Eristalis (Diptera: Syrphidae) Flower Flies are Potential Non-host Vectors of the Common Trypanosome Bee Parasite, Crithidia Bombi

Flowers can be transmission platforms for parasites that impact bee health, yet bees share oral resources with other pollinator taxa, such as ies, that could be hosts or non-host vectors (i.e., mechanical vectors) of parasites. Here, we assessed whether the fecal-orally transmitted gut parasite of bees, Crithidia bombi, can infect Eristalis tenax ower ies. We also investigated the potential for two conrmed solitary bee hosts of C. bombi, Osmia lignaria and Megachile rotundata, as well as two ower y species, Eristalis arbustorum and E. tenax, to transmit the parasite at owers. We found that C. bombi did not replicate (i.e., cause an active infection) in E. tenax ies. However, 93% of inoculated ies defecated live C. bombi in their rst fecal event, and all contaminated fecal events contained C. bombi at concentrations sucient to infect bumble bees. Flies and bees defecated inside the corolla (ower) more frequently than other plant locations, and ies defecated at volumes comparable to or greater than bees. Our results demonstrate that Eristalis ower ies are not hosts of C. bombi, but they may be mechanical vectors of this parasite at owers. Thus, ower ies may amplify or dilute C. bombi in bee communities.


Introduction
Recent analysis of long-term sampling data and biological records have shown that globally, wild insect pollinators, including solitary bees and ies, are experiencing population declines and range contractions [1][2][3][4] . Parasites are key drivers of pollinator health and are associated with declines of several pollinator species [5][6][7] . This fact is concerning for both conservation and economic reasons; pollination services are valued at more than $170 billion/year globally 8 , solitary bees and ies perform a large proportion of the services [9][10][11][12] , and agricultural dependence on pollinators continues to increase each year 13 .
Most information about pollinator parasites is known from social bees 7,14,15 . Honey bees and bumble bees were previously thought to be the only host of certain viruses, microsporidians and trypanosomes, but recent studies have found many of these same parasites are present in and can infect wild solitary bee species, too [16][17][18][19][20][21][22] . The host range of these parasites, however, is understudied, and limited studies have assessed the incidence and infectivity of bee parasites in non-bee pollinators.
Evison et al 23 reported high prevalence of Wolbachia bacteria and Ascosphera fungi, and low prevalence of microsporidian fungi 23 , among wild-caught ower ies and bees. Additionally, Bailes et al 24 found high viral titres of two honey bee viruses (Sacbrood Virus and Black Queen Cell Virus) in wild-caught Eristalis (Diptera: Syrphidae) ower ies 24 . The nucleotide sequences of these viruses were 87-100% similar to those found in co-foraging honey bees, suggesting the viruses were not strains unique to ies and possibly being shared between the two taxa. Similarly, Brettell et al. 25 also found bee-associated viruses in wild-caught ower ies after deep sequencing 25 . While these studies suggest many bee parasites may be broad, multi-host parasites, they do not show whether infection (active replication of the parasites) is occurring in non-bee pollinators, nor how transmission occurs.
The trypanosome gut parasite Crithidia bombi lacks cell-speci c host requirements compared to intracellular parasites such as microsporidian Nosema spp. 26,27,28 , has been found in a wide variety of solitary and social bees 15,21,22 , and despite historically being considered a "bumble bee parasite," has recently been found to replicate in two species of solitary bees, the alfalfa leafcutter bee, Megachile rotundata, and blue orchard bee, Osmia lignaria 21,22 . The rather broad host requirements of C. bombi indicate this parasite may be able to infect, or pass through, the guts of many different ower-visiting insects. However, beyond the studies mentioned above, it is unknown if other ower-visiting insects such as ower ies (Diptera: Syrphidae) can act as hosts of C. bombi. This is important since ower ies occupy a similar ecological niche as bees as a result of their similar morphology, behaviors and foraging habits 29,30 .
In addition to host competence, epidemiology of multi-host parasites is also shaped by the environment.
One particularly relevant factor to pollinator parasites are owers -key hubs of transmission. Graystock, Goulson & Hughes 7 experimentally demonstrated that infected honey bees can transmit common, fecaloral bee parasites including C. bombi to susceptible bumble bees (and vice versa) simply by foraging on the same owers 31 . Furthermore, a recent study that screened nearly 3,000 owers in nature found one in eleven owers harbored pollinator parasites, including C. bombi 15 . Multiple factors such as landscape simpli cation 32 , presence of managed social bees 18 , oral traits 33,34 , and location of parasites on owers 35 can in uence the prevalence and likelihood of transmission of pollinator parasites at owers.
Flies may contribute to transmission by mechanically spreading parasites from contaminated owers, potentially redistributing the parasites on owers during defecation and therefore creating more oral transmission hotspots. However, whether ies can act as mechanical vectors and transmit pollinator parasites at owers, at quantities that can infect bees, is unknown. In addition, how ies compare to bees as parasite transmitters at owers is also unknown.
As adult Eristalis ies visit the same oral resources as bees, they, too, can encounter, ingest and potentially become infected by common fecal-orally transmitted "bee" parasites. Therefore, we assessed: 1) whether the common, fecal-orally transmitted "bee" parasite, C. bombi, could infect the cosmopolitan European drone y, Eristalis tenax, 2) whether the quantity of viable C. bombi cells defecated by E. tenax would be su cient to infect a common host of C. bombi, Bombus impatiens, and 3) whether two species of Eristalis ies (E. arbustorum and E. tenax) as well as two megachilid bee hosts (M. rotundata and O. lignara) defecate on owers and therefore could potentially transmit C. bombi at owers.

Results
3.1 Evaluating whether the European drone y, Eristalis tenax, is a host or non-host vector of Crithidia bombi 3.1.1 Inoculation of Eristalis tenax with Crithidia bombi E. tenax ies were inoculated with C. bombi and both the rst defecation event and gut were screened for the parasite. C. bombi was never found in the gut of the ies 10 days post-inoculation (Table 1). However, all ies defecated within 5 hours of the start of the experiment and the rst fecal event from 93% of the inoculated ies (n = 30) contained live C. bombi (Table 1). The average amount of live C. bombi in these rst fecal events were 239 parasites (95% CI: 174.4-362.7 parasites; Fig. 1a). One fecal event contained a total of 1,080 cells, which was the greatest concentration and equated to roughly one-third of the inoculum fed to the ies. There was no signi cant difference in C. bombi load between males vs. females (LRT, χ 2 1 = 0.19, p = 0.66) and no C. bombi were found in the rst defecation events of the control ies (n = 30; Table 1). The average rst fecal volume among the inoculated ies was 1.21 µl (95% CI: 1.03-1.39 µl; Supplementary Figure S1). Together, these results indicate that C. bombi cells can survive passage through y digestive tracts, although they do not cause active infections in the ies.

Response of Bombus impatiens to varying doses of Crithidia bombi
Bumble bees (B. impatiens workers) from two colonies were inoculated with varying doses (between 12 to 25,000 cells) of C. bombi as shown in Fig. 1b. These doses were chosen to include the realistic range of C. bombi in ower y feces (shown above) and beyond. Infection probability increased with dose, and the slope of the relationship was colony-dependent (dose-colony interaction: LR = 9.6, p = 0.002, Supplementary Figure S2a). Smaller bees were also more likely to become infected (LR = 14, p < 0.001). Among the infected bees, infection intensity increased with dose (t 69 = 2.8, p = 0.007), was also colonydependent (t 69 = -2.122, p = 0.037, Figure S2b), and smaller bees had higher infection intensities (t 69 = -2.8, p = 0.006). Conditional (pseudo-)R 2 of the models were 0.43 for infection probability, and 0.27 for infection intensity.
We generated dose-response curves for the two responses by marginalizing across colony and bee size Generalized linear models assume linear relationships between the link function and predictor. For each response and each colony, we did not nd substantially stronger support for more exible monotonic additive models (infection probability: ΔAIC = -3 × 10 − 4 (colony 1), 1.2 (colony 2); infection intensity: ΔAIC = 0.016 (colony 1), 1.3 (colony 2), suggesting that the linear relationships were adequate at capturing the shape of the dose-response curves. Table 1 Proportion of inoculated and control Eristalis tenax ies with live Crithidia bombi cells in rst fecal events (day 1) and guts 10 days post-inoculation.

Treatment
Live C. bombi in y feces (day 1) Live C. bombi in y gut 3.2 Vectoring potential of two bee species, Osmia lignaria and Megachile rotundata, and two Eristalis y species, E. arbustorum and E. tenax

Fecal volumes and defecation frequency
All pollinators were placed in individual cages lined with lter paper and fed ad libitum sucrose solution with uorescent powder to measure fecal volumes and defecation frequency in a 24-hour period. When comparing fecal volumes between different species and sex using the lter paper data only (which used identical methodology for all pollinators), we found that the species-sex interaction was signi cant (F (3, 457) = 10, p < 0.001; Fig. 2), so main effects were not tested in accordance to principle of marginality. Based on post-hoc pairwise contrasts (Supplementary Table S1), we found that for both sexes, E. tenax had signi cantly larger fecal volumes than all three other pollinator species, while E. arbustorum and O. lignaria both had larger fecal volumes than M. rotundata (Fig. 2). Within each species, only E. tenax showed a signi cant difference between sexes, with females having larger fecal volumes than males.
For E. tenax only, comparing the fecal volumes from the two methods (collected from microcentrifuge tubes in the inoculation experiment vs. estimated from lter paper spot diameters), we found that the method-sex interaction was marginally signi cant (F (1,126) = 3.9, p = 0.052), while both the main effects of method and sex were signi cant (method: F (1,127) = 6.0, p = 0.016; sex: F (1,127) = 22, p < 0.001). In particular, we found that fecal volumes collected from the microcentrifuge tubes were about 25% (95% CI: 4.6%-50%) larger than fecal volumes estimated using standard curves from lter paper spot diameters (Supplementary Figure S4). Nonetheless, since the interaction was not signi cant, this means that any inference about fecal volume differences between groups should not be affected by the choice of methods.
E. tenax ies defecated more frequently than E. arbustorum ies in a 24-hour period (F (1, 137) = 85, p < 0.001; Fig. 3 We found that the inside of the corolla was defecated on most often (compared to the outside of the corolla, sepal, bract, leaf and stem) respectively (p < 0.001 in each case; Fig. 4). In addition, the leaf was defecated on more frequently than the bract (post-hoc Z-test, Z (Inf) = -3.542, p = 0.005).

Discussion
In this study, we found that C. bombi did not replicate and cause an active infection in E. tenax ies. However, 93% of inoculated ies defecated live C. bombi in their rst fecal event, often at levels capable of infecting bumble bees. In addition, we show that Eristalis arbustorum and E. tenax both defecate comparable or larger volumes, respectively, of feces compared to Megachile rotundata and Osmia lignaria, two solitary bees that are recently con rmed hosts of C. bombi 21,22 . Furthermore, E. tenax and E. arbustorum are both shown to defecate on owers, which are indirect transmission routes for C. bombi 31,35,36 . Taken together, these results indicate that while Eristalis ower ies are not hosts of C. bombi, they can potentially be non-host vectors (i.e., mechanical vectors) that contribute to community-wide transmission of this multi-host parasite.
Infected bumble bees are known to shed similar parasite loads to the inoculum we fed to E. tenax ies [37][38][39] . In fact, heavily infected bumble bees can shed concentrations as high as 55,000 cells/µl in their feces we used for inoculum. We found that inoculated E. tenax ies defecated C. bombi in levels lower than those ingested. However, all C. bombi quantities found in the y feces were capable of establishing an infection in bumble bees. Speci cally, we show that an inoculation dosage of only 24 C. bombi cells can establish an infection in bumble bees. This quantity of C. bombi is less than the lowest quantity we found in ower y feces (60 cells) and much lower than the mean C. bombi quantity in ower y feces (239 cells). As susceptible hosts range in size and immune-related traits, the number of parasites required to infect a smaller host, such as M. rotundata or O. lignaria, may be different compared to larger Bombus hosts. Therefore, the infectivity of C. bombi loads shown in this study may vary based on the susceptibility of host species.
While we report that C. bombi does not infect E. tenax ies, uninfected Eristalis ies still possess traits which increase their vectoring potential of the parasite by defecating frequently and in large volumes. E. arbustorum ies defecated comparable volumes to O. lignaria bees, and E. tenax ies defecated the largest fecal volumes of the four pollinators. Although both bees are competent vectors of C. bombi 21 , our results suggest that O. lignaria may be more likely to transmit the parasite, as this bee species defecates larger fecal volumes than M. rotundata. Susceptible pollinators are more likely to acquire fecalorally transmitted parasites from large volumes of infected feces when foraging, as large fecal events take longer to evaporate, thus allowing parasites to survive for a greater period of time outside of a host 35 . Differences in vectoring potential between competent hosts and non-host vectors warrant further investigation, with the possibility to reveal novel factors that may be incorporated into disease modelling of species-rich pollinator communities 40 .
We also demonstrated that bee and y pollinators defecate on certain locations of a shared oral resource more frequently than others. The bee O. lignaria defecated the most on Solidago, which may suggest this known host of C. bombi spent the most amount of time on the oral resource compared to the two y species. However, of the six locations (inside the corolla, outside the corolla, on the sepal, bract, stem and leaves), all three pollinators defecated most often on the inside of the corolla. Goldenrod is an important late-season resource for many pollinators, and since Crithidia is not a vertically transmitted parasite 41 , solitary pollinators foraging on owers for pollen and nectar are likely acquiring the parasite at shared oral resources.
While E. tenax ies can act as mechanical vectors by ingesting and shedding viable C. bombi cells, this does not necessarily imply an ampli cation in disease transmission. When vectors ingest parasites at a oral resource, they also remove them from existing oral hotspots; mechanical vectors redistribute parasites across the landscape by increasing the number of hotspots, but at the expense of decreasing the average parasite load per hotspot. Whether this redistribution leads to ampli cation from the increased number of hotspots, or a dilution from the decreased average load, depends on the doseresponse relationship between host and parasite and the number of non-host vectors capable of spreading the parasite 42 . The successful infection of bumble bees when fed low inoculum doses (e.g., only 24 cells) of C. bombi, however, suggest the potential for ampli cation, although ultimately this will need to be assessed using a quantitative epidemiological model. Furthermore, it is unknown how many non-host vectors can spread bee parasites at owers. Cook et al. 43 found that ies from 86 families have been reported visiting the owers of more than 1,100 different species of plants globally, however, ies are not the only insects sharing oral resources with bees.
When we assessed C. bombi viability in the feces of the E. tenax ies, we deemed the parasites viable if they were still motile. While the parasites were all actively swimming, we anecdotally observed that the parasites found in y feces swam with less vigor compared to those that were harvested from bumble bee guts. Whether this impacts likelihood of infecting new hosts is unknown and beyond the scope of our current study. However, we suggest assessing the infectivity of Eristalis-defecated C. bombi for known hosts, such as Bombus spp., O. lignaria and M. rotundata bees. More broadly, little is known regarding the infectivity of mechanically transmitted fecal-oral parasites that pass through non-host vectors.
Our ndings provide justi cation to look beyond bees to better understand the epidemiology of speciesrich pollinator communities. We show that while E. tenax ies are not hosts of C. bombi, they can defecate viable C. bombi cells. However, Eristalis ies still possess traits that facilitate the dispersal of fecal-orally transmitted "bee" parasites, by defecating frequently and in large volumes inside the corolla of owers where susceptible hosts forage for nectar and pollen which has important implications not only for Crithidia parasite transmission networks, but general plant-pollinator-parasite networks. Also, our results suggest that the vectoring potential of known hosts of C. bombi may vary between species, as M. rotundata defecated signi cantly smaller fecal volumes than O. lignaria bees. Therefore, we recommend investigating whether more non-bee pollinators that share oral resources with bees can be hosts or nonhost vectors of bee parasites so these species can be incorporated in epidemiological models of pollinator communities.

Rearing methodology
Eristalis ower ies were reared in laboratory conditions from egg clutches laid by wild-caught females in the summer of 2019 (see Supplementary Materials for detailed rearing methodology). Only ies that emerged on the same day were used in the experiments. An arti cial diapause protocol (see Supplementary Materials for detailed protocol) was used to prolong the lifespan of lab-reared ower ies, as adult Eristalis ower ies in lab colonies have shorter lifespans than adult Eristalis ies in the wild 44 . Once the adult ies eclosed, all siblings were placed in arti cial diapause in a refrigerator and fed 10% sucrose ad libitum until the experiment began. These Eristalis ower y rearing and arti cial diapause protocols are a modi cation of previously published protocols 44,45 .
Osmia lignaria (n = 50; Crown Bees, Woodinville, WA, USA) and Megachile rotundata (n = 50; Watts Solitary Bees, Bothell, WA, USA) were purchased and allowed to eclose in an incubator kept at 23 degrees C and 65% humidity. Bumble bees (Bombus impatiens) used as C. bombi source colonies or as uninfected sources of bees for the dose-response trials were purchased from Biobest (Biobest, Leamington, Ontario, Canada) and maintained in the lab by feeding sucrose and pollen from a mixture of honey bee-collected poly-oral pollen (Bee Pollen Granules, CC Pollen High Desert, Phoenix AZ, USA).

Evaluating whether the European drone y, Eristalis tenax, is a host of Crithidia bombi
After breaking arti cial diapause, the E. tenax ower ies were allowed to groom, but not feed, for one hour. Each y was then placed abdomen-rst into a 1.5 mL microcentrifuge tube harness to collect defecation events (Supplementary Figure S5). The size of these tubes allowed the ies to feed comfortably, but the tubes were also tapered at the bottom, which prevented the ies from stepping in their feces. Holes were placed along the side of the tubes so the y could respire. One large hole was placed on the lid of the tube so the y could be inoculated directly with a pipette.
Flies were randomly divided into treatment and control groups. E. tenax ies in a roughly 1:2 F:M sex ratio were used in both the treatment (n = 30) and control groups (n = 30), for a total of 60 replicates. The ies that emerged from the same egg clutch with this 1:2 F:M sex ratio were the only siblings that could accommodate the replicates needed for this experiment, which is why this sex ratio was used.
The C. bombi inoculum was made fresh from infected B. impatens (Hymenoptera: Apidae) individuals the morning of the experiment using established protocols. Brie y, we dissected the gut of infected B.
impatiens workers from a laboratory source colony that sustained a strain collected from wild B. impatiens workers from Massachusetts, USA (GPS coordinates: 42.363911 N, − 72.567747 W). We homogenized the bee guts in distilled water and diluted the mixture to 1280 C. bombi cells µl -1 , which we then combined 1:1 with 30% sucrose solution for an inoculum of 640 cells µl -1 , a standard inoculum concentration for infecting bumble bees with C. bombi 35,46 . Control groups were fed 5 µL of a 30% sucrose and blue dye (Butler Extract Co., Lancaster, PA, USA) that in pilot experiments was not found to in uence host or parasite survival. Treatment groups were inoculated with 5 µL (3,200 cells total) of C. bombi, 30% sucrose and blue dye solution. The number of cells used in the inoculum is similar to levels of C. bombi found in the feces of bumblebees with recently established infections 37 . Blue dye was used to better visualize when fecal events occurred and ies that did not drink the entire 5 µL inoculum were not used in the experiment.
After feeding, the ies were monitored continuously until defecation occurred. As these ies recently emerged from arti cial diapause and were starved pre-experiment, every hour post-inoculation the ies were fed a 30% sucrose and blue dye solution ad libitum to encourage defecation. Once a y defecated, the feces were collected via pipette and diluted to a 10 µl solution with deionized water to observe and count parasites using Kova Glasstic slides. The y was then placed in an individual 60 mL plastic portion cup with lter paper (Sigma-Aldrich, St Louis, MO, USA) and a 1.5 mL microcentrifuge tube feeder containing 500 µl of a 30% sucrose and blue dye solution for 10 days. Feeders and lter papers were replaced every three days to prevent mold growth. As C. bombi typically replicates in high numbers after 10 days in the guts of bumble bees 47 , both control and treatment ies were dissected and C. bombi gut counts were performed 10 days post-inoculation. Since actively swimming, and thus live, C. bombi is infective to susceptible bumble bee hosts 35 , only actively swimming C. bombi were counted. The fecal volume, dilution factor and counts of C. bombi were quanti ed for each individual y to calculate the exact amount of C. bombi in the individual's rst defecation event.

Dose-response data
Crithidia bombi inoculum was made from infected B. impatiens (Hymenoptera: Apidae) individuals the morning of each trial using the protocols described above, with two exceptions. First, the C. bombi strain was collected from wild B. impatiens workers from New York, USA (GPS coordinates: 42. To obtain these doses, we homogenized bee guts in distilled water and diluted the mixture to 5,000 C. bombi cells µl -1 with 30% sucrose solution. Serial dilutions were then conducted with a 10% sucrose solution to ensure the same osmolarity of each inoculum. We conducted four replicate dose-response trials over a period of four weeks. Each week, 5 uninfected workers per dose from each of two colonies were administered 5 µl of C. bombi inoculum. The ten highest doses were administered for the rst two weeks, and two additional doses (24 cells, and 12 cells) were added for the nal two weeks. Inoculated bees were kept individually in vials and fed 30% sucrose ad lib for seven days at 23 degrees C and 65% humidity. After seven days, the bees were dissected and C. bombi loads were quanti ed using a hemocytometer as described above. In addition, the right forewing was removed from each bee and marginal cell length was measured as a proxy for size 48 . In total, 220 bees were inoculated (20 replicates for each of the ten highest doses, 10 replicates for the two lowest doses).2.4. Vectoring potential of two bee species, Osmia lignaria and Megachile rotundata, and two Eristalis y species, E. arbustorum and E. tenax 2.4.1. Defecation patterns on a shared oral resource All pollinators (O. lignaria, M. rotundata, E. arbustorum and E. tenax) were placed in individual 60 mL plastic portion cups lined with lter paper. Each pollinator received a 1.5 mL microcentrifuge tube feeder containing 500 µl of uorescent dye via 2.5 g of uorescent powder (Stardust Micas) dissolved into 500 mL 30% sucrose feeders to visualize fecal deposition on owers. After 24 hours, lter papers were collected (for analysis of fecal volume and defecation frequency, see below) and a total of ve, randomly selected pollinators of the same species were placed in 12 x 12 x 12" mesh cages (Bioquip Products, Rancho Dominguez, CA, USA) containing in orescences of similar sized Solidago "Dansolitlem" hybrida, little lemon goldenrod each replicate trial. Little lemon goldenrod was used in this experiment because both bees and ower ies were observed foraging on this abundant oral resource. Only pollinators with lter papers containing uorescing defecation events were released in the mesh cages.
All E. arbustorum cages (n = 10) contained 2:3 F:M sex ratios, except one cage contained a 3:2 F:M sex ratio. All E. tenax cages (n = 20) contained 3:2 F:M sex ratios, except four cages contained 2:3 F:M sex ratios. All O. lignaria cages (n = 10) contained 4:1 F:M sex ratios, except one cage contained a 3:2 F:M sex ratio. For the two y species, sample sizes and F:M sex ratios were determined by the greatest, same-day sibling emergence. For O. lignaria, sample sizes and F:M sex ratios were determined by emergence availability. M. rotundata oral deposition data was not collected, as the F:M emergence was heavily skewed to males that did not interact with, and therefore defecate on, the owers.
After 24-hours, the pollinators were removed and the defecation events on the goldenrod from all cages were counted under a blacklight. The location of the defecation events on the goldenrod was recorded. The plant parts were divided into six categories: 'inside' the ower (inside the corolla), 'outside' the ower (surface of the corolla), on the sepal, on the bract (the lea ike structure beneath the ower), on the stem or on a leaf.

Defecation frequency and fecal volumes
The diameter of the smallest and largest defecation events per lter paper was measured by a digital caliper and an average diameter was calculated from these two values for all pollinators. The average diameter of the defecation events was converted to an average volume (in µl) using a standard curve Supplementary Figure S6). R 2 = 0.99 for the calibration data). The collected fecal volumes defecated by control ies from the E. tenax inoculation experiment (see above) were compared to the average y fecal volumes calculated here. This was done to analyze whether ies in a con ned environment, where they were inoculated with C. bombi, defecated similar volumes to ies allowed to move freely in an individual cup, which the average volumes were estimated from. In addition, the number of defecation events (frequency) over a 24-hour period on the collected E. arbustorum (n = 46) and E. tenax (n = 100) lter papers were counted for each y.

Statistical analyses
For the E. tenax inoculation experiment, we evaluated the amount of C. bombi cells in the rst defecation event using a negative binomial generalized linear model (GLM), with y sex as predictor. We chose negative binomial over Poisson to account for overdispersion, which we evaluated using Pearson residuals. Signi cance of sex was evaluated using a likelihood ratio test (LRT).
Data from the B. impatiens inoculation experiment were used to t two dose-response curves, the rst for infection probability, and the second for infection intensity among infected bees. Infection intensity was de ned using the loads estimated from the hemocytometer. A bee was considered infected if the counts were nonzero. We rst tested whether the dose ingested, wing length (as a proxy for body size) and the colony the bee came from affected its response. For infection probability, this was done using a GLM with log 10 (dose), colony, wing length and their interactions as predictors, and infection status as the Bernoulli response. For infection intensity, this was done using a linear model (LM) with the same predictors, and log 10 (intensity) as response, using only infected bees. Doses were log-transformed in accordance to how the experimental doses were varied, while intensities were log-transformed to achieve normality of the residuals. Signi cance of predictors were tested in accordance with the principle of marginality.
While we found that wing length and colony were signi cant predictors, in practice the colony-speci c response of a wild bee is unknown (since it would not have come from any of the experimental colonies), while the dependence on wing length is only useful in a size-based epidemiological model. Hence, we generated dose-response curves by marginalizing across colony and wing length. Finally, we tested whether linear relationships between the link function and log 10 (dose), assumed in LMs and GLMs, were su cient to capture the shape of the dose-response curves, by tting the data to shape-constrained additive models and then comparing AIC values 49 . SCAMs are generalized additive models (GAMs) on which additional constraints such as monotonicity have been imposed; being more exible, they can better capture the shapes of the dose-response curves should the linear relationships be inadequate.
We evaluated whether fecal volume depended on pollinator species and sex with a linear model (LM), tted using weighted least squares to account for unequal variances between group (detected using Levene's test). Since the transformation from diameter (of feces on lter paper) to volume introduced a noticeable skew to the distribution, we transformed the volume back to diameter and further performed a Box-Cox transformation to achieve normality 50 , which we veri ed using the Shapiro-Wilk and D'Agostino's K 2 test. The transformed volume was used as the response in the abovementioned linear model.
For E. tenax, fecal volume was also manually collected from the 1.5 microcentrifuge tubes during the inoculations experiment. We compared the fecal volume from the two methods using a LM with method and y sex as well as their interaction as predictors. Volumes were log-transformed to achieve normality, while the linear model was tted using ordinary least squares since Levene's test indicated no signi cant deviation from the assumption of equal variance.
We evaluated whether defecation frequency depended on pollinator species and sex with a LM, again tted using weighted least squares to account for unequal variances between groups. While two of the groups showed deviation from normality using the Shapiro-Wilk test, the deviations were only marginally signi cant and hence not expected to qualitatively affect the results 51 .
Finally, we evaluated defecation patterns on Solidago using a negative binomial GLM, with feces counts as the response, and pollinator species, plant location and their interaction as predictors. We did not use a mixed model with cage number as a random effect since there was only one count value per cage per location, so pseudo-replication was not an issue. Signi cance of predictors were evaluated using LRT in accordance to the principle of marginality 52 (i.e., main effects were tested only when their interactions were insigni cant and hence dropped). Post-hoc tests of pairwise contrasts with Tukey corrections were performed for predictors that were signi cant. We recognize that the principle sex ratio and its interactions with other predictors could also be included among the predictors; however, since each species had cages with predominantly one sex ratio (E. tenax 3F:2M; E. arbustorum 2F:3M; O. lignaria mix of 4F:1M and 5F:0M), this meant that species and sex ratios were highly correlated, making it impossible to separate their effects. Nonetheless, since female Eristalis ies do not provision their brood, the differences between sexes (e.g., time spent foraging on plants) may be less pronounced than in bees.