Skip to main content

Novel host plant unmasks heritable variation in plant preference within an insect population

Cite this dataset

Steward, Rachel A; Epanchin-Niell, Rebecca S; Boggs, Carol L (2022). Novel host plant unmasks heritable variation in plant preference within an insect population [Dataset]. Dryad.


Introductions of novel plant species can disturb the historical resource environment of herbivorous insects, resulting in strong selection to either adopt or exclude the novel host. However, an adaptive response depends on heritable genetic variation for preference or performance within the targeted herbivore population, and it is unclear how heritability of host-use preference may differ between novel and historical hosts. Pieris macdunnoughii butterflies in the Rocky Mountains lay eggs on the nonnative mustard Thlaspi arvense, which is lethal to their offspring. Heritability analyses revealed considerable sex-linked additive genetic variation in host preference within a population of this butterfly. This was contrary to general predictions about the genetic basis of preference variation, which are hypothesized to be sex-linked between populations but autosomal within populations. Evidence of sex-linkage disappeared when butterflies were tested on methanol-based chemical extracts, suggesting these chemicals in isolation may not be the primary driver of female choice among available host plants. Although unexpected, evidence for within-population sex-linked genetic variation in preference for T. arvense over native hosts indicates that persistent maladaptive oviposition on this lethal plant must be maintained by alternative evolutionary dynamics such as migration- or drift-selection balance or pleiotropic constraints.


Pieris macdunnoughii butterflies used in this study were collected over multiple summer field seasons (1997, 2006, 2015) from the sites near Gothic, CO. All plants used in the preference assays were collected from sites near Gothic for all three years of the study. Preference for T. arvense was tested in simultaneous choice assays against an abundant, preferred native host, Cardamine cordifolia. Butterflies were allowed to choose between T. arvense and a native host (C. cordifolia), in the form of either whole plants, cut stems bearing undamaged leaves, or filter paper treated with methanol-based leaf tissue extracts and a negative control substrate, and in all cases, preference was measured as the number of eggs laid on T. arvense out of the total eggs laid. Larval offspring were reared in the environmental chambers under the same conditions as the ovipositing females. Pupae were collected after hardening of the cuticle, sex was determined, and pupae were grouped by sex and brood and left to emerge in screen cages in the environmental chamber. Upon adult emergence, the F1 butterflies were numbered individually, and maternal and paternal identity were recorded. Preference tests were repeated on the F1 generation. Their F2 offspring were reared, mated and also tested, creating a three-generation pedigree in which all grandmothers and both parents of the F2 generation were known. The final datasets included 37 P, 34 F1, and 138 F2 females that laid eggs in the whole plant preferences tests and 36 P, 44 F1, and 121 F2 females in the cut stem preference tests.

For preference tested on methanol extracts, butterflies were tested on filter paper disks treated with 80 µL of either T. arvense or C. cordifolia methanol extract. Two other disks included a control (70% MeOH only) and a blank (untreated). Eggs laid on each disk were counted and collected, and the disks were replaced with freshly treated disks daily for up to six days or until the butterfly died. Eggs were sterilized, then hatching larvae were transferred to rearing cages containing C. cordifolia leaves and kept in the environmental chamber (27-31:20-22oC, 16:8 L:D). Mated females in the F1 and F2 generations were tested in the same way as the P generation, and the final dataset comprises 104 P, 41 F1 and 48 F2 females. 

To make the methanol extracts, we flash froze leaves of each plant in liquid nitrogen. once frozen, the leaves were lightly crushed, and boiled in 70% MeOH for several minutes before filtering. Excess MeOH was used to boil the leaves, so the filtrate was left to evaporate for 24 hours. We added a small amount of 70% MeOH to achieve equal concentrations (10g fresh weight/L) in the two extracts. We identified glucosinolates in the methanol extracts using Sephadex columns (DEAE 25), prepared with 50 µL 1 mM Progoitrin [2(R)-Hydroxy-3-butenyl GSL] analytical reference standard (ChromaDex, Inc.). Desulfoglucosinolates were quantified via HPLC using a Chromegabond WR C18 column coupled with a diode array detector (DAD) monitoring absorbance at 229 nm and subsequently with a charged aerosol detector (CAD). Only desulfoglucosinolates appearing in both the DAD and CAD output were included. Both mass spectra and comparative retention times from the literature were used to identify desulfoglucosinolates.

At the time leaves were collected to make the extracts, we also collected fresh leaf samples to ensure the glucosinolate components of our methanol extracts captured the glucosinolate profile of fresh leaf tissue. The sixth leaf from the apical meristem of 15 plants of each species was collected directly into screw-cap microcentrifuge tubes containing 70% MeOH. Leaf samples were kept in a cool, dark location for 8 monthsand glucosinolates in the leachate were desulfated and quantified as described for extracts. Our glucosinolate identification and quantification was sufficient for comparing host plant leaves and extracts but should not be used for comparative analyses or reviews of plant chemical defenses.

 Statistical analyses were conducted using the R statistical plaform. We tested for spatial differences in preference of wild-caught females in the parental generation using Bayesian beta-binomial models (brms package). Models were run with uninformative priors for 40,000 iterations with a warmup of 10,000 and thin of 5. The effect of collection location was assessed using Leave One Out information criteria (LOOIC) and Bayes Factors (BayesTestR package). 

We then used Bayesian beta-binomial multi-level models to evaluate the contribution of Z-linked (Vz), autosomal (Va) and environmental variance (Ve) to the proportion of eggs laid on T. arvense  (phenotypic variance, Vp). Heritability (h2) was calculated as the proportion of Vp attributable to genetic variance (e.g. ha2 = Va/Vp, where Vp = Vz + Va + Ve). The environmental variance (i.e., Ve) comprised the product of the overdispersion parameter (w) and the fixed variance of the binomial distribution which is proportional to p2/3. Models followed the form: response ~ predictors + (1|individual, cov = A) + (1|individual, cov = Z), where the response was the number of eggs laid on T. arvense out of the total eggs laid. Predictors included generation and trial start day. A and Z specified the covariance of the random effects based on the autosomal and Z-linked relatedness matrices, generated from pedigrees using the nadiv package (females are heterogametic). The preference test start day was calculated as an ordinal day from first test day within each generation. All models were run using the same set of partially informative priors for 40,000 iterations with a 10,000-iteration warmup and a thin of 5. Full models tested the interaction of generation and start day and were compared to reduced models using Leave One Out information criteria (LOOIC) and Bayes Factors calculated with BayesTestR.

Autosomal and Z-linked variances were calculated from the standard deviations of the random intercepts of A and Z in the model output. We used the ‘hypothesis’ method in brms to evaluate whether the 95% credible intervals for heritability estimates provided support for heritable variation in oviposition preference.

Usage notes

R version 4.0.1


bayestestR 0.12.1

brms 2.17.0

ggmap 3.0.0

ggplotify 0.1.0

ggpubr 0.4.0

gridExtra 2.3

loo 2.4.1

nadiv 2.17.1

rstan 2.21.2

tidyverse 1.3.0

viridisLite 0.4.0


National Science Foundation, Award: DGE-1450810

Rocky Mountain Biological Laboratory

Stanford University

University of South Carolina