Risk of predation increases susceptibility to parasitism via trait-mediated indirect effects
Data files
Dec 06, 2025 version files 31.29 KB
-
Experiment1_R.csv
3.40 KB
-
Experiment2_R.csv
3.55 KB
-
RawCode.txt
1.76 KB
-
README.md
22.59 KB
Abstract
The presence of natural enemies can cause organisms to change habitat use, foraging behaviour and/or resource allocation in response to a perceived risk; responses that may come at the cost of other fitness-related traits. Since most species encounter multiple natural enemies, defensive behaviours against one attacker may make the focal organism more vulnerable to attack by a different natural enemy. Anti-predator behaviours can lead to trait-mediated indirect effects such as an increased risk of attack by parasites, and vice versa. Few empirical studies have examined the response of a single focal species to the risk of attack by multiple species. Our experiments provided Drosophila nigrospiracula with opportunities to prioritise either anti-predator or anti-parasite behaviour at the cost of increased infection or predation, respectively. When exposed to parasites in the presence of predator cues, D. nigrospiracula experienced increased infection compared to flies without predator cues, but the presence/absence of parasite cues had no analogous effect on predation rates. We suggest that flies perceived parasitic infection to be a lesser threat and responded more strongly to predation risk at the cost of increased infection. In an ecological context, we shows how trait-mediated indirect effects could regulate community structure by increasing susceptibility to infection.
Title - Risk of predation increases susceptibility to parasitism via trait-mediated indirect effects
Summary: The presence of natural enemies can cause organisms to change habitat use, foraging behaviour and/or resource allocation in response to a perceived risk; responses that may come at the cost of other fitness-related traits. Since most species encounter multiple natural enemies, defensive behaviours against one attacker may make the focal organism more vulnerable to attack by a different natural enemy. Anti-predator behaviours can lead to trait-mediated indirect effects such as an increased risk of attack by parasites, and vice-versa. Few empirical studies have examined the response of a single focal species to the risk of attack by multiple species. Our experiments provided Drosophila nigrospiracula with opportunities to prioritise either anti-predator or anti-parasite behaviour at the cost of increased infection or predation, respectively. When exposed to parasites in the presence of predator cues, D. nigrospiracula experienced increased infection compared to flies without predator cues, but the presence/absence of parasite cues had no analogous effect on predation rates. We suggest that flies perceived parasitic infection to be a lesser threat and responded more strongly to predation risk at the cost of increased infection. In an ecological context, we shows how trait-mediated indirect effects could regulate community structure by increasing susceptibility to infection.
File overview: the file Experiment1_R.csv contains the data used in Experiment One described on the manuscript. Column headers are as follows:
Trial_Number - Each trial was conducted on a different day. This variable allowed for Trial Day to be tested as a random and fixed effect. Petri_position - Petri dishes were allocated positions inside the darkened box, and positions were randomised each trial to ensure that there was no effect of distance between dish and door to the box. Treatment - exposed to spider cues (spider) or not (control) SpiderID - ID to track which spiders had been used from our spider collection SpiderMass - Spider mass in mg to be used as a covariate in analysis Flies_Recovered - total number of flies recovered from experimental arena. Some flies were lost in transfer to or from arena. Flies_Inf - number of flies infected with one or more mite. Fly1 to Fly5 - data shows the number of mites attached to each fly. Some columns read NA if less than 5 flies were recovered. Abundance - Parasite abundance. Calculated as number of mites divided by total number of flies. Prevalence - Parasite prevalence. Calculated as the number of infected flies divided by the total number of flies. Infected - number of infected flies. Used in cbind() analysis Uninfected - Number of uninfected flies. Used in cbind() analysis InfInt - Parasite infection intensity. Calculated as the number of mites divided by the number of infected flies.
The file Experiment2_R.csv contains the data used in Experiment Two described on the manuscript. Column headers are as follows:
Date - Date that trials were completed TrialDay - Dates as simple numeric values Trial_No - on each date, multiple sets of 4 trials were completed. Each of these sets was given a numeric value Treatment - Mites present or mites absent treatments Position - position of each of the four petris per trial. Position was randomised for each trial set. Order - indication of the order in which each petri was loaded. Order was randomisd between trial sets top avod any unwanted effect of load order. MediaMass - weight (mg) of media (mite or mite-free) added to lower arena SpiderID - ID to track which spiders had been used from our spider collection SpiderMass - Spider mass in mg to be used as a covariate in analysis t0 start time of each trial Dead - binary indicator of fly status: 1 indicated fly was dead, 0 indicated fly was alive (variable unused in analysis) Time1 - time fly was observed dead in seconds after t0: 0s entered if fly survived until end of trial (variable unused in analysis) Time2 - time fly was observed dead in seconds after t0: 30s entered if fly survived until end of trial status - indicator of fly status: 2 indicated fly was dead, 1 indicated fly was alive dead - binary indicator of fly status: 1 indicated fly was dead, 0 indicated fly was alive (used in glmm cbind() function) alive - binary indicator of fly status: 0 indicated fly was dead, 1 indicated fly was alive (used in glmm cbind() function)
Experiment 1 - Infection Prevalence tests - Text from MS Statistical Analysis Section
“A generalised linear mixed effect model (glmm - glmer function in the package lme4 (Bates et al. 2014)) was used to analyse infection prevalence data. This model used infection prevalence as the response variable, treatment as the fixed effect, and trial number as a random effect to compensate for unknown variables in a given day. A general linear model (glm – glm function in the package stats) was used to test for differences of infection prevalence caused by variability in mite density among trials. This model used infection prevalence as the response variable and trial as fixed effect. Both models used a binomial error family and logit link function, and a dispersion parameter for our glmm was calculated using the dispersion_glmer function in the package blmeco (Korner-Nievergelt et al. 2019).”
setwd("D:/Proceedings")
Exper1<-read.csv("Experiment1_R.csv")
attach(Exper1)
str(Exper1)
## 'data.frame': 67 obs. of 12 variables:
## $ Trial_Number : int 1 1 1 1 2 2 2 2 2 3 ...
## $ Petri_position: int 2 7 8 9 2 4 6 8 10 4 ...
## $ Treatment : chr "Spider" "Spider" "Spider" "Spider" ...
## $ SpiderID : chr "s10" "s3" "s57" "s17" ...
## $ SpiderMass : num 7.31 8.68 7.37 10.65 7.9 ...
## $ Flies_Rec : int 5 4 5 4 5 5 3 4 5 5 ...
## $ Flies_Inf : int 2 2 4 1 4 1 3 2 1 0 ...
## $ Abundance : num 0.6 0.5 2.4 2.25 8.2 0.2 2.33 2 0.2 0 ...
## $ Prevalence : num 0.4 0.5 0.8 0.25 0.8 0.2 1 0.5 0.2 0 ...
## $ Infected : int 2 2 4 1 4 1 3 2 1 0 ...
## $ Uninfected : int 3 2 1 3 1 4 0 2 4 5 ...
## $ InfInt : num 1.5 1 3 9 10.3 1 2.3 4 1 0 ...
library(lme4)
## Loading required package: Matrix
library(blmeco)
## Loading required package: MASS
library(car)
## Loading required package: carData
library(emmeans)
model1<-glmer((cbind(Infected,Uninfected))~Treatment+(1|Trial_Number), family="binomial")
summary(model1)
## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: (cbind(Infected, Uninfected)) ~ Treatment + (1 | Trial_Number)
##
## AIC BIC logLik deviance df.resid
## 192.2 198.8 -93.1 186.2 64
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.6832 -0.8748 -0.2399 0.4993 3.7074
##
## Random effects:
## Groups Name Variance Std.Dev.
## Trial_Number (Intercept) 0.4638 0.681
## Number of obs: 67, groups: Trial_Number, 7
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -1.6569 0.3383 -4.897 9.72e-07 ***
## TreatmentSpider 0.7581 0.2783 2.724 0.00645 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## TretmntSpdr -0.496
Anova(model1)
## Analysis of Deviance Table (Type II Wald chisquare tests)
##
## Response: (cbind(Infected, Uninfected))
## Chisq Df Pr(>Chisq)
## Treatment 7.4198 1 0.006451 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
dispersion_glmer(model1)
## [1] 1.213211
TrialAF<-as.factor(Trial_Number)
model2<-glm((cbind(Infected,Uninfected))~TrialAF, family="binomial")
summary(model2)
##
## Call:
## glm(formula = (cbind(Infected, Uninfected)) ~ TrialAF, family = "binomial")
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -1.8886 -1.0336 -0.5073 0.4723 3.0185
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -0.1823 0.3496 -0.522 0.602011
## TrialAF2 -0.1691 0.4603 -0.367 0.713392
## TrialAF3 -0.6650 0.4663 -1.426 0.153872
## TrialAF4 -2.2600 0.6277 -3.601 0.000317 ***
## TrialAF5 -1.3340 0.5077 -2.628 0.008595 **
## TrialAF6 -1.7636 0.5592 -3.154 0.001612 **
## TrialAF7 -1.3340 0.5077 -2.628 0.008595 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 129.107 on 66 degrees of freedom
## Residual deviance: 99.408 on 60 degrees of freedom
## AIC: 190.49
##
## Number of Fisher Scoring iterations: 5
Anova(model2)
## Analysis of Deviance Table (Type II tests)
##
## Response: (cbind(Infected, Uninfected))
## LR Chisq Df Pr(>Chisq)
## TrialAF 29.7 6 4.483e-05 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
#manually estimate dispersion parameter - residual deviance/residual degrees of freedom
99/60
## [1] 1.65
em <- emmeans(model2, "TrialAF")
## Warning in (function (object, at, cov.reduce = mean, cov.keep = get_emm_option("cov.keep"), : There are unevaluated constants in the response formula
## Auto-detection of the response transformation may be incorrect
contrast(em, "pairwise", adjust = "Tukey")
## contrast estimate SE df z.ratio p.value
## TrialAF1 - TrialAF2 0.169 0.460 Inf 0.367 0.9998
## TrialAF1 - TrialAF3 0.665 0.466 Inf 1.426 0.7879
## TrialAF1 - TrialAF4 2.260 0.628 Inf 3.601 0.0059
## TrialAF1 - TrialAF5 1.334 0.508 Inf 2.628 0.1176
## TrialAF1 - TrialAF6 1.764 0.559 Inf 3.154 0.0269
## TrialAF1 - TrialAF7 1.334 0.508 Inf 2.628 0.1176
## TrialAF2 - TrialAF3 0.496 0.430 Inf 1.153 0.9113
## TrialAF2 - TrialAF4 2.091 0.601 Inf 3.478 0.0091
## TrialAF2 - TrialAF5 1.165 0.475 Inf 2.455 0.1760
## TrialAF2 - TrialAF6 1.595 0.529 Inf 3.013 0.0414
## TrialAF2 - TrialAF7 1.165 0.475 Inf 2.455 0.1760
## TrialAF3 - TrialAF4 1.595 0.606 Inf 2.633 0.1161
## TrialAF3 - TrialAF5 0.669 0.480 Inf 1.393 0.8060
## TrialAF3 - TrialAF6 1.099 0.535 Inf 2.055 0.3797
## TrialAF3 - TrialAF7 0.669 0.480 Inf 1.393 0.8060
## TrialAF4 - TrialAF5 -0.926 0.638 Inf -1.451 0.7737
## TrialAF4 - TrialAF6 -0.496 0.680 Inf -0.730 0.9908
## TrialAF4 - TrialAF7 -0.926 0.638 Inf -1.451 0.7737
## TrialAF5 - TrialAF6 0.430 0.571 Inf 0.752 0.9892
## TrialAF5 - TrialAF7 0.000 0.521 Inf 0.000 1.0000
## TrialAF6 - TrialAF7 -0.430 0.571 Inf -0.752 0.9892
##
## Results are given on the log odds ratio (not the response) scale.
## P value adjustment: tukey method for comparing a family of 7 estimates
Experiment 1 - Infection Intensity and Infection Abundance tests - Text from MS Statistical Analysis Section
“We used a Wilcoxon signed-rank non-parametric tests (wilcox.test
function in the stats package) to analyse infection intensity and
infection abundance data as neither met the assumptions of normality.
Infection intensity or infection abundance were the response variables
and treatment, predator present or predator absent, was the fixed
effect. Due to variable mite density among trials, we used a Kruskal
Wallace non-parametric test (kruskal.test function in the stats package)
to look for any significant degree of variability in infection caused by
trial number, and a Nemenyi pairwise test with a Tukey distribution
(NemenyiTest function in the DescTools package) to find significant
pairwise differences between trials.”
Infection Intensity - Wilcox Test followed by Kruskal Wallace and Nemenyi tests
library(DescTools)
##
## Attaching package: 'DescTools'
## The following object is masked from 'package:car':
##
## Recode
wilcox.test(InfInt~Treatment)
## Warning in wilcox.test.default(x = DATA[[1L]], y = DATA[[2L]], ...): cannot
## compute exact p-value with ties
##
## Wilcoxon rank sum test with continuity correction
##
## data: InfInt by Treatment
## W = 372.5, p-value = 0.01423
## alternative hypothesis: true location shift is not equal to 0
kruskal.test(InfInt ~ TrialAF, data = Exper1)
##
## Kruskal-Wallis rank sum test
##
## data: InfInt by TrialAF
## Kruskal-Wallis chi-squared = 17.247, df = 6, p-value = 0.008418
II = NemenyiTest(x = InfInt,
g = TrialAF,
dist="tukey")
II
##
## Nemenyi's test of multiple comparisons for independent samples (tukey)
##
## mean.rank.diff pval
## 2-1 -3.014286 0.9999
## 3-1 -6.564286 0.9935
## 4-1 -27.464286 0.0641 .
## 5-1 -16.314286 0.6167
## 6-1 -24.314286 0.1478
## 7-1 -14.214286 0.7568
## 3-2 -3.550000 0.9996
## 4-2 -24.450000 0.0744 .
## 5-2 -13.300000 0.7290
## 6-2 -21.300000 0.1802
## 7-2 -11.200000 0.8590
## 4-3 -20.900000 0.1990
## 5-3 -9.750000 0.9226
## 6-3 -17.750000 0.3911
## 7-3 -7.650000 0.9759
## 5-4 11.150000 0.8615
## 6-4 3.150000 0.9998
## 7-4 13.250000 0.7325
## 6-5 -8.000000 0.9698
## 7-5 2.100000 1.0000
## 7-6 10.100000 0.9093
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Infection Abundance - Wilcox Test followed by Kruskal Wallace and Nemenyi tests
wilcox.test(Abundance ~ Treatment)
## Warning in wilcox.test.default(x = DATA[[1L]], y = DATA[[2L]], ...): cannot
## compute exact p-value with ties
##
## Wilcoxon rank sum test with continuity correction
##
## data: Abundance by Treatment
## W = 386, p-value = 0.02352
## alternative hypothesis: true location shift is not equal to 0
kruskal.test(Abundance ~ TrialAF, data = Exper1)
##
## Kruskal-Wallis rank sum test
##
## data: Abundance by TrialAF
## Kruskal-Wallis chi-squared = 19.696, df = 6, p-value = 0.003137
AB = NemenyiTest(x = Abundance,
g = TrialAF,
dist="tukey")
AB
##
## Nemenyi's test of multiple comparisons for independent samples (tukey)
##
## mean.rank.diff pval
## 2-1 -4.542857 0.9992
## 3-1 -10.242857 0.9378
## 4-1 -30.042857 0.0291 *
## 5-1 -20.842857 0.3116
## 6-1 -27.442857 0.0645 .
## 7-1 -18.392857 0.4699
## 3-2 -5.700000 0.9949
## 4-2 -25.500000 0.0533 .
## 5-2 -16.300000 0.5000
## 6-2 -22.900000 0.1176
## 7-2 -13.850000 0.6892
## 4-3 -19.800000 0.2575
## 5-3 -10.600000 0.8881
## 6-3 -17.200000 0.4314
## 7-3 -8.150000 0.9669
## 5-4 9.200000 0.9407
## 6-4 2.600000 0.9999
## 7-4 11.650000 0.8346
## 6-5 -6.600000 0.9888
## 7-5 2.450000 1.0000
## 7-6 9.050000 0.9451
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
detach(Exper1)
Experiment 2 - Fly mortality tests - Text from MS Statistical Analysis Section
“Fly survival was analysed using both a glmm (as described above) and a
cox proportionate hazard model with random effects (coxme function in
the package coxme (Therneau 2012)). The glmm analyzed end point survival
only, with fixed effects of treatment (parasite present and parasite
absent), media mass, spider mass, and trial date as a random effect. The
proportional hazards model incorporated time of death of each treatment
(parasite present and parasite absent), spider mass and media mass and
again with the random effect of trial date. In this experiment, variable
mite density was not considered to be an issue, as mite media provide
only a parasite present signal. However, in the interest of
completeness, we tested for any effect of trial date as a proxy for mite
density using a glm with fly mortality as response variable and trial
date as fixed effect, and a cox proportionate hazard model with time of
death as response variable and trial date as fixed effect.”
End-point mortality - Generalised Linear mixed effect model followed by a glm to test for any effect of Trial Date
setwd("D:/Proceedings")
Exper2<-read.csv("Experiment2_R.csv")
attach(Exper2)
str(Exper2)
## 'data.frame': 63 obs. of 18 variables:
## $ Date : chr "19Dec" "19Dec" "19Dec" "19Dec" ...
## $ TrialDay : int 1 1 1 1 1 1 1 1 1 1 ...
## $ Trial_No : int 1 1 1 1 2 2 2 2 3 3 ...
## $ Trial_No_Ct: logi NA NA NA NA NA NA ...
## $ Petri : int 8 5 4 9 8 5 4 9 8 1 ...
## $ Treatment : chr "Mite" "Mite" "NoMite" "NoMite" ...
## $ Position : int 1 2 3 4 1 2 3 4 1 2 ...
## $ Order : int 1 2 3 4 4 3 2 1 1 2 ...
## $ MediaMass : num 5.5 6.4 4.9 5.4 7 7.2 5.7 6 6.5 7 ...
## $ SpiderID : chr "2" "x25" "57" "36" ...
## $ SpiderMass : num 6.01 6.88 7.37 6.62 10.35 ...
## $ t0 : num 10.2 10.2 10.2 10.2 11.1 ...
## $ Dead : int 1 1 0 0 0 0 0 1 1 1 ...
## $ Time1 : int 10 10 0 0 0 0 0 15 5 5 ...
## $ Time2 : int 10 10 30 30 30 30 30 15 5 5 ...
## $ status : int 2 2 1 1 1 1 1 2 2 2 ...
## $ dead : int 1 1 0 0 0 0 0 1 1 1 ...
## $ alive : int 0 0 1 1 1 1 1 0 0 0 ...
TrialDayAF<-as.factor(TrialDay)
model3<-glmer(cbind(dead,alive)~Treatment+SpiderMass+MediaMass+(1|TrialDayAF), family="binomial")
## boundary (singular) fit: see help('isSingular')
summary(model3)
## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: cbind(dead, alive) ~ Treatment + SpiderMass + MediaMass + (1 |
## TrialDayAF)
##
## AIC BIC logLik deviance df.resid
## 93.9 104.6 -42.0 83.9 58
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -0.9904 -0.8096 -0.7298 1.1571 1.4678
##
## Random effects:
## Groups Name Variance Std.Dev.
## TrialDayAF (Intercept) 0 0
## Number of obs: 63, groups: TrialDayAF, 4
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 1.06158 3.55850 0.298 0.765
## TreatmentNoMite -0.36847 0.52099 -0.707 0.479
## SpiderMass -0.04379 0.09711 -0.451 0.652
## MediaMass -0.15279 0.53771 -0.284 0.776
##
## Correlation of Fixed Effects:
## (Intr) TrtmNM SpdrMs
## TreatmntNMt -0.156
## SpiderMass -0.234 0.051
## MediaMass -0.973 0.079 0.027
## optimizer (Nelder_Mead) convergence code: 0 (OK)
## boundary (singular) fit: see help('isSingular')
Anova(model3)
## Analysis of Deviance Table (Type II Wald chisquare tests)
##
## Response: cbind(dead, alive)
## Chisq Df Pr(>Chisq)
## Treatment 0.5002 1 0.4794
## SpiderMass 0.2033 1 0.6521
## MediaMass 0.0807 1 0.7763
dispersion_glmer(model3)
## [1] 1.154044
model4<-glm(cbind(dead,alive)~TrialDayAF, family="binomial")
summary(model4)
##
## Call:
## glm(formula = cbind(dead, alive) ~ TrialDayAF, family = "binomial")
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -1.1010 -0.9794 -0.9282 1.2557 1.4490
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -0.6190 0.4688 -1.320 0.187
## TrialDayAF2 0.1335 0.6494 0.206 0.837
## TrialDayAF3 0.4367 0.7658 0.570 0.568
## TrialDayAF4 0.4367 0.7658 0.570 0.568
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 84.635 on 62 degrees of freedom
## Residual deviance: 84.124 on 59 degrees of freedom
## AIC: 92.124
##
## Number of Fisher Scoring iterations: 4
Anova(model4)
## Analysis of Deviance Table (Type II tests)
##
## Response: cbind(dead, alive)
## LR Chisq Df Pr(>Chisq)
## TrialDayAF 0.51017 3 0.9167
#manually estimate dispersion parameter - residual deviance/residual degrees of freedom
84/59
## [1] 1.423729
Mortality as a function of time - Cox mixed effect (coxme) model followed by a coxph model to test for any effect of Trial date
library(coxme)
## Loading required package: survival
## Loading required package: bdsmatrix
##
## Attaching package: 'bdsmatrix'
## The following object is masked from 'package:base':
##
## backsolve
model5<-coxme(Surv(Time2,status)~Treatment+SpiderMass+MediaMass+(1|TrialDayAF))
model5
## Cox mixed-effects model fit by maximum likelihood
##
## events, n = 25, 63
## Iterations= 5 22
## NULL Integrated Fitted
## Log-likelihood -98.04112 -97.74864 -97.74787
##
## Chisq df p AIC BIC
## Integrated loglik 0.58 4 0.96472 -7.42 -12.29
## Penalized loglik 0.59 3 0.89965 -5.42 -9.07
##
## Model: Surv(Time2, status) ~ Treatment + SpiderMass + MediaMass + (1 | TrialDayAF)
## Fixed coefficients
## coef exp(coef) se(coef) z p
## TreatmentNoMite -0.28707478 0.7504556 0.40803486 -0.70 0.48
## SpiderMass -0.01653467 0.9836013 0.07204492 -0.23 0.82
## MediaMass -0.12971693 0.8783440 0.40672415 -0.32 0.75
##
## Random effects
## Group Variable Std Dev Variance
## TrialDayAF Intercept 9.039815e-03 8.171825e-05
model6<-coxph(Surv(Time2,status)~TrialDayAF)
summary(model6)
## Call:
## coxph(formula = Surv(Time2, status) ~ TrialDayAF)
##
## n= 63, number of events= 25
##
## coef exp(coef) se(coef) z Pr(>|z|)
## TrialDayAF2 0.1678 1.1827 0.5176 0.324 0.746
## TrialDayAF3 0.2399 1.2711 0.5856 0.410 0.682
## TrialDayAF4 0.1739 1.1899 0.5862 0.297 0.767
##
## exp(coef) exp(-coef) lower .95 upper .95
## TrialDayAF2 1.183 0.8455 0.4289 3.262
## TrialDayAF3 1.271 0.7867 0.4034 4.006
## TrialDayAF4 1.190 0.8404 0.3772 3.754
##
## Concordance= 0.511 (se = 0.057 )
## Likelihood ratio test= 0.2 on 3 df, p=1
## Wald test = 0.2 on 3 df, p=1
## Score (logrank) test = 0.2 on 3 df, p=1
detach(Exper2)
See text file RawCode.txt for code with no accompanying text or outputs
Flies, mites, and spiders: Drosophila nigrospiracula flies were cultured on Drosophila media (Formula 4–24 Instant Drosophila Medium; Carolina Biological Supply Company, Burlington, NC), instant mashed potato, and autoclaved necrotic tissue of Saguaro cacti (Carnegiea gigantea). Macrocheles subbadius mites were maintained in a mix of 2:1 mix of wheat bran to wood chips and co-cultured with free-living nematodes as a food source, hereafter referred to as ‘mite media’. Drosophila nigrospiracula and M. subbadius were both collected from the Sonoran Desert, Arizona, USA, in 2019 and maintained separately in continuous culture at the University of Alberta in climate controlled incubators (Percival Scientific, Perry, IA, USA) at 25 °C, 70 % relative humidity (RH) and a 12 h:12 h light:dark regime. Salticus scenicus spiders were collected by hand on the campus of the University of Alberta, Canada. Spiders were maintained in 100 mL specimen jars and provided with water and one D. nigrospiracula fly twice per week.
Experimental arenas: we custom designed a two-level arena system that could expose flies to parasites (M. subbadius) on the lower level or predators (S. scenicus) on the upper level (see schematic - Figure 1). Our design allowed us to test the effects of indirect contact (odour and visual cues) of one threat type while the focal species was simultaneously and directly exposed to the other threat type. The arena system was composed of two clear plastic Petri dishes (diameter 6 cm, height 1.5 cm) stacked one on top of the other. Mesh-covered holes in the floor of the upper dish and the roof of the lower dish allowed odour cues to flow between arenas. We placed a hole in the roof of the upper dish so that spiders and/or flies could be aspirated into this arena; this hole was plugged with adhesive tac (UHU Tac Adhesive Putty, UHU GmbH & Co. KG, Germany) during trials. The inner surfaces of the upper dish were abraded with course sandpaper so that the spiders could access all areas of the arena when hunting flies. This system allowed us to test our two alternate hypotheses using four separate scenarios. Our first hypothesis - flies prioritise anti-predator behavioural defenses over risk of infection - used scenarios 1) direct contact between flies and mites with indirect contact (visual and odour cues only) from spider and 2) direct contact between flies and mites, no spider cues. Our second hypothesis - flies prioritise anti-parasite behaviour at an increased risk of predation - used scenarios 3) direct contact between flies and spider with indirect contact with mites and 4) direct contact between flies and spider but no mite cues.
In Scenarios 1 and 2, approximately 15 g of mite media from our in-lab mite colonies was packed into the lower arena (Figure 1A-iv) and two intersecting trenches (0.5 mm wide, 0.5 mm deep) were made in the media with a short length of thin wood (Figure 1B-iv). These trenches replicated the small interstices or necrotic pockets of the cactus, where flies would normally encounter mites. The number of mites in the mite media varied among days, and therefore among trials, due to patchy aggregation of mites within colony containers. We assumed that any subsequent variability in infection would affect both treatments equally (see statistical analysis section). In both scenarios, five flies were aspirated into the trenches in the lower arena before attaching the upper arena. In Scenario 1, a spider was added to the upper arena, while in Scenario 2, no spider was added. In Scenarios 3 and 4, one fly and one spider were aspirated into the upper level; the fly was added first to avoid an immediate attack by the spider. In Scenario 3 mite media was packed into the lower level, but no trenches were made, while mite-free media was added to the lower level in Scenario 4, again with no trenches. Mite-free media contained wheat bran, wood chips and nematodes, but no mites, allowing us to control for any fly behavioural responses to the odour cues generated by the media and nematodes.
Experiment One: predator present and predator absent trials were conducted over 7 days in November and December 2023. Flies were directly exposed to mites with (n = 34) and without (n = 33) indirect exposure to spiders; arenas were then placed into a dark box inside the incubator for 1.5 hours. After the trial was complete, the flies were anaesthetised with CO2 and inspected under a dissecting microscope for mites. We calculated three response variables to quantify infection success (Rózsa et al. 2000): infection prevalence - the mean proportion of flies that were infected; infection intensity - the average number of mites among all infected flies, and infection abundance – the average number of mites per fly, including uninfected and infected flies.
Experiment Two: parasite present/absent trials were completed over four days in December 2023. Flies were directly exposed to spiders with (n = 32) and without (n = 32) indirect exposure to mites; trials lasted 30 minutes and were run in replicates of four: two with parasites present and two with parasites absent. Trials were limited to 30 minutes because after that time spiders began to spin webs that inhibited fly movement and confounded the behavioural responses of interest. After the spider and fly were loaded, an opaque plastic container (interior light intensity: 136 lux) was placed over top of the arena unit to limit visual disturbances that might affect spider or fly behaviour. The plastic container had a hole in the top that was covered with a lid, which was removed briefly at five-minute intervals to assess fly mortality. Spider and fly were observed until either the fly was killed or the trial ended, at which time we recorded mortality (yes/no) and time until death. Trials were conducted in-lab, where temperature and relative humidity were relatively consistent (18-19 °C and 22-24% RH).
Statistical Analysis
Experiment One: A generalised linear mixed effect model (glmm - glmer function in the package lme4 (Bates et al. 2014)) was used to analyse infection prevalence data. This model used infection prevalence as the response variable, treatment as the fixed effect, and trial number as a random effect to compensate for unknown variables in a given day. A general linear model (glm – glm function in the package stats) was used to test for differences of infection prevalence caused by variability in mite density among trials. This model used infection prevalence as the response variable and trial as fixed effect. Both models used a binomial error family and logit link function, and a dispersion parameter for our glmm was calculated using the dispersion_glmer function in the package blmeco (Korner-Nievergelt et al. 2019).
We used a Wilcoxon signed-rank non-parametric tests (wilcox.test function in the stats package) to analyse infection intensity and infection abundance data as neither met the assumptions of normality. Infection intensity or infection abundance were the response variables and treatment, predator present or predator absent, was the fixed effect. Due to variable mite density among trials, we used a Kruskal Wallace non-parametric test (kruskal.test function in the stats package) to look for any significant degree of variability in infection caused by trial number, and a pairwise Nemenyi test with a Tukey distribution (NemenyiTest function in the DescTools package (Signorell et al. 2024)) to find significant pairwise differences between trials.
Experiment Two: Fly survival was analysed using both a glmm (as described above) and a cox proportionate hazard model with random effects (coxme function in the package coxme (Therneau 2012)). The glmm analyzed end point survival only, with fixed effects of treatment (parasite present and parasite absent), media mass, spider mass, and trial date as a random effect. The proportional hazards model incorporated time of death of each treatment (parasite present and parasite absent), spider mass and media mass and again with the random effect of trial date. In this experiment, variable mite density was not considered to be an issue, as mite media provide only a parasite present signal. However, in the interest of completeness, we tested for any effect of trial date as a proxy for mite density using a glm with fly mortality as response variable and trial date as fixed effect, and a cox proportionate hazard model with time of death as response variable and trial date as fixed effect.