Fig 1.
Bumblebees prefer volatiles emitted by CMV-infected tomato plants.
(A) Experimental set-up: a bumblebee feeds from a sucrose-providing cup on a container placed over a test plant. (B) In free-choice assays, the volatiles induced by CMV infection bias visitation of bees toward CMV-infected over mock-inoculated tomato plants. This is shown for flowering plants (upper panel; n = 26) infected with CMV strain Fny (CMV-Fny) and non-flowering plants (middle panel) infected with CMV-Fny (n = 26) or CMV isolate PV0187 (PV0187) (n = 24). However, bumblebees showed no preference for plants infected with a viral mutant (derived from CMV-Fny) unable to express the 2b RNA silencing suppressor protein (CMVΔ2b; n = 30 for mock vs. CMVΔ2b; n = 30 for CMVΔ2b vs. CMV). Bumblebees showed no difference in preference for Arabidopsis plants that had been mock-inoculated or infected with CMV-Fny (n = 34) (significant differences indicated; n.s., non-significant: binomial test).
Fig 2.
Bumblebees can be conditioned to disregard their innate preference for CMV-infected tomato plants.
Bumblebees can be trained by differential conditioning (using solutions of 30% sucrose as a ‘reward’ versus 0.12% quinine as ‘punishment’) to distinguish between volatiles produced by different plants. Using the example of CMV-infected versus mock-inoculated tomato (S. lycopersicum, wild-type, labeled ‘Sl WT’), bees initially have a 50% chance of making the ‘correct’ choice (sucrose: placed over mock-inoculated plants in A or over CMV-infected plants in B). Increasing success with each choice made is indicated by a rising learning curve, with overall ability to distinguish between plant-emitted volatiles analyzed after 100 choices. Where bees can perceive a difference in the volatiles emitted by two plants, they learn to identify sucrose rewards based on the association with plant volatiles, even as in Panel A when this opposes their innate preference for odor cues of CMV-infected plants (see Fig 1B). Data are shown pooled over all bees (n = 8 or n = 9) into successive groups of 10 choices, with error bars showing 95% binomial confidence intervals for the proportion of correct choices. In these experiments infected plants were inoculated with CMV strain Fny. The white curve shows the fitted binomial logistic model, with blue shading showing 95% confidence intervals on the fitted response. The χ2 statistic and p-value for the likelihood ratio test assessing whether or not bees are able to learn are given at the bottom left of each panel.
Fig 3.
Bumblebees can perceive differences in volatiles emitted by Arabidopsis plants caused by CMV infection and by mutations affecting the microRNA pathway.
Bumblebees can be trained by differential conditioning (learning curve experiments using solutions of 30% sucrose as a ‘reward’ versus 0.12% quinine as ‘punishment’) to respond to differences in volatiles produced by Arabidopsis plants (Panels A—L). Increasing success with each choice made is indicated by a rising learning curve, with overall ability to distinguish between plant-emitted volatiles analyzed after 100 choices per bee using between 8 and 10 bees as indicated in each panel (A—L). Using differential conditioning it was determined whether bumblebees could distinguish between volatiles emitted by wild-type (WT) Arabidopsis thaliana (At) plants after mock-inoculation (mock) or infection with CMV (strain Fny) (A) or CMV and the CMVΔ2b deletion mutant (B). Bees could readily learn to distinguish between volatiles emitted by transgenic plants expressing the CMV 2b protein gene under control of the cauliflower mosaic virus 35S promoter (At 35S::2b) and volatiles emitted either by non-transgenic, WT plants (C) or by control-transgenic plants expressing an untranslatable 2b gene construct (35S::unt2b) (D). Bumblebees did not learn efficiently to distinguish between volatiles emitted by mock-inoculated versus CMVΔ2b-infected wild-type plants (E), WT versus control-transgenic (35S::unt2b) plants (F), or dcl2/4 double transgenic plants that had been mock-inoculated or infected with CMVΔ2b (G). Bees rapidly learned to distinguish between volatiles emitted by WT plants and plants harbouring mutant alleles for the AGO1 (ago1-25) (H) or DCL1 (dcl1-9) genes (I). (J) Bees showed little or no ability to learn to distinguish between volatiles emitted by ago1-25 versus dcl1-9 mutant plants. (K) Bumblebees readily learned to distinguished between dcl2/4 double transgenic plants and WT plants infected with CMV. (L) Bees could not learn to distinguish between mock-inoculated WT plants. Data are shown pooled over all bees (n = 8 to 10) into successive groups of 10 choices, with error bars showing 95% binomial confidence intervals for the proportion of correct choices. The white curve shows the fitted binomial logistic model, with blue shading showing 95% confidence intervals on the fitted response. The χ2 statistic and p-value for the likelihood ratio test assessing whether or not bees are able to learn are given at the bottom left of each panel.
Fig 4.
Virus infection induced quantitative and qualitative changes in the emission of volatile organic compounds by tomato plants.
(A) Score scatter plot from principal component (PC) analysis of m/z values (binned to 1.0 Da) obtained by gas chromatography-mass spectrometry of samples of volatile organic compounds (VOCs) collected by dynamic headspace trapping from tomato plants that had been mock-inoculated (blue), infected with CMV (strain Fny) (red) or CMVΔ2b-infected (green). The analysis shows discrimination between all three treatments. The percentage of variation of the data explained by PC1 and PC2 is in parentheses (80.3 and 9.8%, respectively). (B) Whole plant total emission rate (ng.h-1) of the combined (most abundant) volatiles is similar for mock-inoculated and virus-infected plants. (C) VOC emission rate (ng.h-1) per gram dry weight of the combined (most abundant) volatiles is highest from CMV-infected plants compared to mock-inoculated and CMVΔ2b-infected plants. (D) Whole plant VOC emission rate (ng.h-1) for the five most abundant volatiles from mock-inoculated, CMV-infected and CMVΔ2b-infected tomato plants. (E) VOC emission rate (ng.h-1) per gram dry weight of the five most abundant volatiles from mock-inoculated, CMV-infected and CMVΔ2b-infected tomato plants show that pinene and cymene emission are significantly higher in CMV-infected plants compared to mock-inoculated and CMVΔ2b-infected plants. The mean VOC emission values for combined or individual volatiles are presented (n = 3 plants per treatment). Error bars represent standard error of the mean. The level of significance is shown by a p-value calculated with one-way ANOVA and post hoc Tukey HSD testing.
Fig 5.
Impacts of CMV infection and artificial buzz-pollination on tomato seed production.
(A, B) Tomato plants infected with cucumber mosaic virus (CMV) (strain Fny) produced fewer seeds (A) than mock-inoculated (Mock) plants but seed mass was not affected (B). However, buzz-pollination significantly enhanced seed production (A) but not seed mass (B). Artificial buzz pollination was achieved by touching flower stalks of matured flowers with an electrical toothbrush. This was done three times just before, during and after apparent flower maturation to ensure efficient buzz-pollination. Successful buzz-pollination was noted by observing pollen release from the anther cone. Letters indicate significant differences. A) Mean seed number per fruit (two-way ANOVA: infection status, F(1,10) = 220.9938, p = 3.811e-08; pollination treatment, F(1,10) = 61.5886, p = 1.393e-05; infection status x pollination treatment, F(1,10) = 0.4701, p = 0.5085). B) Mean mass per seed (two-way ANOVA: infection status, F(1,8) = 0.9291, p = 0.3633; pollination treatment, F(1,8) = 0.0030, p = 0.9577; infection status x pollination treatment, F(1,8) = 0.0825, p = 0.7812). Error bars represent the standard error of the mean; n = 3 plants per experiment.
Fig 6.
Bumblebee-mediated pollination of CMV-infected and mock-inoculated tomato plants.
Within a flight arena under glasshouse conditions bumblebees were allowed to forage directly on flowering plants that had been mock-inoculated or infected with CMV (strain PV0187) and fruits were allowed to develop on these plants and later harvested. Fruits were categorized according to whether they were derived from flowers that had not been buzz-pollinated by a bumblebee (fruit from flowers not visited by bee) or from flowers that had been buzz-pollinated (fruit from bee-pollinated flowers). A further category of fruit was from flowers that had not been buzz-pollinated, but had been adjacent to buzz-pollinated flowers (fruit from flowers adjacent to bee-pollinated flowers). Fruits were also harvested from 8 mock-inoculated and 8 CMV-infected plants that had been placed in the flight arena and had otherwise experienced the same growth conditions but were never exposed to bees (fruit from untouched plants). Numbers of seeds per fruit differed between treatments (one-way ANOVA, F(7,459) = 12.34, p<10−13). In control plants not exposed to bumblebees, CMV significantly lowered the number of seeds per fruit by over 30% (p = 0.013 post-hoc Tukey test). Natural buzz-pollination by bumblebees raised the seed number in fruit from both mock-inoculated and CMV-infected plants and remarkably had a more beneficial effect on CMV-infected plants in that the seed yield per fruit matched that of the mock-inoculated plants. Different letters (A, B, or C) are assigned to significantly different results (post-hoc Tukey tests, p< 0.05, *, p< 0.01, **). Histogram bar labeling: n = number of plants, number of fruits. Error bars are standard error around the mean seed number.
Fig 7.
Persistence of genes for virus susceptibility depends on the balance between positive and negative effects of infection on reproduction.
(A) and (B). Whether or not susceptible genotype plants take over (yellow), coexist with resistant plants (green) or are eliminated in favor of resistance (blue), depends on the balance of the pollinator bias to infected plants (ν), the reduction in seed set by infected plants (δ), and the mean number of pollinator visits per flower (γ). Other parameters are fixed at default values: proportion of flowers that self-fertilize without being visited by a pollinator (σ = 0.25), probability that flowers that are visited by a pollinator are cross-pollinated (φ = 0.75) and the proportion of virus-susceptible plants that are infected (α = 0.75). (C) For fixed pollinator bias (ν = 3) and reduction in fertility of infected plants (δ = 0.5), the long-term genotypic structure of the plant population depends on the number of visits by pollinators (γ). (D) The exact form of the response to γ depends on the proportion of susceptible plants that become virus infected (α).
Fig 8.
Virus susceptible plants persist across a broad range of parameter values.
A full two-way sensitivity analysis of the model, showing the effect of independently changing pairs of parameters (all other parameters fixed). The virus susceptible genotype takes over (yellow) or co-exists with the resistant genotype (green) across a large proportion of parameter space. All pair-wise combinations of two parameters are shown: dots on each axis show default values of each parameter.
Fig 9.
Hypothesis: Pollinator preference for virus-infected plants could provide a payback to virus-susceptible hosts.
Under experimental conditions, bumblebees showed an innate preference for volatiles emitted by tomato plants infected with CMV (upper section of cartoon). We speculate that if similar phenomena occur under natural conditions in wild plant populations, this may pay back susceptible host plants by encouraging pollinator visitation. Mathematical modeling suggests that under some conditions this may result in increased production of virus-susceptible offspring and if pollinator preference for infected susceptible plants was sufficiently strong, this could outweigh underlying strong selection pressures favoring the emergence of virus resistance.