The current geographical distribution of Ester2 could be explained by four non-exclusive hypotheses. First, Ester2 could have been the first resistance gene to occur at the Ester locus, thus having an opportunity to spread widely, before other resistance alleles appeared (historical contingency hypothesis). Alternatively, Ester2 may possess a fitness advantage over the other resistant alleles, either locally (i.e. only in some environmental conditions) or globally (i.e. in all areas), and its present distribution would be the result of its competitive advantage (local or globalselection hypotheses). Third, it may be neutral with respect to other resistance genes, and chance alone explains this distribution (drift hypothesis). It is probably impossible to reject the historical contingency hypothesis, due to the lack of data during the 1960s and early 1970s, i.e. at a time where the first Ester resistance gene occurred. The pattern of replacement in resistance alleles and their global distribution do not suggest that genetic drift between resistance alleles (i.e. neutrality among resistance alleles) plays a predominant role: drift would generally create random fluctuations, which are not observed at a local scale (for example in the Montpellier area). Moreover, due to the large population sizes observed for this species, the time needed for a replacement between two resistance alleles (e.g. 7 years, Guillemaud et al., 1998) would be much longer than that observed if the process was driven by genetic drift only. Longitudinal studies are required to evaluate the selection hypotheses. To our knowledge, only eight cases are documented in the literature (Table 5), and five of them document the interaction of Ester2 with another resistance allele. In California, only EsterB1 was reported before 1979 (Pasteur & Georghiou, 1980). Then, Ester2 was also detected in the same populations in 1984, sometimes at a relatively high frequency (>0.5 at Long Beach in 1985, Raymond et al., 1987), suggesting that the replacement of EsterB1 by Ester2 was going on. Unfortunately no additional published data are available to document the evolution of this situation. In Cuba, the same process has apparently taken place: EsterB1 was first present, then Ester2 was detected, and its frequency rose and it became the most frequent resistance gene (Small, 1996). In Houston (Texas), Ester2 is also replacing EsterB1. In 1994, 20% of analysed mosquitoes were found carrying EsterB1, 40%EsterB1 and Ester2, and 40% only Ester2 (n = 104), whereas in 1998, the frequencies were respectively (n = 84) 11.5%, 11.9% and 68.7%, with 7.1% being susceptible mosquitoes (Pietrantonio et al., 2000). The authors explain the replacement by a decrease, since 1993, in malathion (an OP) use: both the number of areas treated and the number of applications decreased, to very low levels (even zero in some areas). In Martinique, the opposite situation is found: in 1990, Ester2 was at a high frequency (>37% in average, >66% in one population), while EsterB1 was very infrequent (<3%) and not found in three of seven populations sampled. Nine years later, EsterB1 had spread to all the sampled populations, with an overall frequency >30%, whereas the susceptible allele has decreased from >60 to <15%. In the same time, Ester2 decreased in some populations (e.g. from >60 to <30% in Carbet population), while increasing in few others (e.g. from <40 to >90% in Fort de France, alleles frequencies were computed using maximum likelihood from phenotype data provided in Yébakima et al., 2004). At the locations where EsterB1 has replaced Ester2, the authors explain this situation by a change of insecticide usage: during the increase of EsterB1 frequency, the intensity of insecticide treatment also increased (∼10-fold) and the authors show that EsterB1 confers a higher resistance to insecticides used in Martinique (Temephos) than Ester2. In Lucca (Italy), there is a possible replacement of Ester1 by Ester2 around 1985–1988 (Bonning et al., 1991; Villani & Hemingway, 1987) although published data are too scarce to firmly document it. However, Ester2 was present in 1985 in Lucca (Bonning et al., 1991), but it disappeared in 1992, probably because of the stoppage of insecticide treatments (Severini et al., 1994). Overall, this fragmentary dataset suggests that there is no overall trend, and that Ester2 could either replace or be replaced by another Ester allele. Thus, this dataset contradicts the global selection hypothesis, and is more consistent with the local selection hypothesis. Environmental conditions are probably involved in this interaction, such as the type or quantity of insecticide used. In Martinique, Ester2 decreased in frequency when treatments were increased, whereas in Houston its frequency increased when treatments were decreased. Both situations suggest that Ester2 is at a disadvantage when insecticide treatments are intensive. Its fitness advantage in California and Cuba, in conditions where insecticide doses are reduced, could be explained by lower fitness costs rather than better resistance. However, in most places, little information is available on pesticide application practices, precluding any firm conclusions. Nevertheless, this indicates that environmental conditions play a role during the course of adaptation at the local scale. There is an additional situation where a detailed longitudinal study has taken place, allowing a clearer understanding of the selection hypotheses.