Pollinator‐mediated selection is better detected when controlling for resource limitation

The wide variation in floral morphology among the angiosperms is considered to be the result of their interactions with pollinators (Darwin 1862; Fægri & van der Pijl 1979). Much attention has been given to the quantification of selection exerted by pollinators on various floral traits, including flower size, colour and scent (Harder & Johnson 2009). While quantifying selection gradients (b) on a floral trait is well-established (Lande & Arnold 1983; Johnston 1991; Conner et al. 1996), proving that pollinators are the agents of selection on this trait requires experimental test, utilising pollen limitation effect on fertilisation success (Burd 1994). Supplementing pollen to flowers and comparing selection gradients of flowers open to natural pollinators with those that receive supplementary pollen (Dbpoll) enables partitioning the net effect of pollinators on traitassociated reproductive success (Sandring & Agren 2009; Parachnowitsch & Kessler 2010; Lavi & Sapir 2015). Dbpoll is the difference between regression slopes, tested by ANCOVA for a significant interaction between phenotype and pollination treatment – natural pollination or supplementary pollination. This provides quantitative estimation of the degree to which pollinators mediate selection across systems (Sletvold & Agren 2014). However, other agents, either biotic or abiotic, could affect trait-associated fitness in the opposite direction than pollinators and obscure pollinator-mediated selection (Strauss & Whittall 2006) or inflate the selection exserted by pollinators (Caruso, Remington & Ostergren 2006). Even worse, unrelated processes that affect fitness can mask pollinator-mediated selection regardless of pollinator choice of floral trait value. For example, variance in resources may affect fruit production or seed mass (Campbell & Halama 1993; Ashman et al. 2004) and result in statistical “noise” that obscures pollinator-mediated selection. Resource variability affects both plant fitness and pollination-relevant floral traits (Campbell & Halama 1993; Burd 1994; Mattila & Kuitunen 2000) and can thus be a major confound to quantifying the net effect of pollinators on phenotypic selection. Whether resources variation interacts with pollinator choice to affect fitness should be addressed in any pollinator-mediated selection field study, but it is not. A thorough literature search revealed that while numerous studies quantified the effect of pollen limitation or resource limitation on fitness, only a single study connected these effects to pollinator-mediated selection. Caruso, Remington & Ostergren (2006) showed in Asclepias syriaca that resources availability affects female fitness, and that pollinator exclusion does not, but they did it in a single-factor design, precluding testing for the interaction and leaving net pollinators’ effect not quantified. In this issue of Functional Ecology, Sletvold, Tye & Agren (2017) present a study in which they quantified the net selection mediated by pollinators while controlling for resource availability in natural population of the orchid Dactylorhiza lapponica. They quantified phenotypic selection on floral traits as a function of resources limitation, manipulated by adding NPK pellets, and as a function of pollen limitation, manipulated by supplementary hand pollination. A factorial experimental design enabled them also to test for the interactive effect of resources and pollen limitation. Interestingly, not only there was no interaction between pollen and resources variation, pollen limitation, rather than resources, was the factor that changed female fitness and provided the baseline for quantification of pollinator-mediated selection. Similar pollinator-mediated selection regimes between resources treatments suggest that resources are not a selection agent on pollination-related floral traits in this orchid. The importance of this study stretches beyond its specific results; it calls for caution in interpreting studies testing pollinator-mediated selection (including my own). Most of previous studies use female fitness, which depends on resources more than male fitness (Cohen & Dukas 1990), and it is not uncommon that pollinatormediated selection measured with maternal fitness is rather low (Harder & Johnson 2009) or non-apparent (Lavi & Sapir 2015). It is plausible that controlling for resource limitation may unmask the net effect of pollen limitation on female fitness, in relation to floral phenotype. In retrospect, perhaps pollinator-mediated selection is stronger than usually observed, countered by the contrasting effect of resources. Variance in resources should no longer be neglected in studies of pollinator-mediated selection. The study of *Correspondence author. E-mail: sapiry@post.tau.ac.il