Introduction

Plant defenses modify species interactions, and induced defenses in response to herbivory can mediate interactions over three or even four trophic levels (Harvey et al. 2003; Soler et al. 2005). Induced plant defenses typically are studied in the context of effects on herbivores and their natural enemies; however, they also have the potential to affect pollinators and their natural enemies. Herbivory on vegetative tissues often induces changes to volatile organic compound (VOC) emissions and increases secondary compounds in floral tissues and rewards (reviewed in Lucas-Barbosa 2016; Moreira et al. 2019; Rusman et al. 2019), and many studies have examined how such changes affect pollinator visitation. In general, floral and leaf herbivory deter pollinators (Moreira et al. 2019) although root herbivory sometimes increases flower visitation (Barber et al. 2011; Moreira et al. 2019; Poveda et al. 2003, 2005) and effects on pollinator behavior may depend on the pollinator species (Hoffmeister et al. 2016; Rusman et al. 2018).

The impacts of herbivory on plant-pollinator interactions could continue past the immediacy of the pollinator visit, if herbivore-induced changes to floral rewards affect pollinator health (Jacobsen and Raguso 2018). Herbivory can alter the value of pollen and nectar through decreased quantity (Krupnick et al. 1999; Lehtilä and Strauss 1999; Chauta et al. 2017), quality and increased secondary compound content (Adler et al. 2006; Halpern et al. 2010; Kaczorowski et al. 2014), although in some cases herbivory can increase nectar sugar concentration (Bruinsma et al. 2014; Chauta et al. 2017). Because pollinators rely on pollen and nectar to meet their nutritional requirements (Heinrich 1979; Roulston and Cane 2000), changes in the nutritional value of floral rewards could have effects on pollinator health. Thus, herbivory could influence interactions between pollinators and their natural enemies, but to our knowledge this relationship has never been assessed.

Pollinators, like other herbivores, ingest secondary compounds in their diet (Irwin et al. 2014), which could have direct negative effects on pollinator performance and indirect positive or negative effects on pollinator-pathogen interactions. Secondary compounds in floral rewards can be directly harmful to pollinators (Detzel and Wink 1993; Tiedeken et al. 2016). For example, Detzel and Wink (1993) found that 17 out of 29 secondary compounds found in honey were lethal to honey bees at concentrations of 0.6% and under, and noted that alkaloids were particularly toxic. Secondary compounds, such as the alkaloid D-lupanine, can also reduce bumble bee microcolony fitness (Arnold et al. 2014). Pollinators may respond to harmful compounds in their diet via processes such as detoxification (du Rand et al. 2015; Mao et al. 2013). These responses can lead to tradeoffs that hinder pathogen resistance, such as energetic stress or reduced immunocompetence (Roger et al. 2017). For example, the alkaloid anabasine can magnify the negative effects of pathogens on bumble bee and microcolony fitness (Palmer-Young et al. 2017), suggesting possible tradeoffs between compound detoxification and immune function or pathogen tolerance.

Secondary compounds could also benefit pollinators by reducing pathogen infection via increasing host resistance or being directly harmful to the pathogens (Stevenson et al. 2017; Koch et al. 2019). Several nectar secondary compounds, including anabasine and nicotine, reduce pathogen infections in bumble bees (Baracchi et al. 2015; Koch et al. 2019; Manson et al. 2010; Palmer-Young et al. 2017; Richardson et al. 2015). For example, Richardson et al. (2015) tested the effects of eight nectar secondary compounds on infections by Crithidia bombi, a common gut endoparasite, in the bumble bee Bombus impatiens and found that half of those compounds reduced infections. However, these effects can be context-dependent (Palmer-Young et al. 2016; Thorburn et al. 2015) and secondary compounds have mostly been tested at the concentrations found in nectar, but pollinators could be exposed to higher concentrations via consumption of pollen (Cook et al. 2013; Palmer-Young et al. 2019).

Pollen chemical composition has the potential to play a central role in mediating pollinator-pathogen interactions. Pollen is the major source of protein and lipids for many floral visitors, and thus is critical for reproduction and survival (Roulston and Cane 2000). Pollen is also the plant’s male gamete, and can be defended with secondary compounds. Optimal defense theory predicts that plant tissues that are more tightly linked to reproduction should be better defended (McKey 1974), and this has been supported by studies in which female reproductive tissue such as fruits and seeds are highly defended (Ohnmeiss and Baldwin 2000; Zangerl and Rutledge 1996). By the same logic, pollen should be highly defended relative to nectar, because pollen viability is directly tied to plant male reproductive success. Consistent with this hypothesis, pollen contains an array of “anti-herbivore” secondary compounds (Adler 2000; Rivest and Forrest 2020; Stevenson et al. 2017), that are often more concentrated (Cook et al. 2013; Palmer-Young et al. 2019) and diverse (Palmer-Young et al. 2019) than compounds in nectar. Hence, the effects of pollen quality on pollinator health may be dictated not only by macronutrients but also by secondary chemistry. However, secondary compound concentrations in pollen may not be consistent over time (see Davis et al. 2019). Instead, they could be shaped by both herbivory and time since flowering (i.e., days since the onset of flowering). Secondary compound concentrations in roots and leaves change in response to herbivory (Kaplan et al. 2008) and ontogeny (Boege and Marquis 2005). Additionally, secondary compound concentrations in vegetative tissues, flowers and nectar can be positively correlated (Adler et al. 2006, 2012; Manson et al. 2012). Thus, we hypothesize that herbivory and time since flowering will affect the concentrations of secondary compounds in pollen and consequently shape multitrophic interactions with pollinator pathogens.

We conducted a study to examine whether herbivory can induce chemical changes in floral rewards and influence interactions between a pollinator and its gut pathogen, and whether time since flowering influences these effects. Specifically, we asked whether herbivory by the Solanaceous specialist, tobacco hornworm (Manduca sexta L., Sphingidae) induces higher levels of alkaloids in tobacco (Nicotiana tabacum L., Solanaceae) nectar and pollen, and we examined the time course of induction. Additionally, we asked whether herbivory-induced effects on tobacco nectar and pollen affected the gut pathogen Crithidia bombi (Trypanosomatidae) in the common eastern bumble bee, Bombus impatiens Cresson (Apidae). We hypothesized that herbivory would induce higher alkaloid concentrations in floral rewards, and that these changes would reduce pathogen counts in bumble bees.

Materials and Methods

Study System

We assessed the effects of herbivory on nectar and pollen secondary compounds and the time course of induction using domesticated tobacco (Nicotiana tabacum L., Solanaceae), and then assessed the effects of pollen and nectar from damaged tobacco on Crithidia infection in B. impatiens. Tobacco nectar contains the alkaloids anabasine and nicotine (Adler et al. 2006), which are inducible in nectar following leaf herbivory (Adler et al. 2006; Halpern et al. 2010). Tobacco is often damaged by the tobacco hornworm (Manduca sexta L., Sphingidae), a specialist herbivore of Solanaceae (Merrill 1916). Furthermore, nicotine and anabasine reduced infection in the common eastern bumble bee Bombus impatiens Cresson (Apidae) by the gut pathogen Crithida bombi (Trypanosomatidae; hereafter, Crithidia) in some studies (Anthony et al. 2015; Baracchi et al. 2015; Richardson et al. 2015), although not in others (Palmer-Young et al. 2016; Thorburn et al. 2015). Crithidia is transmitted fecal-orally on flowers and within the colony (Durrer and Schmid-Hempel 1994) and reduces individual survival, colony founding success, and foraging abilities in bumble bees (Brown et al. 2000; Gegear et al. 2006; Shykoff and Schmid-Hempel 1991). We note that N. tabacum is an agricultural species that does not grow in the wild, but produces copious nectar and has been a model system for studying defensive chemistry (e. g., Kaplan et al. 2008). Bumble bees have been observed visiting flowers of some wild Nicotiana species (Kaczorowski et al. 2005), many Nicotiana species have nicotine and/or anabasine in nectar (Adler et al. 2012), and mechanisms of nectar production and biochemical action seem to be generally conserved across the genus (Silva et al. 2018), making our study ecologically relevant.

Floral Induction: Herbivory Treatments

Two hundred and fifty tobacco plants were grown from seed in a greenhouse as described in Online Resource 1 (see Plant Propagation). We applied herbivory treatments when the first flower buds appeared on each plant (April 14 until June 23, 2017). Three fifth-instar (occasionally third- or fourth-instar) tobacco hornworms (Manduca sexta; Great Lakes Hornworm, Romeo, Michigan, USA) were placed in mesh bags enclosing the three topmost, fully extended leaves. Caterpillars were removed when they had consumed the entire leaf, usually within three days. When leaves were not consumed within three days, another caterpillar was added until the whole leaf was consumed. Undamaged plants were similarly bagged (three topmost leaves for three days) but without caterpillars, to control for handling effects. Plants were treated with herbivores or bagged (controls) on average 7.3 days (range: 2–12 days) prior to the onset of flowering.

Floral Induction: Chemical Analysis

As plants began to flower, we collected nectar and pollen as described in Online Resource 1 (see Nectar and Pollen Collection). Fifty-three pollen samples from 13 herbivore-damaged plants and 45 pollen samples from 11 undamaged plants were tested for anabasine and nicotine. On average, we assayed 4 pollen samples per plant; each 6-mg sample contained pollen pooled over 5 days. Samples were collected 9–59 days after treatment and samples from the same plant were separated by at least 10 days. We also tested 117 15-μL nectar samples from 29 plants for alkaloids, but none contained nicotine or anabasine.

Analyses of pollen and nectar were performed following the procedure in Davis et al. (2019). Briefly, prior to analyses pollen samples were extracted in 500 μL of dichloromethane. Nectar samples were diluted with 250 μL of water and extracted with 250 μL of dichloromethane. Samples were injected into a gas chromatograph (Agilent 6890) coupled to a mass spectrometer (Agilent 5973) and fitted with a DB-5 capillary column (30 m length, 0.25 mm diameter, 0.25 μm film thickness; Agilent). Helium was used as the carrier gas at a constant flow of 1 mL/min. The column was initially held at 150 °C for 2 min, then increased by 6 °C/min until it reached 240 °C. The data was captured and analyzed using Chemstation (Agilent). The NIST Mass Spectral Database was used to identify the compounds, which were compared to authentic standards (Sigma Aldrich). Alkaloids were quantified against standard curves based on the abundance of the molecular ion (m/z 162).

Bee-Pathogen Bioassays: Diet Treatments

We used nectar and pollen collected from the floral induction experiment to assess effects on pollinator-pathogen interactions. After collecting enough nectar and pollen for chemical analysis, remaining samples were pooled by treatment and collection periods in 2019. We are unaware of data demonstrating that herbivory induces changes in pollen chemistry, including the time it takes for induction to occur or relax. We compared pollen and nectar collected within one month of damage to samples collected more than one month after damage. We chose a one-month period to collect sufficient material for bioassays and because we hypothesized induced responses would last for at least one month after damage but might attenuate after that; Adler et al. (2006) showed that induction in nectar occurred in samples pooled for several weeks after damage, and Halpern et al. (2010) showed that induction in nectar may last up to two months (nonsignificant trend). Because anabasine was not detected in any samples collected before May 25, we conducted an additional analysis with three categories (“within one month [before May 25]”, “within one month [after May 25]” and “after one month”), but results were largely consistent and so we report the simpler analysis (see Online Resource 1; Figures S1–2 and Tables S1–3, for the more complex analysis and results). All pollen treatments were mixed with a wildflower pollen mix (CC Pollen Co., Phoenix, Arizona, USA) in a 1:1 ratio by mass and supplemented with deionized water, as preliminary trials suggested that pure tobacco pollen was toxic to bees (J. K. Davis, unpublished data). Nectar treatments were not diluted. During each bioassay, we included control treatments to determine whether all tobacco pollen and nectar could affect parasite counts relative to a more standard diet. In the pollen bioassay, the control diet was the wildflower pollen mix and in the nectar bioassay, the control diet was a 17% sucrose solution, which was the same sugar concentration as the tobacco nectar. We also chemically analyzed one subsample from each pooled pollen treatment but did not detect alkaloids. Because a high proportion of pollen samples did not contain alkaloids originally, pollen with detectable levels of alkaloids may have been diluted by pollen with alkaloid concentrations too low to detect in our pooled sample (See Statistical Analyses in Methods and Floral Induction in Results; Fig. 1). Alternatively, it is possible that after two years, although the samples were stored at −20 °C, the alkaloids originally present at low concentrations (see Results) had decomposed to below our detection threshold. Nonetheless, some macronutrients remain stable over long periods of time (P.C. Stevenson, personal observation), and changes in macronutrient concentrations could mediate interactions between pollinators and their pathogens.

Fig. 1
figure 1

Alkaloid concentrations over time in Nicotiana tabacum pollen from damaged (red triangles) and undamaged (black circles) plants. a Nicotine concentrations with lines indicating the zero-inflated compound Poisson generalized linear model estimate of the mean (dotted), linear model estimate (solid) and the 95% confidence intervals based on linear model estimate (dashed). b Anabasine concentrations with lines depicting the estimate of the mean for damaged (red triangles) and undamaged (black circles) plants. Parameter uncertainty for anabasine could not be calculated with the current statistical packages in R; all model estimates for nicotine and anabasine are provided in Tables S4–5

Full size image

Bee-Pathogen Bioassays: Experimental Procedure

In 2019, we tested effects of the pooled pollen and nectar on Crithidia infection. Nectar and pollen bioassays were conducted separately but identically, apart from the manipulation of pollen or nectar and the dates of inoculation. The pollen bioassays were conducted with 84 B. impatiens workers inoculated on February 14 and 15 (9–22 bees per treatment combination; sample sizes in Online Resource 1, Table S1). The nectar bioassays were conducted with 175 B. impatiens workers inoculated from April 26 to May 6 (19–40 bees per treatment combination; total of 5 inoculation dates; sample sizes in Online Resource 1, Table S2).

Workers were taken from four commercial colonies (Biobest, Leamington, ON, Canada) for each bioassay; a subsample of five workers per commercial colony was screened for Crithidia before experiments began and weekly thereafter to confirm colonies were Crithidia-free. After a 2-h starvation period, workers were inoculated with 10 μL of inoculum; the infection inoculum was made from a mix of sucrose, Ringer’s solution and infected bee gut contents. The inoculum contained 25% sucrose and 6000 cells of a lab-reared Crithidia strain originated from wild B. impatiens workers collected at Stone Soup Farm in 2015 (42°21′51.93”N, 72°33′55.88”W, Hadley, Massachusetts, USA). Bees were randomly assigned to a pollen or nectar treatment and fed their assigned diets for seven days. In the pollen bioassay, bees were fed ~ 0.5 mg of their pollen treatment in 16 × 19 mm queen rearing cell cap cups and 1.5 mL of a 30% sucrose solution. In the nectar bioassay, bees were fed their nectar treatment in a 0.6 mL microcentrifuge tube and ~ 0.5 mg of a wildflower pollen mix (CC Pollen Co., Phoenix, Arizona, USA) in a 1.5 mL microcentrifuge tube cap. Nectar and pollen were replaced every other day in both bioassays. In the pollen bioassay, pollen was wetted with distilled water on days it was not replaced because it dried out within 24 h. In the nectar bioassay, we did not need to wet the pollen because wildflower pollen retained its moisture over 48 h. During the seven days, workers were kept at 27 °C in darkness in 16-oz individual containers. Bees were then dissected to assess infection levels as in Richardson et al. (2015). Briefly, we ground the bees’ hindguts in Ringer’s solution, homogenized them with a vortex and left them standing for 4 h to allow the tissues to settle. We micro-pipetted 10 μL of the supernatant onto a hemocytometer and counted the number of live Crithidia cells in a 0.02 μL volume. We also measured the radial cell length from the right forewing (hereafter referred to as “body size” for simplicity), which serves as a proxy for bee size (Nooten and Rehan 2020) to account for variability in cell counts due to bee size.

Nectar consumption was measured for all surviving bees in the nectar bioassay during the last day of the experiment to assess whether treatments affected consumption. We weighed a 0.6-mL microcentrifuge tube of nectar at the start and end of a 23-h consumption period. On one occasion we assessed consumption at 24 h; thus, we standardized by calculating consumption per hour. We were unable to measure pollen consumption due to logistical difficulties (e.g., frequent defecation in pollen and dislocation of the pollen from the cap). We also recorded deaths daily to assess whether diet treatments affected bee mortality.

Statistical Analyses

All statistical analyses were performed using R ver. 3.3.3 (R Core Team 2019) and all plots were created using the graphics (base), emmeans (Lenth 2019), ggeffects (Lüdecke 2018) and ggplot2 packages (Wickham 2016). For all analyses, we used Akaike information criterion (AIC) and parsimony to select the best model amongst all possible models. We first fitted a global model with all possible factors including interactions between herbivory treatment and collection period. Then we selected the best model by removing interactions first, and then removing factors until simpler models did not yield lower AIC values. When removals did not yield changes greater than 2 AIC units, we selected the simplest model. For zero-inflated compound Poisson models, the current R statistical packages do not calculate AIC and log-likelihood values. Hence, to select the best models we simply removed the least significant term until all remaining factors were significant at P < 0.05. Because calculating uncertainty for the zero-inflated compound Poisson models using the available R statistical packages was not possible, we provide a complete list of model parameter estimates in Online Resource 1 (Tables S4–5).

We assessed the effects of herbivory on pollen nicotine and anabasine concentrations in separate analyses. We used generalized linear mixed models with a binomial error distribution to determine whether herbivory affected the likelihood of detecting alkaloids in pollen at any time within our sampling period (i.e., a plant was scored as having alkaloids present if alkaloids were detected in at least one sample from that plant), since we did not detect nicotine and anabasine in 30% and 52% of plants, respectively. In the global models for each alkaloid, we included herbivory treatment, treatment date (Julian date) and number of pollen samples analyzed (i.e., higher probability of detection due to greater sampling effort) per plant as fixed effects. The top models for nicotine and anabasine only included treatment date as a predictor. We then analyzed the concentrations of nicotine and anabasine for plants whose pollen contained alkaloids in at least one sample. In this analysis, multiple pollen samples from the same plant were included as separate data points, and all pollen samples were included from a plant if that plant had alkaloids detected in any of its pollen. We fit generalized linear mixed models with Tweedie compound Poisson error distributions for each alkaloid using the cplm package (Zhang 2013). We chose a Tweedie distribution because our data were continuous, highly dispersed and included many zeros (Jørgensen 1992). The global models included herbivory treatment, the number of days after the first flower when sampling occurred (for simplicity, hereafter referred to as “days after first flower”) and a scaled quadratic term for days after first flower as fixed effects, and the plant individual as a random effect. We included a quadratic term for number of days after first flower because an analysis of pollen anabasine and nicotine concentrations over time by Davis et al. (2019) detected a quadratic response, where concentrations peaked mid-flowering season. We scaled the quadratic term by dividing it by the mean days after first flower to avoid model convergence issues. The top model for nicotine included days after first flower and its quadratic term. The top model for anabasine included herbivory treatment, days after first flower and its quadratic term. Both models excluded plant individual as a random effect.

To assess the effect of herbivory on Crithidia counts in the nectar and pollen bioassays, we used generalized mixed linear models with negative binomial error distributions using the lme4 package (Bates et al. 2015). The global models for the pollen and nectar bioassays included herbivory treatment, collection period (“within one month” and “after one month”) and their interactions as fixed effects, and colony of origin and inoculation date as random effects. The global model for nectar also included body size (estimated by wing radial cell length) as a fixed effect; the pollen bioassay did not include body size due to many missing values for the covariate. The top model for the pollen bioassay included collection period (“within one month” and “after one month”) and the interaction between collection period and herbivory treatment as fixed effects. The top model for the nectar bioassay included collection period and body size as fixed effects, and colony of origin and inoculation date as random effects. To adjust for multiple comparisons within each bioassay, we adjusted p-values using the false detection rate (FDR) and reported the corrected p-values. To assess whether nectar diets affected bee survival, we performed a Cox Proportional Hazards test. We did not analyze mortality for the pollen bioassay due to too few deaths (only 5 of 84 bees).

Data and Code Availability

Data and R script are available via Dryad (https://doi.org/10.5061/dryad.bk3j9kd8m

Results

Floral Induction

Herbivory did not increase the likelihood of detecting nicotine or anabasine (i.e., presence vs. absence) in tobacco nectar or pollen. Nectar did not have detectable levels of alkaloids. Anabasine was more likely to be present in pollen collected later (z-value = 2.51, P = 0.01), but collection period did not significantly affect the likelihood of detecting nicotine in pollen (z-value = 1.76, P = 0.08). Herbivory did not affect nicotine concentration in pollen (z-value = −1.26, P = 0.20; Fig. 1a), but significantly increased pollen anabasine by about 250% (damaged: 31.45 ng/mg 95% CI [25.37, 37.74], undamaged: 12.77 ng/mg, 95% CI [6.69, 18.85], z-value = 2.29, P = 0.03; Fig. 1b). Both nicotine and anabasine concentrations changed through time. Pollen nicotine concentrations initially increased before peaking 27 days after the first flower and then declining (time since flowering: z-value = 2.52, P = 0.01, time since flowering (quadratic): z-value = −2.53, P = 0.01; Fig. 1a). Similarly, anabasine concentrations in pollen increased until day 37 and then decreased (time since flowering: z-value = 3.07, P = 0.01, time since flowering (quadratic): z-value = −2.61, P = 0.01; Fig. 1b).

Bee Pathogen Bioassays

In the pollen bioassays, herbivory treatment and collection period both affected Crithidia counts. There was a significant interaction between herbivory treatment and collection period (z-value = −2.685, P = 0.01) such that damage only affected Crithidia counts when pollen was collected after one month of flowering. Contrary to expectations, bees fed late-collected pollen from damaged plants had 15 times higher pathogen counts than bees fed late-collected pollen from undamaged plants (damaged plants: 15.00 cells/0.02 μL, 95% CI [5.42, 41.52], undamaged plants: 1.11 cells/0.02 μL, 95% CI [0.31, 3.97]; P = 0.01; Fig. 2). Time since flowering also affected Crithidia counts within the undamaged treatment. Within undamaged plant diets, bees fed pollen collected within 1 month of flowering had 10 times higher Crithidia counts than bees fed pollen collected after 1 month of flowering (within one month: 10.82 cells/0.02 μL, 95% CI [5.25, 22.29], after one month: 1.11 cells/0.02 μL, 95% CI [0.31, 3.97]; P = 0.01; Fig. 2), indicating differences in pollen quality across the flowering period. However, in bees fed pollen from damaged plants we found no effect of collection period on Crithidia counts (within one month: 10.45 cells/0.02 μL, 95% CI [5.07, 21.55], after one month:15.00 cells/0.02 μL, 95% CI [5.42, 41.52], P = 0.57; Fig. 2). Crithidia counts did not differ in bees fed the control pollen diet compared to bees fed any tobacco pollen diet (Table S6).

Fig. 2
figure 2

Crithidia cell counts in B. impatiens workers fed pollen from herbivore-damaged and undamaged plants from two collection periods (within 1 month of flowering, 1 month after flowering). Means and 95% confidence intervals

Full size image

In the nectar bioassay, Crithidia counts were not affected by herbivore damage or collection period. However, larger bees had a non-significant tendency to have lower infection levels (z-value = −1.83, P = 0.07; Online Resource 1; Figure S3). Larger bees also consumed more nectar (Pearson’s correlation: r = 0.23, df = 130, P = 0.01), but nectar consumption was not affected by herbivory treatment or collection period (Herbivory Treatment: X2(2) = 1.94, P = 0.38; Collection Period: X2(2) = 1.36, P = 0.51). Bee mortality rates were not affected by nectar treatments (Wald-test = 0.71, df = 3, P = 0.9). Crithidia counts did not differ in bees fed the control nectar diet compared to bees fed any tobacco nectar diet (Table S7).

Discussion

We provide the first example of multitrophic effects of herbivory on pollinators and their pathogens via changes in floral rewards. Herbivory-induced responses often mediate bottom-up multitrophic effects on herbivores and their natural enemies (Soler et al. 2005), but studies have not considered effects on pollinators. It has been long recognized that herbivory can affect interactions between plants and pollinators, but work has focused on how herbivory changes floral traits, pollinator behavior and plant reproduction (reviewed in Lucas-Barbosa 2016; Moreira et al. 2019; Rusman et al. 2019). Surprisingly, we found that bumble bees that consumed pollen from tobacco plants damaged by herbivores had more Crithidia cells than bees that consumed pollen from undamaged plants, but only when the pollen was collected after one month of flowering (Fig. 2). This result does not support our original hypothesis that damage would reduce Crithidia infection, but demonstrates that in some contexts consequences of herbivory can extend beyond plant-pollinator interactions to affect higher trophic levels in unexpected ways.

The length of time elapsed between production of the first flower (i.e., time since flowering) and nectar or pollen collection was a surprisingly important mediator of the effects of herbivory on pollinator-pathogen interactions. Bees that consumed pollen from damaged plants had higher Crithidia infection intensity than bees that consumed pollen from undamaged plants, but only when the pollen was collected after one month of flowering (Fig. 2). Surprisingly, this effect is largely because bees fed control pollen collected after one month of flowering had considerably lower Crithidia cell counts than bees fed control pollen collected early (Fig. 2), suggesting some quality of pollen that reduces infection as plants age. Thus, pathogen infections may be greatly affected by the period within which infected individuals exploit a specific resource that fluctuates in quality. Temporal effects on pollen chemistry are largely unexplored, but temporal variation in secondary compound production in other tissues can have repercussions for plant-animal interactions, such as when herbivores’ digestive efficiency and growth rate decrease as plants age (Quintero and Bowers 2018). We found that pollen chemical properties changed over time; concentrations of nicotine and anabasine rose and fell along the progression of the plants’ flowering period (Fig. 1), a pattern similar to that found by Davis et al. (2019). However, further laboratory and field studies will be needed to test the hypothesis that temporal variation in floral rewards is widespread and affects pollinator-pathogen dynamics.

Herbivory did not induce higher concentrations of nicotine but did increase anabasine in tobacco pollen (Fig. 1). While anabasine and nicotine share a biosynthetic pathway (Solt et al. 1960), our results suggest differential regulation in floral rewards. Our work is consistent with previous research showing that herbivory induced higher concentrations of some alkaloids in nectar (Adler et al. 2006; Halpern et al. 2010; Kaczorowski et al. 2014), but it is the first to examine induction in pollen. Moreover, we note that we only focused on alkaloids; it is possible that other primary or secondary compounds that we did not consider could also be affected by herbivory. For example, proteins in tobacco nectar have been characterized as part of a nectar redox cycle involved in antimicrobial defense (Thornburg et al. 2003), but we know less about how herbivory affects these proteins or whether they occur in pollen. Understanding how pollen secondary chemistry is regulated is crucial because its role in plant-pollinator interactions is largely unexplored relative to nectar chemistry (Parachnowitsch and Manson 2015; Stevenson 2019). Optimal defense theory predicts that pollen should be more heavily defended than nectar, and several studies have shown that pollen contains higher concentrations (Cook et al. 2013; Davis et al. 2019) and diversity (Palmer-Young et al. 2019) of secondary compounds than nectar. Consistent with this prediction, we detected alkaloids only in pollen, although we note that both nicotine and anabasine are commonly found in tobacco nectar. Because pollinators can quickly assess some pollen qualities (i.e. taste) (Muth et al. 2016; Ruedenauer et al. 2016), pollinators may be sensitive to variation and rapid changes in pollen chemistry. Thus, our result that herbivory induces higher pollen anabasine provides a new potential mechanism for pollinator deterrence in response to herbivory (Kessler and Halitschke 2009; Moreira et al. 2019).

Herbivory altered pollen enough to modify interactions between bees and their pathogens, but the mechanism remains unclear. Although we predicted that anabasine or nicotine in floral rewards would mediate multitrophic interactions between herbivores, plants, pollinators and pathogens, we did not find evidence to support this hypothesis. Previous studies have shown that anabasine is inducible in nectar (Adler et al. 2006; Kaczorowski et al. 2014) and that anabasine and nicotine can reduce Crithidia cell counts in B. impatiens (Anthony et al. 2015; Baracchi et al. 2015; Richardson et al. 2015), although at higher concentrations than typically found in N. tabacum. However, we did not detect alkaloids in the nectar or pollen fed to bees in the bioassays (see Bee-Pathogen Bioassays: Diet Treatments in Materials and Methods). Furthermore, herbivory increased anabasine in our original pollen samples, but contrary to our prediction, herbivory also increased Crithidia counts when pollen was collected after one month of flowering. This suggests that induced anabasine is not the mechanism underlying the effect of herbivory on Crithidia counts. Nonetheless, the differences in Crithidia levels between bees that consumed control and damaged pollen indicate that herbivory induced some change that mediated interactions between bees and their pathogens, although we do not know whether the changes occurred in nutritive or nonnutritive components of pollen. This is an exciting area for future research.

This work provides an example of the role that biotic factors may play in shaping pollinator diet. We found that Crithidia infections were substantially higher in bees fed pollen from damaged plants than those fed pollen from undamaged plants, but only when pollen was collected after one month of flowering; this could either be because components induced by damage increase pathogen infection, or because components that reduce infection in undamaged plants are degraded by herbivory. Given the detrimental effects of Crithidia on their hosts, such as cognitive reductions, higher mortality and lower colony founding success (Brown et al. 2000; Gegear et al. 2006; Shykoff and Schmid-Hempel 1991), the negative indirect effects of herbivory could potentially reduce pollinator fitness. Environmental impacts on pollinator-pathogen interactions via changes in floral rewards may extend well beyond effects of herbivory. Other recent work in the same system found that both soil nutrients and mycorrhizal fungi can shape floral reward quality (Davis et al. 2019). Soil fertilizer increased pollen alkaloids in N. tabacum, while mycorrhizal association decreased them. Fertilizer and mycorrhizae treatments affected nectar and pollen quality, which in turn influenced bumble bee-Crithidia interactions, but effects did not correlate with alkaloid concentrations. This is very similar to our study, in which herbivory increased pollen anabasine but the effects of pollen diets on Crithidia were not mediated by alkaloids. Both these studies highlight the importance of a plant’s ecological context in shaping pollinator-pathogen interactions through changes in floral rewards.

In conclusion, we demonstrate that pollen secondary chemistry varies with time since flowering and in response to herbivory. We also demonstrate that herbivore damage may affect pollen quality and under some conditions modify interactions between pollinators and their pathogens, a novel example of multitrophic effects resulting from herbivory. Surprisingly, pollen from undamaged plants increasingly reduced Crithidia infection as time since flowering progressed, while in damaged plants this reduction over time did not occur. To understand the mechanisms driving differences in pollen quality due to temporal changes and herbivory, surveys of how broader arrays of nutrients and secondary compounds vary with time and ecological context will be essential. We conclude that biotic factors can impact pollinators beyond pollination events by shaping pathogen infection, with potential consequences for pollinator fitness.