Population size and dispersal of Anopheles coluzzii on São Tomé and Príncipe Islands based on mark-release-recapture Authors: Sureni Wickramasooriya, Melina Campos, Lisa Chamberland, Robert E Ditter, Maria Júlia M Corrêa, Brett Morgan, Danspaid P Mabuka, Ivan Mulongo Mugeni, Lodney Nazaré, Carla A Sousa, Claire M Egan, Robin Oriango, Aires Januário Fernandes da Moura, João Viegas, João Pinto, Anthony J Cornel, Gregory C Lanzaro, Gabriel Hamer Published: 2026 ABSTRACT AbstractMosquito-borne diseases continue to exact a heavy toll on human health, particularly in sub-Saharan Africa. Yet in many environments, fundamental aspects of mosquito behavior and population dynamics remain poorly characterized. Here, we employ a well-established method for directly measuring mosquito dispersal and estimating population size: mark-release-recapture (MRR). We focused on the key malaria vector Anopheles coluzzii and conducted experiments in São Tomé and Príncipe, an island nation in the Gulf of Guinea under consideration as a site for the first field trial using gene drive mosquitoes for malaria elimination. Understanding mosquito dispersal, population size, and responses to environmental factors is essential for planning such releases. To assess these parameters in An. coluzzii, a total of four MRR experiments were conducted across both São Tomé and Príncipe islands during both wet and dry seasons. Population size estimates were higher during the wet season in both study areas, but seasonal fluctuation was more pronounced in São Tomé. Seasonal patterns of mosquito dispersal differed by location, with greater dispersal in São Tomé during the wet season and in Príncipe during the dry season. Mosquito flight direction was biased toward broad-scale wind direction in São Tomé, but not in Príncipe. Together, these results enhance our understanding of An. coluzzii behavior in island ecosystems and support the design of effective vector control approaches in this biogeographical context.Keywords: malaria, gene drive trial, island, biogeography, mosquito dispersal Introduction Despite extensive control efforts, malaria continues to claim over half a million lives annually, with more than 90% of cases and deaths occurring in Sub-Saharan Africa (World Health Organization 2024). Existing control measures including insecticide-treated bed nets (LLINs), indoor residual spraying (IRS), and artemisinin-based combination therapies are insufficient to achieve malaria elimination (Mao et al. 2023, Mulebeke et al. 2025). Moreover, maintaining current levels of control is becoming increasingly challenging due to insecticide resistance and behavioral adaptations in mosquito populations, such as shifts in feeding and resting patterns that reduce the efficacy of LLINs and IRS (Trape et al. 2011, Reiner et al. 2013, Toé et al. 2014, Sougoufara et al. 2020). Thus, innovative alternative strategies are urgently needed to improve malaria control and reach elimination goals. One promising approach under consideration is the use of genetically engineered mosquitoes (GEM). Advances in GEM technology have led to the development of mosquitoes carrying malaria-refractory genes (Corby-Harris et al. 2010, Isaacs et al. 2012, Inbar et al. 2021, Carballar-Lejarazú et al. 2023), or genes designed to suppress mosquito populations (Windbichler et al. 2008, Galizi et al. 2014, Inbar et al. 2021). These GEMs are equipped with gene drive systems, which enhance the spread of engineered traits by biasing inheritance, allowing for their rapid proliferation within wild populations (Windbichler et al. 2011, Gantz et al. 2015, Bier 2022). This strategy offers a potentially sustainable, cost-effective, and environmentally low-impact solution for the elimination of malaria. However, key questions regarding its efficacy and safety remain, necessitating field trials to gather critical data (Lanzaro and Kormos 2025). Reliable estimates of target mosquito population size and dispersal rates are fundamental to the design of such trials. Mark-release-recapture (MRR) experiments have been integral in estimating mosquito dispersal rates and population sizes (Cianci et al. 2013, Guerra et al. 2014), which are valuable parameters in modeling vector transmission and understanding the ecology of mosquito-borne pathogens (Reiner et al. 2013, Tusting et al. 2014). Accurate dispersal modeling in spatially heterogeneous landscapes is essential for optimizing release strategies and predicting outcomes of GEM deployment. While laboratory studies provide insights into mosquito survival, they fail to account for natural environmental variability, limiting their applicability to real-world dispersal and survivorship estimates. MRR experiments, in contrast, yield empirical data under field conditions, helping to parameterize and validate mathematical models (Benedict et al. 2018). MRRs have been applied to estimate dispersal and population size in several Anopheles species in mainland Africa, including Anopheles gambiae Giles (Touré et al. 1998, Midega et al. 2007, Epopa et al. 2017), Anopheles arabiensis Patton (Kaiser et al. 2021), and Anopheles coluzzii Coetzee & Wilkerson (Yao et al. 2022). Following an extensive review, the University of California Malaria Initiative (UCMI) identified São Tomé and Príncipe (STP) as an ideal location for an ecologically confined field trial of a low-threshold gene drive system (Lanzaro et al. 2021, Carballar-Lejarazú et al. 2023). An island nation in the Gulf of Guinea, STP is located approximately 300 km off the West African coast. Anopheles coluzzii, a member of the An. gambiae species complex, is a major malaria vector in Africa. It is one of only two anopheline species present in São Tomé (alongside Anopheles coustani Laveran) and is the sole anopheline species on Príncipe Island (Pinto et al. 2000, Lanzaro et al. 2021). Anopheles coluzzii is well established on both islands after likely being introduced via human-mediated dispersal approximately 500 yr ago (Ditter et al. 2022). It is widely distributed across STP, with particularly high habitat suitability in the northeastern regions of each island, where human population densities are greatest and elevations are low (Chamberland et al. 2025). The An. coluzzii populations found in STP belong to the FOREST chromosomal form (Pinto et al. 2000, Campos et al. 2021) which typically inhabits rainforest and humid savanna environments, and lacks the common paracentric chromosome inversions found commonly across most of this species range (Coluzzi et al. 1985). Here, we present findings from a series of MRR experiments conducted on both São Tomé and Príncipe islands. These experiments, carried out during both wet and dry seasons, employed well-established methods to estimate population size and dispersal rates. We compare these parameters across the two islands, which differ significantly in land area and human population size. The data generated from this study will be integrated into mathematical models to predict GEM dynamics if deployed in STP. Materials and Methods MRR Locations Study areas on each island were selected based on previous research detailing the distribution of An. coluzzii in STP (Chamberland et al. 2025). MRR experiments were carried out a total of four times: twice on each island between 2022 and 2023, with one release during the dry season and one in the wet season (Table 1). Releases were conducted at a single site on each island: the district of Água Grande on São Tomé Island and the town of Santo António on Príncipe Island. Água Grande is on the northeast coast of São Tomé Island, in an area comprised mostly of densely populated urban habitat (eg homesteads, shops, office buildings, and roads) interspersed with smaller agricultural and forest enclaves (Fig. 1A and B). The area features a gradual topographical incline, with recapture sites near the coastline (sites 9 and 17) situated at elevations between 9 and 12 m above sea level and inland recapture sites (sites 1, 25, and 27) located at elevations of 96, 98, and 118 m above sea level, respectively (Table S1). The marked mosquito release site in São Tomé was in a mixed agricultural and forested area at 0.3444° N and 6.7134° E (Fig. 1B). On Príncipe Island, the marked mosquito release site was a small grassy park in the northeast of the island on the coastal side of Santo António at 1.6401° N and 7.4207° E (Fig. 1C and D). Santo António is the largest town on Príncipe and is surrounded by primary and secondary rainforests and small agricultural fields. The urban center is nestled within a small valley, extending from the coastline to an elevation of approximately 23 m above sea level (site 1). The terrain surrounding the town, except the coastal side, rises steeply, with the most distant mosquito recapture sites reaching elevations of 199, 213, and 245 m above sea level (sites 15, 24, and 12, respectively) (Supplementary Table S1). Table 1.Summary of MRR experiment locations, season, release dates, and post-release recapture days (indicated by a ✓ in shaded cells, while white cells indicate no recaptures conducted)Days post-releaseLocationSeasonRelease date (day 0)123456789101112131415 Água Grande São Tomé Island Dry25th Jul 2022✓✓✓✓✓✓✓✓✓✓Wet23rd Oct 2023✓✓✓✓✓✓✓✓✓✓ Santo António Príncipe Island Dry11th Jul 2023✓✓✓✓✓✓✓✓Wet23rd Mar 2022✓✓✓✓✓✓✓✓ Summary of MRR experiment locations, season, release dates, and post-release recapture days (indicated by a ✓ in shaded cells, while white cells indicate no recaptures conducted) Água Grande Água Grande São Tomé Island São Tomé Island Santo António Santo António Príncipe Island Príncipe Island MRR experiment study areas in São Tomé and Príncipe Islands. Maps displaying layers of agricultural and forested area (white), urban area (light gray), main roads (red lines), and elevation (gray lines, contour interval = 200 m); as well as release sites (green stars) and recapture sites (numbered black dots). A) Map of São Tomé Island, with red box delimiting the study area. B) Recapture locations in and around Água Grande District. C) Príncipe Island, with red box delimiting the study area. D) Recapture sites in and beyond the city of Santo António indicated by numbered black dots. Released Mosquito Rearing In the week preceding each MRR experiment, late instar An. coluzzii larvae were collected using nets and dippers at sites within the designated MRR study area. Anopheles coluzzii larvae were easily distinguishable from the only other Anopheles species present, An. coustani (Gillies and De Meillon 1968). In São Tomé, larvae were reared to adulthood in an insectary maintained at 27 to 28 °C with 80% relative humidity and a 12:12 h light: dark cycle. No insectary facility exists on Príncipe Island, so larvae were reared under ambient conditions in plastic trays placed on a shaded balcony roughly 10 m off the ground. All larvae were fed 6 mg per tray of finely ground koi fish food (Drs. Foster and Smith Staple Diet Quality Koi and Goldfish Food, distributed by Petco Animal Supplies, Inc., USA) twice daily, with up to 300 larvae per tray. Pupae were collected daily, counted and transferred to 45 cm3 Bugdorm cages (MegaView Science Co., LTD, Taichung 407008, Taiwan), each supplied with a 10% sucrose-moistened cotton wick as a constant food source for emerging adults. A total of 1,500 pupae were placed in each cage and the date of pupal collection recorded. In São Tomé, adult cages were maintained in the temperature and humidity-controlled insectary, while in Príncipe, each cage was covered with two moist towels and held in a windowed room at 25 °C. Mosquito Marking and Release At noon on the planned release day, all dead and un-emerged pupae were counted and subtracted from the 1,500 originally placed in each cage. The cage, containing 2- to 5-d-old adults, was then taken outdoors for marking. Mosquitoes were dusted inside the cages using a bulb duster for non-toxic insect powders (Dr. Killigan’s, Redding, California, United States) filled with green, fluorescent DayGlo powder (BioQuip Products Inc., Compton, California, United States). Males and females were marked together in the cages to avoid disturbing and irritating the females by removing males. The sex ratio of emerged adults was assumed to be 1:1 (Pollegioni et al. 2023). In São Tomé, 30 min before release, cages were covered with moist toweling and transported in air-conditioned vehicles to the release site. In Príncipe, the release site was within a 3-min walking distance from the location where mosquitoes were held. In both locations, mosquitoes were released shortly before sunset, between 17:00 and 17:30. The number of marked females released in each experiment is provided in Table 2. Table 2.Number of marked females released, recaptured, total captures (marked and unmarked), filtered captures, farthest recapture distance with non-zero recaptures (in meters), recapture rate, and population size estimationsMarked femalesPopulation size (females + males) estimationIslandSeasonReleasedRecapturesTotal CapturesFiltered CapturesFarthest recapture distance (m)Recapture rate (%)LincolnBaileySeber (Chapman)95% CI (Chapman) São Tomé Dry5,864191,1366451,1900.320352,457335,421335,476194,761 to 476,191Wet6,830336,03756131,5630.4832,323,4422,255,5072,255,8351,512,598 to 2,999,072 Príncipe Dry5,000122532523,2430.240210,000194,616194,65295,470 to 293,836Wet3,750351,1607081,3840.933222,214216,250216,306148,155 to 284,567 Number of marked females released, recaptured, total captures (marked and unmarked), filtered captures, farthest recapture distance with non-zero recaptures (in meters), recapture rate, and population size estimations To assess survivorship of marked versus unmarked mosquitoes, controls were established consisting of 25 marked or 25 unmarked females placed in a waxed cardboard cup covered with netting, with four cups per treatment. Each cup was provided with continuous access to cotton soaked in 10% sucrose solution, covered with moist toweling, and kept in a shaded outdoor location for the duration of each MRR experiment. Mortality in each cup was recorded daily. This protocol for controls was identical for all MRRs. Mosquito Recapture Human landing catches (HLC) were used to collect female An. coluzzii during all MRRs. One week prior to each MRR, paid mosquito collectors were recruited from the study area and attended a town hall meeting where the study design was explained. They were trained in mosquito aspiration techniques and provided with consent forms to sign. Ethical and regulatory approval for the MRRs, including the use of HLC collectors, was obtained from the São Tomé and Príncipe Ethical Commission for Health and Scientific Research (Protocol 020-2022). Each collector was equipped with a headlamp, an aspirator, and a paper cup covered with netting. Collectors were assigned to specific recapture sites, where they conducted HLCs for the entire duration of the MRR. 26 collectors participated in Príncipe and 31 participated in São Tomé. HLC sites were selected using a pre-determined radial pattern, with collection points positioned between 236 and 5,349 m away from the mosquito release point (Supplementary Table S1). Homeowners’ permission was obtained for collectors to operate on their property overnight. When permission was not granted at pre-selected locations, the nearest alternative residence was chosen. If a selected site was a business office or other non-residential property, the collection point was relocated to a nearby home with granted permission. The finalized recapture site locations are shown in Fig. 1. Due to the narrower urban layout of Santo António, some recapture sites were placed at houses beyond the town’s limits. This design allowed assessment of mosquito dispersal into surrounding areas, requiring mosquitoes to traverse forested spaces and steep terrain to reach surrounding villages and homes. HLC collections took place from 21:00 to 02:00 over five consecutive nights, beginning approximately 28 h after mosquito release (Day 1). Subsequent collections occurred every second night up to Day 11 for the Príncipe MRRs and Day 15 for the São Tomé MRRs (Table 1). At 02:00, collected mosquitoes were provided with cotton soaked in 10% sucrose and stored in cooler boxes until further processing. Later, on the same day, mosquitoes were euthanized by freezing at −20 °C when a freezer was available, or by exposure to a cotton ball soaked in 91% ethanol placed on top of the container and enclosed in a plastic bag. An. coluzzii individuals were then identified. Marked and unmarked individuals were counted using a blacklight attached to a dissecting microscope and marked mosquitoes were verified by three senior investigators. Data Analysis and Visualization Spatial maps illustrating recapture distributions and dispersal patterns were generated using QGIS v3.42 (QGIS.org 2024), an open-source Geographic Information System software. To visualize recapture trends, the proportion of recaptured mosquitoes was plotted against distance from the release site using ggplot2 in R (Wickham 2011). Survival Probability The survival probability of marked and unmarked females placed in cups as controls for the duration of each MRR were visualized using Kaplan–Meier survival curves (Kaplan and Meier 1958) and compared using log-rank tests. Population Size The population size of An. coluzzii was estimated using three related MRR estimators to assess the robustness of population size estimates under the same closed-population framework. Applying multiple estimators that share the same biological assumptions but differ in their statistical bias corrections is common practice in MRR analyses, as it allows evaluation of the sensitivity of population size estimates to estimator choice and sampling variability. Similar approaches have been used in mosquito MRR studies, where population size has been estimated using several indices (Epopa et al. 2017, Kaiser et al. 2021). In this study, we applied the deterministic Lincoln–Petersen index (Petersen 1896, Lincoln 1930) and its bias-corrected Bailey modification (Bailey 1951), both of which assume a closed population between sampling occasions (ie negligible birth, death, immigration, or emigration), equal probability of capture among individuals, and that marks are not lost and are correctly identified upon recapture. To reduce bias associated with small sample sizes, we also used the Seber–Chapman estimator (Chapman 1951), which provides an approximately unbiased estimate of population size under the same closed-population assumptions. Furthermore, to quantify the plausible range of population size and assess robustness, we derived 95% confidence intervals from Seber’s sampling variance (Seber 1973), computing CIs as N^±1.96Var⁡(N^) where N^ is the estimated population size and Var⁡(N^) follows Seber’s formulation. Because recapture sampling occurred over several days to increase recovery rates, the strict assumption of population closure may not have been fully satisfied; therefore, population size estimates derived from these estimators should be interpreted as approximate estimates. Recapture sites with zero catches or those located far from release points can introduce bias and reduce the accuracy of population size estimates (Cianci et al. 2013, FAO/IAEA 2023). To define an effective sampling domain, we first calculated the distance between the mosquito release point and each recapture location using the haversine function from the R package geosphere (Hijmans et al. 2024) in R version 4.3.1 (R Core Team 2023). We then filtered the recapture counts by identifying the farthest station from the release point at which at least one marked mosquito was recaptured. This distance was used to define the outer boundary of the effective sampling area, corresponding to 1,190 m (São Tomé dry), 1,563 m (São Tomé wet), 3,243 m (Príncipe dry), and 1,384 m (Príncipe wet). Stations located beyond this distance were excluded from the analysis because they were considered outside the effective release influence zone. All stations located within this boundary were retained in the analysis, including those with zero recaptures, to ensure that sampling effort was fully represented and that capture probabilities were not artificially inflated. The total number of recaptures recorded within this filtered recapture zone was then used for subsequent population size analyses. Estimates were calculated on females only (all recaptured individuals were female). Assuming a 1:1 sex ratio in the field population, total population size was obtained by doubling the female-based estimate. Mean Distance Traveled per Day (MDTpD) To accurately estimate mean dispersal rate in a multi-day MRR study, it is necessary to incorporate both the timing of recaptures and a standardized time frame into the analysis. We generally followed the approach of Morris et al. (1991), which employed an annulus-based method that calculates the mean distance traveled (MDT) of mosquitoes by multiplying an expected recapture (ER) count by the distance from the release point, assuming a single-day sampling frame. However, applying this directly to multi-day studies can bias estimates, as it does not account for the variation in time mosquitoes had to disperse. To address this, we adapted the method by dividing the recapture distance by the number of days since release for each individual, yielding a standardized dispersal rate (distance/day). Averaged together over all individuals, we refer to this as the Mean Distance Traveled per Day (MDTpD). MDTpD was calculated by delineating contiguous concentric annuli centered on the mosquito release site, defined in 100-m distance classes (0 to 100 m, 100 to 200 m, 200 to 300 m, etc.). Annuli were extended outward to the farthest recapture location with at least one marked mosquito, defining the outer boundary of the analysis. The outermost annulus (final annulus included in the analysis) ranged from 1,100 to 1,200 m (dry) and 1,500 to 1,600 m (wet) in São Tomé, and 3,200 to 3,300 m (dry) and 1,300 to 1,400 m (wet) in Príncipe. This approach accounts for the temporal component of movement, allowing more accurate comparisons across recapture events. Furthermore, MDTpD calculations were restricted to the first 5 d post-release to standardize the observation period across MRRs and reduce bias introduced by long-distance dispersers that may have exited the recapture area over time. This modification should improve both the precision and comparability of dispersal estimates while retaining sufficient data for robust statistical inference. MDTpD was calculated by first delineating contiguous concentric annuli, each 100 m wide, centered on the mosquito release site. Stations were assigned to annuli based on distances between the release point and each recapture location, calculated using the haversine function from the R package geosphere (Hijmans et al. 2024), which computes great-circle distances based on geographic coordinates. The correction factor (CF) for each annulus was calculated following Morris et al. (1991), by multiplying the area of the annulus by the total number of recapture sites in the study and then dividing the result by the total study area (Equation 1). The expected recapture (ER) count for each annulus was then calculated as the observed number of recaptures multiplied by the CF, and divided by the number of recapture sites within that annulus (Morris et al. 1991). The result was then rounded up to the nearest whole number to represent the number of estimated recaptured mosquitoes (Equation 2). We assumed that each of these estimated mosquitoes, treated as pseudo-individuals, traveled the median distance of their corresponding annulus, divided by the number of days since release. This value represents the distance traveled per day by each estimated mosquito (Equation 3). The individual-level dispersal estimates across annuli were then combined, and the mean distance traveled per day (MDTpD) was calculated by averaging the daily travel distances across all estimated mosquitoes (Equation 4). MDTpD was computed independently for each of the four MRR experiments. Wilcoxon rank-sum tests (McKnight and Najab 2010) were subsequently performed using R version 4.3.1 (R Core Team 2023) to assess whether there were statistically significant differences in mosquito dispersal between seasons and between islands. Correction factor for annulus i Estimated recaptures of ith annulus Distance traveled per day by each estimated mosquito k = 1, 2, …, ERi are pseudo-individuals from annulus i Mean Distance Traveled per Day Where: i – Annulus number Ai – Area of annulus i ATotal – Total area covered by all annuli T – Total number of recapture sites in the entire study area ni – Number of recapture sites in annulus i Ri – Number of observed recaptures in annulus i Di – Median distance from the release point to the center of annulus i dj – Days since release for mosquito j N – Number of estimated mosquitoes across all annuli Flight Range (FR) Flight range (FR) was estimated using a linear regression approach following Muir and Kay (1998), and FAO/IAEA 2023. For each annulus, the median distance from the release point (Di) was calculated and log10-transformed. A linear regression was then fitted between the cumulative percentage of recaptured mosquitoes and these transformed distances. FR50 and FR90 represent the distances corresponding to 50% and 90% cumulative recaptures, respectively, and were derived from the fitted regression model. Because the number of recaptured mosquitoes on individual sampling days was low, all recapture data were pooled across the entire sampling period to obtain more robust estimates rather than calculating flight range on a day-by-day basis. Net Dispersal (ND) In addition to annulus- based metrics, net dispersal was calculated to quantify mosquito movements from the release point using the exact distance to each recapture location (Epopa et al. 2017). Because the number of marked mosquitoes recaptured on individual sampling days were relatively low, net dispersal was summarized as a composite value for each MRR experiment rather than as daily estimates. Composite net dispersal was calculated as the recapture weighted mean distance between the release point and all recapture stations pooled across the entire sampling period, as follows: where di is the distance from the release site to recapture site i, and ri is the total number of marked mosquitoes recaptured at site i. Wind Data and Analysis To assess the influence of wind on mosquito movement, we analyzed the directions of mosquito recaptures relative to station positions and prevailing wind during each MRR experiment using circular statistics. Circular statistics are methods for analyzing directional data that account for the cyclical nature of variables where 0° and 360° represent the same value. All circular analyses were conducted in R version 4.5.0 (R Core Team 2025) using the circular package (Agostinelli and Lund 2024). Each recapture was assigned a compass bearing from the release point, representing the direction of the mosquito’s movement. Hourly wind data for each island were obtained from the ERA5 hourly time-series dataset (Hersbach et al. 2020), which represents wind at 10 m above the Earth’s surface. Wind direction for each day was summarized as the net displacement vector over a 24-h period. Days 0 and 1 were combined due to the 17:00 release on day 0 and lack of recaptures that night. The average wind direction for each MRR was then calculated by taking the circular mean of all daily wind directions using the mean function in R. Because recapture stations were not uniformly distributed around the release site, we first tested whether mosquito movement differed directionally from the spatial arrangement of stations. Watson’s two-sample test of homogeneity was used to compare the angles of all stations relative to the release point with that of mosquito recaptures. This assessed whether mosquitoes’ dispersal direction was non-random in relation to the recapture site layout. To evaluate alignment between mosquito movement and wind direction, we used Rayleigh’s V-test (Jammalamadaka and Ashis 2001), which tests whether movement directions are non-randomly clustered around a specified angle—in this case, the mean wind direction during each MRR. This test allowed us to assess whether directional bias in mosquito dispersal was significantly wind-aligned. Results Survival of Marked and Unmarked Controls Kaplan–Meier survival estimates showed no significant difference in survival between marked (treatment) and unmarked (control) female mosquitoes across all MRRs (Fig. 2). In Príncipe during the wet season, survival did not differ between the two groups (log-rank test: χ2 = 0.01, df = 1, P = 0.93; Fig. 2A), nor was a significant difference observed in São Tomé (log-rank test: χ2 = 1.97, df = 1, P = 0.16; Fig. 2C). Similarly, during the dry season there was no statistically significant difference in survival probability between groups in either Príncipe (log-rank test: χ2 = 0.08, df = 1, P = 0.77; Fig. 2B) or São Tomé (log-rank test: χ2 = 0.10, df = 1, P = 0.75; Fig. 2D). Survival estimates of marked and unmarked controls. Kaplan–Meier survival comparison between marked and unmarked female mosquitoes during MRR trials in São Tomé and Príncipe Islands during wet and dry seasons. Colors indicate treatment groups: blue (Marked Wet), orange (Marked Dry), and green (Unmarked). The gray shading shows the 95% confidence intervals around each survival curve. The dashed horizontal line indicates a reference survival probability threshold (0.5 in most panels and 0.75 in panel C) used to aid visual comparison of survival across treatments. Log-rank tests indicated no significant differences in survival between marked and unmarked groups in either island or season (Príncipe wet: χ2 = 0.01, df = 1, P = 0.93; Príncipe dry: χ2 = 0.08, df = 1, P = 0.77; São Tomé wet: χ2 = 1.97, df = 1, P = 0.16; São Tomé dry: χ2 = 0.10, df = 1, P = 0.75). In Príncipe during the wet season, both marked and unmarked groups fell below 0.5 survival around 10 to 11 d. In São Tomé during the wet season, the decline to below 0.5 occurred earlier, around day 7 to 8. During the dry season, median survival remained above 0.5 until approximately days 10 to 11 in both Príncipe and São Tomé. Releases, Collections and Population Size Estimates The total number of released An. coluzzii females for each MRR experiment ranged from 3,750 to 6,830 (Table 2). In both islands, female HLC numbers were higher during the wet season than in the dry season (Table 2). Marked recaptures were generally concentrated near release points, with some dispersal to surrounding areas, as shown in Figs. 3 (São Tomé) and 4 (Príncipe). In São Tomé, a total of 6,037 An. coluzzii females were collected during the wet season, of which 33 were marked (recaptured). In contrast, 1,136 females were collected during the dry season, with 19 marked individuals recaptured (Table 2). In Príncipe, 1,160 total including 35 marked females were collected in the wet season MRR, compared to 253 total and 12 marked in the dry season. Spatial distribution and flight distances for MRR experiments in São Tomé. Maps displaying layers of agricultural and forested area (white), urban area (light gray), main roads (red lines), release sites (green stars), and proportional abundance of recaptured (orange circles) and unmarked (magenta circles) female mosquitoes during the wet season A) and dry season B). C) Cumulative proportion of marked recaptures by ranked distance from the release site (in meters). Boxplots display the mean, minimum, maximum, and standard errors for D) flight distance from release site and E) daily flight distance from release site of marked females. Seasonal differences were tested using a Welch two-sample t-test (D: t =−2.12, df = 18.98, P = 0.04; E: t = −0.15, df = 31.06, P = 0.88). Recapture counts were filtered to exclude sites located beyond the farthest site with at least one marked recapture, and the remaining counts were used to estimate population size. Population size estimates based on the Seber (Chapman) model varied between islands and seasons (Table 2). In São Tomé, the estimated population was approximately 3.35 × 105 individuals during the dry season and 2.26 × 106 during the wet season, a 6.72-fold difference. In Príncipe, estimates were 1.95 × 105 (dry) and 2.16 × 105 (wet), a 1.11-fold difference. Recapture rates were higher in the wet season on both islands (São Tomé: 0.48% wet vs. 0.32% dry; Príncipe: 0.93% wet vs. 0.24% dry). Flight Distances of Marked Females Release and recapture sites, land cover, and relative proportions of marked and unmarked females collected over the study area in São Tomé are illustrated in Fig. 3A (wet season) and B (dry season). Flight distance patterns of recaptured An. coluzzii females were similar across seasons in São Tomé (Fig. 3C). Half of the recaptured females were found within approximately 500 m of the release site during the dry season and 750 m during the wet season. The maximum recorded flight distances were 1,563 m (site 10) in the wet season and 1,190 m (site 29) in the dry season. There was a significant difference in the mean flight distances from the release site between the two seasons (Fig. 3D), but there was no significant difference in the mean daily flight distances (Fig. 3E). Maps displaying collection sites and land cover type of the study area in and around Santo António on Príncipe Island are illustrated in Fig. 4A and B. Flight distances varied between seasons in Príncipe, with significantly greater dispersal observed in the dry season (Fig. 4C and D). Half of the recaptured females were collected within 1,250 m of the release site during the dry season, compared to 250 m in the wet season (Fig. 4C). The maximum recapture distance was 3,237 m (site 18) in the dry season and 1,384 m (site 8) in the wet season. Notably, site 18, where the farthest recapture occurred, is situated at 185 m above sea level (Supplementary Table S1). The average flight distance per day was 1,258 m in the dry season, compared to 963 m in the wet season. However, this difference was not statistically significant (Fig. 4E). Spatial distribution and flight distances for MRR experiments in Príncipe. Maps displaying layers of agricultural and forested area (white), urban area (light gray), main roads (red lines), release sites (green stars), and proportional abundance of recaptured (orange circles) and unmarked (magenta circles) female mosquitoes during the wet season A) and dry season B). C) Cumulative proportion of marked recaptures by ranked distance from the release site (in meters). Boxplots display the mean, minimum, maximum, and standard errors for D) flight distance from release site and E) daily flight distance from release site of marked females. Seasonal differences were tested using a Welch two-sample t-test (D: t = 2.16, df = 12.10, P = 0.05; E: t = 1.28, df = 18.02, P = 0.21). Mean Distance Traveled per Day (MDTpD) The distribution of distances traveled per day by season and study site are illustrated in Fig. 5. Wilcoxon rank-sum tests indicated no significant difference in dispersal between islands during either season (dry: W = 31, n1 = 5, n2 = 5, P = 0.22; wet: W = 25, n1 = 5, n2 = 5, P = 0.31; Fig. 5). Likewise, no significant seasonal differences were observed within each island (São Tomé: W = 26.5, n1 = 5, n2 = 5, P = 0.68; Príncipe: W = 33.5, n1 = 5, n2 = 5, P = 0.23). The overall MDTpD in Príncipe was 695 m in the dry season and 486 m in the wet season. In São Tomé, MDTpD was 559 m in the dry season and 623 m in the wet season. Boxplots showing the distribution of distance traveled per day (DTpD) by marked mosquitoes across islands and seasons, with cross-island comparisons shown within each season. The white diamond in each boxplot indicates the mean (MDTpD), while the thick black line represents the median of DTpD. Wilcoxon rank-sum test p-values for these comparisons are displayed above the boxplots. Net Dispersal (ND) Table 3 summarizes daily net dispersal (ND) across islands and seasons. ND was calculated for each island and season using all recaptures pooled across the sampling period. In São Tomé, dispersal distances were generally higher during the dry season (1,459.84 m), suggesting greater mosquito movement during the dry period at this site. Table 3.Flight range (FR, in meters) of mosquitoes by island and season based on pooled recapture data across the entire sampling period. The table also includes the parameters of the linear regression model used to estimate flight range (slope, intercept, and R2). The last column reports the composite net dispersal distance (ND, in meters) Flight range Net dispersal (ND) FR50 FR90 SlopeIntercept R 2 São Tomé Dry 464.95902.13 0.0072 2.3076 0.8519 539.39 Wet 721.471208.30 0.0056 2.5783 0.8057 686.80 Príncipe Dry 1128.152605.820.0091 2.5979 0.90141459.84 Wet 399.481241.97 0.0123 1.9857 0.7823 618.39 Flight range (FR, in meters) of mosquitoes by island and season based on pooled recapture data across the entire sampling period. The table also includes the parameters of the linear regression model used to estimate flight range (slope, intercept, and R2). The last column reports the composite net dispersal distance (ND, in meters) Flight Range (FR50/FR90) Distances within which 50% (FR50) and 90% (FR90) of marked individuals were recaptured, estimated from pooled recapture data across the entire sampling period, are summarized in Table 3. In Príncipe, flight ranges during the dry season were substantially greater than during the wet season. The estimated FR50 and FR90 were 1,128 and 2,606 m in the dry season, compared with 399 m and 1,242 m in the wet season. On São Tomé Island, flight ranges were more moderate overall. During the wet season, FR50 and FR90 were estimated at 721 and 1,208 m, respectively, whereas the dry season showed lower values of 465 m for FR50 and 902 m for FR90. The regression models showed good fits for all experiments (R2 = 0.78 to 0.90), indicating a strong relationship between cumulative recapture percentage and log10-transformed distance. Influence of Wind Wind direction, as recorded in the ERA5 dataset, was generally consistent across all MRR trials, with prevailing displacement from north to northeast (Supplementary Fig. S1). Wind displacement was notably lower in the Príncipe wet season MRR compared to the other trials. Mosquito dispersal directions varied among the four MRR experiments (Fig. 6). In São Tomé, mosquito movement was strongly directional during both the wet and dry seasons (Watson’s test P < 0.01, <0.001 respectively) and aligned with prevailing wind (Rayleigh’s V-test P = 0.0005, <0.0001 respectively). In contrast, for Príncipe, the wet and dry season MRRs showed no significant directional structure (Watson’s test P > 0.05, >0.1 respectively) or alignment with wind direction (Rayleigh’s V-test P = >0.99, 0.98 respectively). Full test statistics are available in Supplementary Table S2. Directional patterns of mosquito movement relative to wind during dry and wet seasons on São Tomé and Príncipe islands. Circular plots show average wind direction (dark blue arrow), daily wind directions (light blue arrows), and average mosquito movement direction (green arrow) based on mark-release-recapture (MRR) data. Each black dot represents a recapture site, with green circles indicating locations where marked mosquitoes were recaptured. The size of the green circle corresponds to the number of mosquitoes recaptured at that location. Discussion Mark-release-recapture (MRR) experiments are widely used to estimate mosquito population size and to assess mosquito dispersal. In this study, we estimated An. coluzzii population size, comparing wet and dry season populations on the islands of São Tomé and Príncipe. We also assessed the dispersal characteristics of An. coluzzii, including seasonal differences in distance traveled and the relationship between dispersal direction and prevailing wind patterns. Population Size Estimates Population size estimates derived from the MRR data were higher during the wet season at both sites (Table 2). The most striking difference was observed on São Tomé, where the wet-season population was approximately 572% higher than the dry-season estimate, indicating a strong seasonal increase during the much wetter October compared with August. This is consistent with many regions of mainland sub-Saharan Africa where seasonal shifts are often extreme and malaria vector populations decline markedly during dry periods (Taylor et al. 1993, Oduola et al. 2016, Ould Lemrabott et al. 2018, Epopa et al. 2020). By contrast, in Santo António, Príncipe, the seasonal change was small (∼11%), with overlapping confidence intervals. Dispersal Estimates Our MRR experiments revealed contrasting seasonal patterns of An. coluzzii dispersal between the two islands: dispersal was greater during the wet season on São Tomé Island, whereas on Príncipe it was more extensive during the dry season. Despite these differences, the mean distance traveled per day was consistent across islands and seasons, suggesting that daily dispersal rates remain broadly stable. Composite net dispersal estimates were consistent with flight range patterns, showing higher dispersal during the wet season on São Tomé and substantially greater dispersal during the dry season on Príncipe. Because composite net dispersal represents a recapture-weighted mean distance calculated from all recaptures pooled across the study period, occasional long-distance recaptures may disproportionately influence the overall estimate. Flight range estimates showed similar seasonal contrasts between islands, with greater flight ranges during the dry season in Príncipe and during the wet season in São Tomé. Together, these results indicate that seasonal conditions influence the spatial extent of Anopheles coluzzii movement. Ecological factors may help explain these patterns. The availability of blood hosts likely remains unchanged between wet and dry seasons (Sousa et al. 2001), but lower humidity (drier vegetation) and less standing water may force longer flight distances for female mosquitoes as they disperse to locate resting and oviposition sites (Abiodun et al. 2016). Houses in Príncipe are not built on stilts above ground whereas many houses in São Tomé (Água Grande) are (Charlwood et al. 2003), providing ample year-round shaded and moist resting sites for Anopheles mosquitoes, which may have contributed to the smaller dispersal difference between the wet and dry seasons there. Unlike many other Anopheles species and sub-specific forms, the “forest form” of An. coluzzii typically inhabits areas where more humid, and often forested, environments overlap with human settlements (Bayoh et al. 2001). In this study, results suggest that An. coluzzii is capable of dispersing between forested habitats. This observation is supported by population genomic analyses describing An. coluzzii populations on both islands as being panmictic (Campos et al. 2024). It is plausible that roads, which create corridors between otherwise fragmented habitats, may play a crucial role in facilitating mosquito movement, especially in sparsely populated areas. Our observations from STP and previous publications (Chamberland et al. 2025) suggest that roads can act as corridors for mosquito dispersal, connecting habitats and promoting gene flow. Future research should aim to identify specific dispersal routes, particularly by analyzing the role of human settlements and infrastructure, such as roads, in facilitating or constraining mosquito movement. Influence of Wind Our results show that wind may play a role in shaping An. coluzzii dispersal direction, particularly in the São Tomé trials, where mosquito movement was significantly directional and aligned with prevailing wind (Fig. 6). This is consistent with previous research which found that wind patterns can explain mosquito dispersal (Endo and Eltahir 2018), however it has been found that Anopheles can actively fly upwind to locate hosts (Midega et al. 2012) rather than downwind as observed in São Tomé. In contrast, the Príncipe experiments revealed no significant alignment between mosquito movement and wind direction, nor was any bias detected relative to the layout of recapture stations. Spatial heterogeneity in the influence of wind direction on mosquito dispersal may be explained by local topography. On Príncipe, the release site is located in a valley surrounded by higher terrain, which likely dampens wind speeds and creates more variable wind conditions across the recapture area. Because ERA5 represents broad-scale atmospheric patterns, it may not capture the fine-scale, near-surface wind dynamics specific to this site. So, even if mosquitoes on Príncipe respond to wind, the conditions they experience may differ markedly from the ERA5-derived mean directions. Additionally, the lower wind speeds during Príncipe wet season MRR, and low recapture sample size in the Principe dry season MRR, could further explain the lack of measurable effects of wind on mosquito dispersal during those experiments. The use of both Watson’s two-sample test and Rayleigh’s V-test provided complementary insights into movement behavior. Watson’s test allowed us to rule out inherent spatial biases in the station layout as drivers of directional movement. In São Tomé trials, the significant Watson test results indicate that mosquitoes moved non-randomly even relative to the station layout—supporting the inference that wind alignment was a meaningful signal. In the Príncipe trials, no significant directional bias was found. Together, these findings support the conclusion that mosquito dispersal may be wind-influenced but that this effect is context-dependent, perhaps modulated by local conditions such as terrain. Study Limitations The mosquitoes used in all release trials described here were sourced from larval collections at or near the release points. Artificial rearing conditions can influence multiple life history traits, such as survival and fecundity (Yan et al. 2021) and flight performance (Kaufmann et al. 2013) of mosquitoes. In this study, the larvae were reared out of direct sun exposure, but were reared in the water they were originally collected in to maintain a similar biotic environment. The larvae were, however, provided with a diet of finely ground fish food that they would not obtain in nature. In addition, population size estimates in this study were derived using closed-population mark release recapture models, which assume negligible birth, death, immigration, and emigration during the sampling period. Because the sampling period extended over multiple days, this assumption may not have been fully met, and therefore population size estimates should be interpreted as approximate values. Dispersal and population abundance calculations derived from the low recapture rates of marked females in all MRRs in this study (< 1%) should be viewed with some caution. Low recapture rates of Afrotropical Anopheles mosquitoes in MRR studies are unfortunately common (Touré et al. 1998, Guerra et al. 2014) and may reduce confidence in these estimates. Ideally, fine-scale measurements of near-ground wind speed and direction would be available across the MRR study areas throughout the study periods. In their absence, we used a broad-scale dataset averaged over the study periods. While local wind dynamics are likely more complex than represented here, we nevertheless detected significant directional bias in mosquito movement aligning with wind direction. Due to logistical constraints, the total duration of MRR experiments varied by island (Table 1). However, several aspects of our analytical design should mitigate potential biases associated with this variation. Population size estimates excluded sites located beyond the furthest distance at which marked mosquitoes were recaptured, thereby limiting the analysis to stations within the effective dispersal range of released mosquitoes. Mean distance traveled per day was calculated using recaptures from the first 5 d only. Flight range was estimated using a linear regression-based approach, which relies on cumulative recapture patterns rather than daily counts and is expected to be robust to differences in the total number of recapture days. Although the wind direction analysis could theoretically be biased toward detecting stronger wind effects in longer studies, this is unlikely to have influenced our conclusions: in São Tomé, few recaptures occurred after day 11, suggesting that extending the experiment in Príncipe would not have substantially altered the observed relationships. Implications and Future Directions This study provides valuable insights into the dispersal behavior of An. coluzzii across varying seasonal and environmental conditions in STP. Our findings suggest that resting site and oviposition site seeking are probable drivers of dispersal patterns. Additionally, our results demonstrate that wind may influence mosquito dispersal direction, but this effect is context-dependent and potentially shaped by local environmental conditions. Moving forward, further research into the roles of human infrastructure (eg roads), vegetation, topography, and fine scale breeding site availability will be essential for understanding the full range of factors influencing mosquito dispersal. This knowledge is crucial for designing effective release strategies of genetically engineered mosquitoes across time and space. Supplementary Material Contributor Information Sureni Wickramasooriya, Vector Genetics Laboratory, Department of Pathology, Microbiology, and Immunology, University of California, Davis, CA, USA. Melina Campos, Vector Genetics Laboratory, Department of Pathology, Microbiology, and Immunology, University of California, Davis, CA, USA. Lisa Chamberland, Vector Genetics Laboratory, Department of Pathology, Microbiology, and Immunology, University of California, Davis, CA, USA. Robert E Ditter, Vector Genetics Laboratory, Department of Pathology, Microbiology, and Immunology, University of California, Davis, CA, USA; Department of Biological Sciences, State University of New York at Buffalo, Buffalo, NY, USA. Maria Júlia M. Corrêa, Vector Genetics Laboratory, Department of Pathology, Microbiology, and Immunology, University of California, Davis, CA, USA. Brett Morgan, Vector Genetics Laboratory, Department of Pathology, Microbiology, and Immunology, University of California, Davis, CA, USA. Danspaid P Mabuka, Vector Genetics Laboratory, Department of Pathology, Microbiology, and Immunology, University of California, Davis, CA, USA. Ivan Mulongo Mugeni, Vector Genetics Laboratory, Department of Pathology, Microbiology, and Immunology, University of California, Davis, CA, USA. Lodney Nazaré, Vector Genetics Laboratory, Department of Pathology, Microbiology, and Immunology, University of California, Davis, CA, USA. Carla A Sousa, Global Health and Tropical Medicine, LA-REAL, Instituto de Higiene e Medicina Tropical, Universidade Nova de Lisboa, Lisbon, Portugal. Claire M Egan, Vector Genetics Laboratory, Department of Pathology, Microbiology, and Immunology, University of California, Davis, CA, USA. Robin Oriango, Vector Genetics Laboratory, Department of Pathology, Microbiology, and Immunology, University of California, Davis, CA, USA. Aires Januário Fernandes da Moura, Global Health and Tropical Medicine, LA-REAL, Instituto de Higiene e Medicina Tropical, Universidade Nova de Lisboa, Lisbon, Portugal; Unidade de Ciencias da Natureza da Vida e do Ambiente, Universidade Jean Piaget de Cabo Verde, Praia, Cape Verde. João Viegas, Programa Nacional de Eliminação do Paludismo, Centro Nacional de Endemias, São Tomé, São Tomé and Príncipe. João Pinto, Vector Genetics Laboratory, Department of Pathology, Microbiology, and Immunology, University of California, Davis, CA, USA; Global Health and Tropical Medicine, LA-REAL, Instituto de Higiene e Medicina Tropical, Universidade Nova de Lisboa, Lisbon, Portugal. Anthony J Cornel, Vector Genetics Laboratory, Department of Pathology, Microbiology, and Immunology, University of California, Davis, CA, USA; Mosquito Control Research Laboratory, Department of Entomology and Nematology, University of California, Parlier, CA, USA. Gregory C Lanzaro, Vector Genetics Laboratory, Department of Pathology, Microbiology, and Immunology, University of California, Davis, CA, USA. Author Contributions Gregory C Lanzaro (Conceptualization [lead], Data curation [equal], Funding acquisition [lead], Investigation [equal], Methodology [equal], Project administration [lead], Supervision [lead], Writing—original draft [equal], Writing—review & editing [equal]), Sureni Wickramasooriya (Formal analysis [equal], Writing—original draft [lead], Writing—review & editing [equal]), Melina Campos (Formal analysis [equal], Writing—original draft [lead], Writing—review & editing [equal]), Lisa Chamberland (Formal analysis [equal], Investigation [equal], Writing—original draft [equal], Writing—review & editing [equal]), Robert E Ditter (Formal analysis [equal], Investigation [equal], Writing—review & editing [equal]), Maria Júlia M. Corrêa (Data curation [equal], Investigation [equal], Writing—review & editing [equal]), Brett Morgan (Formal analysis [equal], Writing—original draft [equal], Writing—review & editing [equal]), Danspaid P Mabuka (Investigation [equal], Writing—review & editing [supporting]), Ivan M Mugeni (Investigation [equal], Writing—review & editing [supporting]), Lodney Nazaré (Investigation [equal], Writing—review & editing [supporting]), Carla Alexandra Gama Carrilho da Costa Sousa (Data curation [equal], Investigation [equal], Writing—review & editing [supporting]), Claire M Egan (Data curation [equal], Investigation [equal], Writing—review & editing [supporting]), Robin Oriango (Investigation [equal], Writing—review & editing [supporting]), Aires J Fernandes Da Moura (Investigation [equal], Writing—review & editing [supporting]), João Vegas (Investigation [equal], Writing—review & editing [supporting]), Joao Pinto (Conceptualization [equal], Investigation [equal], Methodology [equal], Writing—review & editing [equal]), and Anthony Cornel (Conceptualization [equal], Data curation [equal], Investigation [equal], Methodology [equal], Writing—original draft [equal], Writing—review & editing [equal]) Supplementary Material Supplementary material is available at Journal of Medical Entomology online. Funding This study was implemented in the framework of the University of California Malaria Initiative (http://stopmalaria.org) with funding from the UC Irvine Malaria Initiative Program (A17-0209) and Coefficient Giving (https://coefficientgiving.org/) (A22-2768-001). Conflicts of Interest None declared. References Associated Data Supplementary Materials