Fitness consequences of altered feeding behavior in immune-challenged mosquitoes

Background Malaria-infected mosquitoes have been reported to be more likely to take a blood meal when parasites are infectious than when non-infectious. This change in feeding behavior increases the likelihood of malaria transmission, and has been considered an example of parasite manipulation of host behavior. However, immune challenge with heat-killed Escherichia coli induces the same behavior, suggesting that altered feeding behavior may be driven by adaptive responses of hosts to cope with an immune response, rather than by parasite-specific factors. Here we tested the alternative hypothesis that down-regulated feeding behavior prior to infectiousness is a mosquito adaptation that increases fitness during infection. Methods We measured the impact of immune challenge and blood feeding on the fitness of individual mosquitoes. After an initial blood meal, Anopheles stephensi Liston mosquitoes were experimentally challenged with heat-killed E. coli at a dose known to mimic the same temporal changes in mosquito feeding behavior as active malaria infection. We then tracked daily egg production and survivorship of females maintained on blood-feeding regimes that either mimicked down-regulated feeding behaviors observed during early malaria infection, or were fed on a four-day feeding cycle typically associated with uninfected mosquitoes. Results Restricting access to blood meals enhanced mosquito survival but lowered lifetime reproduction. Immune-challenge did not impact either fitness component. Combining fecundity and survival to estimate the population-scale intrinsic rate of increase (r), we found that, contrary to the mosquito adaptation hypothesis, mosquito fitness decreased if blood feeding was delayed following an immune challenge. Conclusions Our data provide no support for the idea that malaria-induced suppression of blood feeding is an adaptation by mosquitoes to reduce the impact of immune challenge. Alternatively, the behavioral alterations may be neither host nor parasite adaptations, but rather a consequence of constraints imposed on feeding by activation of the mosquito immune response, i.e. non-adaptive illness-induced anorexia. Future work incorporating field conditions and different immune challenges could further clarify the effect of altered feeding on mosquito and parasite fitness. Electronic supplementary material The online version of this article (doi:10.1186/s13071-016-1392-x) contains supplementary material, which is available to authorized users.

Here we develop a population-scale model that uses individual-scale life table information to estimate a fitness measure for each individual in the experiment. In essence, the model gives the population growth that would result if a population was comprised of individuals with identical life histories based on the empirical life table of a single individual. While the model is developed for mosquitoes in this experiment, the approach is general to organisms with distinct stages. The experiments exposed adult mosquitoes to different treatments and recorded the day of death and the number of eggs laid each day. Let E(t), L(t) and A i (t) denote the density of eggs, larvae and adults respectively at time t. For adults, i denotes the time since pupal eclosion. We assume that all individuals in a stage share the same development, birth and mortality rates. To match the experimental protocol, we assume that i is an integer number of days since pupal eclosion. Specifically, i ∈ {0, 1, . . . , α} where α is the last day the individual was alive. The birth rate b i differs depending on the age since pupal eclosion as observed in the experiments. To account for mortality, the model assumes that individuals in the egg and larval stages have constant daily mortality rates, but that adult mortality occurs exactly as observed in the data. This is done by introducing a recruitment term out of the adult class at a time delay given by the observed longevity for that individual.
The population model is where τ E and τ L are the egg and larvae stage durations, δ E and δ L are the stage-specific mortality rates, and S E and S L are through stage survivorship's. The second term of (3) is recruitment about of the adult class, which describes the death of adults at an exact age since pupal eclosion of α. The index i is used to integrate reproduction over all adult ages, and the index j denotes the dynamics of adult densities at each age. The full system has dimension α + 3.
Since the adult stage equation does not depend on egg or larval densities, we can model the population growth using just the adult equations as for j ∈ {0, . . . , α}. Since the model is density independent, adult densities will asymptotically undergo exponential growth or decay. While it is possible to write the corresponding characteristic equation for this model, there are no analytical methods to solve for the maximum growth rate. To simplify the process of estimating the growth rate, we write the discrete-time approximation to the model as where n t is a vector of the number of mosquitoes in a particular age class at time t, and L is a Leslie matrix given by where the age classes start from the first egg stage and go until the observed day of adult death (τ E +τ L +α). Note that, as in the full model, adult mortality is incorporated explicitly through senescence through the dimension of the transition matrix (L).
The experimental protocol was to work with adult mosquitoes immediately after pupal eclosion. As a result, development times and mortality rates of the egg and larval stages will be the same for all treatments. We use egg and larval development times of τ E = 1 d and τ L = 12 d respectively, which are typical for this lab population at the temperature and humidity conditions of the experiments (1). To estimate egg and larval mortality rates, we use the observation that roughly 70% of eggs hatch (Moller-Jacobs, unpublished data), and 75% of the individuals survive from egg eclosion to pupal eclosion (1), which corresponds to egg and larval mortality rates of δ E = − ln(0.7)/τ E = 0.356 d −1 and δ L = − ln(0.75)/τ L = 0.024 d −1 respectively. The remaining parameters are the age of adult death (α) and daily birth rate (b i ), which are obtained directly from the life-table data for each individual.

Model analysis and evaluation of assumptions
The life-table data for each individual leads to a unique parameterization of the population model, and therefore a unique value of the asymptotic population growth rate. Recognizing that the inference is constrained to these laboratory conditions under an assumption of unlimited resources, the calculated population growth provides an estimate of fitness that properly combines the role of fecundity and survivorship. The asymptotic growth rate for the population model in eqn (11) is given by the right leading eigenvalue using the eigen function in the R statistical environment (2). Fitness for the full continuoustime model was estimated through time simulations using the PBSddesolve library (3) in the R statistical environment (2).
For each individual, we calculated a fitness estimate using the full continuous-time model in eqns (1)-(7) and the discrete-time approximation in eqn (11). The estimates of fitness from both models were in excellent agreement (Figure 1), which indicates that the approximation is valid for these experiments.
We used the full model when reporting statistical results, and for computational speed used the discretetime model to explore the robustness of our conclusions to the values of egg and larval development and mortality (i.e., τ E , τ L , δ E , δ L ). We explored changes from 10% to 500% for each parameter and found the statistical results are robust to variation in these parameters (Figure 2), but the effect size increases with increasing mortality rates and development times (Figure 3).