Maintenance of variation is a classic paradox in evolution because both selection and drift tend to remove variation from populations [1–3]. If one form has an advantage (e.g., being more cryptic), it should replace all others. Likewise, random drift alone will eventually result in loss of all but one form when there are no fitness differences. Maintenance of a stable polymorphism requires either recurrent mutation, a balance between dispersal and divergent selection between populations, or some form of balancing selection within populations [3]. For example, predators are expected to form a search image for the most common form if it is easier to search for one cryptic prey type than to simultaneously look for two [4–6]. If predators switch their search images to whichever prey is most common, the result is frequency-dependent selection against common prey [4–7]. Empirical tests of this frequency-dependent selection hypothesis are rare, and the link between predator behaviour and maintenance of variation in prey has been difficult to confirm [4, 8]. Here, we test whether predatory birds can act as agents of frequency-dependent selection on terrestrial salamanders using a manipulative field experiment.
Multiple species of small, forest floor-dwelling salamanders exhibit polymorphism for the presence of a dorsal stripe (Fig. 1) and what maintains this polymorphism within populations is a long-standing question [9–15]. Polymorphism is widespread both geographically and phylogenetically, occurring in many species of Plethodon throughout North America in addition to the distantly related Batrachoseps of the Pacific coast of North America [16] and Karsenia of Asia [17]. Thus, it is unlikely that this polymorphism is merely a transitional stage in the evolution of colour pattern. Presence or absence of a dorsal stripe appears to have a simple genetic basis in Plethodon cinereus and is not related to sex [12, 18].
What polymorphic salamander species seem to have in common is that they are small, slender, and locally abundant. These species often comprise a substantial fraction of the animal biomass in temperate forests and are subject to predation by ground-foraging birds and other predators that search the leaf-litter for small animals [16]. Studies of polymorphic Plethodon cinereus have shown differences between the morphs in behaviour [11, 14, 15, 19], physiology [11, 13], and geographic variation in relative abundance [10, 20]. However, none directly address how these differences might relate to a mechanism maintaining variation and there is no evidence of similar phenotypic correlations in the other polymorphic species. In fact, in P. ventralis, the frequency of the striped form is less at higher elevations [21] whereas the frequency of striped P. cinereus is greater at higher latitude and cooler microclimates in general [10, 20]. Thus, phenotypic correlations between appearance and physiology are not consistent (also see [13]) and this is not surprising given that dorsal colour pattern is unlikely to have a direct function in temperature tolerance or metabolism (Plethodon always avoid sunlight).
To explain the maintenance of a similar visual polymorphism across many taxa, we propose that selection acts directly on appearance. The most likely function of dorsal colour pattern is crypsis (Fig. 1), though this is an untested assumption. Though there has been speculation that the red stripe of P. cinereus might be an aposematic signal [19], Brodie and Brodie [22] showed that wild birds prey extensively upon striped P. cinereus, and give no sign that the salamanders are unpalatable. Striped P. cinereus were taken just as often as D. ochrophaeus, which were assumed to be cryptic and palatable [22]. A third, completely red form of P. cinereus does appear to be a colour mimic of the toxic Red-spotted Newt eft (Notophthalmus viridescens), but our focus is on the more widespread stripe/no-stripe polymorphism. Polymorphism in cryptic prey populations might be maintained in migration-selection balance if alternative colour forms are adapted to different habitats [23], or might be promoted by frequency-dependent predation causing rare form advantage within populations [4, 8].
Empirical tests of frequency-dependent foraging on cryptic prey are rare; the best-studied examples involve either colored pellets of dough fed to free-ranging birds [7, 24, 25]or digital images of moths selected by trained Blue Jays (Cyanocitta cristata) [4, 26–28]. These studies demonstrate that birds can generate frequency-dependent selection, that individual Blue Jays learn to focus on abundant prey, and that they readily switch focus in response to changes in prey abundance. Apostatic selection by free-ranging wild birds has been shown for small pastries differing in colour [7, 25, 29], and presence or absence of a stripe [30, 31], and also for real Cepaea snail shells differing in banding pattern [32]. However, in other studies, selection changed from negative frequency-dependence (rare form advantage) to positive frequency-dependence (common form advantage) when artificial prey were provided at high densities, showing that rare form advantage is not entirely general and may depend on abundance or aggregation of prey [25, 33]. A recent study of Trinidadian guppies implicated frequency-dependent survival in the maintenance of variation in conspicuously colored males [34]. However, it is not clear whether search images are important for predators of conspicuous prey, and other factors might affect polymorphism in sexually-selected traits [35, 36]. Thus, whether frequency-dependent foraging can explain the maintenance of polymorphism in realistic prey at realistic densities remains a critical question that might need to be addressed on a case-by-case basis. Here, we show empirically that predatory birds can act as agents of frequency-dependent selection on terrestrial salamanders.
To test for frequency-dependent selection by birds, we used standardized models resembling polymorphic Plethodon salamanders (Fig. 2). A food reward (1/2 peanut) was glued to the underside of each model, and models were set out at random in 10 × 10 m plot at the edge of a woodlot in Knox county, Tennessee (see Methods). We manipulated the relative abundance of striped and unstriped prey and quantified survival by counting the number of models with peanuts still attached at the end of each day.