Introduction

Studies on the genetic structure of malaria vector populations can be used to infer the likely success of vector control strategies. They can highlight issues such as the impact of conventional methods (eg, insecticide spraying) in reducing vector abundance, the reintroduction of vectors into formerly controlled areas, the spread of insecticide-resistance genes or the control of vectors by means of transgenic technology (Collins et al, 2000). The latter is a promising novel control strategy, although still subject to intense debate (Curtis et al, 1999). Should transgenic-based control be attempted, a natural first step would be to test its efficacy on islands, where confounding effects such as migration are expected to be lesser problems than on the continent.

The islands of the Gulf of Guinea, in particular São Tomé, are potential sites that might benefit from malaria control strategies based on transgenic vectors. In São Tomé, malaria is responsible for most hospital admissions and is the leading cause of child mortality. The only vector present is the forest cytoform of Anopheles gambiae s.s. (Pinto et al, 2000). Its origin on the island is likely to be contemporary with the first human settlements in the 15th century (Baptista, 1996). The vector is largely confined to coastal areas, being virtually absent at altitudes above 200 m. It is intimately associated with humans, feeding predominately on them or their domestic animals, and breeding in sites located close to houses (Pinto et al, 2000; Sousa et al, 2001). In 1980–1982, the island was subjected to a malaria eradication programme based on intensive intradomicilliary DDT spraying and mass drug (chloroquine) distribution (Ceita, 1986). During this period, both malaria prevalence and A. gambiae densities inside houses were dramatically reduced. The subsequent disruption of the programme led to a severe epidemic with many fatalities and present malariological parameters have reached or even exceeded preintervention levels (Pinto et al, 2000).

In a previous study, populations of A. gambiae from the islands of São Tomé and Príncipe did not show evidence of departures from mutation-drift equilibrium (MDE), in spite of their relatively recent founding and the possible impact of vector control in the 1980s (Pinto et al, 2002). Furthermore, despite the island's relatively small size, there was significant genetic differentiation between two samples from extreme locations within São Tomé. Whether this was due to physical barriers to gene flow or was a consequence of geographic distance between two extremes of a continuous vector population distribution remained unknown.

The aims of the present study were therefore (i) to determine whether the DDT spraying intervention of the 1980s has had an impact on the effective population size of the vector, (ii) to assess the present population stability and effective population size, in order to infer whether there are likely to be optimal periods for transgenic releases and, (iii) to determine patterns of population substructure, in order to infer possible restrictions to gene flow that may interfere with the spread of refractory genes. Results are discussed in relation to geographic, ecological and historical aspects of the vector populations, as well as to current and future prospects of vector control in the island.

Material and methods

Collection sites

The island of São Tomé (836 km2) is part of a volcanic chain located in the Gulf of Guinea, ca. 240 km off the coast of Gabon (Figure 1). It has a mountainous topography, particularly in the west-central and southern parts, where a number of peaks rise over 1500 m (Denny and Ray, 1989). The southern part of the island is heavily forested, while the northern part is flatter and humid-savannah-like landscapes prevail.

Figure 1
figure 1

The island of São Tomé. Geographic location, topography, demography and collection sites. (1) Neves: capital of Lembá district, where the Port for fuel importation is located. (2) Riboque: one of the main suburbs of São Tomé city, the capital of the Rep. Dem. São Tomé and Príncipe. (3) Bobo Forro: peripheral village of São Tomé city, in the most populated Água Grande district. (4) Ribeira Afonso: main town of Cantagalo district. (5) Angolares: capital of Caué, the largest but least populated district of the island. (6) Porto Alegre: the most extreme southern human settlement, a former colonial farm (Roça). With the exception of Porto Alegre, where most houses are ground-level and brick-built, wood-houses built on stilts up to 2 m prevail in all other collection sites.

Full size image

The climate is equatorial with an average annual temperature of 25°C and average relative humidity of 80%. Average annual rainfall varies between 500–2000 mm3 in the north and 3000–7000 mm3 in the south (Denny and Ray, 1989). There are two dry seasons, gravana (June–August) and gravanito (January), when rainfall is reduced but rarely absent.

The human population is unevenly distributed throughout the island with over 60% of the 140 000 inhabitants living within 10 km of the capital city, São Tomé (Figure 1). Most of the remaining population lives to the northeast of the capital in small villages along main roads close to the coast (Pinto et al, 2000). In the mountainous and forested south and southeast, human settlements are scarce or absent, and the southwestern part of the island is almost completely uninhabited.

Mosquitoes were collected by human baited landing catches between March and May 1998, in six localities of São Tomé (Figure 1). Samples of 60 females each were also collected in May 1997 and May 1999 from Riboque using CDC light traps. Both methods are considered comparable and equally representative for A. gambiae sampling (Davies et al, 1995). Female A. gambiae were individually stored in silica-gel-filled tubes and kept at −20°C until further processing.

Species identification and microsatellite genotyping

The DNA from individual A. gambiae was extracted by a phenol–chloroform protocol (Ballinger-Crabtree et al, 1992) and species identification was made by PCR (Scott et al, 1993). Samples were further identified as belonging to the M/S molecular form according to the PCR protocol described in Favia et al (2001). Microsatellite DNA analysis was performed for 11 dinucleotide loci (GT repeats) and one trinucleotide locus (AGC repeat). Details on the cytological location for each locus are given in Table 1. Genotyping procedures with fluorescence technology using an ABI 373 automatic sequencer (Applied Biosystems) were as described in Donnelly et al (1999).

Table 1 Genetic variability of microsatellite loci in A. gambiae s.s. populations of São Tomé island
Full size table

Data analysis

Estimates of expected heterozygosity were made using Nei's unbiased estimator (Nei, 1987). Differences in heterozygosity among samples and mean number of alleles were tested, pairwise, by Wilcoxon signed-ranks tests and overall by the Friedman test. Tests of deviation from Hardy–Weinberg proportions at each locus and of linkage disequilibrium between pairs of loci were performed using GENEPOP v3.3. (Raymond and Rousset, 1995).

Two independent estimates of effective population size (Ne) were produced. Current Ne was estimated for Riboque, based on the temporal variation in allele frequencies for three samples from May 1997, 1998 and 1999, respectively. The standardised variance of allele frequency change, F, was calculated for 1997–1998, 1998–1999 and the two-year interval 1997–1999, according to Pollak (1983), using equation (9) in Waples (1989). Estimates of Ne assumed that individuals were sampled without replacement prior to reproduction (Waples, 1989). A conservative number of 12 generations per year was used as in previous studies (Taylor et al, 1993; Lehmann et al, 1998). Calculations were made using programs written in the SAS language (SAS Institute, 1990).

Long-term Ne was calculated based on the expected heterozygosity at each microsatellite locus, for each locality. Estimates were made using equations described in Nei (1987) based on two mutation models, the stepwise mutation model (SMM) and the infinite alleles model (IAM). These are at the extreme ends of the spectrum of mutation models and therefore provide robust range estimates of long-term Ne (Lehmann et al, 1998). As in previous studies, average mutation rate (μ) was assumed to be 10−4 (Lehmann et al, 1998). For X-linked loci, Ne values were adjusted by a factor of 4/3 based on an assumption of a 1:1 sex ratio.

Two methods were used to assess deviations from MDE. Cornuet and Luikart's (1996) heterozygosity tests compare two estimates of expected heterozygosity, one based on allele frequencies (He) and another based on the number of alleles and sample size (Heq). In a population at MDE, both estimates should be equal (ie, He=Heq). If a population experiences a bottleneck, rare alleles will be lost and therefore Heq will decrease faster than He (ie, He>Heq). This apparent excess of heterozygosity is an indicator of a recent bottleneck event, whereas the converse (ie, He<Heq) may indicate an expansion event. Wilcoxon signed-ranks tests were used to determine if there was a significant number of loci in which He>Heq. Estimates of Heq were calculated under three mutation models, the SMM, the IAM and an intermediate two-phased model (TPM) with fractions of mutations greater than one repeat of 10, 20 and 30%. Tests were performed using BOTTLENECK 1.2.02 (Cornuet and Luikart, 1996).

Kimmel et al's (1998) imbalance index (β) is the quotient of two estimates of the genetic diversity parameter θ (for diploid loci: θ=4Neμ). One estimate is obtained from the variance of allele length (in repeat numbers) at each locus (θV) and the other from expected homozygosity (θP0). At MDE and under the SMM, θV=θP0; so β=θV/θP0=1. Simulations have shown that if β<1, this is an indication of recent population expansion from an original MDE (Kimmel et al, 1998). If β>1, expansion has followed a previous bottleneck event. Two estimators of mean β over loci, which yield normal-like distributions, were calculated as described in King et al (2000). The first is the log ratio of means, given by

and the second is based on the mean of log ratios,

where L is number of loci. Bootstrapped 95% confidence intervals were used to evaluate if the mean over loci β were significantly different from one.

Genetic differentiation was determined by the fixation index FST (Wright, 1978) and the analogous estimator for microsatellite data RST (Slatkin, 1995). Estimates of FST were calculated according to Weir and Cockerham (1984) using FSTAT v2.9.3 (Goudet, 1995). Estimates of RST were produced using RSTCALC software, according to Goodman (1997). The significance of FST and RST estimates was assessed by genotypic permutation tests among samples.

Isolation by distance was investigated by the regression of FST/(1−FST) and RST/(1−RST) on geographic distance (Rousset, 1997). Given the narrow distribution of the vector, a one-dimensional habitat was assumed and map-measured distances along coastal roads were used in the regression analyses (Rousset, 1997). Significance of the Spearman's rank correlation coefficient between variables (differentiation vs distance) was determined by Mantel tests. Calculations were carried out using GENEPOP v3.3 (Raymond and Rousset, 1995).

Whenever multiple tests were performed, the nominal significance level (P<0.05) was adjusted using the sequential Bonferroni procedure (Holm, 1979).

Results

Species identification and genetic diversity

All 448 specimens analysed belonged to the M molecular form of A. gambiae s.s. Microsatellite genetic diversity was moderate to high (Table 1). Overall, allele distribution was similar among localities (data available from authors on request). Numbers of alleles per locus varied from two to 11 (mean 6.3, SD 3.2) and heterozygosity varied from 0.389 to 0.805 (mean 0.623, SD 0.126). There were no significant differences in either variable across samples (Friedman test: mean number of alleles, P=0.300; heterozygosity, P=0.070). There were no significant pairwise differences in the mean number of alleles among samples. However, four of the five pairwise comparisons of mean heterozygosity that included the sample from the southernmost site, Porto Alegre, were significant prior to adjustment for multiple tests (Wilcoxon signed-ranks tests: 0.010<P<0.023). The only nonsignificant comparison was that between Porto Alegre and the next southernmost sample, Angolares (Figure 1).

Within-sample departures from Hardy–Weinberg expectations, as a result of heterozygote deficit, appeared to be clustered at loci AG2H147 and AG3H750 (Table 1). There was, however, no particular association of these two loci with any others in pairwise tests of linkage disequilibrium which were all insignificant after the sequential Bonferroni procedure was applied (16 tests out of 396 significant at the 0.05 level). The absence of genome-wide deviations from Hardy–Weinberg equilibrium and of linkage disequilibrium support that samples were drawn from homogenous and randomly mating populations.

Effective population size and MDE

Mean estimates (1-year) of current Ne in Riboque ranged from 1078 (349–∞), in 1998–1999, to infinity (1049–∞), in 1997–1998. For the 1997–1999 2-year estimate, a bound 95% CI was obtained (mean: 1457; 95% CI: 559–13 677). In single-locus estimates, the upper limit of 95% CIs was infinity in all but two cases (AG2H147: 86–2137; AG3H128: 43–5075) in 2-year estimates.

Estimates of long-term Ne were also in the order of thousands for both mutation models, values not consistent with recent bottleneck events (Table 2). No significant differences in Ne were found across localities (Friedman test: P=0.070, in both estimates). Depending on mutation model, two or three pairwise comparisons, which included Porto Alegre, were significant prior to adjustment for multiple tests (Wilcoxon signed-ranks tests: SMM: two out of five, 0.023<P<0.034; IAM: three out of five, 0.028<P<0.041). Furthermore, overlapping 95% confidence intervals were obtained among all estimates, indicating that long-term Ne is comparable among samples and that they appear to be subjected to similar amounts of genetic drift.

Table 2 Estimates of long-term effective population size (Nei, 1987) and heterozygosity tests (Cornuet and Luikart, 1996) for A. gambiae populations in São Tomé island
Full size table

The results obtained from the heterozygosity tests (Cornuet and Luikart, 1996) are shown in Table 2. Significant numbers of loci with an apparent heterozygote excess were detected for the less suitable mutation model, IAM, in five out of the six samples, and for TPM-70% (two samples) and TPM-80% (one sample). Luikart and Cornuet (1998) point out that these tests under the IAM may wrongly detect heterozygosity excess in nonbottlenecked populations when microsatellite data are used. They suggest using the strict SMM or a TPM model with 5–10% multistep changes with such data. Under these models, no significant proportion of loci with heterozygote excess was found in any sample (Table 2).

Contrasting results were obtained by the two estimators of the imbalance index, as shown in Table 3. While β2 estimates, varying between 0.92 and 1.42, do not support population perturbation, β1 estimates were significantly greater than 1 for all samples, suggesting that A. gambiae is experiencing a demographic expansion following a bottleneck or founder effect (Kimmel et al, 1998). Notwithstanding overlapping confidence intervals, there was an apparent trend for estimates of β1 to increase towards the south of the island, varying between 2.42 and 2.61 in northern localities, to 3.09 and 4.32 in the southern Angolares and Porto Alegre.

Table 3 Estimates of the imbalance index (Kimmel et al, 1998) for A. gambiae in São Tomé island
Full size table

Genetic differentiation

Pairwise over loci estimates of FST and RST showed comparable patterns of population differentiation (Table 4). Significant FST values were obtained for all comparisons involving Porto Alegre and between Bobo Forro and Angolares (Figure 1). Significant RST values were recorded in all comparisons with Porto Alegre, with the exception of the nearest site, Angolares. Patterns of population differentiation did not differ when loci AG2H147 and AG2H175, which exhibited highest heterozygote deficits, were excluded from the analysis, demonstrating that deviations from HWE had little effect on the FST/RST estimates (Table 4).

Table 4 Pairwise genetic differentiation estimates, FST and RST, among populations of A. gambiae s.s. in São Tomé island
Full size table

A significant positive correlation was found between both FST/(1−FST) and RST/(1−RST) and geographic distance (Mantel test; P=0.038 and 0.012, respectively) (Figure 2a). However, pairwise estimates of FST/(1−FST) with Porto Alegre were higher than expected by the regression with distance in four out of five cases and in three out of five cases for RST/(1−RST). When Porto Alegre was excluded from the analysis, no significant correlation was found for the remaining pairs (Mantel test; P=0.359 and 0.284, respectively) (Figure 2b).

Figure 2
figure 2

Isolation by distance of A. gambiae populations in São Tomé island. Regression analysis was made with estimates of FST/(1−FST) and RST/(1−RST) against road distances (see methods). Equations refer to the best fit regression line. White diamonds: comparisons with Porto Alegre.

Full size image

Discussion

This study provided evidences that the vector control campaign of the 1980s did not drastically reduce the effective population size of A. gambiae in the island of São Tomé. Populations, however, are likely to be expanding, possibly reflecting a relatively recent founder effect associated with the human colonization of the island in the 15th century. The genetic differentiation observed between northern and southern localities may be a consequence of north–south environmental heterogeneity, coupled with the synanthropic nature of this vector. In the south, human settlements are scarce and relatively isolated from those in the north.

Estimates of current Ne calculated for Riboque were in the order of thousands. These were in the range of those obtained by microsatellite data for A. gambiae from Western and Eastern Kenya (2455–103 317 and 1671–24 203, respectively) or for A. arabiensis from Senegal (409–2768) (Lehmann et al, 1998; Simard et al, 2000). In Mali, estimates of Ne around 2000 for A. arabiensis were obtained by inversion chromosome frequencies and between 922 and 1922 for A. gambiae using mark-release-recapture methods (Taylor et al, 1993, 2001). Given the equatorial climatic conditions of São Tomé, 12 generations per year is probably a conservative number; hence presented values may be an underestimate of the actual current Ne. Such values are not consistent with strong reductions in effective size occurring between 1997 and 1999. It is therefore unlikely that, in this island, A. gambiae experiences seasonal episodes of drastic reduction of effective population size.

Although robust, inferences on population stability based on current Ne are limited to the period within sampling time points. Estimates of long-term Ne, based on expected heterozygosity assuming equilibrium, however, reflect the long-term history of a species or population (Waples, 1991; Lehmann et al, 1998). Regardless of mutation model, large estimates of long-term Ne were obtained in all locations analysed and are not consistent with a bottleneck having occurred during the 1980's DDT intervention. Estimates are within the range of those found in A. gambiae mainland populations. Long-term Ne values (IAM–SMM) around 6500–21 000 (Senegambia), 7500–23 000 (Western Kenya), 8500–24 000 (Gabon) and 11 500–49 000 (Cameroon) are derived by reanalysing microsatellite data in previous publications (Lehmann et al, 1996, 1998; Pinto et al, 2002; Wondji et al, 2002). In addition, present estimates of long-term Ne may also be underestimated. This could be due to: (i) the conservatively high mutation rate used in calculations, since microsatellite mutation rates in some insect species may be lower than expected (Schug et al, 1997), and (ii) size constraints at microsatellite loci (Garza et al, 1995; Estoup and Cornuet, 1999), that could disrupt the correlation between allele size variation/heterozygosity and Ne.

Should a recent bottleneck have occurred, it would have been detected by heterozygosity tests which are sensitive to such events for a period of less than 4Ne generations but that can be as short as 0.2Ne generations (Cornuet and Luikart, 1996; Luikart and Cornuet, 1998). In the 17 years from the start of the DDT intervention, no more than 400 generations will have passed, even assuming 23 generations/year, an estimate for A. gambiae under optimal laboratory conditions (Lehmann et al, 1998). Heterozygosity tests under the most appropriate mutation models for microsatellite data (SMM and TPM 10%), however, showed no evidence of departures from MDE.

While values of β2 also provided no evidence to reject MDE, values of β1 suggest that populations are expanding. Estimator β2 tends to be seriously downward biased, being less sensitive to situations where expansion is preceded by a bottleneck or founder effect (Kimmel et al, 1998). This is more likely to have been the case for A. gambiae in São Tomé. Since the power of β1 and β2 increases to values above 0.5 after more than 625 generations following expansion (King et al, 2000), it is unlikely that the signature of expansion captured by β1 is associated with a bottleneck occurring in the 1980s. Rather, it may be the consequence of a founder effect associated with colonization of the island in the 15th century. The increasing β1 values towards south may also reflect more recent founder events or range expansions, since increased establishment of farms in the south occurred only in the 19th century (Tenreiro, 1961). In continental A. gambiae populations, β values significantly different from 1 were consistent with a population expansion scenario, possibly associated with the agrarian revolution that occurred in Africa ca. 10 000 years ago (Coluzzi, 1999; Donnelly et al, 2001).

Population substructure was found between northern and the southernmost sites, particularly Porto Alegre. Both FST and RST estimators showed significant differentiation between this locality and the remaining ones, suggesting a degree of genetic isolation between them. Little genetic differentiation was observed among northern localities of the island. The genetic uniformity in northern samples suggests that they were drawn from the same panmictic unit, implying a continuous distribution of A. gambiae along the northern coast. By itself, geographic distance may not be sufficient to explain the observed patterns of differentiation. The significant correlation observed under an isolation by distance model was lost when Porto Alegre was excluded from the analysis. Although removing data points reduces the power of the tests, the observed genetic differentiation of Porto Alegre was consistently higher than expected. In the southern parts of the island, the scarcity of human settlements coupled with the mountainous terrain and dense forest may act as a barrier to gene flow.

The observed patterns of population structure of A. gambiae in the island of São Tomé beg a number of questions related to its control. Conventional vector control measures directed solely to certain aspects of its lifecycle, such as indoor insecticide spraying, may have little long-term effect. The apparently low impact of such a control strategy on the genetic structure of A. gambiae may be related to the vector's exophagic and exophilic preferences in this island (Sousa et al, 2001). The campaign was probably successful in reducing human–vector contact or vector longevity and thus transmission. However, once the programme was disrupted, increased human–vector contact from a still large vector population, combined with loss of acquired immunity and appearance of chloroquine-resistant Plasmodium falciparum strains, led to the subsequent epidemic (Ceita, 1986).

High estimates of current Ne, suggesting little seasonal fluctuations in effective size, may pose difficulties in assessing the timing of transgenic insect releases. In such scenarios, the use of conventional methods aimed at reducing vector populations may be required prior to releases. Restrictions to gene flow may also pose obstacles to malaria control with transgenic insects. Releases undertaken at a limited number of sites might be sufficient to transform northern vector populations, but these might not affect those in the south. On the other hand, an isolated population such as Porto Alegre, located at the extreme end of the vector distribution on the island, may be regarded as a potential site for experimental transgenic releases, a crucial first step prior to the implementation of any large-scale releases.