Predation yields greater population performance: What are the contributions of density- and trait-mediated effects?

Abstract

1. Population responses to extrinsic mortality can yield no change in number of survivors (compensation) or an increase in number of survivors (overcompensation) when the population is regulated by negative density-dependence. This intriguing response has been the subject of theoretical studies, but few experiments have explored how the source of extrinsic mortality affects the response.

2. This study tests abilities of three functionally diverse predators, alone and combined, to induce (over)compensation of a prey population. Larval Aedes aegypti (Diptera: Culicidae) were exposed to predation by Mesocyclops longisetus (Crustacea: Copepoda), Anopheles barberi (Diptera: Culicidae), Corethrella appendiculata (Diptera: Corethrellidae), all three in a substitutive design, or no predation.

3. The number of survivors to adulthood, female size and development time, and a composite index of performance (r′) were analysed. Predator treatment did not have a significant effect on total number of survivors, nor on number of males, suggesting mortality by predation was compensatory. Predation significantly affected number of female survivors, with a trend of more females produced with predation, though no post hoc tests were significant. Predation significantly increased female development rate and r′ relative to no-predator control.

4. A sensitivity analysis indicated that the change in the number of female adults produced was the largest contributing factor to the differences in r′ among cohorts. While predation did not significantly increase overall production of adults, it did release survivors from density-dependent effects sufficiently to increase population performance. This study provides an empirical test of mechanisms by which predation may yield positive effects on a population of victims, a phenomenon predicted to occur across many taxa and food webs.

Introduction

Extrinsic mortality (e.g., due to enemies, harvesting, or other human interventions) impinging on populations has traditionally been predicted to interact additively with intrinsic mortality sources, with greater levels of extrinsic mortality leading to reductions in number of survivors, population density, and population rate of increase. However, populations regulated by strong negatively-density dependent intrinsic effects may demonstrate counter-intuitive responses. By initially reducing the population density, extrinsic mortality may reduce detrimental density-dependent effects on the survivors. This may result in the production of the same number (compensation) or a greater number (overcompensation) of survivors in the next life stage as would have been the case in the absence of extrinsic mortality. Extrinsic mortality that results in an increase in the equilibrium density of a population has been termed the ‘hydra effect’ (Abrams & Matsuda, 2005).

Compensatory and overcompensatory responses to mortality have been demonstrated in both field and laboratory studies (Nicholson, 1954; Agudelo-Silva & Spielman, 1984; Washburn et al., 1991; Moe et al., 2002; Cameron & Benton, 2004; Zipkin et al., 2008; Schröder et al., 2009; Weber et al., 2016; McIntire & Juliano, 2018; Neale & Juliano, 2019). Numerous theoretical studies have explored the mechanisms that can produce this phenomenon (Abrams, 2009; Schröder et al., 2014; see De Roos et al., 2007 for theoretical examination of biomass overcompensation). The timing of extrinsic mortality relative to the onset of density-dependent effects is predicted to influence the likelihood of overcompensation, with mortality occurring prior to density-dependence postulated to lead to overcompensation and increased population sizes (Jonzen & Lundberg 1999, Abrams 2009, Pardini et al. 2009) This hypothesis, known as the ‘temporal separation of mortality and density dependence hypothesis’, was recently supported in an empirical study on container mosquitoes (McIntire & Juliano, 2018). Furthermore, the extrinsic mortality rate (i.e., proportion killed) is expected to have an effect on whether additive, compensatory, or overcompensatory effects are observed (Sandercock et al., 2011; Neale & Juliano, 2019) and these effects appear to be related to competitive abilities of the species involved (Neale & Juliano, 2019). We have found only a few published studies empirically examining the mechanisms of overcompensation and the hydra effect (Slobodkin & Richman, 1956; Cameron & Benton, 2004; Fryxell et al., 2005; Schröder et al., 2009; Sandercock et al., 2011; McIntire & Juliano, 2018), and more empirical studies are needed to determine the conditions under which it occurs.

Predation is a common natural source of extrinsic mortality for animal populations, and mathematical models predict that both density- and trait-mediated effects of predators can lead to the hydra effect in prey populations (Abrams, 2009; Cortez & Abrams, 2016). However, only two of the aforementioned empirical examples included predation as a mortality source, both in container mosquito systems (Nannini and Juliano 1998, Neale & Juliano 2019). In natural food webs, many prey populations are preyed upon by multiple predators (Sih et al., 1998) often with different hunting behaviors. Understanding how predation by multiple predator species differs from a single predator species in production of overcompensation and impact on prey population performance is critical to predicting prey responses to predation in nature. However, we have found no published studies examining the effects of multiple predators on overcompensation and the hydra effect.

Increasing predator functional or phylogenetic diversity can in principle result in emergent multiple predator effects (MPE’s), which are characterized by non-additive effects of combinations of different predators (i.e., risk reduction or risk enhancement) on prey populations. Such non-additive effects result in an increase or decrease in predation rates relative to that observed with single predators (Sih et al., 1998; Schmitz, 2007; Bruno & Cardinale, 2008; Greenop et al., 2018). A recent meta-analysis of studies on terrestrial arthropod systems found predator functional diversity was more important in determining the outcome on prey than predator phylogenetic diversity (Greenop et al., 2018). The degree of overlap in predator foraging spatial domains and hunting modes among different predator functional groups is predicted to determine if MPE’s will occur and whether they will be risk enhancing or reducing, as foraging domain and hunting mode determine the likelihood of intraguild predation and availability of prey refugia (Schmitz, 2007). A functionally diverse assemblage of predators may yield different degrees of overcompensation (including absence of overcompensation) from those produced by single predator species in multiple ways. By changing intensity of extrinsic mortality through risk enhancement or reduction, functionally diverse predators may have a direct, density-mediated effect likelihood and intensity of overcompensation, which is dependent on the intensity of mortality (Sandercock et al., 2011; Neale & Juliano, 2019). A functionally diverse assemblage of predators may also change the pattern of mortality compared to that produced by single predators by changing the temporal partitioning of mortality (i.e., temporal niche complementarity), which is an important factor in determining whether mortality will be overcompensatory (Abrams, 2009; McIntire & Juliano, 2018). Finally, a diverse assemblage of predators may alter the traits of the individuals that become victims, and this too can alter the likelihood of overcompensation (e.g., Abrams 2009), which depends on extrinsic mortality taking the greatest toll on individuals most likely to die due to density dependent effects (e.g., slow developing poor competitors).

As a first step toward understanding the consequences of predator functional diversity for prey overcompensation, we compare the population-level effects of functionally diverse predators to those of constituent single species of predators in a substitutive design. We hypothesize that predation from a single species occurring early in the development of a prey population subject to negative density-dependence leads to overcompensatory mortality, and predation from multiple predator species may increase or decrease the strength of overcompensation due to risk enhancement or reduction.

Material and Methods

Study System

We tested our hypotheses using Aedes aegypti (Diptera: Culicidae) as the prey species. The complex life-cycle and negatively density dependent survival of the larval stage that is largely associated with resource limitation (e.g., Dye 1984, Juliano 1998, Walsh et al. 2013; reviewed by Juliano 2009) are consistent with the assumptions of models of hydra effects (De Roos et al., 2007; Abrams, 2009). Overcompensation has been demonstrated in this species (Neale & Juliano, 2019) as well as its congeners, A. sierrensis (Washburn et al., 1991), A. albopictus (McIntire & Juliano, 2018), and A. triseriatus (Neale & Juliano, 2019). The predators we included were Mesocyclops longisetus (Crustacea: Copepoda), Anopheles barberi (Diptera: Culicidae), and Corethrella appendiculata (Diptera: Corethrellidae). All three are efficient predators of small Aedes larvae under similar conditions (Marten et al., 1994; Nannini & Juliano, 1998; Alto et al., 2009). All are size-selective, feeding primarily on first and second instar larvae (Nannini & Juliano, 1998; Soumare et al., 2004; Alto et al., 2009). This size-selectivity is ideal for inducing overcompensation, as it concentrates the mortality early in prey development and potentially separates mortality due to predation temporally from the density-dependent effects, which are expected to increase as immatures grow. The three predators differ in their within-habitat hunting locations, henceforth referred to as habitat domain (Schmitz, 2007). Mesocyclops longisetus swims throughout the water column, lunging at prey when it passes within ~ 1mm (Marten & Reid, 2007). Anopheles barberi sits in the surface tension and ambushes larvae as they surface (Clements, 1992). Corethrella appendiculata primarily sits at the bottom of the water column and preys on mosquito larvae when they browse in the substrate (Kesavaraju et al., 2007). Anopheles barberi demonstrates a non-asymptotic functional response indistinguishable from a type 1 functional response (Nannini & Juliano, 1998), while C. appendiculata and Mesocyclops demonstrate type 2 functional responses (Griswold & Lounibos, 2005; Cuthbert et al., 2018). Predators exhibiting type 1 and 2 functional responses do not impose negatively density-dependent mortality on prey, at least at short time scales that do not allow numeric responses by the predator (Taylor, 1984).

Organism Collection

Aedes aegypti used in this study were from a laboratory colony originating from pupae and larvae collected from Vero Beach, FL approximately 1 year before the start of this experiment. To maintain the colony, larvae were reared in plastic pans at 25°C and provided bovine liver powder. Adults were given a constant supply of 20% sucrose solution, and blood meals were provided from anaesthetized guinea pigs (IACUC# 842043).

Mesocyclops longisetus were from a laboratory colony maintained at Illinois State University in Normal, IL, which originated from a colony maintained at the Florida Medical Entomology Laboratory (FMEL) in Vero Beach, FL. Corethrella appendiculata were 4^th instars field collected from tree holes on the FMEL grounds. Larvae were housed in water from the tree holes at 25°C until the start of the experiment. Anopheles barberi were collected as larvae in rain-filled buckets at Parklands Merwin Nature Preserve near Lexington, IL. To maximize the number of late-instar larvae available at the start of the experiment, 3^rd and 4^th instars were housed at 22°C to delay pupation, while 1^st and 2^nd instars were housed at 25°C.

Experimental Setup

Four days prior to the beginning of the experiment, 500 ml plastic containers were filled with 400 ml reverse osmosis water, 1 g dried live oak (Quercus virginiana) leaves collected from Vero Beach, FL, 0.05 g dried decorated crickets (Gryllodes sigillatus) from a colony maintained at Illinois State University, and 100 μl microbial inoculum, from rain-filled buckets in Merwin Nature Preserve, Lexington, IL. Lids were placed on the containers with holes punched for ventilation. The containers were housed in an environmental chamber at 25°C until the beginning of the experiment to allow the establishment of a microbial community to serve as food resources for mosquito larvae.

Aedes aegypti eggs were hatched 24 hours prior to the start of the experiment by placing strips of egg papers in 14.79 ml glass vials containing 0.4g/l Difco™ nutrient broth (Becton, Dickinson and Company, Sparks, MD) at 25°C. At the start of the experiment, hatchling larvae were rinsed in ultrapure water and 150 were placed in each experimental container (n=15). Containers were randomly assigned one of five predator treatments: no predator, M. longisetus, C. appendiculata, A. barberi, and diverse. The experiment used a substitutive design wherein single-species treatments received three predator individuals and the diverse treatment received one individual of each predator species. Only non-gravid adult female M. longisetus, 4^th instar C. appendiculata, and 3^rd and 4^th instar A. barberi were used. Multiple A. barberi instars were included because of a limited number of larvae available. Because 4th instars consume greater numbers of Aedes prey than 3^rd instars (Nannini & Juliano, 1998), the A. barberi treatment received one 4^th and two 3^rd instars, and diverse containers each received a 4^th instar. Once prey and predators were added to each container, they were placed in an environmental chamber set to 25°C and a 14:10 light:dark photoperiod.

Containers were checked daily for A. aegypti pupae and survival of predators. A. aegypti pupae were removed, placed in 0.92 ml glass vials with cotton stoppers, and returned to the environmental chamber, and any dead or missing predators were replaced. All predators were removed on day 6 because the replacement stock of A. barberi larvae was depleted. Due to the size selectivity of the three predators and the size and developmental stage of prey by day 6, only minimal amounts of predation would have occurred if the predators remained (Nannini & Juliano, 1998; Soumare et al., 2004; Alto et al., 2009). On days 16 and 30, 0.5 g dried live oak leaves and 0.025g dried decorated crickets were added to replenish resources for bacteria and fungi that are the food of A. aegypti.

Pupae were checked daily for eclosion. Water was removed from vials containing adults and the vial was placed in a drying oven at 70°C for >48 hours. All individuals reaching adulthood were counted as survivors. Female wings were dissected and photographed with a digital camera, and wing lengths were measured in Image J 1.51. The final adult eclosed on day 36 of the experiment.

Index of Performance

Using data collected on female survivorship to adulthood, development time to adulthood, and predicted fecundity based on wing length, Livdahl & Sugihara’s (1984) index of performance r′ was calculated for each container (Equation 1). This index synthesizes information on these variables in a manner analogous to calculations of net reproductive rate (R₀) and cohort generation time (T_c) from a cohort life table. This index was designed to provide an approximation of cohort rate of change (dN/Ndt) (Livdahl & Sugihara 1984) and is strongly and significantly correlated with dN/Ndt calculated from a full life table (Chmielewski et al., 2010; Chandrasegaran & Juliano, 2019). We use it to assess how predator treatments affected population growth for experimental cohorts in each container. We infer that cohorts of 150 larvae from a container are farther below equilibrium density for their environment if their index of performance is farther above 0. It is important that we use the initial density of 150 larvae as our initial cohort size in our estimate of cohort rate of increase because that approach means estimated rate of increase (r′) includes the impact of mortality, so that purely additive mortality would be expected to reduce estimated rate of increase.

Equation 1 Livdahl and Sugihara’s (1984) index of performence

$$r^{\prime }=\frac{\ln \left[\left(\frac{1}{N_0}\right){\Sigma }_x{A}_{x}f\left(w_x\right)\right]}{D+\left[{\Sigma }_{x}x{A}_{x}f\left(w_x\right)/{\Sigma }_x{A}_{x}f\left(w_x\right)\right]}$$

N₀ is the initial number of females (assumed to be 50% of the initial 150 larvae), A_x is the number of new females emerging on day x, w_x is the mean wing length of new females emerging on day x, and D is the estimated days between female eclosion and oviposition, (estimated to be 12 days; Grill & Juliano 1996). Production of female offspring in the first clutch ƒ(w_x) was estimated as a function of wing length using the regression eggs vs. wing length provided by Chandrasegaran & Juliano (2019), multiplied by 0.5: f(w_x)= 0.5(2.5w_x³ - 8.616). When there were missing values for day of emergence (n = 3 of 216 females), we substituted the median day of emergence for the entire experiment (=18). When there were missing values for wing length (n = 8 of 216 females), we substituted the mean wing length for females eclosing from the same container on the same day. These substitutions enabled us to use all females that eclosed from the pupa in the calculation.

Statistical Analysis

One-way ANOVA’s were used to analyse the effects of predator treatment on overall survivorship, female survivorship, male survivorship, r′, mean female size, and mean female development time. For survivorship we used generalized linear models with binomial error distributions (PROC GLIMMIX). For all other variables we used standard ANOVA with normal distributions of error. For all ANOVA we report η² , the proportion of the variance in the dependent variable accounted for by the treatments as a measure of effect size (PROC GLM in SAS 9.4). Contrast statements were used as post hoc tests following significant F-tests in analyses of index of performance and female development time, with sequential Bonferroni adjustment to correct for multiple tests (Holm, 1979). The contrasts we tested were predator versus no-predator control, single-predator versus diverse predators, and pairwise comparisons of each of the three single-predator treatments.

To assess the impact of female adult production (= Σ_x A_x , equation 1), fecundity (= f(w_x)), and days to adulthood ( = x ) on statistical differences in the composite index r′, we used a sensitivity analysis approach. Simulated r′ values were calculated after holding one or two of those variables constant across all cohorts, and the treatment effects were analysed with ANOVA’s on the simulated r′. We held x constant by setting days to eclosion for all females equal to 18, which is the median day of eclosion for all females in the experiment. We held f(w_x) constant by setting predicted female egg production equal to 12.997, which is the predicted f(w_x) for a female of the mean size for all females from the entire experiment (=2.40 mm). We held number of adult females constant by setting the cumulative number of female adults produced in each cohort equal to 15, which is the median number of females produced across all cohorts. To maintain differences among cohorts in days to eclosion we multiplied all A_x within a cohort by a constant factor of 15/Σ_x A_x. This approach has the effect of increasing or decreasing production of females within the simulation proportionally across all days x. It results in fractional females on many days, but this does not change the calculation of r’.

Results

The no-predator treatment produced the lowest number of survivors across both sexes, but the overall treatment effect was not significant (F_4,10=2.12, P=0.1532, η²=0.459; Figure 1). The effects of predator on the number of adult males produced (F_4,10=1.26, P=0.3479, η²=0.335), the number female adults produced was significant (F_4,10=3.16, P=0.0636, η²=0.558). (Table 1, Figure 1) were also not significant.

Fig. 1.

Mean ± SE numbers of adult A. aegypti produced in the presence of no predators, 3 A. barberi, 3 M. longisetus, 3 C. appendiculata, or 1 of each predator species. N=3 in all cases

Table 1.

Results from contrasts comparing effects predator versus no predator, single predator versus diverse, and pairwise comparisons of individual species treatments on A. aegypti index of performance, number of female adults produced, and female days to adulthood.

	Index of performance (r’)		Female number of adults		Female days to adult		Female wing length
Contrast	F1,10	Pr > F	F1,10	Pr > F	F1,10	Pr > F	F1,10	Pr > F
Predator vs no predator	11.63	0.0067*	5.86	0.0362	11.70	0.0065*	6.60	0.0280
Single predator vs diverse	0.04	0.8381	0.0001	0.9894	1.99	0.1899	0.04	0.8365
A. barberi vs C. appendiculata	4.80	0.0535	5.15	0.0234	0.56	0.4725	0.12	0.7320
M. longisetus vs C. appendiculata	1.42	0.2606	3.65	0.0465	0.07	0.7980	1.06	0.3278
A. barberi vs M. longisetus	1.00	0.3428	0.0013	0.7205	0.23	0.6390	1.90	0.1974

^*Indicates significant p-values after correcting for multiple comparisons with sequential Bonferroni procedures

Predator treatment significantly affected index of performance r′ (F_4,10=4.11, P=0.0317, η²=0.622). No-predator treatment produced the lowest value of r′ at 0.0108, whereas C. appendiculata produced the largest r′ at 0.0435 (Figure 2a). Post hoc contrast analyses indicated predation led to a significantly greater r′ compared to no predation (Table 1, Figure 2a). Predation by single predator treatments was not significantly different from the diverse treatment, and there were no significant pairwise differences among the three single-predator treatments (Table 1).

Fig. 2.

Mean ± SE a) Indices of performance (r′), b) wing length, and c) days to adulthood for cohorts of A. aegypti reared in the presence of no predators, 3 A. barberi, 3 M. longisetus, 3 C. appendiculata, or 1 of each predator species. Brackets indicate the significant difference between predator vs. no predator. N=3 in all cases

Predator treatment had no significant effect on average female wing length (F_4,10=2.17, P=0.1457, Figure 2b, η²=0.465). Predator treatment had a significant effect on the mean number of days to adulthood for females (F_4,10=3.57, P=0.0468, η²=0.588), with mean time to adulthood in predator treatments significantly lower than that in no-predator control (Table 1, Figure 2c).

All simulations holding number of survivors constant, either alone or in conjunction with holding any second factor constant, resulted in nonsignificant F tests for the treatment effects on r′ (Table 2). Holding days to adulthood constant likewise resulted in a nonsignificant F test for treatment effects on r′, but holding fecundity constant alone or with days to adulthood yielded significant treatment effects (Table 2). When removing the effect of single contributing factors by holding them constant, effect size η² was greatest when the effect of fecundity was removed and least when the effect of survivors was removed (Table 2). When removing the effect of two factors simultaneously, effect size η² was greatest when effects of days and fecundity were removed and least when effects of survivors and fecundity were removed. Whenever the effect of survivors remained in the calculation, the correlation coefficient for simulated r′ and real r′ was >0.96 (Table 2). The greatest correlation of was >0.99 and when only the effect of days was removed from the simulation. Whenever the effect of survivors was removed from the calculation, the correlation dropped considerably (Table 2).

Table 2.

ANOVA results testing effects of predator treatments on index of performance r′ for real and simulated data holding every combination of one or two components of the index constant.

ANALYSIS	F4,10	P	Effect size η2	Contrast results (Sequential Bonferroni)	Pearson correlation with real r′ (P) N=15
Real r′	4.11	0.0317	0.6219	1. =0.0067 2. P>>0.10 3. P>>0.10 4. P>>0.10 5. P=0.0535	-
Days constant	3.10	0.0669	0.5535	1. P=0.0123 2. P>>0.10 3. P>>0.10 4. P>>0.10 5. P=0.1091	0.99027 (<0.0001)
Fecundity constant	4.97	0.0181	0.6654	1. P=0.0062 2. P>>0.10 3. P=0.0661 4. P>>0.10 5. P=0.0231	0.97722 (<0.0001)
Survivors constant	2.45	0.1146	0.4945	1. P=0.0228 2. P>>0.10 3. P>>0.10 4. P>>0.10 5. P>>0.10	0.61723 (0.0145)
Days & Fecundity constant	3.55	0.0474	0.5868	1. P=0.0152 2. P>>0.10 3. P=0.1028 4. P>>0.10 5. P=0.0488	0.96703 (<0.0001)
Days & Survivors constant	2.00	0.1703	0.4446	1. P=0.0624 2. P=0.1234 3. P>>0.10 4. P>>0.10 5. P>>0.10	0.54188 (0.0360)
Fecundity & Survivors constant	1.07	0.4196	0.3003	1. P=0.1560 2. P>>0.10 3. P>>0.10 4. P>>0.10 5. P>>0.10	0.27335 (0.3319)

Notes: Results of hypothesis tests significant at α=0.05 are shown in bold face. The effect size η² quantifies the magnitude of the differences among treatment means as the proportion of total variation in the dependent variable that can be accounted for by treatments. Maximum η² = 1.0. We included the same five contrasts that we used for the real analysis of r’ (1. Control vs. Predation; 2. Diverse vs. Single predators; 3. M. longisetus vs. C. appendiculata; 4. M. longisetus vs. A. barberi; 5. A. barberi vs. C. appendiculata).

Discussion

Population performance

Predation led to a significantly greater index of performance (r′) compared to control, indicating that mortality due to predation had the counter-intuitive effect of increasing this approximation of population rate of increase compared to the no-predator control (Livdahl & Sugihara, 1984). For this measure of population performance, predation and density dependent mortality had nonadditive effects. All of the overall F tests on total, female, and male. Because females require more time to reach adulthood and emerge as larger adults, they have greater resource demands than do males (Wormington & Juliano, 2014a, 2014b). We had thus expected females would receive a greater benefit from conspecific mortality, as they are the individuals most subject to resource limitation. McIntire & Juliano (2018) found such a sex-specific difference in how extrinsic mortality affects adult production in another Aedes species, A. albopictus. We did not find significant evidence of this sex-specific effect in survivorship to adulthood; however in the context of estimating the effects of predation on population rate of increase, females are the demographically important sex for life tables and population growth models (Livdahl & Sugihara, 1984) which raises the important question of whether the density mediated effect of predation on number of survivors, or the trait mediated effects of adult female size (and associated fecundity) and development time contribute to predation effects on population rate of increase as estimated by r’.

Female survivorship, days to adulthood, and fecundity directly contribute to inter-cohort differences in Livdahl & Sugihara’s (1984) index of performance in a nonlinear fashion (See eq. 1), regardless of whether these contributing variables are significantly affected by treatments. These contributions are further complicated because these variables are strongly correlated across cohorts. Our sensitivity analysis enabled us to assess the impact of each of the three variables on statistical differences in r′ . Removing the effect of treatments on number of female survivors (adults produced) led to lower effect sizes, higher treatment effect p-values, and a much lower correlations of simulated r′ with observed r′ . This result strongly suggests the number of surviving females was the most important contributor to r′ and treatment effects on r′ in this experiment, despite the nonsignificant effects of predator and no-predator treatments on female adult production. This emphasizes the synthetic nature of this index of performance calculation. Even variates that by themselves are not significantly distinguishable among treatments can make large contributions to differences in r′ that result from the cumulative nonlinear impacts of the three variables.

All three components of the index of performance displayed trends consistent with the observed differences in r′ (Figure 1, 2). Predation led to greater mean number of surviving females, greater mean female size and predicted fecundity, and lower mean female development time to adulthood. Significantly faster development of females and the trends in the other variables were likely caused by the weakening of density-dependent effects by reductions in population density caused by predation, but behavioural or life history changes by larvae in response to cues to the threat of predation are also a possible cause of greater adult size and shorter development time. Chandrasegaran & Juliano (2019) tested for cue-induced, non-lethal effects of a different predator, Toxorhynchites rutilus, on size at and time to adulthood of A. aegypti larvae and found no evidence for either effect, suggesting that reduction in density due to predation is ultimately the cause of all the demographic contributions to greater r′ with predators in our experiment. The difference we observe in r′ with predation suggests that equilibrium population densities could be greater for cohorts exposed to these predators (i.e., the hydra effect; Abrams 2009), compared to no predator if the experimental populations were allowed to persist for multiple generations.

Compensatory response to predation

The absence of significant differences in adult production among the predator treatments and the control indicates overall mortality from these predators yielded compensation in the overall A. aegypti cohorts. All three of these predators are effective at killing and eating first and second instar larvae (Nannini & Juliano, 1998; Griswold & Lounibos, 2005; Neale & Juliano, 2019), so it is extremely unlikely that the statistically indistinguishable numbers of adults produced between control and predation treatments result from lack of predation. We did not observe significant overcompensation in adult production in any treatment, and the adult production in the diverse predator treatment was not significantly different from any single-species predator treatment; therefore, our results do not support our hypothesis that predation from a diverse assemblage of predators leads to increased or decreased strength of overcompensation compared to any single-species of predator. The compensatory response suggests predation removed a number of individuals similar to the number that would have otherwise died from density-dependent effects, but this removal did not sufficiently release the surviving larvae from density-dependent effects to increase significantly production of adults.

Significant overcompensation was induced in A. aegypti by predation from M. longisetus in a substantially larger experiment (Neale & Juliano, 2019). However, the present study failed detect the same effect despite the tendency for all of our predator treatments to produce more adults, particularly more females (Figure 2). The compensatory response to A. barberi predation is consistent with results of Nannini & Juliano (1998), in which predation by A. barberi yielded compensation in A. triseriatus, a congener to the A. aegypti tested in this study. However, comparisons of (over)compensatory responses between species should be made with caution, as interspecific differences in responses to population density and competitive abilities are associated with interspecific variation in the level and likelihood of (over)compensation (Neale & Juliano, 2019).

Single- vs multi- predator treatments

The single-species predator treatments yielded the very similar production of adults as the diverse predator treatment; therefore, we have no evidence for emergent MPE’s at the level of adult production. Compensation or overcompensation in adult production are products of the interaction of mortality induced by predators and responses of prey to conspecific density (Abrams, 2009), and because this was our focus, we did not quantify predation on a daily basis. The typical approach to MPEs is to quantify short-term (e.g., 1 day or less) direct effects of predator-induced mortality. Detecting MPEs is complicated by the dependence of predicted predator-induced mortality on the nonlinearity of the functional responses of predators and by prey depletion (McCoy et al., 2012). Prey depletion certainly occurred in our experiment. Limited previous work suggests that our predators show functional responses that are often linear. Anopheles barberi has a functional response that is indistinguishable from linear, with no asymptote, over a range of densities from 10–250 larvae, suggesting partial consumption of prey at higher prey densities (Nannini & Juliano, 1998). Corethrella appendiculata has a more typical asymptotic type II functional response (Griswold & Lounibos, 2005), but its expected number of prey eaten per day in our experiment would fall approximately at the asymptote (~17-18 larvae in 24 h), even with prey depletion, for the first 6 days of our experiment (Griswold & Lounibos 2005, their Figure 2). Functional responses of Mesocyclops to Aedes larvae are not well studied, but existing functional responses to other mosquitoes are asymptotic and kill a maximum of 25-40 larvae in 24 h at the density used in our experiment (e.g., Cuthbert et al. 2018). McCoy et al. (2012) noted that detecting MPEs may best be accomplished using a range of prey densities to characterise the functional responses of the predators, and such experiments with predation on A. aegypti would be useful. We believe the best interpretation of the lack of differences in adult production in our predator treatments is that death rates induced by these predators are similar for the densities present in this experiment.

Our results suggest that effects of the three predators on adult production were substitutable. Predator substitutability is predicted to occur when the predators exhibit non-overlapping within-habitat domains and prey exhibit broad within-habitat domains (Schmitz, 2007). In our experiment, within-habitat domains for A. barberi and C. appendiculata have little or no overlap, but the domains of each overlap with the uppermost (A. barberi) and lowermost (C. appendiculata) portions of the domain for M. longisetus, which hunts throughout the entire water column (Clements, 1992; Kesavaraju et al., 2007; Marten & Reid, 2007). Scenarios in which the within-habitat domains of multiple predators overlap can lead to emergent MPE’s, the nature of which depend on the respective hunting modes of the predators and the degree of overlap with prey within-habitat domain (Schmitz, 2007). Because the degree of domain overlap between M. longisetus and either of the other two predators is small, the chances for interactions between predators may have been minimal. We saw no evidence of intraguild predation in our experiment, one mechanism that can lead to risk reduction when predator habitat domains overlap (Schmitz, 2007).

Conclusion

We have demonstrated predation on larval A. aegypti by three predator species, alone and in multi-species assemblages, can yield compensatory production of adults. Our evidence suggests predation by at least some predators may relieve A. aegypti populations of a sufficient level of density-dependent effects to produce the counter intuitive effect of increased population rate of increase with predation. Our data further indicate that greater survival of female mosquitoes when predators are present as opposed to absent is the main contributor to predator-induced increases in estimated population rate of increase. As such counterintuitive effects of predation are predicted to occur in a variety of food web structures (Cortez & Abrams, 2016), these results provide insight on a phenomenon likely to affect many taxa. Further work should be conducted to elucidate the mechanisms mediating compensation, predator enhanced rate of increase, and the possibility of the hydra effect to better predict their occurrence in nature, allowing more effective pest management, conservation, and harvest strategies (Abrams, 2002; Ratikainen et al., 2008; Zipkin et al., 2009; Sandercock et al., 2011).