Using sibship reconstructions to understand the relationship between larval habitat productivity and oviposition behaviour in Kenyan Anopheles arabiensis

Wild blood-fed Anopheles females were collected in May 2015 by aspiration from a cattle-baited trap [22] set up in Kirindo village, Mbita sub-county (0°26′38.46″S 34°15′36.95″E), western Kenya. Anopheles gambiae sensu lato (s.l.) females were identified morphologically, and other mosquito species discarded. Individual females were held in paper cups covered with a fine net. Moist cotton wool covered with filter paper was provided at the base for egg laying. Food was provided ad libitum by a disc of cotton wool soaked with 10% glucose solution on top of the cup. After egg laying, single legs from individual females were genotyped to identify species [23] to ensure that only An. arabiensis families were reared. Anopheles arabiensis eggs from each mother were raised separately in trays and the mothers preserved individually at − 80 °C in Eppendorf tubes containing absolute ethanol. After hatching, larvae were fed twice daily on TetraMin fish food (Tetra, Germany) and pupal stages from each family were separated into females and males prior to emergence to prevent mating. Following emergence, single virgin females and three males from different families were held in 15 × 15 cm mosquito cages for 48 h to copulate. Females were then offered two blood meals and held in paper cups to lay eggs, with resulting offspring reared to early instar stage. All the early instar larvae offspring were preserved at − 80 °C in Eppendorf tubes (20 larvae/tube) containing absolute ethanol for subsequent DNA analyses.

Field surveys and larval sampling from natural aquatic habitats was conducted between March and May 2017 within a rice irrigation scheme (00°28.473′S 034°32.786′E) (see Additional file 1). Two common mosquito larval habitat categories in the area—puddles and drainage ditches—were identified. Six puddles and five ditches of a perimeter size between 0.64 and 11 m² were the selected, and exhaustively sampled using larval sweep nets. The estimated pairwise distance between the habitats ranged from 2 m between drainage 2 (D2) and puddle 7 (P7) to 1264 m between puddle 1 (P1) and puddle 8 (P8) Anopheles larvae were separated from the collections morphologically and preserved at − 80 °C in 15 ml Falcon tubes containing absolute ethanol.

Genomic DNA was extracted from: (i) the two An. arabiensis families (Family A—adults n = 4, larval offspring n = 46; Family B—adults n = 4, larval offspring n = 36) collected from cattle-baited traps; (ii) the An. gambiae s.l. larvae collected from the artificial pond habitats (n = 466); and (iii) the natural field habitats (n = 702). Samples were extracted using the Nexttec DNA isolation kit (Biotechnologie, GmbH) following manufactures instructions. Following PCR-based species identification [23], but prior to further analysis, An. arabiensis samples from different artificial ponds and natural habitats were randomized on DNA extraction plates to mitigate any bias from plate batch effects.

Fifteen An. arabiensis microsatellite DNA markers located on chromosomes 2 and 3 were used in this study (Table 1). Markers were allocated into four multiplexes based on compatible primer annealing temperatures and non-overlapping expected allele size ranges using Multiplex Manager software [24]. Forward primers were fluorescently labelled with either NED, VIC, 6-FAM, or PET dyes. Amplification was carried out using the Qiagen Type-it microsatellite PCR kit with each reaction consisting of 6.25 μl Type-it master mix, 1.25 μl primer mix, 3 μl water, and 2 μl DNA template. The thermocycling conditions were as follows: initial activation at 95 °C for 5 min, 30cycles of denaturation at 95 °C for 30 s, annealing at 57–63 °C for 90 s, extension at 72 °C for 30 s with a final extension at 60 °C for 30 min. The PCR fragments were separated on an ABI 3730XL (Applied Biosystem) sequence analyser using the GeneScan™–500LIZ™ size standard. The allele sizes were scored using GeneMarker software v.2.6.7 (SoftGenetics) with each allele size score checked manually.

Genotyping errors possibly due to null alleles, large allele dropouts and mis-scoring of stutter peaks were identified using MICRO-CHECKER v.2.2.3 [25].

The capacity of the markers to determine sibships within known pedigrees was first tested using the two An. arabiensis families, A and B, with known parents. Family reconstruction was performed using two programs, COLONY [26] and BAPS 6 [27], that differ fundamentally in their methodologies and thus provide independent analyses. COLONY implements a likelihood algorithm in which individuals are simultaneously inferred into groups with the aim of finding the most likely grouping. In COLONY a full-likelihood (FL) analysis algorithm was used without a sibship prior and using a pool of allele frequencies from the natural habitats as a reference population. BAPS implement a Bayesian clustering algorithm treating both sample allele frequencies and the number of larval family groups as random variables when inferring clusters. In BAPS an individual clustering analysis was performed with the number of possible clusters ranging from 5 to 100.

A Kolmogorov–Smirnov test was used to check for normality of the distributions for numbers of larvae and reconstructed families, followed by a Grubbs test to check for outliers in the data, which were removed if detected. Pearson correlation and linear regression were used to test the relationship between the number of families and larval abundance. A partial Mantel test was used to test whether large families were more likely to be split across multiple habitats and if split families were more likely to occur in closer habitats in the software ZT [28]. For this test, a dummy (1/0) variable indicated whether habitats were direct neighbours or not, whilst controlling for sample size (as the minimum sample size per pair of habitats). The different experimental designs and analysis methods used in this study have been summarized in Fig. 1.

The capacity of 15 microsatellite markers, developed for An. arabiensis, to determine relatedness within known pedigrees was tested using two families of known full-siblings (A: n = 46 offspring; B: n = 36 offspring). The genotype from the mothers in Family B were inconclusive because of lack of informativeness of the loci, with some genotypes missing and therefore, only genotypes from family A were used for determination of parent–offspring Mendelian segregation but both families were used for sibship analysis. Twelve of the 15 markers yielded segregation patterns between the parents and amongst full-sibs in family A or amongst full sibs in family B consistent with Mendelian inheritance (see Additional file 2), and these markers were retained for subsequent analyses. Clustering analysis using either COLONY or BAPS provided entirely concordant results and correctly identified a single cluster of siblings within each family with no cross-family assignment.

Eighteen of the twenty artificial ponds were colonized, with a total of 466 Anopheles larvae collected. The total number of larvae collected from each pond ranged from 4 to 62 (median = 16.5, inter-quartile range = 31) (Table 2). From the species identification genotyping, 96.8% (n = 451) of the larvae were identified as An. arabiensis and were subsequently genotyped using microsatellites, 1.9% (n = 9) An. gambiae sensu stricto (s.s.), with 1.3% (n = 6) failed reactions.

COLONY estimated on average more females per pond compared to BAPS (Table 2), but the results were well correlated across families (r = 0.715, p < 0.001). Overall, BAPS estimated a total of 28 gravid females to have laid eggs across the 18 ponds while the COLONY analysis suggested 70 gravid females. Despite the difference in the total number of families estimated, 95% of the pairwise assignments of individuals as siblings in COLONY were also identified by BAPS, but the latter then appears to reconstruct many additional family pairings within its much larger clusters. In both clustering methods, the total number of larvae in each pond significantly predicted the number of families reconstructed (BAPS R² = 0.318, p = 0.01; COLONY R² = 0.476, p < 0.001) (Fig. 2). This corroborated the hypothesis that a relatively higher number of larvae in a pond is indicative of a higher number of females laying eggs in the pond as compared to a pond with a lower number of larvae in close vicinity.

Average numbers of larvae per family were quite low but very variable (median, IQR-BAPS 4.3, 4.9; COLONY 3.0, 2.9) suggesting high mortality of unsampled first instar larvae or eggs. Overall, BAPS estimated that 57% (95% CI 39, 75) females to have deposited larvae in multiple ponds and COLONY 43% (95% CI 30, 56), suggesting a high frequency of skip-oviposition by the mothers (Figs. 3, 4). Given the high skip-oviposition estimates, two linked questions were asked: are larger families more likely to be split across ponds, and are split families more likely to occur in closer ponds? There was a significant correlation between the number of larvae from each mosquito family and the number of ponds in which they were found using either estimator (BAPS r = 0.46, p = 0.015; COLONY r = 0.35, p = 0.003). Dependence of family sharing on pairwise proximity of pond pairs, while controlling for the minimum sample size per pair of ponds, was also evaluated in a partial Mantel test. There was a non-significant result for BAPS (r = 0.107 p = 0.11) and a marginally significant correlation for COLONY (r = 0.15 p = 0.04). This suggests that over the small spatial scale examined in the experimental pond design (Figs. 3, 4), larger families were more likely to be split but the preference of females to lay their eggs in physically closer habitats was weak.

A total of 702 larvae were sampled from across 11 natural habitats with the total number of larvae in each habitat ranging from 28 to 109 (median = 41, IQR = 41) (Table 3); 79% (n = 556) were identified as An. arabiensis, which were genotyped using microsatellites, 9.7% (n = 68) An. gambiae s.s., and 11.1% (n = 74) were failed reactions, some of which may have been non-An. gambiae s.l. Owing to much greater variability in genotyping success in the natural habitat samples, mean family size (number of larvae genotyped divided by number of families) was used as a metric for analysis rather than the number of mosquito families as in the ponds. Larval counts from habitat 8 was found to be an outlier compared to the other 11 natural habitats (p < 0.05) and was excluded from the analysis. After this exclusion all the larval counts from all habitats were normally distributed, Kolmogorov–Smirnov test p > 0.05.

Once again COLONY estimated many more families (N = 171) compared to BAPS (N = 41) but the results for mean family sizes estimated using each method were marginally correlated (r = 0.634, p = 0.049). Despite this correlation, only 56% of the pairwise sibship assignments in COLONY were also detected by BAPS suggesting a more moderate agreement between the two software than in the pond experiment. Mean family sizes from the BAPS estimation predicted larval abundance but those estimated by COLONY did not (BAPS R² = 0.829, p = 0.017; COLONY R² = 0.218, p = 0.68) (Fig. 5). Similarly, only results from the BAPS analysis supported the hypothesis that a higher number of larvae in one habitat compared to another at the same time is indicative of a higher number of females choosing to lay eggs. Nevertheless, both estimators suggested once again a high frequency of skip-oviposition (BAPS = 73%; 95% CI 59, 87; COLONY = 60%; 95% CI 53, 67.

Analysis to answer the same two linked questions as before, found that the mean family size was significantly positively correlated with the number of habitats among which they were distributed using BAPS (r = 0.902, p < 0.001) or COLONY (r = 0.451, p < 0.001). The tests of whether family sharing was dependent on proximity of habitat pairs were both non-significant (BAPS partial Mantel test r = 0.104, p = 0.22; COLONY r = − 0.027, p = 0.43).