Divergence in threat sensitivity among aquatic larvae of cryptic mosquito species


  • Olivier Roux,

    Corresponding author
    1. Institut de Recherche pour le Développement, UMR IRD224-CNRS5290-UM1-UM2 MiVEGEC (Infectious Diseases and Vectors: Ecology, Genetics, Evolution, and Control), Montpellier, France
    2. Institut de Recherche en Sciences de la Santé (IRSS), Direction Régionale de l'Ouest (DRO), Bobo Dioulasso, Burkina Faso
    Search for more papers by this author
  • Abdoulaye Diabaté,

    1. Institut de Recherche en Sciences de la Santé (IRSS), Direction Régionale de l'Ouest (DRO), Bobo Dioulasso, Burkina Faso
    Search for more papers by this author
  • Frédéric Simard

    1. Institut de Recherche pour le Développement, UMR IRD224-CNRS5290-UM1-UM2 MiVEGEC (Infectious Diseases and Vectors: Ecology, Genetics, Evolution, and Control), Montpellier, France
    Search for more papers by this author


  1. Predation is a major evolutionary force driving speciation. The threat-sensitive response hypothesis predicts that prey adjust and balance the time spent on a costly antipredator response with other activities that enhance their fitness. Thus, prey able to develop an antipredator response proportional to risk intensity should have a selective advantage.
  2. Knowledge on how evolution has shaped threat sensitivity among closely related species exposed to different predation pressures is scarce, prompting investigations to better predict and explain its effect on communities.
  3. We explored and compared the antipredator response of aquatic mosquito larvae in three sibling species of the Anopheles gambiae complex, with contrasting larval biologies in Burkina Faso. Anopheles arabiensis and An. gambiae sensu stricto breed in temporary water collections where predator densities are low, whereas Anopheles coluzzii is able to thrive in permanent pools where the predation pressure is much higher. We hypothesized that the increase and decline of behavioural antipredator responses might differ between the three species over time. To test this hypothesis, progenies of field-collected mosquitoes were experimentally exposed to a range of soluble predation cues and their response was monitored for up to 48 h.
  4. The three species were all threat sensitive but their reaction norms differed. For the range of concentrations tested, An. coluzzii larvae gradually increased in antipredator response, whereas An. gambiae larvae readily displayed antipredator behaviour at low concentrations leading to a saturation of the response for high cue concentrations. An. arabiensis displayed a narrower reaction norm with low response intensity. Larval instars did not differ in their threat sensitivity. The antipredator behaviour of the three species waned after about 1 h of exposure. Early instars tended to express antipredation behaviour for longer than did older instars.
  5. This study provides information on how aquatic prey species with an aerial adult stage manage larval predation risk over time according to cue concentrations and suggests that different predation pressures might play a role as a disruptive selective force fostering habitat segregation and speciation within the An. gambiae complex. The evolution of phenotypic plasticity is further discussed in the light of divergent predation pressures.


Predation is a major evolutionary force driving community structure and species diversification (Nosil & Crespi 2006). In turn, plastic antipredator behavioural responses allow prey to cope with variable predation risks (West-Eberhard 1989; Agrawal 2001). The threat-sensitive response hypothesis predicts that prey adjust the time spent on a costly antipredator response according to the threat level to allow as much time as possible for foraging and development (Helfman 1989). Behavioural responses to a predation risk can take on different shapes, from a relative indifference wherein prey are unreceptive or poorly receptive to the threat to a hypersensitive response in which prey develop a high level of antipredator response for low threat levels. Intermediate plastic responses whereby prey adjust their antipredator behaviour proportionately to the threat intensity are believed to be more adaptive (Mathis & Vincent 2000; Mirza & Chivers 2003; Brown, Poirier & Adrian 2004; Brown et al. 2006; Ferrari et al. 2009). Hence, prey might fine-tune their antipredator response by adjusting the amount of time allocated to that response through trade-offs with other activities affecting its fitness, such as foraging and mating, minimizing costs (Lima & Dill 1990). Thus, prey able to develop an antipredator response as a function of risk intensity should be at a selective advantage (Helfman 1989; Mirza & Chivers 2003).

In aquatic systems, predation risk information is generally chemically mediated (Ferrari, Wisenden & Chivers 2010; but see Roberts 2012). Chemical cues can be related to different threat levels ranging from the perceived presence of a resting predator to realized predation acts (i.e. prey consumption or digestion). Cue concentrations may provide information on the amount of time that has elapsed since the last predation event and the distance and/or number and identity of the predator(s) and prey. Because higher cue concentrations are related to acute predation risk, prey should perceive greater risk with increasing cue concentrations (Chivers & Smith 1998; Kesavaraju, Damal & Juliano 2007; Fraker 2008; Ferrari et al. 2009). Cue concentrations may vary in time due to dilution or degradation (Sih, Ziemba & Harding 2000; Peacor 2006; Ferrari, Messier & Chivers 2008). The prey itself can be a source of variability in the interpretation of chemical information. For example, continuous exposure to high cue concentrations may induce a long period of food deprivation incompatible with prey fitness. Hence, in such a situation, prey might actively decrease their level of antipredator behaviour and redirect their efforts towards foraging and resource harvesting in order to survive and develop. This process requires prey to be able to adequately gauge predation risk, something that may involve a complex cognitive process in which they balance risks and benefits (Ferrari et al. 2010). The waning of antipredator behaviour can also be due to sensory habituation linked to an uncontrollable decrease in the sensitivity of the chemical receptors whereby prey passively lower their level of antipredator behaviour because of a physiological incapacity to assess risk (Ferrari et al. 2010).

If threat-sensitive responses have frequently been studied for a range of cue concentrations, their expression in their simplest form over time has rarely been investigated, making it difficult to generalize (but see Peacor 2006; Ferrari, Messier & Chivers 2008). In addition, knowledge on how evolution has shaped threat-sensitive responses by species exposed to different predation pressures is scarce, prompting investigations to find the best way to better predict and explain their effect on communities. Here, we investigated the ability of mosquito larvae within the Anopheles gambiae species complex to adjust their antipredator behavioural responses according to the threat level over time, both in their intensity and sensitivity.

The An. gambiae complex is particularly well suited to investigations on local adaptations and the evolution of antipredator behaviours because cryptic species within the complex are exposed to different predation pressures at the larval stage. The complex consists of seven closely related species that are reproductively isolated and genetically and eco-ethologically different, albeit morphologically indistinguishable (White, Collins & Besansky 2011). Anopheles gambiae s.s. and Anopheles arabiensis are the most widespread throughout sub-Saharan Africa, populating highly diverse environments and frequently sharing larval as well as adult habitats. Extensive genome scans have further uncovered an additional lineage split within the nominal species of the complex, An. gambiae s.s., leading to the description of the M and S ‘molecular forms’ (della Torre et al. 2001; White, Collins & Besansky 2011). Most recently, the taxonomic status of the M form was elevated to species level with the name Anopheles coluzzii, while the S form retains the original name An. gambiae (Coetzee et al. 2013). These two cryptic species are now widely recognized as biologically relevant, assortatively mating reproductive units (Lehmann & Diabaté 2008; White, Collins & Besansky 2011). Speciation and further radiation within the An. gambiae complex are believed to be the products of a process of ecological speciation putatively triggered by divergent selection at the larval stage (Coluzzi et al. 2002; Costantini et al. 2009). The system therefore provides excellent opportunities for improving our understanding of the process of divergent selection and the dynamics of ecological speciation (Costantini et al. 2009; White, Collins & Besansky 2011). The current hypothesis is that man-made hydrological schemes such as agricultural irrigation ditches and dams have created new ecological larval niches and thus new opportunities for specialization and expansion into marginal habitats (Ayala & Coluzzi 2005; Costantini et al. 2009).

Anopheles gambiae, the ancestral species, is widely distributed across sub-Saharan Africa, whereas the derived species An. coluzzii is found only in West and Central Africa (della Torre, Tu & Petrarca 2005; Lehmann & Diabaté 2008). However, although the adults are commonly found in sympatry, larvae might use diverse habitats. The two species are able to develop in temporary waters (e.g. puddles, quarries, ruts) during the rainy season where they frequently coexist with An. arabiensis larvae (Edillo et al. 2002; Gimonneau et al. 2012a). However, only An. coluzzii tends to exploit more permanent freshwater habitats (e.g. irrigated fields such as rice paddies, urban reservoirs) that persist across seasons (Gimonneau et al. 2012a; Kamdem et al. 2012). Hence, An. coluzzii is able to breed and transmit Plasmodium, which causes malaria, all year long in areas where permanent breeding opportunities exist, whereas An. gambiae and An. arabiensis typically die out at the onset of the dry season (Simard et al. 2000; Baldet, Diabaté & Guiguemde 2003).

In western Burkina Faso, the ecological segregation of larvae is strong with the total dominance of An. coluzzii that breeds in rice paddies, whereas An. gambiae and An. arabiensis predominate in the temporary waters in the surrounding savannas (Gimonneau et al. 2012a). In spite of recent studies showing that the temporal nature of the water habitat (i.e. temporary vs. permanent), predation and interspecific competition possibly contributed to this ecological divergence (Diabaté et al. 2005, 2008; Gimonneau et al. 2010), the way in which these selective forces interact and operate remains poorly understood. It has been shown, both in the field and through laboratory experiments, that An. coluzzii larvae from Burkina Faso are more capable of surviving an acute predation risk than An. gambiae (Diabaté et al. 2008; Gimonneau et al. 2010).

It is widely accepted that permanent water habitats shelter higher densities of predators than do temporary water habitats (Sunahara, Ishizaka & Mogi 2002; Diabaté et al. 2008). It follows that a better ability to gauge the predation risk and to produce an adequate antipredator response to balance other fitness traits should provide evident benefits in permanent water habitats (Tollrian & Harvell 1999). Moreover, it is assumed that heterogeneous and fluctuating environments should favour phenotypic plasticity with large reaction norms, whereas more stable environmental conditions should favour a loss of plasticity and produce flat reaction norms through the genetic assimilation of traits (Sultan & Spencer 2002; Hollander 2008). Furthermore, if species are highly mobile between heterogeneous environments, the gene flow should favour plasticity (Crispo 2008). In this context and keeping in mind the predation heterogeneity between permanent and temporary freshwater collections, how have Anopheles larvae evolved an antipredator behaviour across species of the complex and between freshwater habitats? Compared to An. gambiae and An. arabiensis, did An. coluzzii evolve a more plastic response with regard to its ubiquity within permanent and temporary habitats or did it evolve a selected narrower reaction norm with regard to the consistently high predation level in permanent habitats?

To address the aspect of these questions, we analysed the intensity and sensitivity (i.e. the ability of individuals to detect small differences in threat intensity) of the antipredator response of the larvae of An. coluzzii, An. gambiae and An. arabiensis from wild populations sampled in Burkina Faso. Because differences in plasticity between species are predicted to occur when the heterogeneity of their environment differs, we expected to observe divergences in cue sensitivity between species naturally exposed to different predation pressures (Brown et al. 2009). Our recent work showed that both An. coluzzii and An. gambiae were sensitive to chemical cues issuing from predation acts, while An. arabiensis seemed to respond only to physical cues betraying the presence of a predator (Roux, Diabate & Simard 2013). Nevertheless, because we found the lack of sensitivity to chemical cues in An. arabiensis surprising, we decided to include this species in the study to examine its sensitivity to high cue concentrations. Furthermore, extending the analyses to the sympatric sibling species, the results concerning An. arabiensis allowed us to further assess and compare the use of predation cues in mitigating antipredator responses among taxa within the An. gambiae complex with an increasing level of reproductive isolation (Ayala & Coluzzi 2005; Costantini et al. 2009; Simard et al. 2009).

Overall, we predicted that (i) the larvae would respond more readily and intensively to higher cue concentrations in keeping with the threat-sensitive response hypothesis; and (ii) the intensity of the antipredator behaviour would decrease over time as a result of stimulus degradation and/or habituation in the mosquito larvae. More precisely, we expected that (iii) An. coluzzii would display a large reaction norm and a proportional response to the threat reflecting its ability to thrive in both types of habitats and its greater sensitivity in gauging predation risk, respectively, whereas An. gambiae and An. arabiensis would express a constrained reaction norm (i.e. threshold-dependent), being naturally exposed to lower predation risk in temporary waters. (iv) Predictions about instar sensitivity were based on two mutually exclusive hypotheses. On the one hand, as prey–predator relationships are generally size dependant, late instars should be less subject to predation and so should be less sensitive and raise their antipredator behaviour faster than early instars. On the other hand, late instars are less conspicuous than early instars and thus should be more exposed to predation and could be more sensitive to chemical cues. Our predictions are summarized in Table 1.

Table 1. Specific predictions for the dynamics of behavioural antipredator responses in three sympatric species of the Anopheles gambiae complex in Burkina Faso
  Anopheles coluzzii An. gambiae Anopheles arabiensis
Larval ecology
HabitatPermanent (e.g. irrigated fields, dams) + Temporary (e.g. quarries, puddles)Temporary (e.g. quarries, puddles)Temporary (e.g. quarries, puddles)
Predation riskLow–highLowLow
Expected antipredator response
ShapeGraded (proportional to cue concentrations)Non-Graded (with threshold and/or ceiling)Non-Graded (with threshold and/or ceiling)
Width of reaction normLargeSmallSmall

Materials and methods

Insect Collections


Experiments were conducted with larvae obtained from wild gravid An. gambiae s.l. females collected while at rest in inhabited human dwellings in villages in south-western Burkina Faso (West Africa) during the 2011 rainy season (May–October). The village of Bama (11°23′14″N, 4°24′42″W), surrounded by 1200 ha of irrigated rice fields, predominantly shelters An. coluzzii all year round. In Soumousso (11°00′46″N, 4°02′45″W), a village in the humid savanna located 57 km away, An. arabiensis, An. coluzzii and An. gambiae are sympatric. In Bama, gravid females lay their eggs in the rice paddies, whereas in Soumousso, females only have access to temporary, rain-filled puddles and quarries that permit larvae to develop from June to November. The field-collected gravid females were placed individually in oviposition cups containing spring water and maintained under controlled conditions (28 ± 1 °C, 80 ± 10% relative humidity, 12 L/12 D). After oviposition, females were identified to species by routine PCR-RFLP based on segregating SNP polymorphisms in the X-linked ribosomal DNA intergenic spacer region (Fanello, Santolamazza & della Torre 2002). The larvae were reared in spring water exposed to ambient conditions in the insectaries (28 ± 1 °C, 80 ± 10% relative humidity, 12 L/12 D) and fed with Tetramin® Baby Fish Food ad libitum. The three populations used in our experiments consisted of An. gambiae s.l. females collected in Bama and all identified An. coluzzii, whereas An. gambiae and An. arabiensis were all obtained from Soumousso.


The backswimmer, Anisops jaczewskii (Hemiptera: Notonectidae), is the most abundant and widespread predatory bug in both permanent and temporary mosquito larval habitats in our study area (Diabaté et al. 2008; Gimonneau et al. 2010). Notonectids have been shown to be very efficient at reducing larval mosquito populations, and their impact on aquatic invertebrate communities has also been demonstrated (Blaustein 1998). Predators were collected in both locations, pooled (n = 80), fed daily ad libitum with a combination of the larvae of the three populations of Anopheles and maintained in the same controlled conditions as the mosquito larvae.

Experimental Design

Predation cue preparation

To investigate the effect of cue concentrations on the intensity and sensitivity of the antipredator behavioural response, we prepared a stock solution that was subsequently diluted to make concentrations of 100 (no dilution), 75%, 50%, 25%, 10%, 5%, 1% and 0% of the stock solution. The stock solution was prepared with the spring water in which the mosquito larvae had been reared for 5 days (hereafter ‘rearing water’). Four predators were kept in 120 mL of rearing water in plastic cups (Ø = 65 mm; h = 85 mm; hereafter ‘preparation cup’) together with 30 mosquito larvae offered as prey (a random mix of 2nd to 4th instars). The larvae were counted daily, and any missing larva was replaced. After 5 days, live larvae and predators were removed. This stock solution has been demonstrated to trigger an antipredator behaviour in larvae (Roux, Diabate & Simard 2013). For each mosquito population, instar and cue concentration tested, 10 stock solutions were simultaneously used and pooled to neutralize inter-replicate variation before sequential dilution. The solutions were diluted with rearing water. For each combination of population, instar and cue concentration, 20 aliquots of 60 mL were placed into plastic cups (Ø = 55 mm, h = 60 mm; hereafter ‘test cup’) and one larva was tested in each aliquot (=20 larvae per combination).

Behavioural tests

The larvae were fed 12 h before experimentation and used only once. The experiments began between 9:00 and 9:30 a.m. and ran for 48 h under controlled insectary conditions (see above). One 2nd, 3rd or 4th instar larva was gently placed in the middle of each test cup, and observations started after the larvae were allowed a 5-min period of acclimation. The antipredator behaviour in An. gambiae s.l. larvae was characterized by the larva positioning itself on the walls of the container (Gimonneau et al. 2012b; Roux, Diabate & Simard 2013). Data were therefore recorded as either ‘1’ for location on the walls (safe behaviour) or ‘0’ (larva on the surface, in the middle of the water column or at the bottom of the container = risky behaviour). The location of the larva was recorded after 5, 15, 30 min, 1 h, 1 h 30 min, 2 h, 2 h 30 min, 3 h, 3 h 30 min, 5 h, 6 h, 8 h, 10 h, 24 h, 36 h and 48 h leading to 320 observations per combination of population (n = 3), instar (n = 3) and cue concentrations (n = 8; =total of 23 040 observations of which 1440 were mutually independent because repeated measurements were taken for each larva over 48 h). For each population, instar and concentration combination, the 20 biological replicates (i.e. larvae) were tested simultaneously and data recordings were separated by c. 5 s between replicates. Three to four treatment combinations were randomly chosen and run at the same time, with data recordings being separated by c. 2 min.

Statistical Analysis

Pattern of the response to increasing cue concentrations

First, to determine the impact of cue concentrations on the antipredator behaviour, the larval location (Binary; wall vs. other) at t = 5 min was analysed using the generalized linear model (GLM) procedure with binomial errors and logit link. To test for the differences between populations and between instars, we considered populations, instars and the log of cue concentrations and all their interactions as fixed effects.

Secondly, to determine the pattern of the larval response to threat intensity (i.e. proportional or not), the larval location at t = 5 min within species subsets was analysed using the GLM procedure with binomial errors and logit link. Instars and the log of cue concentrations and their interactions were used as fixed effects. To test for linearity between threat intensity and antipredator responses, we included the quadratic term of cue concentrations. Significance of quadratic terms in the minimal model (P < 0·05) should reveal a curvilinear relationship. When curvilinearity was observed, data were additionally analysed with a nonlinear regression procedure with a four-parameter logistic S-shaped function (nls function in R with self-starting four-parameter logistic model function ‘SSfpl’). This procedure should allow both the upper and lower horizontal asymptotes, indicative of non-sensitivity or saturation, respectively, to be identified (Crawley 2007, p 678).

For model selection, we used the stepwise removal of terms, followed by likelihood ratio tests. Term removals that significantly reduced explanatory power (P < 0·05) were retained in the minimal adequate model (Crawley 2007).

Dose–response analyses

Because previous GLM analyses did not reveal an instar effect (see Results), data on instars were pooled in the dose–response analysis. The concentration inducing an antipredator behaviour in 50% of the larvae tested (Effective Concentration: EC50) during the first observation (at t = 5 min) was hence computed at the population level using the dose.LD50 function in the doBy package in R. An anova followed by a Tuckey's post hoc test was used to compare the EC50 between species.

Effect of time on the waning of the antipredator response

The analysis of changes in larval location over time was carried out using the generalized linear mixed model (GLMM) procedure with binomial errors and logit link (glmer function in the lme4 package). Populations, instars and length of exposure [log(time)] and interactions between populations and time as well as between instars and time were used as fixed effects. Individuals were assigned as a random effect due to repeated measures, and the factor ‘time’ was nested in concentration levels.

To test for a curvilinear relationship between time and the antipredator response, we included the quadratic term of time. As the introduction of the quadratic term into the model was highly significant (= 67·223; d.f. = 1, < 0·001), we investigated the curvilinearity of the relationship within species and instar subsets. When curvilinearity was observed in the subsets, data were analysed with a curvilinear regression procedure with a four-parameter logistic S-shaped function (SSfpl) to identify significant asymptotes (Crawley 2007, p 678). Models were simplified as described above. All analyses were conducted using the R statistical package (version 2·12·1; R Development Core Team 2011).


Pattern of the Antipredator Response to Increasing Cue Concentrations

The analysis of larval position at t = 5 min showed that the three species were all sensitive to soluble chemical cues reflecting a predation threat. The higher the cue concentrations, the higher the level of antipredator behaviour recorded, highlighting threat sensitivity (Fig. 1, Table 2). There was no significant effect of instars on the antipredator response to increasing cue concentration. However, species adjusted their antipredator behaviour differently as the cue concentration increased (i.e. significant interaction effect). Anopheles arabiensis (intercept in Table 2) increased its antipredator behaviour more slowly than did An. coluzzii and An. gambiae (Fig. 1, Table 2). Anopheles coluzzii and An. gambiae did not differ from each other in the rate at which they increased their antipredator response according to cue concentration, although An. gambiae displayed a significantly higher level of antipredator behaviour than did An. coluzzii (Tukey's post hoc test: z-value = 2·82; = 0·013, Fig. 1).

Table 2. Effect of cue concentrations on larval antipredator behaviour at t = 5 min
 EstimateSEZ-value P
  1. Results for the minimal adequate model obtained by a GLM procedure with binomial errors and a logit link function.

Anopheles coluzzii 0·150·200·77NS
Anopheles gambiae 0·710·193·65<0·001
3rd instar−0·150·14−1·02NS
4th instar0·120·150·81NS
An. coluzzii × log(concentration)0·190·062·99<0·01
An. gambiae × log(concentration)0·250·073·68<0·001
Figure 1.

Proportion of larvae exhibiting antipredator behaviour after 5 min of exposure to increase cue concentrations. Anopheles coluzzii: red triangles and red line; Anopheles gambiae: black diamonds and black line and; Anopheles arabiensis: blue circles and blue line. Curves are predicted lines obtained from the GLM. The curve for An. gambiae is obtained from the model including the quadratic term of log(concentration).

In analyses by species subsets, we found a significant effect of the quadratic terms for log(concentration) only in An. gambiae (Likelihood ratio test: G1 = 4·24; = 0·039) suggesting a curvilinear relationship between the anti-predator response and cue concentrations in this species only (An. arabiensis: χ² = 2·23; d.f. = 1; = 0·13 and An. coluzzii: χ² = 1·19; d.f. = 1; = 0·27). The curvilinear regression revealed that the antipredator behaviour levelled off at higher concentrations (upper asymptote: t-value = 6·812; d.f. = 4; = 0·002).

Larval Sensitivity to Predation Cues: A Dose–Response Analysis

A dose–response analysis revealed that the three species differed in their EC50 (anova: F2,6 = 9·13; = 0·015 and Table 3). Anopheles arabiensis was the less sensitive compared to An. coluzzii (Tukey's post hoc test: z-value = 4·51; = 0·04) and to An. gambiae (Tukey's post hoc test: z-value = 5·73; = 0·01). Anopheles coluzzii and An. gambiae did not differ from each other (Tukey's post hoc test: z-value = 1·21; = 0·68).

Table 3. Mean EC50 for chemical predation cue concentrations inducing location on the walls after 5 min of exposure
 EC5095% CI
  1. EC50, Mean effective concentration at 50%; CI, Confidence interval. Different letters indicate significant differences at < 0·05 (anova: F2,6 = 9·13; = 0·015; followed by a Tukey's post hoc test for pairwise comparisons).

  2. a

    Values obtained from the model fitted with quadratic term for log(concentration).

Anopheles arabiensis 7·6 a2·127·2
Anopheles coluzzii 1·9 b1·03·5
Anopheles gambiae a 0·7 b0·51·2

Waning of the Antipredator Response

The results of the GLMM showed that time has a negative effect on the expression of the antipredator behaviour as the proportion of larvae on the walls decreased significantly over time (Table 4, Figs 2 and 3). Species–time interactions showed that the different species managed their antipredator behaviour level differently over time. Anopheles coluzzii and An. gambiae had similar rates of decline in their antipredator behaviour (similar interaction with log time in Table 4). Nevertheless, An. gambiae expressed a higher level of antipredator behaviour during the entire waning process (Tukey's post hoc test; z-value = 4·87 and < 0·001, Fig. 2 and Table 4). Anopheles arabiensis decreased its antipredator behaviour more slowly over time than did the two other species (Fig. 2, Table 4). The inclusion of the quadratic term of ‘time’ into the minimal model was highly significant (G1 = 67·223; < 0·001) as it was in the analysis of the subsets, revealing that the waning in all species was curvilinear (An. coluzzii: G1 = 511·8; < 0·001; An. gambiae: G1 = 32·15; < 0·001; An. arabiensis: G1 = 278·34; < 0·001). The curvilinear regression obtained with the SSfpl function revealed that the level of the antipredator behaviour in the three species stayed constant until it reached a threshold in time at which it started to decrease (significant asymptotes for low value of log(time): An. coluzzii:= 38·09; d.f. = 12; < 0·001; An. gambiae:= 68·12; d.f. = 12; < 0·001; An. arabiensis:= 54·55; d.f. = 12; < 0·001). This threshold was about 1 h after the beginning of exposure for all three species. The three species also had an asymptote for a high value of log(time) meaning that they reached their baseline behaviour before the end of the experiment (about 12 h after the beginning of exposure for An. gambiae and An. arabiensis and, later, for An. coluzzii, Fig. 2) with significant asymptotes for a high value of log(time) (An. coluzzii:= 6·48; d.f. = 12; < 0·001; An. gambiae:= 28·79; d.f. = 12; < 0·001; An. arabiensis:= 28·75; d.f. = 12; < 0·001).

Table 4. Effect of time on larval antipredator behaviour
 EstimateSEZ-value P
  1. Results for the minimal adequate model obtained by a GLMM procedure with binomial errors and a logit link function.

Anopheles coluzzii 1·180·138·92<0·001
Anopheles gambiae 1·820·1313·38<0·001
3rd instar−0·440·13−3·24<0·01
4th instar0·010·140·07NS
An. coluzzii × log(time)−0·180·02−7·88<0·001
An. gambiae × log(time)−0·190·02−8·46<0·001
3rd instar × log(time)−0·010·02−0·06NS
4th instar × log(time)−0·070·02−3·03<0·01
Figure 2.

Effect of time on the proportion of larvae exhibiting antipredator behaviour. Anopheles coluzzii: red triangles and red line; Anopheles gambiae: black diamonds and black line and; Anopheles arabiensis: blue circles and blue line. Curves are predicted lines obtained from a curvilinear regression with the SSfpl function in R.

Figure 3.

Effect of time on the proportion of larvae exhibiting antipredator behaviour in the three instars. 2nd instars: circles and dotted line; 3rd instars: triangles and dashed line and; 4th instars: diamonds and full line. Curves are predicted lines obtained from a curvilinear regression with the SSfpl function in R.

The GLMM also revealed significant interactions between instars and time (Table 4). The level of the antipredator behaviour of 4th instar larvae decreased significantly faster than for both 2nd and 3rd instar larvae over time (Table 4, Fig. 3). The waning rates of the antipredator responses of 2nd and 3rd instars were similar. All instars presented a curvilinear waning of their antipredator behaviour (2nd instars: G1 = 28·79; < 0·001; 3rd instars: G1 = 9·12; = 0·002; 4th instars: G1 = 341·6; < 0·001). The curvilinear regression revealed that the level of the antipredator behaviour in the three instars stayed constant until it reached a threshold in time at which it started to decrease (significant asymptotes for a low value of log(time): 2nd instars: = 93·79; d.f. = 12; < 0·001; 3rd instars: = 39·04; d.f. = 12; < 0·001; 4th instars: = 43·85; d.f. = 12; < 0·001). This threshold was about 1 h after the beginning of exposure for 2nd instar larvae and about 30 min for 3rd and 4th instar larvae. The three instars also have an asymptote for a high value of log(time) meaning that they reached their baseline behaviour before the end of the experiment (about 12 h after the beginning of the exposition in all instars, Fig. 3) with significant asymptotes for a high value of log(time) (2nd instars: = 28·86; d.f. = 12; < 0·001; 3rd instars: = 17·28; d.f. = 12; < 0·001; 4th instars: = 16·11; d.f. = 12; < 0·001).


Knowledge of how prey gauge and manage a predation threat is important to the understanding of how predation pressures drive species divergence and structure communities. However, evidence for such mechanisms is scarce because, from an ecological standpoint, they act over the long term. Our research on incipient species offers an opportunity to report the early outcomes of such mechanisms and provides empirical support for the hypothesis that divergence in predation pressure may drive ecological divergence in the An. gambiae complex.

The results of this study provide estimation of how aquatic prey species with an aerial (terrestrial) adult stage manage larval predation risk both over time and according to cue concentrations. We show that An. gambiae s.l. larvae are threat sensitive, increasing their antipredator responses as predation cue concentrations increase. The antipredator behaviour waned over time for all three species. However, the different species and instars displayed different rates and intensities in their antipredation behaviour highlighting divergent adaptations to predation risk.

Threat Sensitivity

Anopheles arabiensis larvae were less sensitive than both An. coluzzii and An. gambiae larvae. This might explain why in a previous study, where chemical cue concentrations were much lower (about less than the 25% dilution to the present work, Roux, Diabate & Simard 2013), An. arabiensis larvae appeared to rely mostly on physical rather than chemical cues to assess and mount an antipredator response. The response was gradual (proportional to the threat) but constrained to a narrow reaction norm because of a low sensitivity to high cue concentrations (i.e. weak slope in the range of concentrations tested). An. gambiae larvae, which share temporary waters with An. arabiensis, expressed an antipredator response that was higher in this range of concentrations. The response was also gradual but with a stronger slope, leading the antipredation response level to reach its maximum (i.e. c. 100% of the larvae on the walls) for moderate to high cue concentrations (i.e. the upper asymptote was reached for concentrations above 50%). The antipredator response of the two species is then limited to a narrow reaction norm but for two different reasons. In contrast, An. coluzzii larvae living in permanent waters gradually engaged in an antipredator response across the entire range of cue concentrations tested and never reached a plateau in our experimental conditions.

The weak sensitivity of An. arabiensis to low cue concentrations (this study; see also Roux, Diabate & Simard 2013) may actually not be maladaptive (Brown et al. 2001; Mirza & Chivers 2003). Indeed, Brown et al. (2006) showed that cichlids had covert and imperceptible changes in behaviour when exposed to subthreshold cue concentrations. In this situation, the prey waits for complementary risk indicators such as visual cues or novel predation cues to mount a complete antipredator response. Such a threshold, when associated with a subsequent graded response, could be adaptive and may limit the cost of antipredator behaviour allowing prey to feed at a normal or subnormal rate until additional cues confirm the risk. Weak sensitivity to low cue concentrations might therefore provide a selective advantage in temporary waters, allowing larvae to maintain optimal foraging and growth and thus escape early from short-lived habitats (Lima & Dill 1990). In Burkina Faso, An. arabiensis is found in temporary waters and occurs in higher numbers at the end of the rainy season when temporary ponds are drying out (Gimonneau et al. 2012a). However, in East Africa, An. arabiensis is found in permanent fresh water characterized by high predatory pressure suggesting the existence of different ecotypes (Coluzzi et al. 2002) or different local environmental pressures that may or may not allow permanent habitats to be colonized.

Previous transplantation experiments have shown that An. coluzzii out-competes An. gambiae in rice paddies in the presence of predators (Diabaté et al. 2008). Our results show that this higher performance by An. coluzzii could be due to better threat sensitivity to chemical predation cues. Indeed, our results show that at a high risk level, such as is found in permanent fresh water (Diabaté et al. 2008), An. gambiae mounted a maximal antipredator response, whereas An. coluzzii displayed a lower and gradual level of antipredator behaviour and then potentially spent more time feeding. Such differences may explain the ecological divergence between the two species in which An. coluzzii takes advantage of permanent fresh water (Diabaté et al. 2005, 2008).

Waning of Antipredator Behaviour

Fraker (2008) found that young Rana clamitans tadpoles exhibited a higher level of antipredator behaviour and recovered their baseline activity after a longer time-lag than did their older and bigger counterparts. In our study, the baseline recovery had the same time-lag in the three instars but the rate at which the 4th instar larvae resumed their antipredator behaviour was significantly higher than the 2nd and 3rd instars. The 2nd instars maintained their antipredator behaviour for longer than did older instars. This suggests that early instars (i.e. smaller individuals) perceive a high risk level for a longer period after their initial exposure to predation cues than do older (larger) instars, pointing towards a prey size effect. Notonectids are known to engage in size-selective predation and to structure aquatic communities by reducing the density of larger individuals (Scott & Murdoch 1983; Blaustein 1998). Larger individuals may be preyed upon as a consequence of their shorter period of sensitivity over time and/or because they are more conspicuous than smaller prey.

The decrease in a response to predation cues over time is common to most organisms (Lima & Dill 1990). Several non-exclusive hypotheses can be suggested to explain the waning of an antipredator response. First, chemical cues are not stable over time. Compounds can be quickly diluted or degraded by micro-organisms and ultraviolet (UV) radiation. Such a loss of biological activity can occur from within a few hours to several days (Peacor 2006; Ferrari, Messier & Chivers 2008). Moreover, when predators move around, low concentrations of predation cues are less reliable indicators of acute predation risk than are higher concentrations (Fraker 2008). In our study, the waning of an antipredator response took about 12 h. However, the cue degradation process is probably more rapid in natural conditions, whereby biological (i.e. microbial) and abiotic (e.g. UV radiation, adsorption) processes contribute to the destruction of cues (Ferrari, Messier & Chivers 2008).

Sensory habituation cannot be excluded in explaining the waning of the antipredator response. However, the 2nd instar larvae seemed to maintain their maximal response longer than did later instars. In the case of sensory habituation, this would imply that the turnover of odorant binding proteins and the reactivation of the sensory receptors of 2nd instar larvae are faster than for later instars allowing 2nd instars to be receptive longer. This difference between early and late instars allows us to partially exclude the dominant effect of cue degradation in the waning of antipredator responses.

Finally, according to the risk allocation hypothesis, prey might actively resume foraging in the face of a predation threat because long periods of food deprivation may seriously hamper growth and survival, therefore speeding up the waning of antipredator responses. It is therefore possible that prey rely on additional cues to prompt such behaviour.

Knowing which threat level induces an antipredator response is good, but, in addition, knowing for how long and why the response is maintained is better. Most studies, like this one, dealing with threat-sensitive responses investigated the effect of independent exposure to a range of cue concentrations. However, few of them attempted to quantify the dynamics of the waning of the behavioural response over time (but see Peacor 2006; Ferrari, Messier & Chivers 2008). Identifying and quantifying the chemical compounds involved in an antipredator response as well as their degradation derivatives over time hence presents a challenge for future research. Indeed, if low cue concentrations are not informative because it is impossible to distinguish between a recent small predation event and a larger but past event, detecting degradation derivates could be a selective advantage. By comparing the concentrations of both the original predation cue and its derivatives, prey could obtain more complete information on the temporal risk and then gradually adjust their antipredator behaviour over time.

Evolution of Phenotypic Plasticity

Phenotypic plasticity provides a selective advantage in fluctuating environments. However, the impact on fitness traits remains to be tested in order to determine which patterns of response to the threat are the best adapted. Nevertheless, Diabaté et al. (2008) have shown that in the presence of predators, An. coluzzii out-competes An. gambiae. Its response is thus probably the closest to the optimal, allowing it to best balance antipredator behaviour with other fitness traits. Here, phenotypic plasticity was shown in all species, whatever their habitat of origin, suggesting that in the two habitats, larvae are exposed to fluctuating levels of predation pressure. Permanent freshwater habitats are, by definition, more stable than temporary ones and thus should favour a loss of plasticity and generate flat reaction norms (Sultan & Spencer 2002; Hollander 2008). However, as expected, our results showed that An. coluzzii displayed a highly plastic response proportional to the threat level, suggesting an acute response over a large range of risk. Such a high phenotypic plasticity could be explained by two non-exclusive hypotheses. First, An. coluzzii is found in both temporary and permanent habitats suggesting the potential exposure of natural mosquito populations to heterogeneous environments with a high dispersal rate between habitats. Secondly, Notonectids are winged aquatic insects at both the adult and nymphal stages and thus, as predators move between habitats in search of prey, their densities can fluctuate over time. Phenotypic plasticity in an antipredator response might have been acquired and evolved from the ancestral species, An. gambiae, which also displays a large reaction norm, with the same slope as An. coluzzii but with a different intensity. However, An. gambiae is restricted to temporal habitats and therefore experiences lower predation risk; its strong antipredator behaviour to high concentration cues may prevent it from developing in permanent habitats due to a poor match with other traded-off fitness traits.

The reaction norm of An. arabiensis is narrower than the two other species. As An. arabiensis is mainly found at the end of the rainy season in our study area (Gimonneau et al. 2012a), we hypothesise that its abundance may be negatively correlated with predator density all year around, allowing larvae to fully develop only at the end of the rainy season when temporary ponds are drying up.


It is well known that high levels of predation pressure over generational time may determine the intensity of antipredator responses (Giles & Huntingford 1984). However, little is known about the impact of such pressure on threat sensitivity. Brown et al. (2009) have shown that Trinidadian guppy populations exposed to a high level of predation exhibit a more intense and more proportional antipredator response than do populations exposed to low predation pressure. Our results agree with these findings as An. coluzzii mounts a more proportional antipredator response than either An. gambiae or An. arabiensis larvae. Nevertheless, the similarity of the rates at which An. coluzzii and An. gambiae increased their antipredator response is consistent with An. gambiae being considered ancestral to the derived An. coluzzii and is in line with previous scenarios of ecological speciation within the An. gambiae complex (Ayala & Coluzzi 2005; Costantini et al. 2009). Hence, our results provide further evidence that differences in the predation pressure exerted in permanent and temporary fresh water likely contributed to disruptive selection promoting ecological divergence between An. coluzzii and An. gambiae and between some species of the An. gambiae complex (Diabaté et al. 2008; Roux, Diabate & Simard 2013).


We would like to thank Boubacar Nikiema for the PCR, Thierry Lefèvre and Carlo Costantini for helpful discussions and Andrea Yockey-Dejean for proofreading the paper. We also thank two anonymous Referees and the Associate Editor for valuable comments. Financial support for this study was provided through IRD/MIVEGEC in-house funding. OR received financial support through a post-doctoral fellowship from the IRD.