Intensifying drought conditions across the western United States due to global climate change are altering plant-insect interactions. Specialist herbivores must find their host plants within a matrix of nonhosts, and thus often rely upon specific plant secondary chemistry for host location and oviposition cues. Climate-induced alterations to plant chemistry could thus affect female selection of larval food-plants. Here, we investigated whether host-plant water limitation influenced oviposition preference in a threatened invertebrate: the monarch butterfly (Danaus plexippus). We found that females deposited more eggs on reduced-water than on well-watered narrowleaf milkweed plants (Asclepias fascicularis), but we could not attribute this change to any specific change in plant chemistry. Specialist herbivores, such as the monarch butterfly, which are tightly linked to specific plant cues, may experience a shift in preferences under global-change conditions. Understanding oviposition preferences will be important to directing ongoing habitat restoration activities for this declining insect.

Seeds of A. fascicularis were collected from Reno, NV (39.49361, -119.85459) in 2018 and 2019 and germinated in May 2020. Plants were grown in 164 mL treepots with 50% peat moss: 34% vermiculite: 16% perlite. To manipulate water availability, soil saturation was maintained at 70% field capacity in control (well-watered) plants and at 30% in reduced-water plants for 1–3 weeks using a gravitational dry-down method following Diethelm et al. (2022). The variation in dry-down time is due to initially high dieback in the reduced-water group, which led us to add more plants to that treatment group, with the treatment maintained for 1 week. The 70% control level reflects what plants typically experience at agricultural field edges, whereas 30% represents a dry treatment that does not induce wilting (Diethelm et al. 2022). We avoided wilting because females may discriminate against wilted plants, independent of plant chemistry (Aikins et al. 2023).

To allow female monarchs to select a mate, each female was initially kept in a mesh caging (40 cm x 40 cm x 61 cm) with three male butterflies and two other females. Male-female pairs that were observed mating were moved, still linked, to separate cages, and mated females were isolated the following day. Once males were observed to mate twice, they were removed from the study. Each oviposition preference trial presented a single, mated female butterfly (n = 15) with a choice between one control and one reduced-water plant box within a flight cage. In an attempt to isolate the effects of plant chemistry and account for potential oviposition bias toward larger plants (Cohen and Brower 1982), we selected experimental plants of similar size and presented only the top 10 cm of the stem to the butterflies. Each oviposition trial occurred in a 20 m³ flight cage for 3 h between 9:30 and 18:30. To allow females to eat ad libitum, a cotton pad of 1:1 ratio of red Gatorade®:deionized water was placed at the middle of the flight cage. After each trial, the exposed biomass of milkweeds per treatment was clipped, the number of monarch eggs per treatment was recorded before eggs were removed, and the plant section was weighed. All but two of the females were used twice, with trials > 6 d apart. At the end of each 3-hour trial, the exposed sections of the plants were transferred to a -80 ℃ freezer for storage until chemical analysis.

To investigate how our water treatments affected plant secondary metabolites, we performed a non-targeted analysis of UV-absorbent metabolites following Diethelm et al. (2022). We estimated the concentration of each metabolite using ultrahigh-performance liquid chromatography (UPLC; Waters Corporation, Milford, MA).

Methods for processing the data: To calculate the concentrations of plant secondary compounds in digitoxin equivalents, we used a digitoxin internal standard (Sigma Chemical Company, St. Louis, MO) and corrected peak areas by sample dry mass and the 0.15mg/ml concentration of the digitoxin standard. We also calculated metabolite diversity, using the exponential term of the Shannon index (q = 1; Chao et al. 2014). 

For female choice response variables, we used generalized linear mixed models (GLMMs) with Gaussian or negative binomial distributions in R version 3.6.1.  For secondary chemistry response variables, we modeled the data with linear regressions. We report beta coefficients (β) with standard errors as effect sizes (Bischl et al. 2017). Marginal and conditional R2 values were calculated in the MuMIn R package (Barton 2019).

To establish the predictors of the number of monarch eggs, we started with a saturated model treating water and the concentration of flavonol glycosides as fixed effects and the number of leaves, female age, time since female mated, and female identity as covariates. Initially, our global model had two predictors and four co-variates. The co-variates were used based on other research that suggested they could be important. However, to prevent over-fitting, we used backward model selection with the adjusted Akaike Information Criterion and likelihood ratio tests. We selected this method because it penalizes models with more parameters. We were able to eliminate all of the co-variates based on this method.

We included both flavonol glycosides and water treatment as fixed effects in the model because preliminary tests suggested that the water treatment did not affect chemistry (t = -0.5, df = 50, P = 0.6). To account for repeated trials of a given female, we included female identity as a random intercept effect. To control for non-independence between plants within a single trial, the trial number was also as a random intercept effect, nested within female identity. We then used backward selection (Zuur et al. 2009) in MuMIn, and the best-fit model was selected based on the Akaike Information Criterion (AICc). Marginal predictors were evaluated using log-likelihood ratio tests (LRT) in lmtest (Zeileis and Hothorn 2002). To evaluate the predictors of plant secondary metabolites, including the total concentration of UV-absorbent metabolites, the concentration of flavonol glycosides, and the exponential of Shannon's entropy index (q = 1) for metabolite diversity, we modeled water as a fixed effect and the duration of dry-down as a covariate. To measure the strength of the relationship between total concentration and the concentration of flavonol glycosides, we calculated a Pearson’s correlation value.

Standards and calibration information: We used a digitoxin internal standard (Sigma Chemical Company, St. Louis, MO).

Environmental/experimental conditions: We used a flight cage inside a greenhouse under controlled temperatures but with natural lighting.

Quality-assurance procedures performed on the data: We ensured data accuracy by verifying values and identifying any missing or erroneous entries.

Instrument- or software-specific information needed to interpret the data: 
-R Core Team (2023). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL https://www.R-project.org/.
-RStudio (2023) URL http://www.rstudio.com/
-Rstudio packages: tidyvers, ggplot, ggforce (for jitter plot), Hmisc, Rmisc (for SE summary), lmtest (for model comparisions), glmmTMB (for mixed models), MuMin, DHARMa, BBmisc, hillR, ggpubr