Manipulation of multiple floral traits demonstrates role in pollinator disease transmission

Abstract

Plants modulate multitrophic ecological interactions, and variation in plant traits can affect these interactions. Pollinators are exposed to pathogens at flowers and acquire or transmit pathogens at different rates on different plant species, but the traits mediating those interactions are almost entirely unknown. We experimentally manipulated five plant traits that span scales including flower, inflorescence, and plant, to determine their effects on pathogen transmission between foraging bees. Specifically, we manipulated two morphological traits (corolla lip length and flower orientation within an inflorescence) and three resource distribution traits (inflorescence nectar, plant patch nectar and plant aggregation) in tents to test how plant traits affect bee pathogen transmission. We also quantified foraging behavior and fecal deposition patterns as potential mechanisms driving differences in transmission, and assessed trait manipulation consequences for bee reproduction. We found that pathogen transmission was reduced when we trimmed the corolla lip, evenly dispersed nectar distribution within an inflorescence, or aggregated plants in space. Some traits also affected bee reproduction; tents with trimmed corollas had more larval production than control tents, and tents with evenly distributed nectar across plant patches had more larval production than tents with clumped resources. Thus, some trait manipulations both reduced transmission and increased bee microcolony reproduction, although our design does not allow us to discern whether these are related or separate effects. Taken together, our results demonstrate causal effects of several floral traits on pathogen transmission and pollinator reproduction, indicating the importance of intraspecific plant trait variation for pollinator health and population dynamics.

Introduction

Plants are basal food resources that mediate diverse multitrophic interactions. Flowers can serve many beneficial functions for pollinators, providing not only nutrition (Di Pasquale et al. 2013), but also locations for finding mates (Eickwort and Ginsberg 1980), and habitat via overnight sheltering or materials for nest construction (e.g. Armbruster 1984). However, flower visitation includes risks to bees through exposure to antagonists, and we know surprisingly little about how floral traits affect pollinator interactions with their own community of natural enemies. Pollinators are at risk of predation while foraging by crab spiders and beewolves that attack bees at flowers (e.g. Abbott 2006), and a range of pathogens can be encountered during floral visits including viruses (Singh et al. 2010, Alger et al. 2019a), the trypanosomatid pathogen Crithidia bombi (Durrer and Schmid-Hempel 1994), the parasite Apicystis bombi, and fungi Nosema spp. (Graystock et al. 2015). Thus, plants mediate the transmission of bee pathogens via shared floral resources (Adler et al. 2018, 2020), and flowers can act as disease transmission hubs within communities.

Pathogen transmission via flowers can differ between plant species, but few specific traits have been identified that mediate transmission. A large observational study quantified fecal deposition, C. bombi survival, and pathogen acquisition by Bombus impatiens across 16 plant species, and found that plants with short, wide corollas were more likely to transmit pathogens (Pinilla‐Gallego et al. 2022). By contrast, a previous study that examined C. bombi acquisition rates from flowers of 14 plant species found no such relationship; the only trait that consistently correlated with acquisition was the number of reproductive structures per inflorescence (Adler et al. 2018). Evidence from several other studies suggests that plant species affects pathogen deposition, acquisition or transmission, but most include only 2–3 species (Graystock et al. 2015, Figueroa et al. 2019, Alger et al. 2019, but see Bodden et al. 2019). To our knowledge, only one study has manipulated floral traits to determine their causal role in pathogen transmission (Durrer and Schmid-Hempel 1994). Bombus terrestris that foraged on C. bombi-contaminated flowers arranged in a linear formation along an Echium vulgare inflorescence were more likely to become infected than bees that foraged on a spiral arrangement (Durrer and Schmid-Hempel 1994). This foundational work indicates the potential for floral traits to shape pathogen transmission, but there is a lack of studies to assess a wider range of traits.

Floral traits could change pollinator behavior in ways that affect pathogen transmission via many mechanisms (McArt et al. 2014, Adler et al. 2021). For pathogens like Crithidia with fecal-oral transmission, traits could alter transmission by affecting the location of contaminated feces. Plant species with longer or narrower corollas may be less likely to have fecal deposits in the flower than species with short, wide corollas (Figueroa et al. 2019, Bodden et al. 2019), although pathogens may desiccate more quickly in short, wide flowers (Pinilla‐Gallego et al. 2022). The arrangement of flowers in an inflorescence could affect transmission by changing how bees move (Durrer and Schmid-Hempel 1994), altering the likelihood of encountering feces. A higher amount or density of nectar rewards could increase foraging times (Harder 1983, Goulson 2000); longer total foraging times on an inflorescence can increase infection probability in bees (Adler et al. 2018) and longer flower handling times may increase fecal deposition on flowers (Alger et al. 2019a, Bodden et al. 2019). At a larger scale, resource density may also affect transmission, as flower density was positively correlated with Deformed Wing Virus prevalence on flowers and active infections in bumble bees (Alger et al. 2019b). Thus, variation in traits at the level of the flower, plant or patch all have the potential to shape dynamics between pollinators and pathogens.

Understanding how floral traits affect pollinator-pathogen dynamics is important because pathogens have been implicated in the decline of several pollinator species (Goulson et al. 2015), but ultimately, it is essential to also consider impacts on pollinator performance. Although there is a vast literature assessing how floral traits affect pollinator foraging decisions and plant reproduction (e.g. Essenberg 2012), assessing impacts of floral traits on measures of pollinator fitness is more challenging, and is typically only done in the context of pollen and nectar nutrition and chemistry (e.g. Di Pasquale et al. 2013, Stevenson 2020). A range of floral traits may affect both pathogen transmission and colony reproduction; such effects could be mediated by changes in pathogen infection (Malfi et al, in revision), or by other impacts of floral traits on pollinator performance, such as changes in foraging efficiency (Harder 1983, Goulson 2000).

To assess the effect of floral traits on disease transmission and bee performance, we manipulated multiple floral traits on different plant species in separate experiments and measured Crithidia bombi (hereafter ‘Crithidia’) transmission from infected to uninfected B. impatiens, adult worker survival, and colony larval production. Specifically, we asked how corolla lip length, flower orientation within an inflorescence, inflorescence nectar distribution, nectar distribution between plant patches, and plant aggregation between patches affected pathogen transmission and bee performance. We hypothesized that longer corolla lips would increase transmission because bees would be more likely to defecate inside the corolla, and that concentrating supplemental nectar resources (at the inflorescence or plant level) and creating architectures with higher floral densities would increase transmission due to greater likelihood of depositing and contacting feces. Isolating the effect of floral trait variation on the outcome of pollinator-microbial interactions is important for understanding how variation within and between plant species shapes disease transmission dynamics in pollinator populations.

Methods

We tested the effect of five plant traits on pollinator pathogen transmission inside large, enclosed foraging arenas (tents). All plants inside a single tent were of the same species (except for plant aggregation manipulations; see below) and subjected to the same treatment. In addition, for the three traits manipulated at the flower level (corolla lip length, inflorescence nectar distribution, and flower orientation), we quantified mechanistic drivers of transmission by measuring foraging behavior in the tents and tracking fecal deposition on manipulated inflorescences in separate choice tests using small flight cages.

Study system

Bombus impatiens is a eusocial bumble bee native to the United States commonly used for commercial pollination. Bombus impatiens can be infected by Crithidia, a trypanosomatid pathogen found in both wild and commercial colonies (Colla et al. 2006). Crithidia is transmitted horizontally via fecal-oral transmission within and between colonies through the shared use of flowers (Durrer and Schmid-Hempel 1994). Crithidia can be found in up to 80% of bumble bees in western Massachusetts, USA, where this study took place (Gillespie 2010). Under resource-stressed conditions Crithidia infection reduces colony founding success, colony size and male production (Brown et al. 2003). We assessed effects of floral traits on pathogen transmission and bee performance using microcolonies, which are groups of related workers where one worker establishes dominance and begins egg-laying; other workers forage and provision drone offspring (Tasei and Aupinel 2008). Bumble bee microcolonies can be used to estimate whole-colony reproductive performance (Tasei and Aupinel 2008).

Experimental design

We conducted this experiment in 2019 inside tents with known densities of bees and floral resources, using arrays of flowering perennial plants established in 2018 and selected for seasonal phenology and attractiveness to bumble bees. All species are native to the northeastern United States. We grew 36 blooming plants per tent arranged in three rows of 12, in one field (University of Massachusetts Center for Agriculture; South Deerfield, MA, USA, 42°28.60 N, 72°34.80 W). These consisted of: Penstemon digitalis (used for the corolla lip length experiment; 20 tents, June 13 to 27), Physostegia virginiana (flower orientation within inflorescences: 20 tents, July 12 to 26), Monarda didyma (inflorescence nectar distribution: 12 tents, July 9 to 18), Eupatorium perfoliatum (plant nectar distribution: 20 tents, August 8 to 21), and Solidago canadensis grown with Symphyotrichum novae-angliae blooming together (plant aggregation: 20 tents, September 11 to 24). Each tent had the same number of plants; the number of open flowers per tent was recorded for 3 days during each transmission experiment as a measure of floral abundance. A single floral trait was manipulated for each plant species, with half the tents in each experiment assigned to each treatment.

During peak bloom, we constructed tents using sealed 4.75 m L × 2.13 m W × 1.83 m H polytunnels to prevent bees from entering or leaving. Tents had transparent and UV-protected covers and four mesh windows for temperature regulation (manufactured by Palm Springs via Shop 274, Las Vegas, Nevada, USA). We deployed marked donor (infected) and recipient (uninfected) microcolonies together, assessed infection in recipient microcolonies at the end of each experiment to quantify transmission, and measured microcolony reproduction.

Creating microcolonies to quantify transmission

Transmission was measured as pathogen acquisition by recipient bees after foraging with donor bees in tents for 10–14 days. Bombus impatiens colonies were obtained commercially (Biobest, Leamington, Ontario, Canada). We constructed up to eight queenless microcolony replicates of 15 workers per parent colony from 26 parent colonies. To quantify transmission, two microcolonies were deployed into each tent, one inoculated with Crithidia (donor) and one uninfected (recipient). We painted thoraxes with acrylic paint (POSCA Uni-Mitsubishi Pencil, Tokyo, JP) to identify bees as donors or recipients. Within each tent, we paired donor and recipient microcolonies from the same parent colony, but due to uneven sample sizes this was not possible in 3 of the 92 tents. At the end of each experiment, we dissected bees and assessed pathogen infection.

We inoculated bees in the donor microcolonies on the day they were created, following a standard protocol (Van Wyk et al. 2021). We counted Crithidia cells according to the protocol below (Quantifying pathogen transmission) in the hindgut of bees from infected ‘source’ colonies maintained in the lab. Source colonies were originally infected from wild-caught local B. impatiens foragers in 2014 (Hadley, Massachusetts, USA: 42.3639 N, −72.5677 W). We diluted supernatant with Ringer’s solution to 1200 cells/ μL and added equal parts 50% sucrose solution to make a final inoculum with 600 cells/ μL and 25% sucrose. Donor bees were starved for 2 hr, provided 15 μL of inoculum (9000 Crithidia cells), and only entered the experiment if they consumed the entire droplet. Recipient microcolonies were not inoculated.

Plant trait manipulations

Corolla lip length

Penstemon digitalis (Plantaginaceae) produces clusters of 3–10 tubular flowers, with an extended lower lip (mean corolla length 20.9 mm). We manipulated corolla length in all flowers per tent by removing the lower lip with small dissecting scissors, leaving sexual organs intact but reducing corolla landing area (trimmed, Figure 1, A1). In a previous study, smaller B. impatiens workers were more likely to defecate inside the corolla tube than larger workers, suggesting corolla length might be a key trait mediating likelihood of pathogen transmission via contact with feces (Figueroa et al. 2019). In the control treatment, we inserted sharp forceps through the outer surface of the corolla along the upper center vein to create six small holes to control for the effect of tissue damage (control, Figure 1, A1). In total, we trimmed 105,190 and pierced 103,770 corollas.

Figure 1.

Schematic of plant trait treatments and main results for all experiments. Trait treatments in each experiment are lettered column headings: (A) Corolla lip length (Penstemon digitalis), (B) flower orientation within inflorescence (Physostegia virginiana), (C) inflorescence nectar distribution (Monarda didyma), (D) plant nectar distribution (Eupatorium perfoliatum), and (E) plant aggregation (Solidago canadensis and Symphyotrichum novae-angliae combined). Schematic and responses are numbered rows: (1) treatment schematic, (2) probability of infection of recipient bees, (3) infection intensity (cells/ 0.02 μL) of recipient bees, (4) reproduction (larvae produced at the end of transmission experiment), and (5) fecal deposition on inflorescences in small flight cages for plant traits manipulated at the inflorescence level. Note that A5 depicts deposition on flowers and leaves combined, B5 depicts deposition on leaves only, and C5 depicts deposition on flowers only. Error bars (95% confidence limits); asterisks indicate significant differences at p < 0.05 * and p < 0.01 **.

Flower orientation within an inflorescence

Physostegia virginiana (Lamiaceae) is a perennial forb that produces multiple racemes of small white tubular flowers (mean corolla length 7.7 mm). Colloquially known as the ‘obedient plant,’ manually repositioned flowers will remain in place until senescence. We capitalized on this unusual characteristic by manipulating the 3-dimensional orientation of flowers within a raceme to create two inflorescence architecture arrangements. Racemes grow with flowers arranged in four vertical lines of flowers set at 90 degrees. Linear arrangements were created by manipulating the flowers from opposite rows to create only two rows with the helical angle between flowers at 180 degrees (linear, Figure 1, B1). For the control treatment, flowers were pressed into the linear arrangement and then returned to the original perpendicular arrangement (control, Figure 1, B1). Tents were treated every two days to reposition all newly opened flowers, resulting in approximately 160,700 manually repositioned flowers across all tents.

Inflorescence nectar distribution

Monarda didyma (Lamiaceae) has dense, capitulum-like inflorescences with 1–25 flowers open at a time (mean corolla length 19.7 mm, mean 24-hr nectar volume 2.8 μL; range 0.5 to 7 uL). To manipulate the distribution of nectar resources within an inflorescence, we added 30% sucrose to mimic nectar resources inside one out of every five inflorescences, 3 times per day. We dispensed 2 μL drops to ten flowers (ten droplets 2 μL, Figure 1, C1), or 10 μL drops to two flowers (two droplets 10 μL, Figure 1, C1) using an Eppendorf repeating pipette. To control for the effect of physically inserting the pipette tip inside each flower, we performed dry insertions in eight additional flowers of the ten-droplet treatment by inserting and removing the pipette without adding sucrose. This resulted in 6490 pipette insertions into flowers.

Plant nectar distribution

Eupatorium perfoliatum (Asteraceae) has inflorescences (5–10 cm diameter) composed of small white flowers (mean corolla length 6.2 mm). We manipulated between-plant nectar distribution by spraying 30% sucrose solution on inflorescence surfaces. All tents received equal volumes of supplemental sucrose, dispensed from a 300 mL fine mist nozzle spray bottle (approximately 15 μL per spray). To create clumped resource patches, we marked three locations in the tent with open flowers using pin flags to receive three sprays each (clumped, Figure 1, D1). In even resource tents, 9 sprays of sucrose were dispersed across the tent to open flowers, with no inflorescence receiving more than 1 spray (control, Figure 1, D1). We treated all tents three times per day for 10 days.

Plant aggregation

Solidago canadensis and Symphyotrichum novae-angliae (both Asteraceae) can grow over 2 m tall with abundant flowering stalks composed of small compound flowers (mean corolla length 3.8 mm). The two species were planted in equal abundance together in each tent, interspersed such that mature plants grew to fill the tent completely. We manipulated plant aggregation by creating denser, more bunched flowering stalks. To do this we bound all stalks on a plant with twine 1 m from the ground and below the flowers prior to bloom (bunched, Figure 1, E1), to create 12 distinct clusters of inflorescences within each tent. We controlled for the effect of manipulation by gathering stalks and then releasing them (control, Figure 1, E1).

Quantifying pathogen transmission and microcolony reproduction

After bees had spent 10–14 days co-foraging in tents, we collected all marked donor and recipient bees. We assessed infection in both donor and recipient colonies because floral resources can affect infection intensity in previously infected colonies (Adler et al. 2020), which is biologically important in its own right and could also affect transmission to uninfected colonies. We anesthetized each bee to dissect the hindgut, then ground and vortexed it in 300 μL of Ringer’s solution. After waiting 4 hours for the supernatant to settle, we counted live Crithidia in a 0.02 μL subsample of a 10 μL aliquot on a hemocytometer at 400x magnification. We measured the marginal cell length (mm) of the forewing to estimate bee body size (Nooten and Rehan 2020), which can influence pathogen infection intensity (Van Wyk et al. 2021). Because floral manipulations could affect bee reproduction directly, as well as indirectly via pathogen transmission, we also dissected donor and recipient brood structures at the end of each experiment (except for inflorescence nectar distribution) to count number of larvae produced.

Mechanism of transmission: Foraging behavior

To assess whether plant treatment or bee infection status affected colony level foraging behavior, we recorded the behavior of donors and recipients in tents. Three times daily, we visited each tent and immediately recorded the number of bees foraging from each microcolony, but found no effect of any floral trait manipulation on this response (Appendix S1: Table S1). To assess individual behavior, we observed one foraging bee at a time, identified by paint marks as a donor or recipient, and timed all transitions between flowers and time spent in flowers. Observation of a single bee concluded when the bee stopped foraging or after 10 minutes to see if floral trait manipulation affected time spent inside of and transitioning between flowers. Individual observations on the nectar supplement and plant aggregation experiments were not conducted due to logistical constraints; the location of supplemental sucrose (a clear liquid) was difficult to see, marking flowers could influence bee behavior, and the flowers in the plant aggregation experiment had grown taller than the tent windows, obscuring observations.

Mechanism of transmission: Fecal deposition

We tested the effect of flower manipulations on patterns of fecal deposition for corolla lip length (Penstemon), flower orientation in inflorescences (Physostegia), and inflorescence nectar distribution (Monarda) using uninfected workers in small flight cages. We fed bees fluorescent paint, allowed them to forage for 3 hours, then illuminated defecated paint with ultraviolet light to count fecal droplets on flowers and leaves. We fed workers fluorescent paint powder dissolved in 30% sucrose (Day-Glo, Cleveland, OH) and pollen ad libitum a minimum of 24 hr before experiments. Bees were maintained in plastic 500 mL deli cups in an incubator at 27 °C in complete darkness. Each replicate consisted of a single flight cage (45 cm × 71 cm × 56 cm) with three related workers and two inflorescences per trait manipulation located on opposite sides of the cage. Flowers were bagged 24 hr prior to experiments and placed in water picks with preservative (30 mL/L FloraLife, Walterboro, SC, USA) to optimize nectar availability and reduce wilting.

Statistical Methods

General approach.

All statistical analyses were performed with R v.4.0.5 (R Core Team 2021). Statistical models were built for each trait manipulation experiment (glmmTMB::glmmTMB; Brooks et al. 2017) and the fixed effects were evaluated with Type II Wald χ² tests (car::Anova; Fox and Weisberg 2019). We assessed variance inflation factors (VIF) between all predictors (vif::car; Fox and Weisberg 2019); values indicated low collinearity (VIF ≤ 2.7 for all). The best fit model was chosen via information criterion evaluation (bbmle:: ICtab; Bolker 2016).

Transmission.

We tested treatment effects on pathogen transmission separately for each trait manipulation experiment using hurdle models (glmmTMB::glmmTMB; Brooks et al. 2017) to determine the probability of transmission to recipient bees (binary response of infection presence/absence) and infection intensity in infected bees only (negative binomial model with count response of Crithidia cells/ 0.02 μL). We also calculated total infection of donor bees (i.e., summed cells/ 0.02 μL for all surviving donors) for each tent. The initial full model to analyze infection in recipient bees included the fixed effects of trait manipulation, and the covariates of total infection of donor bees, floral abundance, and recipient body size. Tent was included as a random intercept in all full models for transmission. We also analyzed treatment effects on donor infection intensity, and report these methods and results in Appendix S1.

Reproduction, behavior and survival.

To assess whether trait manipulations affected bee reproduction for both donor and recipient colonies, we used generalized linear mixed models with a Poisson distribution with the number of larvae produced as the response, trait manipulation and donor status (i.e., donor or recipient) of the microcolony as the fixed effects and covariates of total tent infection, number of surviving bees in that colony and floral abundance. We conducted similar but separate analyses on number of surviving bees per donor and recipient colony, including total donor infection for recipient colonies only. Trait manipulations did not significantly affect bee survival in any experiment (Appendix S1: Table S2). To analyze foraging behavior, we used a generalized linear mixed model with the number of flowers visited per bout, time inside and between flowers as responses, trait manipulation and donor status as the fixed effects, floral abundance and total tent infection as covariates, with the random effect of tent.

Fecal deposition.

We used a linear mixed model with fecal deposition (drops of feces detected on flowers or leaves in small cages) as the response, trait manipulation as a categorical fixed effect, number of flowers per cage as a covariate, and cage replicate as the random effect.

RESULTS

Corolla lip length (Penstemon digitalis)

Transmission.

The best fit transmission model retained trait manipulation as a fixed effect, total infection of donor bees as a covariate, and tent as a random effect. At the end of the 14-day experiment, 18.7% of recipient bees acquired infection (n = 34/182). Trimming the corolla lip decreased infection probability and intensity of recipient bees (probability: χ² = 9.06, df = 1, p = 0.002, Figure 1, A2; intensity: χ² = 9.53, df = 1, p = 0.002, Figure 1, A3; n = 182 bees). Total infection intensity of donor bees increased infection of recipients (probability: χ² = 9.78, df = 1, p = 0.002; intensity: χ² = 25.3, n = 182, df = 1, p < 0.001).

Reproduction.

More larvae were produced in trimmed than control tents (χ² = 3.58, n = 139 larvae across 20 tents, df = 1, p = 0.05; Figure 1, A4). The number of surviving donor bees (χ² = 6.44, n = 139, df = 1, p = 0.01), floral abundance (χ² = 67.2, n = 139, df = 1, p < 0.001), and total infection in the tent (χ² = 7.3, n = 139, df = 1, p = 0.006), were positively correlated with the number of larvae produced.

Behavior.

We observed 3207 floral visits (trimmed = 1534, control = 1673) by 547 bees (trimmed = 296, control = 251) during 750 minutes of observation. Trimming the corolla lip increased time spent inside flowers for donor bees (mean (s) [95% CI] control = 1.39 [1.34, 1.43]; trim = 1.45 [1.40,1.50]; χ² = 4.19, n = 1614 flower visits, df = 1, p = 0.04), but not recipient bees (mean (s) [95% CI] control = 1.48 [1.42, 1.54]; trim = 1.47 [1.42,1.53]; χ² = 0.04, n = 1787 flower visits, df = 1, p = 0.82). Trimming the corolla lip did not change the time spent between flowers for all bees (χ² = 1.4, n = 1673 trimmed visits, 1534 control visits, df = 1, p = 0.23), but donor (infected) foragers spent more time traveling between flowers than recipient foragers (mean (s) [95% CI] donor = 5.27 [0.44, 10.1], recipient = 4.91 [0.09, 9.7]; χ² = 8.1, n = 1535 recipient transitions, n = 1202 donor transitions, df = 1, p = 0.004).

Fecal deposition (small cages).

Feces landed on inflorescences (flowers and leaves) with untrimmed corollas more often than the trimmed corolla inflorescences, although this effect was not significant (χ² = 3.48, n = 27 cages, df = 1, p = 0.06, Figure 1, A5). There were more fecal droplets on inflorescences with more flowers (χ² = 10.5, n = 27, df = 1, p = 0.001), regardless of treatment.

Flower orientation within an inflorescence (Physostegia virginiana)

Transmission.

The best fit transmission model retained trait manipulation as the fixed effect (but this was not a significant predictor) and tent as the random effect. At the end of the 16-day experiment, only 4.4% of recipient bees acquired infection (n = 8/180). There was no significant effect of floral orientation on infection transmission (probability: χ² = 0.01, df = 1, p = 0.99, Figure 1, B2; intensity: χ² = 2.96, df = 1, p = 0.08, Figure 1, B3; n = 180 bees across 20 tents).

Reproduction.

Manipulating floral orientation had no significant effect on microcolony reproduction (χ² = 0.25, n = 21 larvae, df = 1, p = 0.6; Figure 1, B4). Larvae produced did not correlate with the number of surviving bees, floral abundance or total tent infection (χ² < 1.8, n = 21, df = 1, p > 0.2).

Behavior.

We observed 3206 floral visits and transitions (linear = 1618, control = 1588) by 201 bees (linear = 108, control = 93) during 450 minutes of observation. Manipulating flower orientation did not affect the time spent inside or transitioning between flowers for any bees (inside: χ² = 0.64, n = 1603 flower visits, df = 1, p = 0.4; between: χ² = 0.35, n = 1466 timed transitions, df = 1, p = 0.56).

Fecal deposition (small cages).

Manipulating floral orientation did not affect the number of fecal drops on flowers (χ² = 0.12, n = 57 cages, df = 1, p = 0.73), but feces were more likely to land on leaves of control compared to linear inflorescences (χ² = 9.6, n = 57, df = 1, p = 0.001, Figure 1, B5). There were more fecal droplets on inflorescences with more flowers (χ² = 27.4, n = 57, df = 1, p < 0.0001).

Inflorescence nectar distribution (Monarda didyma)

Transmission.

The best fit transmission model retained trait manipulation as the fixed effect, total infection of donor bees and recipient body size as covariates, and tent as the random effect. At the end of the 10-day experiment, 13.7% of recipient bees acquired infection (n = 13/95). Tents with clumped resources (two 10 μL drops) had higher recipient infection intensity than tents with dispersed resources (probability: χ² = 0.35, df = 1, p = 0.55, Figure 1, C2; intensity: χ² = 3.59, df = 1, p = 0.05, Figure 1, C3; n = 95 bees across 12 tents).

Fecal deposition in small cages.

Feces landed on the flowers of plants with dispersed resources (ten 2-μL drops) more often than clumped (two 10-μL drops) (χ² = 5.5, n = 47, df = 1, p = 0.01, Figure 1, C5), but inflorescence nectar distribution had no effect on fecal deposition on leaves (χ² = 0.38, n = 47, df = 1, p = 0.54). There was no significant relationship between fecal droplets and the number of flowers in an inflorescence (χ² = 0.8, n = 47, df = 1, p = 0.35).

Plant nectar distribution (Eupatorium perfoliatum)

Transmission.

The best fit model retained trait manipulation as the fixed effect (but this was not a significant predictor), recipient body size as a covariate, and tent as the random effect. At the end of the 14-day experiment, 8.8% of recipient bees acquired infection (n = 14/158). There was no significant effect of the trait manipulation on transmission (probability: χ² = 0.65, df = 1, p = 0.41, Figure 1, D2; intensity: χ² = 2.9, df = 1, p = 0.08, Figure 1, D3; n = 158 bees across 20 tents), but there was a positive effect of recipient body size on infection intensity (χ² = 10.19, n = 158, df = 1, p = 0.001).

Reproduction.

More larvae were produced when supplemental nectar was evenly distributed than when it was clumped (χ² = 6.81, n = 36 larvae, df = 1, p = 0.01; Figure 1, D4). The number of larvae produced was positively correlated with the number of surviving bees (χ² = 19.66, n = 36, df = 1, p < 0.001), but not floral abundance (χ² = 0.36, n = 36, df = 1, p = 0.55) or total tent infection (χ² = 2.1, n = 36, df = 1, p = 0.15).

Plant aggregation (Solidago canadensis with Symphyotrichum novae-angliae)

Transmission.

The best fit transmission model retained trait manipulation as a fixed effect, floral abundance, total infection of donor bees, and recipient body size as covariates, and tent as the random effect. By the end of the 13-day experiment, 20% of recipient bees acquired infection (n = 26/132) The bunched treatment reduced the probability and intensity of infection transmission compared to the control treatment (probability: χ² = 9.06, df = 1, p = 0.002, Figure 1, E2; intensity: χ² = 10.3, df = 1, p < 0.001, Figure 1, E3; n = 132 bees across 20 tents). Floral abundance was positively correlated with infection intensity (χ² = 10.7, n = 132, df = 1, p < 0.001).

Reproduction.

There was no effect of manipulating plant aggregation on microcolony reproduction (χ² = 0.09, n = 59 larvae, df = 1, p = 0.77; Figure 1, E4). The number of larvae produced was positively correlated with the number of surviving bees (χ² = 7.66, n = 59, df = 1, p = 0.006) and floral abundance (χ² = 4.51, n = 59, df = 1, p = 0.03), and negatively correlated with total tent infection (χ² = 4.1, n = 59, df = 1, p = 0.04).

Discussion

Pollinators encounter a complex network of natural enemies and pathogens while visiting flowers, but we know relatively little about how variation in floral traits mediates these interactions. We manipulated traits at the flower, inflorescence, and plant level in tents, and found that traits at all of these scales can alter pathogen transmission in Bombus impatiens workers. Because each trait manipulation was conducted in only one set of species, it is possible that effects would vary in different species contexts; an interesting future direction would be to manipulate the same trait across many plant species to assess generality. Our manipulative results complement those of a new observational study which found that floral shape variation across 16 plant species affected several aspects of Crithidia transmission in B. impatiens. Short, wide flowers were more likely to have feces deposited and acquired, but pathogen survival was lower on these more exposed flowers; the net effect was that plant species with short and wide corollas were more likely to transmit pathogens (Pinilla‐Gallego et al. 2022). Both observational and manipulative approaches are important to demonstrate the potential for floral traits to influence pathogen dynamics at landscape levels. Although landscape studies will involve much greater complexity of species interactions, two studies have found that the presence of a single plant species can reduce pathogen infection in bumble bees across sites (Vanderplanck et al. 2019, Malfi et al, in revision), suggesting that floral traits or resources can have landscape-level impacts. Taken together, this research demonstrates the potential for floral traits to play causal roles in shaping pathogen transmission to bees.

Trimming the corolla lip of Penstemon flowers significantly decreased transmission probability and infection intensity. Although we cannot conclusively determine the mechanism underlying this result, it is consistent with our hypothesis that removing the “landing area” would result in the bee abdomen protruding from the corolla and reduce fecal deposition inside the flower. Previous work found that larger B. impatiens were less likely to defecate inside corolla tubes than smaller bees (Figueroa et al. 2019), and in our fecal deposition experiments feces tended (p = 0.06) to land inside control flowers more often than trimmed flowers. Although we found that trimming corolla tubes increased flower handling times by infected bees and thereby the potential to deposit feces, this effect may be negated by the decreased surface area for feces to land on. Thus, the manipulation of a single trait may affect not only transmission, but also interact with infection to shape bee behavior and pollination services.

Surprisingly, floral orientation within a Physostegia inflorescence did not affect transmission rates. We expected this manipulation to affect transmission because the only prior experiment to experimentally assess how floral traits shape transmission found that Bombus terrestris were more likely to become infected with Crithidia when foraging on linearly-arranged flowers compared to the natural spiral arrangement (Durrer and Schmid-Hempel 1994). The authors hypothesized that transmission differences were due to bees skipping more flowers on the spiral inflorescence, which could decrease encounters with feces. Similarly, we predicted the linear arrangement would result in higher transmission than the natural ‘cross’ because flowers are more densely arranged within the linear rows and bees are more likely to travel vertically as the angle between flowers increases (Iwata et al. 2012), increasing the likelihood of following the same path and contacting feces. However, in our fecal deposition experiments, manipulating floral orientation did not affect fecal deposition on flowers. The contrast between our results and those of Durrer and Schmid-Hempel suggests that even similar manipulations can have distinct effects with different plant and bee species; work should be done with more species to confirm the generality of plant trait effects in shaping pathogen transmission.

We manipulated resource density at the level of inflorescences and plants, and found contrasting effects at different scales. Concentrating nectar resources as larger rewards in a few Monarda flowers resulted in five times higher transmission intensity than smaller resources dispersed across many flowers. Based on the threshold departure rule (Hodges 1985), bees are more likely to remain at a plant when nectar rewards are large, and thus could increase chances of depositing or contacting feces. However, we found more feces on flowers when nectar was dispersed in many small drops compared to few large drops. This suggests that care should be taken when extrapolating results at different scales or between species. At the plant patch scale, dispersing nectar resources in Eupatorium did not affect transmission. At the plant scale, bunching flowers together decreased transmission compared to more evenly distributed natural arrangements. This contrasts with the results at the inflorescence level, where more clustered resources increased transmission, and with our general hypothesis that higher density floral resources would increase transmission due to increased contact between foraging bees and infectious feces. The effect of floral density on pollinator visitation can be inconsistent across studies, and depends on background floral density and traits of both plants and pollinators (Essenberg 2012). The few studies asking whether floral density affects pathogen transmission in the field also had contrasting results; an observational study found that pathogen transmission between bees was lowest when floral density was high, suggesting dilution effects (Graystock et al. 2020), while a manipulative study found that bee pathogen prevalence was highest in plots with the most floral resources, suggesting transmission hotspots (Piot et al. 2019). These contrasting results indicate that the effect of floral density on both visitation and pathogen transmission can be context-dependent and affected by scale. Experiments at the inflorescence level are useful for isolating mechanisms, but ultimately, broader experiments will be needed to determine how these mechanisms affect transmission at a field-realistic scale.

Manipulating floral traits also affected bee colony reproduction, and our results showed that trait variation can have consequences for pollinator populations that are distinct from effects on pathogen transmission. Trimming Penstemon corolla lips decreased transmission and increased larval production compared to control flowers. However, evenly dispersing nectar on Eupatorium inflorescences did not affect transmission but did increase larval production. Conversely, bunching Solidago and Symphyotrichum plants decreased transmission, but did not affect reproduction. Floral trait manipulations may therefore affect bee reproduction via mechanisms other than pathogen transmission. For example, more clumped nectar resources in Eupatorium may require increased foraging effort and energy expenditure, decreasing reproduction. We are aware of surprisingly few studies that have directly assessed how variation in floral traits other than nectar or pollen quality affect bee reproduction. Our experiments demonstrate that variation in multiple floral traits can have consequences not only for pathogen transmission, but also bee reproduction.

Our experiments also suggest that increased floral resources can shape transmission. In our fecal deposition experiments, the number of open Penstemon and Physostegia flowers was positively correlated with fecal deposition. Bees visit more flowers on inflorescences with more flowers (Ohashi and Yahara 2002, Grindeland et al. 2005) and therefore may deposit and/or encounter more feces. Previous transmission experiments found that pathogen transmission was correlated with the number of reproductive structures (including flowers, buds, and fruits) on an inflorescence, and also with the number of open flowers in Monarda, although inoculum was manually added to flowers, rather than allowing natural deposition of feces (Adler et al. 2018). Thus, plants with more open flowers may offer more resources for bees, but also have the potential to be hubs of increased transmission.

Using a manipulative approach, we demonstrated that a wide range of floral and plant traits affect pollinator pathogen transmission and pollinator reproduction. B. impatiens is the most common bumble bee in the Eastern United States, driving pathogen dynamics in the rest of the pollinator community (Figueroa et al. 2020). Identifying traits also allows us to take a trait-based approach to predicting the roles of species in multitrophic interactions, which has greater power than describing the role of individual species (Westoby and Wright 2006, Adler et al. 2018). Given that pollinators are declining in part due to pathogens (Goulson et al. 2015) and that governments, industry and individuals are investing in pollinator-friendly options for plantings, understanding the effects of floral traits on pollinator pathogen transmission will allow us to more efficiently identify plant species that are maximally beneficial for pollinator health.