Frequency-dependent selection by wild birds promotes polymorphism in model salamanders

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.

We observed at least five individual Blue Jays (Cyanocitta cristata) foraging on model salamanders. Other birds observed foraging on the ground in the study site included Northern Cardinals (Cardinalis cardinalis), American Crows (Corvus brachyrhynchos), American Robins (Turdus migratorius), and Eastern Towhees (Pipilio erythrophthalmus). Only Blue Jays were observed taking peanuts from salamanders.

Fig. 3 shows that striped models had higher relative fitness when at low frequency and lower relative fitness when at high frequency. Multiple logistic regression supported a significant relationship between frequency and survival on non-test days (likelihood ratio G² = 6.37, df = 1, P = 0.0116). There was also a strong effect of "day" (G² = 58.27, df = 12, P < 0.0001), which probably reflects variation in predator activity (predation was lighter on rainy days) and a delay in learning. There was no detectable association between survival and morph (likelihood ratio G² = 0.225, df = 1, P = 0.635). Whichever form was rare had a survival advantage within one or two days after the relative abundances were reversed (Fig. 3). Thus, when wild birds were "trained" to expect one form to be most common, they preyed disproportionately on that form. This occurred regardless of which form was common.

Selection by birds can be heterogeneous and these results should be tested at other sites. However, decades of research have consistently found that captive and free-ranging wild birds cause frequency-dependent selection on all manner of food items differing in appearance [4, 6, 7, 25, 26, 28]. We show that this very general process can operate on the kind of phenotypic polymorphism seen commonly in small salamanders.

We have not tested the perceptual and behavioural mechanisms underlying frequency-dependent foraging, therefore we have avoided invoking search image formation as the specific ultimate cause of rare form advantage. Nevertheless, this study joins a growing body of literature showing that birds often switch foraging behaviour in response to the relative abundance of alternative cryptic prey and this can promote the maintenance of colour polymorphism [4, 8]. The cognitive processes underlying this phenomenon are important subjects for further research.

In addition, a crucial question for the maintenance of polymorphism is to what extent frequency-dependent foraging by visual predators is counteracted by other selective processes, including non-visual predation. Other sources of mortality and other phenotypic differences between morphs might interfere with or enhance the stable coexistence of distinct colour morphs. The conceptual link between promoting polymorphism and promoting coexistence of similar species [37] merits greater attention.

Rare form advantage owing to frequency-dependent selection by visual predators has been proposed to explain maintenance of polymorphism in many cryptic species [4]. The experiment reported here provides evidence that the key mechanism (frequency-dependent foraging) operates with realistic prey at realistic densities in the wild. The maintenance of colour variation in terrestrial salamanders might be explained by the oldest and most obvious hypothesis – rare form advantage arises because predators tend to overlook rare prey.

Model salamanders were constructed with modelling clay (Sculpey™). One model was sculpted by hand and used to create a flexible mould. This mould was then used to make models of consistent size and body position. Models were either grey-brown or grey-brown with an ochre stripe from the base of the head to the base of the tail (Fig. 2a). A food reward was associated with each model salamander by adhering one peanut half to the bottom of the model using edible library paste (40 g corn starch, 1 L water, boiled and cooled).

The field site was an open area adjacent to a woodlot in Knox County, TN. Polymorphic P. ventralis are known from within 2 km. To avoid spurious loss of models, we used as foraging units 100 black plastic trays (0.185 m²) placed in a 10 × 10 array and filled with partially decomposed leaf-litter (natural forest floor substrate). One m of open lawn was left between adjacent trays. Each day, we used a random number generator to assign model salamanders to 50 trays. The ordering of striped and unstriped models was also random; as a result, each tray position had a 50% chance of having a model salamander each day and the probability that the model was striped was the frequency of striped models for that day. The resulting density of 1/2 m^-2 was realistic for small Plethodon and Batrachoseps salamanders [16]. Models were placed haphazardly within the trays and arranged so that the attached peanut could not be seen without moving the model. This design ensured that foraging birds had the opportunity to learn where to look for salamanders and that all "surviving" salamanders would be recovered at the end of each day. In addition, it was important to make sure that each tray was not guaranteed to provide a salamander, so that sometimes giving up and moving to another was the right foraging decision [27].

The experiment was divided into two training periods of six days each and two test days, similar to protocols used with green and brown pastry pellets[7, 25]. First, 5 striped and 45 unstriped salamander models were presented each day from 8:00 am until 7:30 pm (it usually took this entire time for the birds to attack roughly half of the salamander models). At the end of the day, the flats were gathered and the number of "surviving" salamanders counted (those with the peanut still attached). Six days of training were followed immediately by a test day in which 25 of each morph were presented. After the first test day, another six-day training period was started, this time with frequencies of 45 striped and 5 unstriped salamander models. The second test day (25 of each morph) immediately followed the second training period (Table 1).

We used 2 × 2 contingency tables to test the null hypothesis of equal predation rates for striped and unstriped models on the two test days. For the 12 "training" days, we tested for an effect of relative abundance on survival using logistic regression with individual survival as the response variable, day as a random effect, and morph and frequency category as fixed effects. The goodness-of-fit of the model was adequate (likelihood ratio test relative to a saturated model: G² = 4.03, df = 11, P = 0.708) and adding interaction terms did not significantly improve the fit. For all 14 days of the experiment, we estimated fitness (survival value) of striped relative to unstriped models as w _S = [(S _A + 0.5)/(S _B + 0.5)] ÷ [(U _A + 0.5)/(U _B + 0.5)], where S _A and U _A are the numbers of surviving striped and unstriped models, and S _B and U _B are the numbers before selection. This estimates the ratio of the survival probability of striped relative to unstriped models [38]. For example, when the two forms have equal fitness, w _S = 1.0 and when a striped individual is twice as likely to survive as an unstriped individual, w _S = 2.0.