DISCUSSION
The findings of this study make a significant contribution to the understanding of plant-herbivore-pollinator systems in demonstrating, firstly, the relative importance of pollinators as selective agents on both plant-attractive and defence traits, which here even surpassed that of herbivores; and secondly, the extent to which net selective outcomes were dynamic, and context-specific.
Herbivores have long been considered the primary drivers of defence trait evolution in plant populations (Johnson et al. 2015). However, in line with our predictions, pollinators were found to impose positive selection on direct defence-related traits in one instance (for dihydroxybenzoic acid). Although negative selection was also surprisingly apparent (for shikimic acid) (Fig. 1). Dihydroxybenzoic acid is a phenolic that is strongly associated with resistance against the strawberry leaf beetle (Weber et al. 2020a). Given that pollinators were previously found to avoid damaged flowers of woodland strawberry (Muola et al. 2017), it appears logical that pollinator selection on this compound was exerted only in the presence of herbivores, and not in their absence. Hence the suggested link is that when herbivores are present, plants with higher direct defences receive less damage and are preferred by pollinators.
While this mechanism is intuitive for dihydroxybenzoic acid, the opposite pattern was found for shikimic acid – where pollinators selected against this defensive compound in the presence of herbivores. Shikimic acid is probably a common constituent of floral nectar (Hölscher et al. 2008), and like other nectar secondary compounds may be rapidly upregulated in response to herbivore attack (Adleret al. 2006; Kaczorowski et al. 2014). Thus, one explanation is that herbivore-induced changes in this compound in nectar (or other correlated derivatives of the shikimic acid pathway) could have led to gustatory deterrence of pollinators, similar to other nectar phenolics (see Stevenson et al. , 2017 and references therein). This explanation is also consistent with our finding that pollinator-mediated selection on shikimic acid was diffuse, and disappeared in the absence of herbivores (Fig. 1). However, regardless of the underlying mechanisms, these findings nonetheless establish the capacity of pollinators to impose both positive and negative selection on direct defence traits of relevance to herbivores.
Indirect defence-related traits were in contrast selected on by both agents (Fig. 1). Herbivores selected for higher levels of dehydroascorbic acid, meaning that plants with lower levels of this compound suffered greater fitness damage by herbivores. This carbohydrate is strongly associated with indirect defence in woodland strawberry owing to its positive link with parasitoid development in the herbivore host (Weber et al. 2020a). Why herbivores appeared to prefer plants with lower levels of dehydroascorbic acid could relate to a natural deterrence effect of this compound (Felton & Summers 1993), or innate avoidance as a behavioural adaptation against parasitism (as per the concept of ‘enemy free space’ (Stamp 2001)). Pollinators made only a minor contribution to selection on this compound (Fig. 1), but did clearly positively select on dihydroxybenzoic acid; a phenolic that – in addition to serving as a direct defence, as discussed – is associated with strawberry leaf beetle vulnerability to parasitism (Weber et al. 2020a). This selection was most likely imposed via herbivore deterrence effects (Muola et al. 2017). These findings build on past work by Kessler and Halitschke (2009) indicating potential pollinator selection on volatile signals associated with indirect defence, and establish the potential for pollinators to also mediate selection (coincidentally or otherwise) on non-volatile chemical traits linked to tritrophic interactions.
Of the four plant attractive traits examined in this study, significant herbivore- and pollinator-mediated selection was observed only for inflorescence density (Fig. 1). Selection on this trait was both diffuse and conflicting; a pattern that is likely to arise when a trait is shared as a positive host-plant selection cue by both pollinators and herbivores (Knauer & Schiestl 2017; Ramos & Schiestl 2019). Use of this cue by the strawberry leaf beetle and another present herbivore (see Methods) appears logical given that both are florivores, and that the former is thought to occupy dense inflorescences as a means of enemy escape in its primary host plant, meadowsweet (Filipendula ulmaria (L.) Maxim.) (Stenberg 2012). Shared use of this cue would hence explain why this trait was negatively selected against by herbivores only when the influence of pollinators was controlled for, and positively selected for by pollinators when herbivores were absent. Hence the combined effect of both agents was for intermediate selection, which was neither significantly negative nor positive overall (Fig. 1), nor appeared to contribute to the ‘total’ section acting on this trait (Table S2).
As a whole, this study permits greater insight into the eco-evolutionary dynamics of plant-herbivore-pollinator interactions, and how multiple selective forces can act to shape the microevolution of traits that link pollination and herbivory. Our findings suggest that dynamic patterns of selection may be a common feature of these tripartite interactions, given that all significant selection on traits was notably context-specific. Diffuse interactions between pollinators and herbivores may thereby explain why the phenotypic optima of plant attractive and defence-related traits could be expected to fluctuate in populations across time, in accordance with changing biotic interaction strengths. Furthermore, our study demonstrates the highly significant role that pollinators can play in selecting for increased direct and indirect defence-related traits. While the evolution of increased selfing has been proposed as one way for plants to overcome herbivore-induced pollinator limitation (Kessler & Halitschke 2009; Adler et al. 2012; Johnson et al. 2015), pollinator selection for increased anti-herbivore defences could offer another potentially commonplace route (Egan 2015), and one that does not risk the disadvantages of increased inbreeding. However, ours and a previous study (Egan et al. 2016) also show that when defence-related traits themselves are associated with pollinator limitation, then negative selection can also be expected. Together these findings highlight the complexity of potential selection pressures that can act on plant attractive and defence-related traits. The predictions generated from this study can hence serve as valuable hypotheses to test in future studies in wild plant populations.
ACKNOWLEDGEMENTS We wish to thank Sara Janbrink for technical assistance in the field. This study was funded by The Swedish Research Council Formas (Grant Nos. 2018-01036 and 2016-00223) and the SLU Centre for Biological Control.