have more and/or darker scales than those from the lab colony(Fig. 6).

Fig. 6
figure 6

Age model derived from colonized, marked Anopheles gambiae reared in laboratory mesocosms to known ages. Mean pixel intensities of colonized mosquitoes reared to known ages (n = 169) were plotted by their true age and fitted by stepwise comparisons to a linear and multiple non-linear models. The best fit model (depicted) was a variable slope sigmoidal model. The best fit line of the final model is in black and red dotted lines represent 95% confidence bands.

Full size image
Fig. 7
figure 7

Interpolated ages of wild Anopheles gambiae s.l. captured in the RIMDAMAL II trial using the pixel intensity-based age model derived from colony mosquitoes. Dots are individual wild mosquitoes placed on the best-fit-line of age model; 30% (n = 578) of the wild mosquitoes’ pixel intensity (PI) values fit within the model’s lowest and highest range of 95% CIs. PIs corresponding to the interpolated ages of 5 and 10 were plotted (dotted lines) using the model.

Full size image

A simple linear regression model was first applied to the lab mosquitoes, which had relatively poor fit to the data (r2 = 0.4011) and predicted the wild mosquitoes to be between − 24.31 and 163.17 days old. We then compared various non-linear models that may plausibly fit the data by reflecting the biology of scale loss over time. The best fit model was a four parameter logistic equation (difference in AICc from the linear model = 90.33; r2 = 0.6577), from which the best-fit values were PIs between 128.4 (95% CI 128.3–128.5) and 129.0 (95%CI 129.0–129.1). With this model, PIs from 128.30 to 128.55 were interpolated for mosquitoes between < 1–4 days old, PIs from 128.56 to 128.96 were interpolated for mosquitoes 5–9 days old, and PIs from 128.97 to 129.45 were interpolated for mosquitoes ≥ 10 days old (Fig. 7). However, given the variance of wild mosquito PI data compared to the model, predicted age could only be interpolated from approximately 30% (578/1928) of the wild mosquitoes (Fig. 8). Nonetheless, using this model, we binned all the mosquitoes into these 3 pixel intensity groups, the highest and lowest containing both interpolated and extrapolated wild mosquito ages, and observed that in 2019, wild mosquitoes caught from the placebo arm had fewer ‘middle aged’ mosquitoes with PIs between 128.57 and 128.96 and more ‘old’ mosquitoes with PIs between > 128.96 compared to those caught from the ivermectin arm. However, the difference in PI bin structures in 2020 compared to 2019 were noticeable, whereby all mosquitoes tested were predicted to be ‘young’ (PI < 128.57) in both arms (Fig. 8; Table 1).

Fig. 8
figure 8

Wild Anopheles gambiae s.l. captured in the RIMDAMAL II trial grouped in into three, pixel intensity (PI) bins corresponding to interpolated and extrapolated age classes. (A) mosquitoes from the first year of the trial (2019) and caught in Placebo clusters. (B) mosquitoes from the second year of the trial (2020) and caught in Placebo clusters. (C) mosquitoes from the first year of the trial (2019) and caught in Intervention clusters. (D) mosquitoes from the second year of the trial (2020) and caught in Intervention clusters.

Full size image
Table 1 Proportions and totals number of wild Anopheles Gambiae s.l. Captured from each RIMDAMAL II cluster and arm, and during each trial year, corresponding to their pixel intensities and associated with their predicted age class in days.
Full size table

Discussion

Here we demonstrate that this revised PI analysis of wing photos enables normalization of the data collected. We then showed that the PI distributions, either from individual mosquitoes or with binned data, are similar to anopheles mosquito age structures observed using the gold standard Polovodova method that counts ovarian dilations and is sometimes combined with analysis of Christophers stages11,23,29,30,31,32. In both methods, the bulk of wild caught Anopheles gambiae (~ 70–85%) are skewed towards values suggestive of young ages; for example, using the Polovodova method the bulk of An. gambiae populations caught in Tanzania and Benin were nulliparous or 1- and 2-parous mosquitoes, while using the PI method, the bulk of the An. gambiae population caught in Burkina Faso were in the 3 lowest PI bins meaning they have darker wings due to little wing scale loss)4,5,30. Furthermore, we show how the PI method can be used to observe changes in mosquito population structure in response to two different vector control methods that were performed during the RIMDAMAL II trial, specifically ivermectin mass drug administrations and use of IG2 ITNs. Lastly, we made a mosquito age model with lab mosquitoes and used it to predict the ages of the wild caught mosquitoes from the trial.

The most prominent changes in PI distributions from An. gambiae s.l. captured at intervals during the RIMDAMAL II trial were observed in the significant differences between intervention seasons in 2019 and 2020. These changes were associated with the distribution of new IG2 ITNs at the end of the 2019 season, just prior to the last mosquito sampling period of that year when the rains had ended and the populations of An. gambiae s.l. were nearly gone from the study site. The efficacy of IG2 ITNs on wild An. gambiae has been thoroughly examined, and is known to strongly effect mosquito populations33,34,35,36. The entomological indices we previously reported from the trial suggested a similarly strong effect on malaria vectors from both arms of the trial and that occurred primarily in the in trial’s second season (2020)28,37. This effect included, (a) significant mortality (~ 50%) of indoor-captured, blood fed mosquitoes held in survival bioassays from both arms, and (b) significant reductions in the number of the An. gambiae s.l captured in 2020, which were approximately half of those captured in 2019, and a near absence of An. funestus captured at the study site in 2020 (n = 49) relative to 2019 (n = 1148) despite similar rainfall and climate at the site across these two years. Similarly, the 3-bin PI and predicted age observations from wild mosquitoes captured in 2020 relative to 2019 (Fig. 8) supports the strong mosquito killing effect from these ITNs and demonstrates how PI can be used to track the efficacy of malaria vector control methods.

The ability of this age-grading method to detect mosquito population changes from the effect of ivermectin mass drug administrations was more nuanced in the generalized linear mixed model, but the ivermectin intervention itself was not successful in reducing malaria incidence in the trial, nor in reducing the classical entomological indices of density and entomological inoculation rate28. The primary entomological index that was affected in treatment clusters compared to placebo clusters was the wild blood fed An. gambiae s.l. survival rate, but only in 2019 and only from blood fed mosquitoes captured in the week following the MDAs. Regarding the PI of these mosquito populations, there were modest detectable differences between treatment and control arms at the beginning of the trial in 2019 observed from the mosquito sampling periods that occurred three weeks after MDAs 1–3 were administered. It is interesting that these PI trends mirror the trends of infectious An. gambiae s.l. bites per person per night in each arm shown over the trial, suggesting that lower median PIs may correlate with fewer infectious mosquitoes in the population28. The observation that the strongest effect of ivermectin MDA on mosquito PI seems to be observed 3 weeks after the MDAs occurred, when the direct mosquitocidal effects are inapparent and the drug is at very low pharmacokinetic levels or undetectable in most of the blood samples tested from the populace, suggests a delayed action of the drug’s effect on mosquito population structure and infectiousness28,38. Perhaps this delayed effect is due to increased probability of killing the fewer, older mosquitoes that were actively blood feeding immediately after the MDA was administered, and leaving unharmed most of the younger, newly emerged mosquitoes with lower PIs and that have yet to blood feed.

The age model developed using PIs obtained from a mesocosm-aged laboratory colony of An. gambiae was somewhat limited due to potential differences in the mosquitoes’ wings, as well as the laboratory vs. field conditions experienced by the mosquitoes. Overall, the lab mosquitoes had a very narrow PI range so that we could only interpolate ages of wild mosquitoes with PIs residing inside the range of our model, and we extrapolated the remainder of ages with values below or above the range of the model. Nonetheless, the best fit model suggests relatively small differences in the PIs among very young (1–4 days old) and very old (≥ 10 day old) mosquitoes, respectively, but a rapid shift between these young to old classes in between the ages of 4 to 10 days, presumably due to extensive flying and loss of wing scales during this timeframe. Importantly, wild-type Anopheles gambiae do not become infectious until they complete the minimum EIP, which typically correlates with them reaching the 2- or 3-parous state, or approximately 8–11 days old, thus our model’s ability to predict proportions of ‘old’ mosquitoes with cut-off PIs ≥ 129.7 may be a simple method to classify potentially infectious An. gambiae in a population4,5,30. Unfortunately, there was no way to validate the accuracy of the model’s predictions as we did not perform other independent means of age grading these wild mosquitoes. We assume a better age model could be made from marked, wild type mosquitoes reared in large outdoors mesocosms in Africa.

While our newer analysis method allowed for better normalization of wing photos, further work could be done to refine the use of wing pictures to age mosquitoes. Technologies which auto-crop images to only include the wing, that incorporate wing size into the data analysis, or which focus PI or machine vision analyses on specific wing areas that are more sensitive to scale loss from flying (such as the wing fringe), could lead to a more refined analysis of mosquito age structures. Future segmentation analyses may be able to directly count wing scales and quantify those lost over time from flying and aging. Additionally, further standardization of photo backgrounds and microscope light intensities would likely add more control to the method and allow for easier comparisons across space and time. We believe that these refinements would abrogate the need for regionally calibrated age models at each place and time such analyses were to occur. Finally, due to sample preservation and costs, we were not able to speciate most of the mosquitoes from the entire Anopheles gambiae sensu lato complex from RIMDAMAL II. As such, there may be differences in An. gambiae sensu strictu, An. coluzzi and An. arabiensis PIs28,37. Therefore, further investigations into the An. gambiae species complex are needed to determine if specific species have differing PI ranges.

Overall, we have demonstrated how using PI on wing images can be used to understand the structure of wild Anopheles gambiae mosquito populations, and how their population structures change from vector control interventions. This method might easily be expanded to allow investigators and vector control experts across the world, who are implementing and evaluating control of malaria, helminth and arbovirus vectors, to expediently determine mosquito population ages from their surveillance efforts. Such efforts may be highly valuable for evaluation of their control applications and estimations of the risk of mosquito-borne disease spread in their communities.

Methods & materials

Colony mosquito rearing

Anopheles gambiae (G3 strain) were reared at 27–30 °C with 60–80% relative humidity. A standard photoperiod of 16 h light:8 h dark was utilized. Larvae were raised in open-top plastic bins and fed a diet of TetraMin fish food (Spectum Brands Pet, LLC). Pupae were transferred into closed-top plastic bins. 24 h after emergence, adults were aspirated and placed into a closed-top plastic adult bin. Adults were grouped based on their respective date of emergence to ensure adults were the same age. If adult collection was limited, groups were mixed but never exceeded more than two days in age difference.

Mesocosm experiment

These experiments were described in Gray et al., 202227. In brief, 200 An. gambiae G3 mosquitoes were marked using florescent pigment powder (Shannon Luminous Materials, Inc.) placed into a large (122 cm long x 61 cm wide x 96.5 cm high) plastic bin in which a fake plant was placed in the middle to force the mosquitoes navigate around as they sought water and bloodmeals that were placed on opposite sides of the bin. Plastic dishes with holes cut into them covered the oviposition papers and sugar cubes. These obstacles were used to replicate a ‘wild’ environment for the lab-reared mosquitoes to force flying behaviors. Twice a week, 15 mosquitoes were removed from the mesocosm, and their wings were detached for analysis.

RIMDAMAL II mosquito collection and photo acquisition

During the RIMDAMAL II clinical trial, entomological sampling was conducted to compare entomological indices, such as bioassay survival, entomological inoculation rate, and age structure, between the control and treatment arms of the trial. Mosquitoes were collected from 6 (3 treatment; 3 placebo) clusters (villages or village sectors), 1 and 3 weeks after each mass drug administration using a Prokopac aspirator inside predetermined households. Mosquitoes were then transported to the field station, speciated, and Anopheles mosquitoes were retained. A portion of An. gambiae s.l. mosquitoes were dissected to remove their wings via the same procedure described in Gray et al. Briefly, wings were removed with forceps by grasping the base where it attaches to the thorax and gently pulling them off. Removed wings were gently adhered to a glass slide by placing them on a 2 µL drop of sugar water, allowed to dry, and then photographed using a Leica EZ4 W stereoscope (Leica Microsystems, USA) with the following settings: magnification at 25X; overhead lights off, underlit light on and set at an intensity of ‘5’. Photos were taken with the integrated microscope’s 5 MPixel HD digital camera and using the Leica Imaging software called LAS EZ App using the following settings: automatic exposure = on; auto white balance = on; exposure adjust = on; brightness = 100%; gamma = 0.55; saturation = 70.00; input options: camera automatically selected; capture format: 2.0 MP (1600 × 1200) 4:3; shading = none; sharpening = robust; flip = optional; Line = off. All photos were then electronically transmitted to Colorado State University for pixel intensity (PI) analyses using the Big Picture program (Viden Technologies, LLC).

Digital wing photo analysis

Each mosquito had photos taken of both wings and PI was calculated for each wing using the Big Picture program developed by Viden Technologies, LLC. Big Picture converts the image into greyscale which consists of values from 0 to 255 using OpenCV cv2.cvtColor() function. To account for variations in brightness, contrast, and imaging quality we used histogram normalization by scaling the respective image to the histograms’ maximum and minimum values. Code used for this analysis can be found at https://github.com/BlueHephaestus/malaria-project. Subsequently, the mean of the left- and right-wing PIs were calculated to establish a PI per individual mosquito (a single PI from the mean of both wings). If a mosquito had only one wing image, or if one wing was folded or torn during the slide mounting process, the individual mosquito was given a PI that came from only one of its wings.

Statistical and mathematical methods

Raw PI values from the trial were modeled using a lognormal generalized linear model with a log link using the glmmTMB package in R version 4.5.0. Left- and right-wing PI values were input into the model individually, with a mosquito-level random effect (grouping) term and random effects for household nested within village. Fixed effects of treatment arm (binary variable; IVM arm compared to control), year (binary variable; 2020 compared to 2019), time since MDA (binary variable; 3 weeks-post-MDA vs. 1 week-post-MDA), day of the year (continuous variable; scaled and centered to aid model fitting) and their interaction terms were included. To delineate population PI range differences, mosquitoes were arbitrarily placed into bins with a difference of 0.3 PI. Furthermore, mosquitoes were separated based on the arm location they were captured in to determine differences in PI bins. To establish an age-to-pixel intensity model, mean PIs from mesocosm-reared colonized mosquitoes were first analyzed with a linear model using GraphPad Prism (v10.1.1), and then a suite of non-linear models in the program that fit the biological process of increased PI values over time due to wing scale loss during aging were compared using Akaike’s Information Criterion. The best fit model emerging from this process was four parameter logistic equation. The ages of the wild mosquitoes captured in the RIMDAMAL II trial were then interpolated and extrapolated using this best-fit model.