Drought disrupts volatile-mediated predator foraging and oviposition, weakening trait-mediated top-down control

We thank Abby Seltzer for her invaluable assistance throughout the 31 course of this study, particularly for her support with data collection, insect rearing, and 32 greenhouse logistics. We also thank Nate McCartney for his help with volatile sample processing 33 and data extractions. This research was supported by the Department of Entomology at the 34 Pennsylvania State University, the National Science Foundation, Division of Environmental 35 Biology award (#2440876), and by the Pennsylvania Department of Agriculture (#C940001869). 36 CONFLICT OF INTEREST: Authors declare no conflict of interest. 37 38

39 1. Drought is a major abiotic stressor that can restructure trophic interactions by limiting 40 herbivore success and disrupting chemical signaling between plants and natural enemies. In 41 tritrophic systems, plant volatiles guide natural enemy foraging and reproductive investment, 42 often scaling with herbivore density; however, it is unclear whether drought alters this 43 relationship and weakens top-down control. 44 2. Using a tomato-aphid-ladybeetle system, we tested how drought and herbivore density jointly 45 affect plant VOC emissions, predator behavior, and aphid dynamics. We manipulated water 46 availability (well-watered vs. drought) and aphid density (low vs. high), and measured plant 47 physiology, volatile profiles, predator visitation and oviposition, and aphid responses. 48 3. Drought reduced stomatal conductance, plant biomass, and both total and compositional 49 output of VOCs. Emission of key predator-attracting compounds (e.g., methyl salicylate, β -50 myrcene) peaked in well-watered, high-density plants but was suppressed under drought. 51 4. Ladybeetle visitation increased with aphid density but declined under drought, reflecting 52 conserved shifts in volatile cues. Oviposition was concentrated on well-watered, high-density 53 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 3 plants and associated with specific compounds (e.g., methyl salicylate, carvacrol), while 54 others (e.g., cymene-7-ol, para, 1-octanol) were negatively associated. 55 5. Aphid suppression by predators occurred only under well-watered, high-density conditions. 56 Under drought, aphid growth was already constrained, and predators had little additional 57 effect on their abundance. However, both drought and predator presence influenced aphid 58 demography, increasing production of dispersive alates. 59 6. These findings underscore the sensitivity of chemically mediated trophic interactions to 60 environmental stress. Increased drought disrupts plant signaling, reducing natural enemy 61 effectiveness, weakening biocontrol, and shifting herbivore population structure. 62 Understanding how stress alters cue reliability is key to predicting community dynamics and 63 managing ecosystem functions under stress. 64

Hippodamia convergens, Macrosiphum euphorbiae, Non-consumptive predator 65 effects, Plant volatiles, Predator foraging and oviposition, Solanum lycopersicum, Top-down 66 control, Tritrophic interactions 67 1. Introduction 68 Natural enemies play an important role in regulating herbivore populations, often producing 69 cascading effects on plant communities (Schmitz et al., 2000). However, the strength of top-70 down control often declines under abiotic stress (Barton & Schmitz, 2009; Lin et al., 2023). This 71 weakening does not always result from natural enemy loss or herbivore escape (Clavijo 72 McCormick, 2016; Schmitz & Barton, 2014). Observed declines in natural enemy efficacy have 73 been linked to reductions in foraging, prey encounter rates, and reproductive investment; 74 behavioral shifts that collectively undermine herbivore suppression despite apparent trophic 75 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 4 continuity (Barton & Schmitz, 2009; Lin et al., 2022). These findings suggest that abiotic stress 76 may disrupt not just species presence but also the behavioral mechanisms that sustain effective 77 natural enemy-prey interactions. One emerging hypothesis is that stress degrades the 78 informational channels, such as chemical cues, that natural enemies rely on to locate, evaluate, 79 and commit to prey-rich habitats, severing top-down control through a breakdown in ecological 80 communication rather than demographic loss (Holopainen et al., 2025; Pinto-Zevallos & Blande, 81 2024). 82 In plant-insect systems, natural enemies locate prey using plant volatile organic 83 compounds (VOCs), including herbivore-induced plant volatiles (HIPVs), which are released in 84 response to herbivory (Clavijo McCormick, 2016). HIPVs often convey information about prey 85 presence, abundance, and quality, guiding natural enemy foraging, patch choice, and oviposition 86 (Shiojiri et al., 2010; Turlings & Erb, 2018). However, volatile production can be metabolically 87 costly and are tightly regulated by plant physiological status (Holopainen & Gershenzon, 2010). 88 Drought stress disrupts hormonal signaling pathways, such as abscisic acid, jasmonic acid, and 89 salicylic acid, that govern VOC biosynthesis (He et al., 2025; Holopainen & Gershenzon, 2010; 90 Weldegergis et al., 2015). Such stress-driven disruptions may alter the composition, timing, or 91 intensity of VOC emissions, thereby diminishing the reliability of these cues for predators, even 92 when herbivore damage persists. 93 Many natural enemies respond to specific compounds or ratios within a blend, and slight 94 changes in composition can disrupt recognition (Turlings & Erb, 2018). Changes in both major 95 or minor components can compromise signal identity (Bruce et al., 2010), leading natural 96 enemies to reduce searching, abandon prey-rich patches, or refrain from oviposition (Ali et al., 97 2023; De Rijk et al., 2016). Such shifts may reduce top-down control independently of natural 98 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 5 enemy abundance. More broadly, natural enemy responses are likely shaped by the odor 99 environment available when a patch is first encountered, which reflects an integrated plant 100 response to ongoing herbivory rather than a momentary snapshot of damage. Oviposition is 101 particularly sensitive: for many carnivorous insect, including ladybugs, VOC cues guide 102 reproductive investment (Riddick, 2020; Verheggen et al., 2008; Xiu et al., 2019). Plant VOCs 103 help them assess whether a site has sufficient prey to support offspring development (Peñaflor et 104 al., 2011; Riddick, 2020). If drought reduces plant VOC reliability, natural enemies may forego 105 oviposition, weakening top-down control across generations. 106 While drought often suppresses VOC production, some studies suggest that high 107 herbivore densities may partially compensate by intensifying feeding damage, potentially 108 enhancing HIPV output or shifting blend composition (Horiuchi et al., 2003; Shiojiri et al., 109 2010). This raises the possibility that increased herbivore density, by intensifying feeding 110 damage, could, in theory, mitigate signal suppression under stress. However, the outcome likely 111 depends on whether plants under stress maintain the capacity to scale plant VOC output with 112 damage, and whether the resulting blends remain behaviorally relevant to natural enemies. In 113 some cases, stress may favor deterrent or non-informative compounds that are less effective in 114 guiding natural enemies (Rahman et al., 2025). Natural enemies often rely on a limited subset of 115 volatiles, so-called “keystone infochemicals”, to make foraging and oviposition decisions (Ali et 116 al., 2023; Turlings & Erb, 2018). This raises a critical question: does drought selectively reduce 117 these key compounds, even when total emission or herbivore pressure remains high? If so, top-118 down control may fail not due to signal loss, but loss of important signaling compounds. 119 Understanding how stress shapes keystone volatiles is therefore essential for evaluating the 120 resilience of chemical communication under environmental stressors. 121 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 6 Despite growing evidence that both abiotic and biotic factors shape plant VOC signaling, 122 few studies test how both abiotic and biotic stressors interact to shape chemical signaling, 123 predator behavior, and ecological outcomes. Most examine emission or behavior in isolation, 124 limiting insight into full trophic pathway from cue production to herbivore suppression. 125 Crucially, the potential for herbivore density to modulate signal output under stress is rarely 126 examined in tandem with predator decision-making. Even fewer address reproductive decisions, 127 like oviposition, which may offer strong indicators of signal perception and predator 128 commitment. Moreover, volatile blends are often treated as uniform signals, overlooking 129 predators reliance on a limited subset of behaviorally active compounds. As a result, it is 130 unknown whether density-driven amplification of plant VOCs under stress preserves the 131 functional components needed for predator engagement, or if abiotic stress selectively 132 undermines chemical communication even at high herbivore pressure. Understanding this 133 interaction is key to predicting whether prey density can buffer, or is ultimately overridden by, 134 abiotic constraints on top-down control. 135 Here, we test whether drought and herbivore density jointly influence trophic interactions 136 by altering plant VOC signaling and predator responses. Specifically, we test whether: (1) 137 drought reduces or alters plant VOC emission, (2) high herbivore density can restore signal 138 quality or relevance, (3) predators adjust foraging and oviposition in response to these cues (4) 139 specific plant VOCs are predictive of predator behavior, reflecting their role as ecologically 140 relevant cues that guide predator foraging and oviposition, and whether these compounds are 141 selectively suppressed under drought and (5) herbivore suppression by predator depends on both 142 signal reliability and prey availability. By reframing predator-prey interactions as signal-143 contingent processes, this study proposes a new hypothesis in multitrophic ecology: one where 144 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 7 the reliability, composition, and behavioral interpretability of cues determine the strength of top-145 down ecological regulation under stress. 146 147 2. Materials and methods 148 2.1 Study system 149 The tri-trophic model system comprised tomato (Solanum lycopersicum cv. Moneymaker) as the 150 host plant, the potato aphid (Macrosiphum euphorbiae (Homoptera: Aphididae)) as the 151 herbivore, and the convergent lady beetle (Hippodamia convergens (Coleoptera: Coccinellidae)) 152 as a generalist predator. We conducted a greenhouse experiment between May and August 2025 153 to investigate how drought and herbivore density interact to influence plant physiology, volatile 154 emissions, predator behavior, and trophic dynamics in a tritrophic system. 155 156 2.2 Plant cultivation and water treatments 157 Tomato seeds were sown in 10-cm pots containing Sunshine Mix #1 (Sungro Horticulture, USA) 158 and maintained in a greenhouse under controlled environmental conditions (22 ± 2/i2 °C, 16:8 h 159 light:dark photoperiod) (Fig. 1). After germination, seedlings were thinned to one per pot and 160 fertilized using a 15-9-12 NPK slow-release formulation (Osmocote Plus, Scotts Miracle-Gro). 161 Plants were allowed to grow for four weeks until they reached the four-leaf stage, at which point 162 water treatments were initiated (Fig. 1). 163 To impose drought conditions, we manipulated soil volumetric moisture content 164 beginning five days prior to aphid introduction. Well-watered plants were maintained at 75-80% 165 of pot capacity, while drought-stressed plants were maintained at 10-15% (Fig. 1). We verified 166 and adjusted moisture levels every other day using an ECOWITT WH0291 soil moisture probe 167 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 8 (ECOWITT, Shenzhen, China), and calculated irrigation volumes using a gravimetric calibration 168 curve. 169 170 Fig. 1 Experimental timeline and schematic overview of the experimental sequence showing key 171 phases. The timeline illustrates when each major activity occurred, including plant trait 172 measurements (stomatal conductance and volatile collection), aphid and predator introductions, 173 behavioral observations, and final harvest for plant biomass. 174 175 2.3 Insect colonies and inoculation protocol 176 We maintained M. euphorbiae colonies on tomato plants in growth chambers set to 22 °C with a 177 16:8 h light:dark cycle. For experimental inoculations, we used a mixture of apterous adults and 178 late-instar nymphs. Each plant received either 10 (Low) or 100 (High) apterous aphids of mixed 179 age on its 3rd fully expanded leaf, which was enclosed with fine mesh sleeve to allow the plants 180 to emit HIPVs reflective of treatment intensity. 181 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 9 Predators were adult H. convergens beetles, which were reared on tomato plants with 182 potato aphids at 23 °C with a 16:8 h light:dark cycle. Prior to release, we selected newly 183 emerged, mated females for use in assays. To standardize foraging motivation, we starved 184 individuals for 24 hours before introducing them into the experimental cages. 185 186 2.4 Greenhouse mesocosm setup 187 To assess predator foraging and oviposition, and aphid suppression across treatments, we 188 established mesh cages mesocosms (40 × 60 × 100 cm; BioQuip Amber Lumite) in the 189 greenhouse. The experiment followed a fully factorial design, manipulating two independent 190 variables at plant level: water (well-watered vs. drought) and aphid density (Low (10 aphids per 191 plant) vs. High (100 aphids per plant); Fig. 1). These treatments were fully crossed, resulting in 192 four unique plant-level treatment combinations: well-watered with low aphid density, well-193 watered with high aphid density, drought-stressed with low aphid density, and drought-stressed 194 with high aphid density. Each mesocosm (cage) contained one plant from each of these four 195 treatment groups, arranged in randomized positions. A third factor, predator presence (present vs. 196 absent), was manipulated at the cage level, with replicate cages assigned to either predator-197 present or predator-absent treatments (Fig. 1). 198 2.5 Plant traits 199 We measured stomatal conductance (gsw) using a LI-600 Porometer (LI-COR Environmental, 200 Lincoln, NE, USA) on the fourth fully expanded leaf of each plant prior plant volatile collection 201 (N = 14 replicates across two trials) (Fig. 1). At the end of the experiment, aboveground fresh 202 biomass was quantified by harvesting and immediately weighing all shoot tissue using a 203 precision balance (N = 14 replicates across two trials) (Fig. 1). 204 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 10 205 2.6 Volatile collection and chemical analysis 206 We collected plant VOCs 72 hours after aphid inoculation to test how aphid-density and 207 water treatments shaped volatile emissions prior to predator introduction. Our VOC analysis was 208 designed to compare treatment-specific odor environments, rather than to partition constitutive 209 and aphid-induced emissions. We enclosed the third fully expanded leaf of each plant in a 1.2 L 210 polyethylene cup and collected volatiles for six hours (airflow rate: 500 mL/min) using a custom-211 built headspace sampling system (Figure 4-1). Volatile compounds were trapped on HayeSep-Q 212 filters (Supelco, Bellefonte, PA, USA), eluted with 200 μ L of dichloromethane, and spiked with 213 nonyl acetate (2 ng/μ L) as an internal standard. Volatile extracts were analyzed using gas 214 chromatography-mass spectrometry (GC-MS; Agilent 7890A/5975C). One microliter of each 215 sample was injected in splitless mode at 250 °C, with helium as the carrier gas at 0.7 mL min/i2 1. 216 Compounds were separated on an HP-5MS column (30 m × 0.32 mm, 0.25 µm film thickness; 217 Agilent Technologies) using an oven program of 40 °C for 2 min, followed by an increase of 10 218 °C min/i2 1 to 300 °C, with a final hold of 4 min. The MS was operated in positive EI mode. Peaks 219 were deconvoluted in MassHunter Unknowns Analysis (Agilent Technologies, Santa Clara, CA, 220 USA), tentatively identified by comparing mass spectra with the NIST17 and Adams libraries 221 spectra and confirmed by matching retention index with published sources, including The 222 Pherobase (Adams, 2007; Trase et al., 2025). The relative amounts of the detected compounds 223 were then determined by referencing their total ion chromatogram (TIC) peak areas to those 224 obtained for the internal standard, nonanyl acetate, at 2 ng/µL. 225 226 2.7 Predator behavior and aphid suppression 227 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 11 Following plant volatile collection, we released a single 24h-starved H. convergens female adult 228 at the center of each cage (Fig. 1). Each treatment had 14 replicates across two trials. To evaluate 229 sustained predator orientation over time, we monitored ladybeetle location in cage across six 230 consecutive days. We recorded the plant occupied, or visitation behavior, of each individual at 231 multiple timepoints across the trial period. Observations were made twice on days 2, 4, and 6, 232 and three times on days 3 and 5, for a total of twelve observations per individual. In both trials, 233 ladybug began ovipositing by Day 3. We recorded the presence and location of eggs at 72-, 88-, 234 and 110-hours post-release to determine where ladybeetle chose to oviposit (i.e., oviposition 235 behavior). 236 To assess predator-mediated aphid suppression, we conducted aphid counts at day 2, day 237 4, and day 6 after H. convergens release (Fig. 1). At each time point, we recorded the total 238 number of aphids per plant, including both apterous and alate morphs, and counted the number 239 of newly produced nymphs. Because aphids were free to move among plants within each cage, 240 predator effects on individual plants could reflect a combination of consumptive (lethal) and non-241 consumptive (behavioral or trait-mediated) effects, including deterrence or spatial displacement. 242 243 2.9 Statistical analysis 244 All data were analyzed in R (v4.3.1). 245 Plant traits: Stomatal conductance was measured prior to predator introduction and thus was 246 modeled as a function of water availability and aphid density (water × density). In contrast, fresh 247 biomass was an endpoint measurement collected after predator exposure, and its model included 248 all main effects and their interactions (water × density × predator). Each trait was analyzed 249 using a generalized linear mixed model (GLMM) fitted with the glmmTMB package (Brooks et 250 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 12 al., 2017). Fixed effects included all main effects and their interactions, while plant ID nested 251 within cage and block (1| Block/Cage/Plant ID) was included as a random effect to account for 252 non-independence within experimental units. For stomatal conductance, we specified a Tweedie 253 distribution to account for overdispersion and non-Gaussian error structure. For fresh biomass, a 254 Gaussian distribution was appropriate based on residual diagnostics. Model fit and dispersion 255 were assessed using simulated residuals via the DHARMa package (Hartig & Hartig, 2017). 256 Type III Wald χ 2 tests were conducted using the Anova() function from the car package (Fox & 257 Weisberg, 2018) to evaluate the significance of fixed effects. We performed post-hoc 258 comparisons using estimated marginal means (emmeans package) (Lenth & Lenth, 2018), and 259 visualized group differences with compact letter displays using Tukey-adjusted comparisons via 260 the multcomp (Hothorn et al., 2016). 261 262 Plant volatile organic compounds (VOCs): To examine plant volatile profiles in response to 263 aphid density and water availability, we collected VOCs prior to predator introduction (see 264 Section 2.6). We analyzed VOC data across two complementary approaches: 265 (a) Volatile composition: We evaluated multivariate differences in plant VOC 266 composition using permutational multivariate analysis of variance (PERMANOVA) via the 267 adonis2() function in the vegan package (Oksanen et al., 2013). The analysis was based on 268 Jaccard distances calculated from presence/absence-transformed VOC data with 999 269 permutations, testing the effects of water treatment, aphid density, and their interaction. We 270 tested the effects of water treatment, aphid density, and their interaction, and we assessed 271 homogeneity of dispersion with betadisper(). We further explored significant treatment contrasts 272 with pairwise PERMANOV A and applied FDR correction to the resulting P-values. To visualize 273 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 13 VOC composition patterns across treatments, we used CAP/dbRDA (capscale()), and to identify 274 individual VOCs associated with treatment separation, we fitted compounds to the CAP 275 ordination with envfit(), using 999 permutations and FDR correction across compounds. 276 (b) Volatile analysis by individual compounds: A multivariate analysis of variance 277 (MANOVA) was not used due to violations of its core assumptions. All response variables 278 (VOCs) deviated significantly from normality (Shapiro-Wilk P<0.001). Given the zero-inflated 279 and non-normal nature of VOC data, we analyzed the emission rates of individual plant VOCs 280 using GLMMs (glmmTMB package) with tweedie distribution and zero inflation. Fixed effects 281 included water treatment (well-watered vs. drought), aphid density (Low (10) vs. high (100)), 282 and their interaction, with plant ID nested within cage and block included as a random effect. 283 Type II Wald χ 2 tests (car package) were used to assess fixed effects. 284 285 Predator behavioral responses: To capture the full spectrum of ladybeetle behavioral responses, 286 we conducted separate analyses for two distinct behaviors: visitation (i.e., plant choice) and 287 oviposition. These behaviors were modeled independently using statistical frameworks 288 appropriate for each response type, as described below. 289 Predator visitation behavior: We analyzed the ladybeetle visitation behavior (i.e., visits 290 or choices) using a conditional logit model (clogit, survival package (Therneau, 2015)) with 291 plant visitation (chosen = 1 vs. 0) as the binary response. We included water treatment (well-292 watered vs. drought), inoculation density (low (10) vs. high (100)), and their interaction as fixed 293 effects, and stratified model by choice set (cage × timepoint) to account for repeated measures 294 within cages. We assessed the significance of model terms using Type III Wald χ 2 tests (car 295 package). 296 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 14 Predator oviposition behavior: We analyzed ladybeetle oviposition behavior using a 297 Bayesian logistic regression framework implemented in the brms package (Bürkner, 2017). The 298 binary response variable indicated whether or not an individual plant received any oviposition (1 299 = oviposition, 0 = none). Each plant was classified by its water availability treatment (well-300 watered or drought) and aphid density (low (10) or high (100)). We modeled these effects across 301 three time points (72, 88, and 110 hours after introduction) and incorporated a random intercept 302 for each unique choice set (defined by trial, cage, and time point) to account for non-303 independence within trials. The primary model included fixed effects for time point, water 304 treatment, aphid density, and their two-way interactions (time × water, time × density). This 305 model was preferred over the full three-way interaction model (time × water × density), which 306 showed poor convergence (high R-hat and exceeded tree depth), likely due to 307 overparameterization relative to sample size. A second model, collapsed across time points, 308 included only water × density and a random intercept for choice set (i.e., cage). All models used 309 a Bernoulli distribution with a logit link function. We ran four chains of 4000 iterations each 310 (2000 warmup), using adapt_delta = 0.99 and max_treedepth = 12 for robust sampling. Posterior 311 summaries were derived using median estimates and 95% highest posterior density (HPD) 312 credible intervals. 313 314 Role of plant volatiles in ladybeetle visitation and oviposition: We evaluated how plant volatiles 315 influenced two binary behavioral responses by the ladybeetle H. convergens: (1) visitation and 316 (2) oviposition. We modeled predator visitation and oviposition as a function of VOC profiles 317 measured following sustained herbivory. Because herbivore-induced VOC blends reflect 318 integrated plant responses to herbivory and provide a biologically relevant snapshot of the odor 319 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 15 environment available to predators at the onset of the assay, these profiles were used as proxies 320 for the cues underlying predator decision-making despite non-synchronous sampling. In both 321 cases, the response variable was coded as 1 if the event occurred (i.e., ≥ 1 choice or egg laid) or 0 322 otherwise. To reduce dimensionality and identify informative predictors, we first used a random 323 forest classifier (randomForest package) (Liaw & Wiener, 2002) to rank volatile compounds by 324 importance. We retained the top 19 compounds that contributed most to classification accuracy 325 (for oviposition) or node purity (for choice). 326 We then fit Bayesian logistic regression models using the brms package, specifying a 327 Bernoulli likelihood with a logit link and weakly informative priors (normal(0, 5)) on the 328 regression coefficients. Predictors were z-score standardized prior to model fitting. Model 329 performance was assessed using Bayesian R2, posterior predictive checks, and leave-one-out 330 cross-validation (loo package) (Vehtari et al., 2021). To further identify the most predictive 331 volatiles, we applied projection predictive variable selection using the projpred package 332 (Piironen et al., 2023). Final reduced models were refit using only the selected subset of 333 volatiles. For each reduced model, we extracted posterior summaries using the bayestestR 334 package (Makowski et al., 2019) and interpreted effects as significant when 95% credible 335 intervals excluded zero. To evaluate binary classification performance, we used the posterior 336 epred() function to compute posterior mean probabilities for each observation. Predictions had a 337 threshold at 0.5, and classification metrics including accuracy, sensitivity, specificity, precision, 338 and F1 score were calculated against the observed binary responses. Receiver operating 339 characteristic (ROC) curves and area under the curve (AUC) values were generated using the 340 pROC package (Robin et al., 2021). 341 342 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 16 Predator suppression of aphid population and demographics: To evaluate the effects of predator 343 presence, water availability, and aphid density on aphid population dynamics, we analyzed four 344 response variables: aphid population growth rate (calculated as: dN/Ndt) = ln (N2- N1)/(t2-t1), 345 where N1 and N2 are initial and final aphid densities at time t2 and t1, respectively (Gotelli, 346 1995)), final density, nymph count, and alate count. Each response was modeled separately using 347 GLMMs implemented in the glmmTMB package. Growth rate was treated as a continuous 348 variable and modeled using a Gaussian distribution. The count-based responses were modeled 349 using a negative binomial distribution to account for overdispersion. Models initially included all 350 main effects and interactions among water, inoculation density, predator, and day (representing 351 the three time points: day 2, 4, and 6), with final models selected by comparing full and reduced 352 versions based on AIC and model diagnostics. The variable ‘day’ was retained in all models to 353 account for temporal structure, and its interactions with other predictors were considered when 354 relevant. To account for the nested experimental design, we included plant ID nested within cage 355 and block as a random effect in all models. Residual diagnostics were performed using the 356 DHARMa package to assess model fit, dispersion, and distributional assumptions. Significance 357 of main effects and interactions was assessed using Type III Wald chi-square tests via the 358 Anova() function (car package). For post hoc comparisons, we estimated marginal means using 359 the emmeans package and visualized group differences using compact letter displays (multcomp 360 package) with Tukey-adjusted comparisons. 361 362 3. Results 363 3.1 Plant traits 364 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 17 Stomatal conductance was reduced by approximately 96% under drought conditions 365 (χ 2 [1]=118.93, P<0.001), while aphid density (χ 2 [1]=0.09, P=0.76) and its interaction with water 366 availability (χ 2 [1]=0.004, P=0.95; Fig. 2a) were not significant. Because stomatal conductance 367 was measured prior to predator introduction, predator presence was not included in this model. 368 For aboveground fresh biomass, drought also led to approximately 74% reduction (χ 2 [1]=249.32, 369 P<0.001), while aphid density (χ 2 [1]=0.01, P=0.91), predator presence (χ 2 [1]=0.15, P=0.70), and 370 their two-way interactions (water × density: χ 2 [2]=1.27, P=0.26; water × predator: χ 2 [2]=0.44, 371 P=0.51; density × predator: χ 2 [2]=0.10, P=0.75; Fig. 2b) were non-significant. The three-way 372 interaction among water availability, aphid density, and predator presence was also not 373 significant (χ 2 [3]=0.004, P=0.95; Fig. 2b). 374 375 Fig. 2 Effects of aphid density (low vs. high) and water availability (well-watered vs. drought) 376 on (a) stomatal conductance (mmol m-2 s-1) and (b) aboveground fresh biomass (g). In (b), 377 columns show predator presence (left: absent; right: present). Stomatal conductance was 378 measured before predator introduction. Points show mean ± SE. Blue = well-watered; orange = 379 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 18 drought; lighter shades = low aphid density, darker = high. Different letters indicate significant 380 treatment differences (Tukey-adjusted, P<0.05). 381 382 3.2 Plant VOCs 383 (a) Volatile composition: 384 PERMANOVA based on Jaccard dissimilarities of presence–absence volatile profiles 385 revealed a significant water treatment × aphid density interaction (F[1,106]=0.96, R2=0.0087, 386 P=0.042; Fig. 3), indicating that treatment combinations differed in VOC composition. Tests for 387 homogeneity of multivariate dispersion were not significant (betadisper: F[3,106]=0.123, P=0.947; 388 permutation test P=0.949), showing that differences among treatments were not driven by 389 unequal within-group dispersion. Pairwise PERMANOVA comparisons revealed several 390 significant differences in VOC profiles across specific treatment combinations. Pairwise 391 PERMANOVA indicated significant differences in community composition among all treatment 392 combinations after FDR correction. The strongest differences involving the well-watered low 393 density treatment were observed relative to drought high density (F[1,53]=1.80, R2=0.0329, 394 P=0.0012; Fig. 3), drought low density ((F[1,53]=1.67, R2=0.0306, P=0.0012), and well-watered 395 high density (F[1,52]=1.17, R2=0.0220, P=0.0012). Well-watered high density also differed 396 significantly from drought low density (F[1,53]=1.82, R2=0.0332, P=0.0012) and drought high 397 density (F[1,53]=1.61, R2=0.0295, P=0.0012; Fig. 3). The contrast between drought low and 398 drought high density was also significant, though weaker in magnitude than the other pairwise 399 comparisons (F[1,54]=1.08, R2=0.0196, P=0.009). Environmental fitting (envfit) analysis of 400 NMDS ordination identified 37 volatile compounds that were significantly associated with the 401 ordination axes (P<0.05), indicating that their emission patterns were structured by treatment 402 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 19 effects. Full results, including vector coordinates, R2 values, and significance levels, are 403 presented in Supplementary table S1. 404 (b) Volatile analysis by individual compounds: Volatile composition differed among treatment 405 combinations at the multivariate level. A principal components MANOVA on the first 20 PCs 406 detected a significant Water × Density interaction (Pillai’s trace = 0.347, F[20,87]=2.31, P=0.004), 407 Fig. 3 Constrained analysis of principal coordinates (CAP/dbRDA) of plant VOC composition based on Jaccard dissimilarities calculated from presence/absence data across water and aphid density treatments. Points represent individual plants, with orange indicating drought and blue indicating well-watered treatments. Shapes denote treatment combinations: low aphid density as circles or diamonds and high aphid density as squares or triangles. Large symbols indicate treatment centroids, and gray line segments connect individual samples to their respective centroids. Letters identify treatment groups for visual reference. .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 20 demonstrating that the effect of water regime on the volatile blend depended on aphid density. 408 To attribute this interaction to individual compounds, we combined PCA-derived interaction 409 driver scores with compound-specific generalized linear mixed models. Driver scores quantify 410 each compound’s contribution to the multivariate Water × Density separation, while mixed 411 models provide inferential tests of Water, Density, and their interaction. All Wald χ 2 statistics 412 (df=1) and P values are presented in Table S2. 413 Eight compounds exhibited significant Water × Density interactions, directly supporting 414 the non-additive multivariate response. An additional 20 compounds showed significant main 415 effects in the absence of interaction: 13 exhibited a significant effect of Water only, 2 exhibited a 416 significant effect of Density only, and 5 exhibited significant effects of both factors. In total, 28 417 compounds displayed at least one significant fixed effect and are reported in Table 1 and 418 Supplementary Table S2. 419 Table 1: Compounds exhibiting significant fixed effects in generalized linear mixed models. For each compound and significant factor (Water × Density interaction, Water, or Density), the table reports the factor-specific multivariate driver score, its absolute magnitude (|driver score|), and the corresponding Wald χ 2 statistic (Df = 1) with P-value. Driver scores were derived by projecting the principal components effect vectors for each factor back into compound space using PCA loadings and quantify each compound’s contribution to multivariate treatment separation. Only statistically significant effects (P < 0.05) are shown. Compound Factor Driver score Driver score (absolute) X 2 Df P-value p-Cymene Water x Density -0.511 0.511 4.805 1 <0.05 Pentanol, 3-methyl- Water x Density 0.369 0.369 15.527 1 <0.0001 Heptanal Water x Density -0.179 0.179 4.723 1 <0.05 Myrcenone Water x Density 0.152 0.152 75924.612 1 <0.0001 Artemisia ketone Water x Density 0.082 0.082 5.686 1 <0.05 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 21 3.3 Predator preference and oviposition behavior 420 1-Hexanol Water x Density 0.079 0.079 4.882 1 <0.05 Penten-1-al, 2E- Water x Density 0.067 0.067 6.911 1 <0.01 2(5H)-Furanone Water x Density 0.045 0.045 56.793 1 <0.0001 Octen-3-one, 1- Water 0.342 0.342 94.704 1 <0.0001 Unknown 1 (RI 860.7) Water -0.311 0.311 4.959 1 <0.05 Nonanal, n- Water -0.311 0.311 4.679 1 <0.05 Sabinene Water 0.29 0.29 4.611 1 <0.05 Cumene Water -0.286 0.286 16.075 1 <0.0001 Hexenyl Isobutanoate, 3Z- Water 0.271 0.271 17.822 1 <0.0001 Ocimene, (Z)-beta- Water 0.204 0.204 6.76 1 <0.01 Benzene acetaldehyde Water -0.165 0.165 32.426 1 <0.0001 Longicyclene Water 0.152 0.152 11.464 1 <0.001 Octane, 4-methyl- Water -0.146 0.146 4.303 1 <0.05 Decane, n- Water -0.141 0.141 5.12 1 <0.05 Santalol acetate, (Z)- epi-beta- Water -0.118 0.118 4.649 1 <0.05 Pinocampheol Water -0.109 0.109 13.832 1 <0.001 Dihydroisojasmone Water 0.107 0.107 4.109 1 <0.05 1-Hexanol Water -0.101 0.101 45.215 1 <0.0001 heptane, 4-methyl- Water -0.097 0.097 6.169 1 <0.05 Dodecene, 1- Water 0.073 0.073 6.95 1 <0.01 Butyl acetate Water 0.069 0.069 11.423 1 <0.001 Indanol, 5- Water -0.052 0.052 4.805 1 <0.05 Myrcenone Water -0.029 0.029 53585.704 1 <0.0001 Methyl salicylate Density -0.335 0.335 17.712 1 <0.0001 Lavandulyl, tetrahydro- Density 0.256 0.256 9.828 1 <0.01 heptane, 4-methyl- Density 0.221 0.221 8.705 1 <0.01 Longicyclene Density -0.208 0.208 25.25 1 <0.0001 1-Hexanol Density -0.187 0.187 51.6 1 <0.0001 Octen-3-one, 1- Density -0.151 0.151 19.614 1 <0.0001 Artemisia ketone Density 0.119 0.119 5.986 1 <0.05 Dodecene, 1- Density 0.059 0.059 7.709 1 <0.01 Benzene acetaldehyde Density -0.022 0.022 26.694 1 <0.0001 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 22 (a) Predator visitation behavior: Predator were approximately 4.5 times more likely to select 421 plants with high-aphid density compared to those with low-aphid density (χ 2 [1]=10.79, P=0.001; 422 Fig. 4). Drought stress reduced visitation, with ladybeetles 69% less likely to choose drought-423 stressed plants over well-watered ones (χ 2 [1]=22.01, P<0.001; Fig. 4). There was no significant 424 interaction between aphid density and water availability (χ 2 [1]=0.12, P=0.725; Fig. 4), indicating 425 that their effects were additive. 426 427 Fig. 4 Predicted probability of predator visitation across aphid density and water availability 428 treatments. Points show estimated marginal means (±95% CI) from a conditional logistic 429 regression. Blue = well-watered; orange = drought; lighter = low aphid density, darker = high. 430 Letters denote significant differences (Tukey-adjusted, P<0.05). 431 432 (b) Predator oviposition behavior: The probability of oviposition by female ladybeetle varied 433 with plant water availability, aphid prey density, and time (Fig. 5). Across all time points, 434 oviposition was highest on well-watered plants with high aphid density (probability = 0.477; 435 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 23 95% HPD: 0.331–0.635). In contrast, drought-stressed plants with low aphid density had near-436 zero oviposition (0.000; 95% HPD: 0.000–0.013). Intermediate probabilities occurred on well-437 watered low-aphid (0.160; 95% HPD: 0.064–0.282) and drought high-aphid plants (0.161; 95% 438 HPD: 0.057–0.277), indicating independent and interactive effects of water and prey cues. 439 Temporal patterns revealed peak selectivity early in the experiment, before any eggs were 440 present. At 72 hours, oviposition was strongly concentrated on well-watered/high-aphid plants 441 (0.727; 95% HPD: 0.492–0.935), with little to none on other treatments. At 88 hours, this 442 preference persisted (0.604; 95% HPD: 0.353–0.835), though oviposition on drought/high-aphid 443 plants increased slightly (0.105; 95% HPD: 0.005–0.272). By 110 hours, oviposition 444 probabilities converged across treatments (range: 0.092–0.242), with overlapping intervals 445 suggesting reduced selectivity as eggs accumulated. 446 Model estimates confirmed strong effects of aphid density and water status: oviposition 447 was less likely on low-aphid plants (estimate = −19.40; 95% CI: −105.01 to −2.09), and more 448 likely under well-watered conditions (estimate = 1.59; 95% CI: 0.55–2.71). A positive 449 interaction between well-watered plants and low aphid density (estimate = 17.81; 95% CI: 0.36–450 103.67) suggests flexible decision-making based on cue combinations. 451 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 24 452 Fig. 5 Predicted oviposition probabilities of female ladybeetles across water (well-watered vs. 453 drought-stressed) and aphid density (low vs. high) treatments, shown cumulatively and by time 454 point. Posterior means and 95% HPD intervals are based on a Bayesian logistic regression. Panel 455 (a) shows estimates across all time points; panels (b)–(d) show predictions at 72, 88, and 110 456 hours. Circles/squares = well-watered (low/high density); diamonds/triangles = drought-stressed 457 (low/high). Lighter shades = low, and darker = high aphid density. 458 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 25 459 3. 4 Predator visitation behavior in response to plant volatiles: Bayesian logistic regression 460 identified four volatile compounds with credible associations with H. convergens olfactory 461 choice behavior. The model included 19 candidate compounds selected from the random forest 462 importance ranking and evaluated in a Bayesian framework. Overall model fit was moderate, 463 with a Bayesian R-squared of 0.44 and a 95% credible interval of 0.33 to 0.52, and leave-one-out 464 cross-validation indicated reasonable predictive performance with a LOOIC of 116.3. Four 465 compounds had 95% credible intervals that excluded zero, indicating credible effects on 466 ladybeetle choice probability (Fig. 6a-d). Three compounds were positively associated with the 467 probability of ladybeetle choice: methyl salicylate (β = 4.51, 95% credible interval = 1.04 to 468 9.06; Fig. 6a), β -myrcene (β = 4.50, 95% credible interval = 1.30 to 8.09; Fig. 6b), and valencene 469 (β = 2.77, 95% credible interval = 0.67 to 5.06; Fig. 6c). In contrast, 1-octanol was negatively 470 associated with choice probability (β = -2.69, 95% credible interval = -5.41 to -0.17; Fig. 6d). 471 Model discrimination was strong, with an area under the ROC curve of 0.903. Using a 0.5 472 classification threshold, the model achieved an accuracy of 0.833, sensitivity of 0.943, specificity 473 of 0.538, precision of 0.846, and an F1 score of 0.892. Together, these results indicate that adult 474 ladybeetles respond selectively to particular plant volatiles, with methyl salicylate, beta-myrcene, 475 and valencene acting as potential attractant-associated cues, whereas 1-octanol appears to reduce 476 the likelihood of choice. 477 478 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 26 479 Fig. 6 Predicted probability of ladybug preference from Bayesian logistic regression as a 480 function of compound concentration (Z-score standardized). Shaded areas represent 95% 481 credible intervals. Labels above each panel show the posterior median and 95% credible interval 482 for the estimated effect (β ). 483 484 3.5 Predator oviposition behavior in responses to volatile compounds: Bayesian logistic 485 regression identified 12 volatile compounds with credible associations with H. convergens 486 oviposition behavior. The reduced model included 19 candidate compounds and showed strong 487 explanatory power, with a Bayesian R-squared of 0.69 and a 95% credible interval of 0.58 to 488 0.79. Leave-one-out cross-validation indicated good apparent predictive performance, with a 489 LOOIC of 87.0, although several high Pareto k values suggest that these cross-validation results 490 should be interpreted cautiously. Twelve compounds had 95% credible intervals that excluded 491 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 27 zero, indicating credible effects on oviposition probability (Fig. 7a to i). Five compounds were 492 positively associated with ladybeetle oviposition: methyl salicylate (β = 6.22, 95% credible 493 interval = 3.08 to 10.08; Fig. 7a), n-heptanal (β = 6.13, 95% credible interval = 2.75 to 9.79; Fig. 494 7b), p-cymene (β = 6.31, 95% credible interval = 2.01 to 11.15; Fig. 7c), carvacrol (β = 4.64, 495 95% credible interval = 1.32 to 8.76; Fig. 7d), and 1-hexanol (β = 3.15, 95% credible interval = 496 0.98 to 5.49; Fig. 7e). Seven compounds were negatively associated with oviposition: 1-octanol 497 (β = -3.29, 95% credible interval = -6.98 to -0.45; Fig. 7f), n-hexanal (beta = -6.15, 95% credible 498 interval = -10.13 to -2.64; Fig. 7g), dihydroisojasmone (β = -5.39, 95% credible interval = -9.35 499 to -1.96; Fig. 7h), n-nonanal (β = -3.47, 95% credible interval = -6.58 to -0.62; Fig. 7i), α-pinene 500 (β = -4.09, 95% credible interval = -7.40 to -0.98; Fig. 7j), cryptone (β = -5.83, 95% credible 501 interval = -9.85 to -2.16; Fig. 7k), and para-cymen-7-ol (β = -3.36, 95% credible interval = -6.41 502 to -0.73; Fig. 7i). Model discrimination was extremely strong, with an area under the ROC curve 503 of 0.990. Using a 0.5 classification threshold, the model achieved an accuracy of 0.958, 504 sensitivity of 0.955, specificity of 0.962, precision of 0.955, and an F1 score of 0.955. Together, 505 these findings indicate that adult ladybeetle oviposition potentially responds strongly and 506 selectively to particular volatile compounds, with some compounds increasing the probability of 507 egg laying and others acting as deterrent cues. 508 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 28 509 Fig. 7 Predicted probability of oviposition from Bayesian logistic regression as a function of 510 compound concentration (Z-score standardized). Each panel shows the marginal effect of a 511 selected compound or treatment, with 95% credible intervals (shaded). Posterior coefficient (β ) 512 and its 95% CI are shown in panel titles. 513 514 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 29 3.6 Predator suppression of aphid population and demographics: Per capita aphid population 515 growth rate varied significantly over time and was shaped by interactions among water 516 availability and predator presence (water × predator × day: χ 2 [2]=6.85, P=0.032; Fig. 8a). Aphid 517 inoculation density alone did not significantly affect per capita growth rates (χ 2 [1]=0.0629, 518 P=0.802). Under well-watered and high aphid density, predator presence reduced aphid per 519 capita population growth rate by approximately 70%, whereas under drought, growth remained 520 near zero regardless of predator presence. 521 Final aphid density was driven largely by initial aphid density, time, and their interactions 522 with water availability (density × day: χ 2 [2]=153.47, P<0.001; water × day: χ 2 [2]=248.43, 523 P<0.001; Fig. 8b). A significant three-way interaction (water × predator × day: χ 2 [2]=19.11, 524 P<0.001) confirmed that predator effects depended on both resource supply and time. Predator 525 presence alone did not significantly affect overall aphid densities (χ 2 [1]=0.03, P=0.86). However, 526 in well-watered, high-density treatments, predators reduced final aphid numbers by roughly 40% 527 by the end of the experiment. Under drought, predator effects were negligible. 528 For nymph densities, significant interactions occurred for water × day (χ 2 [2]=163.50, 529 P<0.001) and day × density (χ 2 [2]=91.28, P<0.001; Fig. 8c), indicating that nymph production 530 was both temporally dynamic and resource dependent. Predator effects were smaller but 531 significant (χ 2 [1]=5.61, P=0.017), leading to an average reduction of around 15–20% in nymph 532 densities. 533 Alate formation exhibited strong responses to both abiotic and biotic factors, with 534 significant main and interaction effects (day: χ 2 [2]=160.63, P < 0.001; water × predator: 535 χ 2 [1]=65.30, P<0.001; density × predator: χ 2 [2]=64.29, P<0.001; Fig. 8d). Under well-watered, 536 high-aphid density conditions, predator presence reduced alate numbers by about 55%, whereas 537 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 30 under drought, predators increased alate density by over 100%, suggesting an induced dispersal 538 response under combined initial aphid density and drought stress. 539 540 Fig. 8 Effects of water availability, initial aphid density, and predator presence on aphid 541 population growth and demography over time. Panels show (a) per capita growth rate, (b) final 542 density, (c) nymph density, and (d) alate density under combinations of water (well-watered vs. 543 drought), aphid density (low vs. high), and predator presence (No vs. Yes). Lines show model-544 predicted means ± 95% CI; points are individual replicates. Blue = well-watered; orange = 545 drought; lighter = low aphid density, darker = high aphid density. 546 547 4. Discussion 548 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 31 Drought is increasingly recognized as a disruptor of trophic interactions, but the mechanisms 549 weakening top-down regulation are not well characterized (Reinecke et al., 2024); here we tested 550 whether drought alters predator-mediated suppression via changes in chemical information that 551 predators use to locate and evaluate high and low prey density patches. Drought reduced plant 552 physiological function and growth, reshaped volatile emission blends, lowered predator 553 visitation, and sharply reduced their reproduction on drought-stressed plants. Under well-watered 554 conditions with high prey density, predators suppressed aphid population growth, suggesting 555 density-mediated top-down effects, and also reshaped aphid demography. Under drought, aphid 556 population growth was constrained by bottom-up limitation, reducing density-mediated predator 557 effects and causing predator influences to be expressed primarily through changes in aphid 558 demographic structure. Thus, drought weakens top-down regulation by disrupting volatile 559 signaling and reducing cues predators use to locate prey. 560 561 Volatile emissions shift with water availability and herbivore density, with consequences for 562 information transfer 563 Multivariate and univariate compound analyses indicated that volatile composition differed 564 among treatments and that water stress interacted with herbivore density, implying that drought 565 changes both the proportional abundance and functional identity of plant volatile blends (Lin et 566 al., 2022; Rahman et al., 2025). Compound such as methyl salicylate, which is known to increase 567 predator attraction and oviposition (Ayelo, Yusuf, et al., 2021; Russavage et al., 2024; Salamanca 568 et al., 2017; Zhu & Park, 2005), was suppressed under drought, particularly when aphid density 569 was low. For instance, drought suppressed emission of predator-attracting volatiles like methyl 570 salicylate and (E,E)-4,8,12-trimethyltrideca-1,3,7,11-tetraene in tomato, weakening parasitoid 571 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 32 attraction and disrupting VOC-mediated indirect defenses (Lin et al., 2022). Similarly, 572 compounds such as (E,E)-α -farnesene and α -terpineol in aphid-infested plants (Truong et al., 573 2014), and methyl salicylate and (E)-β -ocimene in cassava infested by mites (Pinto-Zevallos et 574 al., 2018), were only significantly emitted at high herbivore densities. This density-dependent 575 induction suggests that at lower infestation levels, VOC blends may be too weak or inconsistent 576 to attract natural enemies. More than just reduced volatile output, shift in blend composition, i.e., 577 altered ratios and compound identities, may render the odor profile less recognizable for natural 578 enemies (Turlings & Erb, 2018), undermining predator attraction and weakening indirect 579 defenses even under high density prey. 580 581 Predators integrate compound-level information, and different compounds may matter for 582 attraction versus commitment 583 Our findings show that natural enemies do not respond to plant VOCs as a uniform signal but 584 instead exhibit selective, compound-specific responses, consistent with the idea that predators 585 parse volatile blends into functionally distinct cues (Salamanca et al., 2017). Ladybeetles 586 consistently preferred plant VOC profiles associated with well-watered, high-prey density plants, 587 and this preference aligned with elevated levels of methyl salicylate and myrcene. Notably, 588 methyl salicylate was associated with both visitation and oviposition and was reduced under 589 drought, providing a mechanistic link between abiotic stress, altered blend composition, and 590 reduced predator commitment. Importantly, these relationships were identified using VOC 591 profiles measured following sustained herbivory, suggesting that temporally integrated volatile 592 signals are sufficient to capture the biologically relevant odor environment underlying predator 593 decision-making. In contrast, several compounds that were negatively associated with predator 594 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 33 responses, including α -pinene, 1-octanol, cryptone, and cymene-7-ol, para, illustrate how 595 deterrent volatiles can shape behavioral outcomes. For instance, Yu et al. (2018) reported that α -596 pinene reduced prey-searching behavior in insect predators, reinforcing its role as a context-597 dependent deterrent. These compounds did not necessarily increase in absolute abundance under 598 drought, but shifts in their ratio relative to key candidate attractants may be sufficient to reduce 599 overall blend attractiveness. 600 This underscores the importance of blend composition and proportionality, as predator 601 responses may depend on a hierarchical sensory evaluation, where certain volatiles function as 602 attractant and others act as deterrents within a blend, making even subtle shifts in their ratios 603 critical to how VOC signals are perceived (Lin et al., 2022; Riddick, 2020; Verheggen et al., 604 2008). We also observed stage specificity in cue use: for instance, carvacrol was neutral for 605 initial foraging choice but positively predicted oviposition, suggesting that some volatile 606 compound may function as post-settlement cues that reinforce or modulate initial patch 607 assessments (Ayelo, Yusuf, et al., 2021; Verheggen et al., 2008). 608 Together, these results suggest that predator responses to plant volatiles may be shaped by 609 blend identity and contextual reliability rather than on VOCs presence alone. Such behavior 610 likely reflects evolved decision rules that help predators avoid ovipositing in low-quality or risky 611 patches in heterogeneous environments (Ayelo, Pirk, et al., 2021). Under drought, shift in blend 612 composition, particularly altered ratios and reduced attractant volatiles, may increase signal 613 ambiguity, leading to a higher risks of inflated false negative. As a result, predators may 614 underexploit suitable prey patches when volatile cues no longer exceed the threshold needed to 615 signal patch quality (Tariq et al., 2013). 616 617 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 34 Oviposition reveals a stronger threshold response than visitation, and it changes over time 618 Ladybeetles consistently laid more eggs on well-watered plants with high prey density. In 619 contrast, oviposition on drought-stressed plants, even at high prey density, was rare and only 620 occurred once egg numbers on other treatments were high. This decoupling of prey abundance 621 from predator reproductive investment suggests a behavioral threshold: unless plant VOC blends 622 cross a certain attractant threshold, shaped by their chemical composition and modulated by 623 contextual cues such as prey density and plant water status, predators do not commit to a patch 624 (Verheggen et al., 2008; Xiu et al., 2019). While olfactory cues alone can trigger physiological 625 readiness, such as reduced oosorption and oocyte maturation observed in Harmonia axyridis 626 exposed to aphid-induced volatiles even without prey (Rondoni et al., 2017), actual oviposition 627 likely requires the convergence of plant VOCs with other ecological signals that confirm patch 628 reliability. 629 Importantly, these decisions were temporally structured. Early in the experiment (72-88 630 hours), oviposition was highly selective and tightly aligned with the most attractive blend. By 631 110 hours, selectivity weakened, and eggs appeared more often on initially less preferred plants, 632 indicating that patch evaluation is dynamic rather than fixed (Singh et al., 2019). Future studies 633 should extend VOC sampling across the oviposition window to determine whether shifts in blend 634 identity and ratios correlated with this behavioral transition and to identify the volatile cues that 635 maintain commitment over time. In parallel, as egg loads and predator activity accumulated on 636 the most preferred patches, further oviposition there may have increased risks of egg cannibalism 637 or conspecific predation, potentially favoring a broader distribution of eggs across plants to 638 reduce offspring loss (Singh et al., 2019). 639 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 35 Notably, drought-stressed plants often received fewer or no eggs even at high aphid 640 density, suggesting that prey availability cannot fully compensate for lack of reliable chemical 641 signaling. This highlights a critical asymmetry: while herbivore pressure may amplify plant VOC 642 output in some cases, if drought suppresses the specific attractants necessary for decision-643 making, predators may overlook resource-rich habitats. These results suggest that predator 644 retention and reproductive investment are governed by the convergence of prey abundance, blend 645 identity, and offspring risk, and drought disrupts this convergence by altering VOC blends in 646 ways that suppress predator commitment, even in prey rich patches. 647 648 Bottom-up limitation constrains the expression of top-down control 649 Drought appears to limit top-down regulation by jointly constraining aphid population growth 650 and weakening attractant cues for predators. Under water stress, aphid abundance was primarily 651 shaped by bottom-up limitation, which reduced the potential for predator to exert additional top-652 down control, even when predators were present (Kansman et al., 2022). Simultaneously, when 653 water stress altered plant volatile emissions, natural enemies were less likely to invest in foraging 654 and oviposition in prey rich patches (Rahman et al., 2025). In addition to altering prey abundance 655 and signaling, drought may also reduce aphid quality, through changes in size, nutritional 656 content, or chemical composition (Quandahor et al., 2022; Subedi & Kersch-Becker, 2025), 657 which can further discourage predator foraging or reproductive investment. These findings imply 658 that under drought, weak aphid suppression may not be sufficient evidence of weak predator 659 influence because water stress can limit prey population growth and its nutritional quality, and 660 disrupt the chemical cues that structure predator foraging behavior. 661 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 36 While these mechanisms shaped overall aphid suppression, their effects also extended to 662 aphid phenotypic responses, particularly the production of dispersive morphs. However, 663 interpretation of these effects requires caution given the experiment design. Predator presence 664 increased alates on plants under drought but reduced alates under well-watered, high-aphid 665 density conditions, a pattern consistent with predation risk cues shifting aphids toward dispersal 666 phenotypes, including increased production of winged morphs (Hermann et al., 2021). At the 667 same time, because all plant treatments occurred within the same cage and aphids could move 668 among plants, alate counts may also reflect within cage redistribution and aggregation, especially 669 because crowding is itself a potent driver of wing induction (Yuan et al., 2025). As plant VOC 670 blends drew predators to well-watered plants and increased residence time there, predator 671 activity could have displaced aphids toward less visited plants or increased local crowding, either 672 of which can elevate alate production through density-dependent cues (Khallaf et al., 2023). 673 Future studies that pair repeated VOC sampling with plant specific aphid density and movement 674 tracking will be needed to distinguish developmental induction from redistribution and to test 675 whether odor guided predator attraction and retention drives within cage aggregation patterns. 676 677 Signal contingent trophic regulation under drought and implications 678 Our results reveal limits to prey-abundance-based models of top-down control by showing that 679 drought undermines trophic cascades not only through bottom-up limitation, but by altering 680 volatile-mediated signaling in ways that reduce predator responsiveness, even when prey are 681 present. We show that predators exhibit compound-specific and behaviorally distinct responses 682 to plant volatiles: methyl salicylate and myrcene increased the likelihood of visitation, while 683 methyl salicylate, n-heptanal, p-cymene, and carvacrol predicted oviposition. Drought 684 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 37 suppressed these compounds and shifted blend composition, reducing predator foraging and 685 reproductive investment. As a result, top-down effects shifted from direct consumption to non-686 consumptive pathways, including changes in aphid alate production and dispersal traits. These 687 findings introduce a signal-contingent framework for understanding trophic regulation: effective 688 control depends not only on prey density, but on the reliability and composition of chemical cues 689 that elicit predator foraging and reproductive responses. When chemical cues are degraded or 690 unreliable, predator populations may decline despite prey availability, weakening trophic 691 cascades and reducing the stability of food webs under environmental stress. Integrating volatile 692 signaling and cue thresholds into multi-species interactions networks will improve predictions of 693 trophic cascades under environmental stress. 694 695 5. References 696 Adams, R. (2007). Identification of essential oil components by gas 697 chromatography/mass spectrometry. Allured Publishing Corporation. 698 Ali, M. Y., Naseem, T., Holopainen, J. K., Liu, T., Zhang, J., & Zhang, F. (2023). 699 Tritrophic interactions among arthropod natural enemies, herbivores and plants 700 considering volatile blends at different scale levels. Cells, 12(2), 251. 701 Ayelo, P . M., Pirk, C. W., Yusuf, A. A., Chailleux, A., Mohamed, S. A., & Deletre, E. 702 (2021). Exploring the kairomone-based foraging behaviour of natural enemies to 703 enhance biological control: A review. Frontiers in Ecology and Evolution, 9, 704 641974. 705 Ayelo, P. M., Yusuf, A. A., Pirk, C. W., Chailleux, A., Mohamed, S. A., & Deletre, E. 706 (2021). Terpenes from herbivore‐ induced tomato plant volatiles attract 707 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 38 Nesidiocoris tenuis (Hemiptera: Miridae), a predator of major tomato pests. Pest 708 Management Science, 77(11), 5255–5267. https://doi.org/10.1002/ps.6568 709 Barton, B. T., & Schmitz, O. J. (2009). Experimental warming transforms multiple 710 predator effects in a grassland food web. Ecology Letters, 12(12), 1317–1325. 711 Brooks, M. E., Kristensen, K., Van Benthem, K. J., Magnusson, A., Berg, C. W., Nielsen, 712 A., Skaug, H. J., Mächler, M., & Bolker, B. M. (2017). glmmTMB balances speed 713 and flexibility among packages for zero-inflated generalized linear mixed 714 modeling. 715 Bruce, T. J. A., Midega, C. A. O., Birkett, M. A., Pickett, J. A., & Khan, Z. R. (2010). Is 716 quality more important than quantity? Insect behavioural responses to changes in 717 a volatile blend after stemborer oviposition on an African grass. Biology Letters, 718 6(3), 314–317. https://doi.org/10.1098/rsbl.2009.0953 719 Bürkner, P .-C. (2017). brms: An R package for Bayesian multilevel models using Stan. 720 Journal of Statistical Software, 80, 1–28. 721 Clavijo McCormick, A. (2016). Can plant–natural enemy communication withstand 722 disruption by biotic and abiotic factors? Ecology and Evolution, 6(23), 8569–723 8582. 724 De Rijk, M., Zhang, X., Van Der Loo, J. A. H., Engel, B., Dicke, M., & Poelman, E. H. 725 (2016). Density‐ mediated indirect interactions alter host foraging behaviour of 726 parasitoids without altering foraging efficiency. Ecological Entomology, 41(5), 727 562–571. https://doi.org/10.1111/een.12325 728 Fox, J., & Weisberg, S. (2018). An R companion to applied regression. Sage 729 publications. 730 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 39 Gotelli, N. J. (1995). A primer of ecology. 731 Hartig, F., & Hartig, M. F. (2017). Package ‘dharma.’ R Package, 531, 532. 732 He, X., Wang, Y ., Munawar, A., Zhu, J., Zhong, J., Zhang, Y ., Guo, H., Zhu, Z., Baldwin, 733 I. T., & Zhou, W. (2025). Manipulating stomatal aperture by silencing StSLAC1 734 affects potato plant–herbivore–parasitoid tritrophic interactions under drought 735 stress. New Phytologist, 245(5), 2133–2149. https://doi.org/10.1111/nph.20391 736 Hermann, S. L., Bird, S. A., Ellis, D. R., & Landis, D. A. (2021). Predation risk 737 differentially affects aphid morphotypes: Impacts on prey behavior, fecundity and 738 transgenerational dispersal morphology. Oecologia, 197(2), 411–419. 739 Holopainen, J. K., & Gershenzon, J. (2010). Multiple stress factors and the emission of 740 plant VOCs. Trends in Plant Science, 15(3), 176–184. 741 Holopainen, J. K., Himanen, S. J., Yuan, J. S., Chen, F., & Stewart, C. N. (2025). 742 Ecological Functions of Terpenoids in Changing Climates. In K. G. Ramawat & J.-743 M. Mérillon (Eds.), Natural Products (pp. 1–44). Springer Berlin Heidelberg. 744 https://doi.org/10.1007/978-3-642-36202-6_129-1 745 Horiuchi, J.-I., Arimura, G.-I., Ozawa, R., Shimoda, T., Takabayashi, J., & Nishioka, T. 746 (2003). A comparison of the responses of Tetranychus urticae (Acari: 747 Tetranychidae) and Phytoseiulus persimilis (Acari: Phytoseiidae) to volatiles 748 emitted from lima bean leaves with different levels of damage made by T. urticae 749 or Spodoptera exigua (Lepidoptera: Noctuidae). Applied Entomology and 750 Zoology, 38(1), 109–116. 751 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 40 Hothorn, T., Bretz, F., Westfall, P., Heiberger, R. M., Schuetzenmeister, A., Scheibe, S., 752 & Hothorn, M. T. (2016). Package ‘multcomp.’ Simultaneous Inference in General 753 Parametric Models. Project for Statistical Computing, Vienna, Austria, 1–36. 754 Kansman, Basu, S., Casteel, C. L., Crowder, D. W., Lee, B. W., Nihranz, C. T., & Finke, 755 D. L. (2022). Plant water stress reduces aphid performance: Exploring 756 mechanisms driven by water stress intensity. Frontiers in Ecology and Evolution, 757 10. 758 Khallaf, M. A., Sadek, M. M., & Anderson, P . (2023). Predator efficacy and attraction to 759 herbivore-induced volatiles determine insect pest selection of inferior host plant. 760 Iscience, 26(2). 761 Lenth, R., & Lenth, M. R. (2018). Package ‘lsmeans.’ The American Statistician, 34(4), 762 216–221. 763 Liaw, A., & Wiener, M. (2002). Classification and regression by randomForest. R News, 764 2(3), 18–22. 765 Lin, P .-A., Kansman, J., Chuang, W.-P ., Robert, C., Erb, M., & Felton, G. W. (2023). 766 Water availability and plant–herbivore interactions. Journal of Experimental 767 Botany, 74(9), 2811–2828. 768 Lin, P .-A., Paudel, S., Bin Zainuddin, N., Tan, C.-W., Helms, A., Ali, J. G., & Felton, G. 769 W. (2022). Low water availability enhances volatile-mediated direct defences but 770 disturbs indirect defences against herbivores. Journal of Ecology, 110(11), 2759–771 2771. 772 Makowski, D., Ben-Shachar, M., & Lüdecke, D. (2019). bayestestR: Describing Effects 773 and their Uncertainty, Existence and Significance within the Bayesian 774 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 41 Framework. Journal of Open Source Software, 4(40), 1541. 775 https://doi.org/10.21105/joss.01541 776 Oksanen, J., Blanchet, F. G., Kindt, R., Legendre, P., Minchin, P. R., O’hara, R. B., 777 Simpson, G. L., Solymos, P., Stevens, M. H. H., & Wagner, H. (2013). Package 778 ‘vegan.’ Community Ecology Package, Version, 2(9), 1–295. 779 Peñaflor, M. F. G. V., Erb, M., Miranda, L. A., Werneburg, A. G., & Bento, J. M. S. 780 (2011). Herbivore-induced plant volatiles can serve as host location cues for a 781 generalist and a specialist egg parasitoid. Journal of Chemical Ecology, 37(12), 782 1304–1313. 783 Piironen, J., Paasiniemi, M., Catalina, A., Weber, F., & Vehtari, A. (2023). Projpred: 784 Projection predictive feature selection. R Package Version, 2(0). 785 Pinto-Zevallos, D. M., Bezerra, R. H., Souza, S. R., & Ambrogi, B. G. (2018). Species-786 and density-dependent induction of volatile organic compounds by three mite 787 species in cassava and their role in the attraction of a natural enemy. 788 Experimental and Applied Acarology, 74(3), 261–274. 789 Pinto-Zevallos, D. M., & Blande, J. D. (2024). Challenges of climate change and air 790 pollution for volatile-mediated plant–parasitoid signalling. Current Opinion in 791 Insect Science, 66, 101290. 792 https://www.sciencedirect.com/science/article/pii/S2214574524001329 793 Quandahor, P., Gou, Y., Lin, C., & Liu, C. (2022). Potato (Solanum tuberosum L.) Leaf 794 Extract Concentration Affects Performance and Oxidative Stress in Green Peach 795 Aphids (Myzus persicae (Sulzer). Plants, 11(20). (WOS:000875171000001). 796 https://doi.org/10.3390/plants11202757 797 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 42 Rahman, S., Rostas, M., & Vosteen, I. (2025). Drought aggravates plant stress by 798 favouring aphids and weakening indirect defense in a sugar beet tritrophic 799 system. Journal of Pest Science, 98(1), 549–564. (WOS:001250531600001). 800 https://doi.org/10.1007/s10340-024-01799-6 801 Reinecke, A., Flaig, I. C., Lozano, Y. M., Rillig, M. C., & Hilker, M. (2024). Drought 802 induces moderate, diverse changes in the odour of grassland species. 803 Phytochemistry, 221, 114040. 804 Riddick, E. W. (2020). Volatile and non-volatile organic compounds stimulate oviposition 805 by aphidophagous predators. Insects, 11(10), 683. 806 Robin, X., Turck, N., Hainard, A., Tiberti, N., Lisacek, F., Sanchez, J.-C., Müller, M., 807 Siegert, S., Doering, M., & Billings, Z. (2021). Package ‘pROC.’ Package “pROC. 808 Rondoni, G., Ielo, F., Ricci, C., & Conti, E. (2017). Behavioural and physiological 809 responses to prey-related cues reflect higher competitiveness of invasive vs. 810 Native ladybirds. Scientific Reports, 7(1), 3716. 811 Russavage, E. M., Hewlett, J. A., Grunseich, J. M., Szczepaniec, A., Rooney, W. L., 812 Helms, A. M., & Eubanks, M. D. (2024). Aphid-induced volatiles and subsequent 813 attraction of natural enemies varies among sorghum cultivars. Journal of 814 Chemical Ecology, 50(5), 262–275. 815 Salamanca, J., Souza, B., Lundgren, J. G., & Rodriguez-Saona, C. (2017). From 816 laboratory to field: Electro-antennographic and behavioral responsiveness of two 817 insect predators to methyl salicylate. Chemoecology, 27(2), 51–63. 818 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 43 Schmitz, O. J., & Barton, B. T. (2014). Climate change effects on behavioral and 819 physiological ecology of predator–prey interactions: Implications for conservation 820 biological control. Biological Control, 75, 87–96. 821 Schmitz, O. J., Hambäck, P. A., & Beckerman, A. P . (2000). Trophic Cascades in 822 Terrestrial Systems: A Review of the Effects of Carnivore Removals on Plants. 823 The American Naturalist, 155(2), 141–153. https://doi.org/10.1086/303311 824 Shiojiri, K., Ozawa, R., Kugimiya, S., Uefune, M., van Wijk, M., Sabelis, M. W., & 825 Takabayashi, J. (2010). Herbivore-specific, density-dependent induction of plant 826 volatiles: Honest or “cry wolf” signals? PLoS One, 5(8), e12161. 827 Singh, S., Mishra, G., & Omkar. (2019). Oviposition in aphidophagous ladybirds: Effect 828 of prey availability and conspecific egg presence. International Journal of Tropical 829 Insect Science, 39(2), 107–114. 830 Subedi, B., & Kersch-Becker, M. (2025). Drought restructures multitrophic food webs 831 through cascading eco-physiological constraints. bioRxiv, 2025–12. 832 Tariq, M., Wright, D. J., Bruce, T. J., & Staley, J. T. (2013). Drought and root herbivory 833 interact to alter the response of above-ground parasitoids to aphid infested plants 834 and associated plant volatile signals. PLoS One, 8(7), e69013. 835 Therneau, T. (2015). A package for survival analysis in S. R Package Version, 2(7), 836 2014. 837 Trase, O. W., McCartney, N., & Ali, J. G. (2025). Suppression of Belowground Volatiles 838 in Maize Depends on Cover Crop Legacy and Genotype. Journal of Chemical 839 Ecology, 51(5), 85. https://doi.org/10.1007/s10886-025-01632-z 840 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 44 Truong, D.-H., Delory, B. M., Vanderplanck, M., Brostaux, Y ., Vandereycken, A., 841 Heuskin, S., Delaplace, P., Francis, F., & Lognay, G. (2014). Temperature 842 regimes and aphid density interactions differentially influence VOC emissions in 843 Arabidopsis. Arthropod-Plant Interactions, 8(4), 317–327. 844 Turlings, T. C. J., & Erb, M. (2018). Tritrophic Interactions Mediated by Herbivore-845 Induced Plant Volatiles: Mechanisms, Ecological Relevance, and Application 846 Potential. Annual Review of Entomology, 63(1), 433–452. 847 https://doi.org/10.1146/annurev-ento-020117-043507 848 Vehtari, A., Gelman, A., Gabry, J., & Yao, Y. (2021). Package ‘loo.’ Efficient Leave-One-849 out Cross-Validation and WAIC for Bayesian Models. 850 Verheggen, F. J., Arnaud, L., Bartram, S., Gohy, M., & Haubruge, E. (2008). Aphid and 851 Plant Volatiles Induce Oviposition in an Aphidophagous Hoverfly. Journal of 852 Chemical Ecology, 34(3), 301–307. https://doi.org/10.1007/s10886-008-9434-2 853 Weldegergis, B. T., Zhu, F., Poelman, E. H., & Dicke, M. (2015). Drought stress affects 854 plant metabolites and herbivore preference but not host location by its 855 parasitoids. Oecologia, 177(3), 701–713. 856 Xiu, C., Zhang, W., Xu, B., Wyckhuys, K. A., Cai, X., Su, H., & Lu, Y . (2019). Volatiles 857 from aphid-infested plants attract adults of the multicolored Asian lady beetle 858 Harmonia axyridis . Biological Control, 129, 1–11. 859 Yu, H., Zhang, Y., Li, Y., Lu, Z., & Li, X. (2018). Herbivore-and MeJA-induced volatile 860 emissions from the redroot pigweed Amaranthus retroflexus Linnaeus: Their 861 roles in attracting Microplitis mediator (Haliday) parasitoids. Arthropod-Plant 862 Interactions, 12(4), 575–589. 863 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint 45 Yuan, Y ., Wang, Y ., Ye, W., Xie, L., Yuan, E., Guo, H., Wang, S., Dong, F., Zhu-Salzman, 864 K., & Ge, F. (2025). The maternal vGluT2 and embryonic mGluR3 signaling relay 865 system controls offspring wing dimorphism in pea aphid. iScience, 28(6). 866 Zhu, J., & Park, K.-C. (2005). Methyl salicylate, a soybean aphid-induced plant volatile 867 attractive to the predator Coccinella septempunctata . Journal of Chemical 868 Ecology, 31(8), 1733. 869 870 871 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 19, 2026. ; https://doi.org/10.64898/2026.04.16.718996doi: bioRxiv preprint