Species abundance, composition, and nocturnal activity of female Anopheles (Diptera: Culicidae) in malaria-endemic villages of Papua New Guinea: assessment with barrier screen sampling

This study was conducted in five rural villages (Dimer, Kokofine, Matukar, Mirap and Wasab) in Madang province of PNG (Fig. 1) from 2012 to 2016. These villages are located in three ecogeographic environments of the malaria-endemic region of northern PNG (coastal plain, hilly inland terrain, and inland alluvial plain) and are inhabited by various Anopheles species, including the seven vector species listed above [13, 16‐18, 23, 29‐32]. The villages Matukar and Mirap share similar features; both are situated on coastal plain along the northern coastline just above sea level. Land cover consists of coconut plantation, secondary forest, vegetable gardens, brackish swamps, houses, foot trails, and exposed soil. The villages border sand beach and shore of the Pacific Ocean. Wasab and Dimer villages are situated several km inland from this coastline, on elevated hilltops about 150 m above sea level, with land cover and topography of steep-sided, forested hills and numerous streams flowing to rivers in nearby valleys. The fifth village, Kokofine, is situated on the alluvial plain of the Ramu river, 39 km from the coast and 400 m above sea level. Land cover there consists primarily of swamps, cocoa plantations and secondary forests. All villages have the same tropical climate condition of hot and wet with average atmospheric temperature of 28 °C.

The structure and set up of barrier screens were similar to those described elsewhere [26‐28]. In this study, each barrier screen consisted of a 20 m long and 2 m wide polyethylene shade cloth (forest green, 70% shading grade) fastened on wooden poles or metal reinforcement bars and erected vertically to a height of 2.15 m. The length of the barrier screen was chosen for efficiency of setting up the screen and the amount of time spent searching the barrier screen surface for resting mosquitoes. The height of the barrier screen was consistent with the ability of the collectors to reach for mosquitoes just above their average height (160 cm). The barrier screens were placed between the village perimeter and surrounding environment or bush. Each barrier screen was positioned with one side of the screen facing the village, hereafter referred to as “village side”, and the other facing away from the village, hereafter referred to as “bush side”. The barrier screen intercepts host-seeking and blood-fed mosquitoes as they commute into and out of the village.

Mosquitoes were sampled using two or ten barrier screens per village per night (see Additional file 1: Table S1 for the number of barrier screens deployed in a particular night in each village). Each barrier screen was assigned two trained mosquito collectors. One collector worked from 6 pm to midnight and was replaced by the second collector who worked from midnight to 6 am. The collectors, who were stationed 20 m from the screen, visited both sides of the screen three times within each hour with approximately 20 min sampling interval, which involved 5 min commuting between the barrier screen and collector station, another 5 min searching and aspirating resting mosquitoes on the barrier screen, and 10 min break before next visit to the barrier screen. As this study was among the first to test the BSS method, no prior information was available to guide the sampling strategy for this study. Mosquitoes were believed to rest only temporarily on the barrier screen, thus searching the barrier screen three times per hour was intended to maximize the number of mosquitoes collected. The collector walked along the barrier screen and collected resting mosquitoes with the aid of a flashlight and a mouth aspirator. The collectors were provided with and instructed to apply mosquito repellents on their bodies to deter mosquitoes from biting them. Captured mosquitoes were placed into screened cups labelled with the hour of the night and the side of the barrier screen (i.e., bush or village side) on which the mosquitoes were captured. Information for each mosquito, including date and hour of collection, the screen side it rested on, and the blood meal status (blood-fed or unfed) were recorded. Mosquitoes were sampled for seven nights in Matukar, 58 in Mirap, 49 in Wasab, 12 in Dimer and six in Kokofine during the years 2012, 2013, 2015 and 2016 (see Additional file 1: Table S1). To minimize sampling biases associated with fixed locations, each barrier screen was changed to a new location on each night of sampling.

With the aid of a light microscope, mosquitoes were sorted by sex. Males which were very few (ca. 1% of the mosquito sample) were identified and discarded. Female Anopheles were identified to their morphological species [33, 34], assigned a unique serial number, and stored dry on silica gel desiccant. Mosquitoes morphologically identified as member of the Anopheles punctulatus sensu lato (s.l.) group were subjected to a PCR assay [35] to identify the species.

Similarity in species composition between pairs of villages was estimated by performing Bray–Curtis index analysis [36] on mosquito abundance matrix for year 2012 only when mosquitoes were sampled in all five villages (Table 1). The resulting Bray–Curtis dissimilarity index matrix (Additional file 2: Table S2) was used in principal coordinates analysis (PCoA) to produce an ordination plot of the study villages. The plot was further modified as a biplot of villages and Anopheles species by a posteriori projection of the species onto the PCoA axes based on their weighted average scores. The weighted average score of a species at PCoA axis 1 and axis 2 was calculated by averaging the product of its abundance (Table 1) and the PCoA axis (1 or 2) score for all villages [37]. Villages clustered together in the biplot are similar whereas those further apart are dissimilar in their species composition. The position of a village in the PCoA biplot was influenced by the abundance of the Anopheles species closest to it [37]. The Bray–Curtis index matrix and PCoA biplot were computed using the functions vegdist, cmdscale and wascores of the package vegan [38] in R software (version 3.4.1). For each of eight Anopheles populations, generalized linear model (GLM) with negative binomial distribution was used to compare number of mosquitoes in collections at each of the 12 hourly periods of the night, from the village and bush side of the barrier screen, and by blood-fed and unfed status. In the GLM regression equation ln(μ) = β₀ + β₁(Time) + β₂(Side) + β₃(Status) + ln(S), the expected mosquito number μ was modelled as a linear function of the categorical predictor variables Time, with 12 levels representing the 12 hourly periods of the night; Side, with two levels representing the two sides of the barrier screens; and Status, with two levels representing blood-fed or unfed status of the mosquitoes. Number of barrier screen-hours S was included as the offset term. The GLM was performed using function glm.nb from R package MASS [39]. The proportion of blood-fed relative to unfed mosquitoes on the bush side of the barrier screen was compared to that of the village side using Chi square test of equality of proportions for each Anopheles population. Significance level for all statistical tests was based on type I error rate of 5%.

A total of 7146 female Anopheles mosquitoes of seven different species (An. bancroftii, An. farauti s.s., An. farauti no. 4, An. hinesorum, An. koliensis, An. longirostris and An. punctulatus s.s.) were collected, including 2611 (36.5%) blood-fed and 4535 (63.5%) unfed; gravid females were very low in numbers (< 0.1%, and not considered further). The distribution of mosquito numbers for each of the Anopheles species over the five study villages is shown in Table 2. The mean proportion (± se) of each Anopheles species (excluding An. hinesorum, due to low sample size) collected per barrier screen per night in each village is shown in Fig. 2a. The results presented in Tables 1 and 2 and Fig. 2a show that these Anopheles species were not evenly represented within and among villages. In Matukar, where five different species were found, An. farauti s.s. constituted 69.0 ± 7% of the mosquitoes sampled per barrier screen per night (Fig. 2a). In Mirap, where six species were found, An. farauti s.s. constituted 82.8 ± 3% of the sample (Fig. 2a). In Kokofine, where four species were found, An. farauti no. 4 predominated (83.0 ± 11%) the sample and in Dimer, where five species were found, An. punctulatus s.s. predominated (66.5 ± 9%) the sample (Fig. 2a). In Wasab, where five species were found, An. longirostris (39.3 ± 6%) and An. punctulatus s.s. (37.3 ± 5%) each constituted a similar, high proportion of the sample while the rest of the species were relatively less abundant (Fig. 2a).

For Kokofine, Mirap and Wasab where mosquitoes were sampled for more than 1 year, variation in species composition between the years was observed (Fig. 2b). In Kokofine, the species composition changed from solely An. farauti no. 4 in 2012 to a more diverse community of four species in 2016. The proportions of these four species in 2012 and 2016 was significantly different (general Chi square test: χ² = 825.8, df = 3, P < 0.0001). In Wasab, An. farauti s.s., An. koliensis and An. punctulatus s.s. increased in proportion whereas An. longirostris decreased from 2012 to 2015 and this change was statistically significant for five Anopheles species (An. farauti no. 4 was excluded due to zero data in all 3 years) (χ² = 463.0, df = 8, P < 0.0001). In contrast, although the proportions of the five Anopheles species (An. farauti no. 4 was excluded due to zero data) in Mirap differed statistically (χ² = 66.8, df = 8, P < 0.0001), between the 3 years, this change in species proportion was not visually dramatic as in Wasab and Kokofine because the change happened only in the less abundant vector species but the proportion of the primary vector An. farauti s.s. remained generally steady over the 3 years (Fig. 2b). Comparison of mosquito abundance among the villages based on data from the year 2012, when mosquitoes were sampled in all five villages (see Additional file 1: Table S1), showed great variation in the number of Anopheles mosquitoes (regardless of species) captured per barrier screen per night among the villages. Kokofine had the highest abundance with an average of 170 Anopheles per barrier screen per night followed by Mirap (36 per night), Matukar (13 per night), Wasab (11 per night) and Dimer (10 per night) in decreasing order.

Ordination of the villages based on the Bray–Curtis dissimilarity matrix (Additional file 2: Table S2) showed that Dimer and Wasab closely clustered together in the PCoA biplot; Matukar and Mirap also clustered together but less closely; and Kokofine was separate from the other villages (Fig. 3). The close grouping of the two inland villages Dimer and Wasab was influenced by the similar presence and abundance of An. koliensis, An. punctulatus s.s. and An. longirostris in the year 2012 (other years were excluded from the analysis due to lack of sampling in some villages in those years). Similarly, the grouping of the two coastal villages Matukar and Mirap was influenced by the similar abundance and dominance of An. farauti s.s. in both villages. However, their cluster was not as tight as the two inland villages because of An. bancroftii, which was present and abundant in Mirap, but absent in Matukar. Kokofine differed greatly from the other villages, owing to the presence of An. farauti no. 4, which was the most abundant and exclusively found in that village (Fig. 3).

For selected species in each village (An. farauti no. 4 in Kokofine, An. farauti s.s. in Matukar, An. bancroftii and An. farauti s.s. in Mirap, An. punctulatus s.s. in Dimer, and An. koliensis, An. longirostris and An. punctulatus s.s. in Wasab), the mean number (± se) of blood-fed and unfed mosquitoes captured on the bush and village side per barrier screen per night were plotted for each of the 12 hourly periods of the night (Fig. 4). Full statistical output of the GLM analysis for these Anopheles populations are presented in Additional file 3: Table S3 and summarized graphically in Fig. 5. Except for An. punctulatus s.s. in Dimer and An. koliensis in Wasab which had a statistically similar number of mosquitoes collected throughout the night, the other six populations had statistically different numbers of mosquitoes for one or more of the hourly periods of the night compared to 19:00 h (Fig. 5, Additional file 3: Table S3). The GLM coefficient plot (Fig. 5) showed that except An. punctulatus s.s. in Dimer and An. koliensis in Wasab, the other six populations had higher number of mosquitoes in the hourly periods before midnight and less mosquitoes in the hours after midnight, after controlling for side of barrier screens and feeding status. This trend was strongly expressed in the unfed host-seeking subgroup of the six populations (Fig. 4; bush side, unfed panels). The number of mosquitoes captured on the village side of the barrier screens was significantly higher than the bush side for An. farauti no. 4 in Kokofine, An. farauti s.s. in Matukar, An. farauti s.s. in Mirap and An. punctulatus s.s. in Wasab, but not the other four populations (Fig. 5, Additional file 3: Table S3). For all populations except An. koliensis in Wasab, significantly more unfed than blood-fed mosquitoes were collected by the BSS method (Fig. 5, Additional file 3: Table S3). The total number of blood-fed and unfed mosquitoes captured on the village and bush side of the barrier screens for each of the eight Anopheles populations is shown in Fig. 6. Chi square tests for equality of proportions applied to the data in Fig. 6 showed that the proportion of blood-fed relative to unfed mosquitoes was significantly higher on the village than bush side of the barrier screen whereas the proportion of unfed relative to blood-fed was higher on the bush than village side for the populations in Kokofine, Mirap and Dimer but not those in Matukar and Wasab (Fig. 6).