Transmission traits of malaria parasites within the mosquito: Genetic variation, phenotypic plasticity, and consequences for control

Abstract Evaluating the risk of emergence and transmission of vector‐borne diseases requires knowledge of the genetic and environmental contributions to pathogen transmission traits. Compared to the significant effort devoted to understanding the biology of malaria transmission from vertebrate hosts to mosquito vectors, the strategies that malaria parasites have evolved to maximize transmission from vectors to vertebrate hosts have been largely overlooked. While determinants of infection success within the mosquito host have recently received attention, the causes of variability for other key transmission traits of malaria, namely the duration of parasite development and its virulence within the vector, as well as its ability to alter mosquito behavior, remain largely unknown. This important gap in our knowledge needs to be bridged in order to obtain an integrative view of the ecology and evolution of malaria transmission strategies. Associations between transmission traits also need to be characterized, as they trade‐offs and constraints could have important implications for understanding the evolution of parasite transmission. Finally, theoretical studies are required to evaluate how genetic and environmental influences on parasite transmission traits can shape malaria dynamics and evolution in response to disease control.

Despite the public health importance of these pathogens, many fundamental aspects of transmission remain unexplored. In particular, the sources of variation in traits that predict transmission from vectors to vertebrate hosts have been largely overlooked (Box 1). Like any vector-borne parasite, malaria parasites must exploit patchy resources, encountering different environments with varying resources and selective forces as they make their way between the human host and insect vector. Parasite transmission traits can thus be influenced by multiple interacting factors including the direct influence of parasite genetic characteristics, the within-vertebrate or within-vector environment (vertebrate/vector genotype, immune responses, resource availability, presence of co-infecting parasites, age, etc.), and the indirect influence of the external environment (temperature, humidity, host's predators, competitors, etc.). In recent years, a great deal of effort has been invested in studying transmission traits of malaria parasites in their vertebrate host (Cameron, Reece, Drew, Haydon, & Yates, 2013;Greischar, Mideo, Read, & Bjornstad, 2016;Neal & Schall, 2014;Reece, Ramiro, & Nussey, 2009). As we would predict, studies have shown that both genetic and environmental factors are important in determining parasite transmission from vertebrate hosts to mosquitoes. Like any other phenotypic trait, transmission traits can respond to environmental changes either plastically or evolutionarily (Box 2).
For example, work using rodent malaria models suggests that parasite genotype can predict virulence and transmission success (De Roode et al., 2005). Furthermore, studies have shown that the investment of malaria parasites in gametocyte transmission stages can vary in response to environmental conditions, such as the presence of drugs, the availability of resources, the host immune response, coinfection with different strains, and the presence of vectors (Cornet, Nicot, Rivero, & Gandon, 2014;Mideo & Reece, 2012;Pollitt et al., 2011). While some of these responses may illustrate cases of passive susceptibility to environmental changes, others are likely examples of adaptive plasticity (Box 2). For example, Plasmodium chabaudi can detect the presence of unrelated conspecifics and adjust the proportion of male and female gametes in a way that supports sex ratio theory (Reece, Drew, & Gardner, 2008). This research demonstrates that unicellular parasites can evolve finely tuned mechanisms to detect information about their within-host environment and plastically adjust some of their transmission traits.
In comparison with explorations of within-host factors that affect transmission from hosts to vectors, little work has been performed on the other half of the parasite transmission cycle: from vectors to vertebrate hosts. We propose that a complete understanding of factors that shape the evolution of transmission strategies must consider not only the within-vertebrate host factors contributing to transmission, but also those factors within the vector (Box 1). We use vectorial capacity (C), one of the most common metrics of transmission for vector-borne diseases, to establish a framework for investigating genetic and environmental variation in transmission traits within the mosquito vector. C is defined as the potential intensity of vertebrate-to-vertebrate Compared to the effort devoted to explore transmission traits in vertebrate hosts, few studies have quantified the genetic and environmental influences on these traits within the mosquito. Beside work on molecular, genetic and environmental determinants of vector competence and a study on density-dependent costs on parasite development (i.e. sporozoite density was negatively associated to oocyst density, variability in the following strategies remain, to our knowledge, unknown: How fast to grow? How much to invest in sporozoites? How much damage to impose to the mosquitoes? How much and for how long the parasite should manipulate the vector's feeding behavior? The answers to these questions are certainly complex and depend on many parameters such as costs and benefits associated with variation in the traits, possible trade-offs between different traits, and evolutionary constraints parasite transmission by mosquito vectors and can be described by the formula: where m is the density of vectors per vertebrate hosts, a is the vector biting rate and host preference, V is vector competence, p is the daily probability of adult vector survival, and n is the duration in days of the parasite's extrinsic incubation period (EIP; Dye, 1992). Four of these critical components of transmission-the biting rate, mosquito competence, mosquito survival, and EIP-are traits that could potentially be determined directly or indirectly by parasites (Table 1). The vectorial capacity equation predicts that parasites could enhance transmission by influencing vector physiology to increase competence (V), altering the timing and propensity of mosquito biting (a), shortening EIP (n), or by increasing vector longevity (p).
Box 2 Genetically fixed responses and (adaptive vs. nonadaptive) phenotypic plasticity Like any other organism trait, changes in parasite phenotypic traits can occur through two nonmutually exclusive processes: genetically fixed responses and/or phenotypic plasticity (Pigliucci, 2005). First, there may be genetic variation underlying transmission traits, and natural selection will favor the genetic variants which produce the phenotypes most fitted to the current conditions. This is the classic evolutionary response whereby some genetic variants can spread through the population over generations. Genetic variation is the raw material for evolution; therefore, characterizing genetic variability in transmission traits is key to understanding how control interventions can drive evolutionary changes in the parasite. As one hypothetical example, reduced vector longevity following insecticide exposure might select individuals with shorter EIP in the parasite population.
Second, a given parasite genotype may be able to produce different phenotypes in response to different environmental conditions, that is, phenotypic plasticity. In contrast to genetic changes over generations, modifications in phenotypic traits through plasticity can occur within a generation. Many examples of phenotypic plasticity are clearly adaptive such as some immune responses, antipredator defenses, and diapauses allowing individuals to adjust to environmental variation in real time (Whitman & Agrawal, 2009). In this case, organisms possess mechanisms to detect cues that predict environmental changes and induce adaptive plasticity. Such plasticity does not necessarily involve changes in gene frequencies in the parasite population and can provide a more rapid response to unpredictably changing environments. Using the above hypothetical example, parasites could detect cues associated with imminent death of their vectors (e.g., directly through the presence of insecticides or indirectly through modifications of vector physiology) and adaptively accelerate their sporogonic development to achieve transmission prior to vector death.
In contrast to adaptive plasticity, other environmentally induced changes in phenotype may illustrate mere susceptibilities to environmental stresses with no adaptive value (Ghalambor, McKay, Carroll, & Reznick, 2007). In this case, the phenotypic changes can arise from a "passive" disruption of physiological processes and do not require any mechanisms for how cues are detected. For example, a longer EIP in mosquitoes exposed to insecticides and hence with reduced potential for transmission compared to mosquitoes with greater longevity would indicate that environmental variation (here a reduction in mosquito longevity) does influence this trait, but this would also be intuitively interpreted as a case of phenotypic plasticity with maladaptive value. However, it is often difficult to conclude whether or not altered phenotypes are adaptive or nonadaptive (Pigliucci, 2005).
In any case, determining the extent to which parasite transmission traits are genetically fixed or plastic will help predict the consequences of control interventions on parasite evolution. Experimental designs with some form of genetic structure (clones, family lines) and environmental treatments are extremely powerful for studying genetic effects and phenotypic plasticity (Whitman & Agrawal, 2009). Measuring transmission trait (EIP, virulence, manipulation, infection level) variation among different genetic backgrounds or environmental conditions will help to quantify the relative importance of phenotypic plasticity and genetic variation. The statistical measure of variation is variance, which quantifies the deviation of values around a mean. The variance of a phenotypic trait can be partitioned as follows: where V P = Total phenotypic variance for a trait; V G = Genetic variance (proportion of phenotypic variation attributable to genes); V E = Environmental variance (proportion of variation caused by the environment); V G×E = Genotype × Environment interaction (genetic variation for phenotypic plasticity); V error = Unexplained variance, including developmental noise.
Quantifying phenotypic variation across different parasite clones or mosquito genotypes in controlled conditions will minimize environmental variance, and the phenotypic variance will be close to the genetic variance. Similarly, randomly assigning mosquito genotypes infected with single parasite clones (monoclonal infections) to different environmental treatments will lead to a robust estimate of phenotypic plasticity (Whitman & Agrawal, 2009).
The degree to which variation in any one of these parameters affects transmission outcomes depends both on how sensitive vectorial capacity is to perturbations in a given parameter and the extent to which a given parameter can vary. Sensitivity analyses can evaluate the relative effect small changes in one parameter have on the outcome of what the model is predicting. Previous sensitivity analyses on the vectorial capacity equation have indicated that vectorial capacity is highly sensitive to adult mosquito survival (Brady et al., 2016;MacDonald, 1957;Smith & McKenzie, 2004). This has consequently led to suggestions that interventions targeting adult survival may be the most effective means of vector control, even when weighted by the relative effort of implementing an intervention (Brady et al., 2016). Using similar analyses to weight sensitivity by the capacity of a trait to vary cannot currently be conducted on key vector traits (V, a, n, and p) because variation in traits is poorly characterized. Control strategy design and transmission predictions could be improved by understanding the extent of variation in these parameters. Here, we explore each of these traits, review the extent of observed and predicted genetic and environmental variation, and discuss how variation in any one of these components of vectorial capacity impacts parasite transmission.

| MOSQUITO COMPETENCE (V )
Mosquito competence is the ability of mosquitoes to support malaria development and transmission. It can be measured in the laboratory by exposing mosquitoes to a given dose of parasite gametocytes during   (Vaughan, 2007), see also Box 1). Studies using the Plasmodium berghei-Anopheles stephensi experimental system found that these developmental transitions experienced negative density dependence, possibly due to resources and space limitation and/or to an elevated mosquito immune response Sinden et al., 2007).
An important assumption of this hypothesis is that there must be a positive relationship between sporozoite burden in the salivary glands and infection of the vertebrate host, something that has long been disputed (Beier, Davis, Vaughan, Noden, & Beier, 1991;Beier et al., 1992;Ponnudurai, Lensen, Vangemert, Bolmer, & Meuwissen, 1991;Sinden, 2016). A recent study using rodent parasites provides strong support for this relationship by showing that mosquitoes with higher numbers of sporozoites in salivary glands are indeed more likely to transmit malaria (Churcher et al., 2017).
It has also been proposed that self-restriction strategies based on programmed cell death may reduce the mosquito immune re-

| VECTOR BITING RATE AND HOST PREFERENCE (a ): PARASITE MANIPULATION OF THE VECTOR'S FEEDING BEHAVIOR
The vectorial capacity equation predicts that, when ready to be transmitted from either vertebrate to vector or vector to vertebrate, malaria parasites able to increase the vector's biting rate on suitable vertebrate hosts species would increase their probability of transmission (Dobson, 1988 There is a reason to think that both parasite and host genetics should be selected upon to shape these phenotypes. The altered patterns in feeding behavior observed in malaria-infected mosquitoes have been empirically demonstrated to have negative impacts on mosquito fitness (Anderson, Knols, & Koella, 2000;Ohm et al., 2016). This suggests that there is selection for both the parasite to alter mosquito behavior and the vector to resist being manipulated (Daoust et al., 2015). Historically, there has been a large emphasis on identifying specific parasite traits that in isolation lead to altered mosquito behavioral phenotypes. Recent work suggests that some components of manipulation may relate to the mosquito's own immune response (Cator et al., 2013(Cator et al., , 2015 and that the transmission phenotype observed is likely dependent on the genotype and condition of the vector, as well as the parasite (Cator et al., 2015). How these phenotypes can vary with the environment (e.g., mosquito age or vector density) is unknown and is critical for our understanding of how they affect transmission.

| THE EXTRINSIC INCUBATION PERIOD (n)
Natural selection will theoretically favor a developmental schedule for each parasite stage which maximizes transmission between successive hosts (Poulin, 2007). Once in the insect vector, a major challenge facing the parasite is to reach its infective stage before the insect takes its last blood meal. The extrinsic incubation period (EIP) is the duration of the parasite's development within the mosquito that starts with the ingestion of infective malaria parasites, gametocytes, in a blood meal and ends with the sporozoite invasion of the salivary glands when the mosquito becomes infectious (Box 1). For many mosquito-Plasmodium associations, this period is as long as the insect vector's average lifespan (Charlwood et al., 1997;Killeen, Mckenzie, Foy, Peter, & Beier, 2000). Plasmodium falciparum, for example, has an extremely variable EIP, but generally ranges from 10 to 14 days in high-transmission settings (WHO, 1975). The question of why this period is so long relative to the vector lifespan has been discussed elsewhere (Cohuet, Harris, Robert, & Fontenille, 2010;Koella, 1999;Ohm et al., 2016;Paul, Ariey, & Robert, 2003).
Both mosquito and parasite are ectothermic, and the impact of temperature on the rate of sporogonic development has long been recognized (Boyd, 1949;Detinova, 1963;Murdock, Paaijmans, Coxfoster, Read, & Thomas, 2012). In general, warming temperatures speed up parasite development, although above a certain threshold (30°C in P. falciparum), this can reduce infection level (Noden, Kent, & Beier, 1995;Okech et al., 2004). Evidence that EIP can vary in response to other environmental factors is limited. Plasmodium falciparum EIP can be modified by the quantity of food received by An. stephensi larvae (Shapiro et al., 2016) or by the source of plant sugar taken by adult Anopheles coluzzii (Hien et al., 2016). Of particular interest would be to test the extent to which malaria parasites can plastically speed up their EIP when their transmission potential is compromised by the imminent death of their vector. Such condition-dependent developmental strategies, described in other parasite species (Donnell & Hunter, 2002;Poulin, 2007) and in blood-stage malaria parasites (Mideo & Reece, 2012), deserve consideration in infected mosquitoes. Besides mosquito age, other environmental factors, including exposure to insecticides (Viana, Hughes, Matthiopoulos, Ranson, & Ferguson, 2016) or presence of other parasite species/genotypes (Blanford et al., 2005;Lorenz & Koella, 2012), are associated with mosquito survival and could induce an adaptive plastic shift in parasite EIP. Similar to withinvector conditions, the extent to which parasite and/or mosquito genetic variation can influence EIP merits exploration.
At the interspecific level, some studies suggest that parasites may adapt to vector lifespan, as demonstrated by Plasmodium species with shorter EIPs associating with shorter lived vectors, such as Plasmodium mexicanum that is vectored by short-lived sandflies. Only about 2% of sandflies capable of transmitting P. mexicanum live long enough to take a second blood meal (Fialho & Schall, 1995). Compared to other Plasmodium species, P. mexicanum has a rapid development time that ensures transmission despite the vector's high mortality, which is likely an evolved response.
At the intraspecific level, there has been no study on the influence of parasite and/or mosquito genetics on EIP duration. A recent study investigating the evolutionary potential of dengue virus EIP in Aedes aegypti demonstrated that genetic variation among a range of mosquito genetic lines can modulate the length of EIP (Ye et al., 2016). Because vectorial capacity is highly sensitive to changes in EIP, it becomes urgent to investigate the evolutionary potential of EIP in malaria parasites using family-based breeding (Ye et al., 2016) and/or experimental evolution design (Nidelet, Koella, & Kaltz, 2009).

| MOSQUITO LONGEVITY (P ) AND OTHER DAMAGES INFLICTED TO THE MOSQUITO
Whether malaria parasites cause fitness costs to their mosquito hosts has received much attention and has long been disputed (Ferguson & Read, 2002b;Hurd, 2009;Vézilier, Nicot, Gandon, & Rivero, 2012).
The parasite should first exhibit a low level of virulence during parasite development to prevent the death of both partners. Once the development is completed and sporozoites are in the salivary glands, parasite genotypes able to increase the biting rate of their mosquito vector could be favored (Koella, 1999;Schwartz & Koella, 2001).
Consistent with these predictions, some studies reported greater survivorship in infected than in uninfected mosquitoes during the oocyst infection phase and the opposite when sporozoites have reached maturity (Anderson et al., 2000;Lyimo & Koella, 1992;Roux et al., 2015).
Whether the increased survivorship observed in infected individuals during oocyst growth resulted from an active manipulation of the parasite or reflects a compensatory response of the mosquitoes to energy depletion remains unknown. Investigating the importance of parasite genetic variability and interactions with mosquito strain would also deserve consideration. In the P. chabaudi-An. stephensi model, there is evidence that different parasite genotypes vary in their effects on mosquito survival and fecundity (Ferguson & Read, 2002a;Ferguson et al., 2003). There also is evidence that some mosquito strains can suffer higher cost of infection by a given parasite genotype than others (Vézilier et al., 2012). Future studies are required to test whether Plasmodium genotype by mosquito genotype interactions impact mosquito longevity and fecundity.

| POSSIBLE ASSOCIATIONS AND TRADE-OFFS AMONG TRANSMISSION TRAITS
It is important to remember that it is the emergent properties of a given set of competence, biting rate, EIP, and survival values that determine transmission and that these parameters do not operate in isolation ( Figure 1). For example, Plasmodium may modify resource allocation of their insect vectors in a way that changes the optimum trade-off between reproduction and longevity, which, in turn, could favor either longer or similar vector survivorship than uninfected counterparts (Hurd, 2001(Hurd, , 2003(Hurd, , 2009. In a study using avian malaria and allowing mosquitoes to lay their eggs, infected mosquitoes were less fecund but lived longer than uninfected counterparts (Vézilier et al., 2012). This emphasizes the need to concomitantly quantify mosquito longevity and fecundity, which is rarely performed in studies on mosquito-parasite interactions. Finally, there is, to our knowledge, no study that investigated the effect of malaria infection on both mosquito longevity and fecundity over multiple gonotrophic cycles.
Beside the existing links between mosquito infection, fecundity, and longevity, an intriguing possibility is that EIP, the parasite's ability to manipulate mosquito biting rate, and mosquito survival are also correlated. For example, reduced longevity in infected mosquitoes or long parasite development duration will limit the time period for parasite transmission, but this could be compensated by increased mosquito biting rate (Koella, 1999(Koella, , 2005. In turn, increased biting rate can also increase the probability of mosquito mortality (Anderson et al., 2000).
Similarly, the reduction in transmission opportunities due to long parasite development duration could be compensated by increased mosquito lifespan. In other words, fast-developing parasites might also be those that induce high level of virulence in their mosquito hosts. A recent study using dengue virus-infected A. aegypti revealed that mosquito family lines allowing fast EIP were also those that died faster supporting the existence of a genetic trade-off between mosquito lifespan and EIP (Ye et al., 2016). To explore these trade-offs, future work should concomitantly quantify multiple mosquito traits. •How much trait variation is there among Plasmodium species, clones, and drug resistance status within a given vector genotype?

| KEY STEPS TO APPLIED VALUE
•How much trait variation is there among vector species, genotype, and insecticide resistance status when infected with a given parasite clone?
•How consistent are these effects across mosquito and parasite genotype/species combinations?
•How does within-mosquito environment (mosquito age, size, nutrition, and condition) affect transmission traits ?
•How do inoculum size (density effects), and intraspecific (other malaria infections) and interspecific (other microorganisms in the mosquito gut) interaction affect transmission traits?
•How do external environmental factors (sublethal doses of insecticide, temperature, resource availability) affect transmission traits?
Possible trade-off between transmission traits?
Are fast-developing parasites also those that are the most virulent i.e. those that induce greatest longevity reduction?
Are fast-developing parasites also those that produce the least transmissible stages?
Are highly infectious parasites also those that induce greatest longevity reduction?
Are highly virulent parasites also those that manipulate mosquito biting rate the most?
Are slow-developing parasites also those that manipulate the most?
Trade-offs Sources of variation Key transmission traits selects for rapid insecticide resistance, but the evolutionary and epidemiological impact of evolved resistance traits in vectors on transmission traits of parasites is still not well understood (Alout, Djègbè, et al., 2014;Alout, Yameogo, et al., 2014;Rivero, Vezilier, Weill, Read, & Gandon, 2010). More work evaluating the consequences of insecticide-resistant mosquitoes on parasite transmission traits will help determine how changing vector traits can influence traits of their co-evolved parasites. In addition to physiological resistance, how mosquito behavioral resistance in response to LLINs and IRS affects parasite transmission traits is unclear. For example, some studies show that Anopheles mosquitoes can shift their host-feeding behavior from night-biting to day-biting following bed net introduction (Moiroux et al., 2012). As diel rhythm shapes mosquito immune responses Rund, Hou, Ward, Collins, & Duffield, 2011;Rund, O'Donnell, Gentile, & Reece, 2016), day-biting may also alter parasite infection prevalence and intensity. Finally, for interventions not yet deployed, such as late-life-acting insecticides or genetically modified mosquitoes, differences in the within-vector environment parasites experience will also provide potentially different selective forces. Whether parasites can evolve or plastically change transmission traits in response to these interventions needs to be evaluated if we are to responsibly deploy these technologies and prepare for possible evolutionary responses.
While all of these interventions are primarily aimed at and assessed by measuring vector traits, they may have important consequences for parasite evolution. Central to understanding how variation in parasite traits will ultimately influence our approach to control is quantifying how the transmission traits identified in the vectorial capacity equation vary by vector and parasite genotypes, and the plasticity of these traits in the face of selection. Any characterization of these effects should include an estimation of trait heritability across parasite generations, within-host environments, and external environments. Central to this will be measuring the responses of multiple traits to the withinvector environment to determine how trade-offs between them may constrain evolution and dictate parasite transmission.

ACKNOWLEDGEMENTS
We thank the NIH and the BBSRC for sponsoring the "Vector Behavior in Transmission Ecology" Research Coordination Network (VectorBiTE RCN) and the RCN steering committee for organizing the 2016 meeting in Florida. We thank Pierre Echaubard, Philip Birget, and Sarah Reece for their valuable comments that improved the manuscript. TL is funded by ANR grant 16-CE35-0007. The content of this manuscript is solely the responsibility of the authors and does not necessarily represent the official views of the NIH.