Laboratory oviposition assays
To examine the interaction between larval breeding sites’ conditions and
oviposition evolution, we performed laboratory oviposition assays in a
common-garden setup. The goal was to examine whether forest and villageAe. aegypti have different oviposition preferences towards a
subset of environmental variables that differed between forest and
village larval breeding sites.
We established a forest colony and a peridomestic colony from La Lopé
using Ae. aegypti collected from larval breeding sites,
supplemented with oviposition traps and human landing capture (approved
by the National Research Ethics Committee of Gabon under the protocol
0031/2014/SG/CNE). In Rabai, we created six independent village colonies
(four domestic colonies and two peridomestic colonies) from the four
villages and four forest colonies from the Kaya Bomu forest. We
blood-fed the mosquitoes in the field and brought the eggs (i.e., the
second generation) back to our lab at Yale University and the McBride
lab at Princeton University. The detailed information of the mosquito
colonies and protocols for maintaining these colonies are in the
Appendix. The Rabai forest colonies correspond to KBO1 and KBO2 in Rose
et al. (2020). All laboratory oviposition assays were performed at Yale
University. We used the fourth to the sixth generation of mosquitoes in
these assays. For simplicity, we refer to the peridomestic and domestic
colonies as village colonies.
\begin{equation}
OAI=\ \frac{N_{1}-\ N_{2}}{N_{1}+\ N_{2}},\nonumber \\
\end{equation}where N1 and N2 are the
number of eggs deposited in the two cups, respectively. OAIranges from -1 to 1, representing a complete preference for the second
choice to a complete preference for the first choice. We performed
beta-binomial models in the R package glmmTMB (Brooks et al.,
2017) to examine whether colonies differ in their oviposition
preference, using the two egg counts in each cage as the dependent
variable (Rose et al., 2020). We added the batch/trial IDs as random
effects if data testing a specific condition were generated from more
than one experimental batch. The statistical significance of colony or
habitat effects were determined by comparing the full model with a null
model that excludes colony or habitat (Table S9). We extracted mean OAI
with a 95% confidence interval from the model using the R packageemmeans (Lenth et al., 2018; Rose et al., 2020).
Using this assay, we compared the oviposition preference of forest
versus village colonies from La Lopé and Rabai towards several
environmental variables. We first focused on a pair of Rabai forest
versus domestic colonies. The conditions tested include: 1) water
samples collected from tree holes and artificial containers in Rabai, 2)
pH, 3) shading, 4) larval density, 5) a combined effect of pH,
conductivity, and shading, and 6) bacterial community composition. We
selected these conditions as they showed significant differences between
forest and village larval breeding sites in the field. In each assay,
the two choices (i.e., two cups) roughly represent the median value of
the focal variable measured in forest and village larval breeding sites
(described in more detail in the Appendix). For example, in the
experiment on water pH, we adjusted the pH in the two cups to the median
pH values of all Rabai forest versus village larval breeding sites.
Lastly, for the experiment of bacterial community composition, we
expanded it to include all Rabai colonies as well as the two La Lopé
colonies (Table S2 in the Appendix).
In addition to the above two-choice assays, we tested the oviposition
preference of all mosquito colonies to five bacterial densities. This
experiment was inspired by the large variation in bacterial density
among field larval breeding sites (more than two orders of magnitude)
and that previous laboratory experiments with Ae. aegypti found
density-dependent ovipositional responses to bacteria (Ponnusamy et al.,
2015; Ponnusamy et al., 2010). We used a similar experimental design as
the two-choice assays but provided each cage of mosquito females five
cups instead of two. The cups contained bacterial cultures at densities
ranging from zero to nearly the maximal bacterial density in field
larval breeding sites. The bacterial culture was generated from an even
mixture of forest and domestic water samples (Table S2 in the Appendix).
We counted the numbers of eggs laid in the five cups and fitted a
negative-binomial model using the R package lme4 (Bates et al.,
2014), with bacterial density, habitat/colony, and their interactions as
predictors. If mosquitoes from different colonies or habitat types have
different oviposition choices, the interaction term would be
significant, which was tested by comparing the full model with a null
model excluding the interactive term (Table S9). We added cage ID as a
random effect. Lastly, we used the emmeans package to estimate
the expected number of eggs in each bacterial density with 95%
confidence intervals.