1. Selective pressures exerted on the feeding behavior of animals have been extensively studied to understand their foraging patterns. In herbivores, specific within-plant patterns of resource exploitation have been reported, but their determinants remain poorly understood.

2. Here, we describe and decipher the determinants of the foraging pattern of the pollen beetle, Brassicogethes aeneus, a pollinivorous insect that is a pest of oilseed rape (Brassica napus). This insect feeds from flowers for almost all of its life cycle, except for a couple of weeks preceding blossoming. During this period, only flower buds are available and the insect destroys them to feed from the pollen they contain, causing serious yield losses.

3. We found that during this critical period, pollen beetles exhibit a stereotypic intra-inflorescence feeding pattern that depends on flower bud maturity. To explain this pattern, we first deciphered the selective pressures driving pollen beetles’ feeding behavior. Using a set of manipulative laboratory experiments, including behavioral experiments on plant tissues and artificial substrates, chemical characterization of plant tissues and performance experiments, we show that the pollen beetles’ feeding behavior does not seem to be driven by specialized metabolites or an attempt to reach an optimal nutrient balance, but rather by a process of maximization of total macronutrient intake. Next, using Optimal Diet Choice models, we found that one aspect of the intra-inflorescence feeding pattern, the preference for young over old flower buds, could be well explained through the lens of total macronutrient intake maximization per unit of time to access the resource.

4. Our study provides new insights into small-scale foraging patterns, and highlights the need to characterize and assess the relative influence of several components of diet quality when deciphering selective pressures driving foraging patterns.

Feeding tests

2.1. Tests on entire plants

To describe the feeding pattern of the pollen beetle, one individual was placed on the main inflorescence of an entire plant at the green-yellow bud stage as described in Hervé et al. (2014a). After three days, the number of buds damaged by feeding was counted. Buds damaged by feeding are recognizable as they show irregular holes of usually large sizes and located anywhere on the buds, while buds damaged by oviposition show stereotypic small holes at their basis. The length (i.e. maturity) of damaged buds was measured under a binocular microscope (0.1 mm precision). Ten replicates were performed.

Data analysis – All statistical analyses were performed with the R software v. 4.0.2 (R Core Team, 2020). The mean number of buds damaged was compared between maturity classes (young, intermediate and old buds) using a Wald test applied on a Generalized Linear Mixed Model (GLMM) including the individual plant as random factor (distribution: Poisson, link function: log) (R package ‘lme4’ (Bates et al., 2015) and ‘car’ (Fox & Weisberg., 2019)). Here and in the following experiments, pairwise comparisons of Estimated Marginal Means (EMMeans) were systematically performed using the 'emmeans' package (Lenth, 2019) and p-values adjusted using the False Discovery Rate correction (Benjamini & Hochberg 1995).

2.2. Tests on plant organs

To confirm the preference for flowers over flower buds and to assess the contribution of resource accessibility and resource chemical composition to this preference, dual-choice tests were performed. One individual was placed in a Petri dish (Ø = 55 mm) for 2 h with two different food sources, and it was then recorded whether each food source had been damaged or not. Three choice tests were conducted: one flower vs. one old bud (to confirm preference for flowers), one old bud vs. six anthers just excised from one old bud (to assess influence of resource accessibility), and six anthers just excised from one flower vs. six anthers just excised from one old bud (to assess influence of resource chemical composition). Old buds rather than young buds were chosen to decipher the factors responsible for the preference for flowers for practical aspects and to avoid any bias related to a physical factor, the anthers of old buds and flowers having the same size. Twenty to 27 replicates were performed per choice test.

Data analysis – For each choice test, the probability of being damaged was compared between the two food sources using a Cochran’s Q test, which considered the replicate as pairing factor (R package ‘RVAideMemoire’, Hervé 2021).

2.3. Tests on artificial substrates

An experimental setup based on agar disks was designed to assess the respective and relative contributions of macronutrients (either their quantity and ratio) and defense metabolites in the preference of pollen beetles for flowers. Artificial substrates consisted of 3 % agar disks (Ø = 5 mm, thickness = 2 mm) supplemented with macronutrients (casein:whey 80:20 as protein source and sucrose as carbohydrate source) and/or pure standards of defense metabolites, depending on the experiment. Macronutrient and defense metabolite concentrations used in experiments varied depending on the hypothesis tested (see Results and Table S1). In all experiments on artificial substrates, two individuals were placed in a Petri dish (Ø = 35 mm) for 3 h, with two to three disks depending on the experiment. Insects were filmed during the experiment and their movements tracked with the Ethovision XT software v. 15 (Noldus, Wageningen, Netherlands). The feeding behavior was estimated as the cumulative duration spent on each disk. Following preliminary observations, it was assumed that individuals were feeding when on disks and that the feeding speed was constant. Thirty replicates were performed per experiment.

Data analysis – For each experiment, the total time spent on disks was compared between treatments using a Wald test applied on a Linear Mixed Model (LMM) that included the replicate as random factor (R packages ‘lme4’ and ‘car’). The response was systematically square-root transformed to ensure model fitting.

3. Chemical characterization of plant organs

3.1. Macronutrients

Macronutrient quantification was performed on five samples of three different plant organs: anthers from flowers, old buds and young buds. Each sample comprised organs collected from 14-16 plants (four organs per plant), immediately frozen into liquid nitrogen then freeze-dried. Plant organs were sampled from the same plants to ensure unbiased comparisons. Total soluble proteins (P) were extracted from 10 mg of dried powder that was agitated for 15 min at room temperature in 1 mL of acidified phosphate buffer (0.2 M, pH = 6.8), then centrifuged at 12,000 g for 30 min at 4 °C. Quantification was performed using the Bradford’s method (Bradford, 1976) and standard solutions of bovine serum albumin as references. Total digestible carbohydrates were extracted from 10 mg of dried powder in 2 mL of phosphate buffer (0.2 M, pH = 6.5) for 20 min at 95 °C. After centrifugation at 14,000 g for 5 min at 4 °C, the supernatant was collected, the extraction steps were repeated and both supernatants were pooled. Total digestible carbohydrates (C) were quantified using the hot anthrone test (van Handel, 1967) and standard solutions of glucose as references.

Data analysis – Macronutrient composition was compared between pairs of plant organs (flower vs. old buds and old buds vs. young buds) using Welch t-tests. Tests were performed for P, C and P+C content as well as for the P:C ratio.

3.2. Defense metabolites

Compounds were chosen according to previous studies (Hervé et al. 2014a,b; Seimandi-Corda et al. 2019): S-methylcysteine sulfoxide (SMCSO) and flavonols that were suspected to be phagodeterrent for the pollen beetle (Hervé et al. 2014b), and glucosinolates that are known defense compounds of Brassicaceae (Hopkins, van Dam & van Loon, 2009) although their effects on the pollen beetle was not clear (Hervé et al., 2014a,b). Material (i.e. anthers from flowers and old buds) was collected as described for macronutrients, and metabolites were extracted and quantified as described in Hervé et al. (2014a) and Seimandi-Corda et al. (2019). 

Data analysis – Metabolic profiles of plant organs were compared multivariately using a redundancy analysis (RDA) on centred and scaled data, and an associated permutation test with 9 999 permutations (R packages ‘vegan’; Oksanen et al., 2018). Univariate Welch t-tests with FDR-adjusted p-values were also performed to compare plant organs separately for each compound.

4. Performance experiments

4.1. Performance on plant organs

Here and in the following experiments, performance was assessed as the survival time following a defined feeding period. Survival was chosen over fecundity as performance estimator since pollen beetles produce eggs continuously as they feed (Hervé et al., 2014a; Ekbom & Ferdinand, 2003) and lay eggs only in flower buds. Providing them with flower buds where to lay eggs would have led to an impossibility to prevent them from feeding, which would have biased the experiments. One starved individual was placed for 24 h in a Petri dish (Ø = 55 mm) with a non-limiting food source consisting of two flowers or two old buds. After the feeding period, individuals were placed in new Petri dishes (same size) with a moistened filter paper humidified every day and their survival time was recorded through daily observations. Forty-one to 43 replicates were performed per treatment.

Data analysis – The survival time was analyzed using a likelihood ratio test (LRTest) applied on a survival regression (distribution: Weibull, link function: log), which included the treatment, the sex and their interaction as explanatory variables (R packages ‘survival’ (Therneau, 2021) and ‘car’).

4.2. Performance on artificial diets

Twenty-eight artificial diets differing in their P+C content and P:C ratio were designed, consisting of all combinations of four total macronutrient concentrations (P+C 45, 90, 180 and 270 g.l-1) and seven macronutrient ratios (P:C 1:0.5, 1:1, 1:2, 1:3.5, 1:5, 1:8 and 1:15). All diets were presented to insects in 3 % agar disks (Ø = 5 mm, thickness = 2 mm). All disks included casein:whey 80:20 as P source and sucrose as C source, as well as a constant amount of additional nutritional resources (i.e. dry yeasts and vitamins) and antimicrobial agents (i.e. Tegosept and propionic acid) (see Appendix S1). The P and C content of dry yeasts was included in the calculations following Lihoreau et al. (2016). After starvation, one individual was offered two identical disks for 48 h in a Petri dish (Ø = 35 mm) that also contained a moistened filter paper. After the feeding period, disks were removed and performance was estimated as above. Fifteen replicates were performed per diet.

Data analysis – The survival time was analyzed using a LRTest applied on a survival regression (distribution: Weibull, link function: log) (R packages ‘survival’ and ‘car’). The model included the total P+C concentration, the P:C ratio and their interaction as explanatory variables.

5. Optimal Diet Choice modeling

To determine whether pollen beetles’ feeding pattern at the inflorescence scale could be explained by a maximization of the total nutrient intake per unit of time, a model of Optimal Diet Choice (ODC) was used. As other optimal foraging models, ODC models are based on the profitability of each food source, which is dependent on the amount of nutrients acquired from the food sources and the handling time needed to locate, capture, manipulate and ingest these food sources. Since pollen beetles can be considered encountering both young and old buds simultaneously on a given inflorescence, the ODC model of Waddington & Holden (1979) was used. This model allows predicting the optimal proportion of each food source to be included in the diet, which maximizes total macronutrient intake per unit of time: 

where p_y is the predicted proportion of young buds to be included in the diet, M.C is the ratio of total nutrient amounts ingested per bud (M: ratio of bud masses ingested, C: ratio of total nutrient concentrations per mass unit), H_y and H_o are handling times of young and old buds respectively, and D_y and 1-D_y are the relative densities of young buds and old buds on an inflorescence respectively. Distributions of the model parameters were estimated from a series of dedicated experiments described below and summarized in Table 1. Values of p_y were computed through simulations since most of the model parameters, i.e. M, C and H, were random variables. For that purpose, the central tendency as well as the distribution of each parameter were computed at the mean point of the two bud maturity classes, i.e. 1.75 mm long for young buds and 6 mm long for old buds (Table 1). Values of p_y were calculated for all values of D_y ranging from 0 to 100 %, by steps of 1 %. For each value of D_y, 10 000 simulations were performed in each of which parameter values were randomly sampled from their theoretical distributions. These 10 000 values allowed estimating a central tendency of p_y (considered as the median value) and an associated 95 % credibility interval. Predicted proportions of young buds to be used could then be compared with real values observed in the feeding experiment on entire plants. The real relative density of young buds at the green-yellow bud stage was estimated by measuring all buds of the main inflorescence of 15 intact plants and calculating the proportion of young buds among the total amount of young and old buds. This proportion was estimated as 0.596 using a Generalized Linear Model (GLM, distribution: binomial, link function: logit).