THE GEOGRAPHIC SELECTION MOSAIC FOR PONDEROSA PINE AND CROSSBILLS: A TALE OF TWO SQUIRRELS

Authors


Abstract

Recent research demonstrates how the occurrence of a preemptive competitor (Tamiasciurus) gives rise to a geographic mosaic of coevolution for crossbills (Loxia) and conifers. We extend these studies by examining ponderosa pine (Pinus ponderosa), which produces more variable annual seed crops than the conifers in previous studies and often cooccurs with tree squirrels in the genus Sciurus that are less specialized than Tamiasciurus on conifer seed. We found no evidence of seed defenses evolving in response to selection exerted by S. aberti, which was apparently overwhelmed by selection resulting from inner bark feeding that caused many developing cones to be destroyed. In the absence of S. aberti, defenses directed at crossbills increased, favoring larger-billed crossbills and causing stronger reciprocal selection between crossbills and ponderosa pine. However, crossbill nomadism in response to cone crop fluctuations prevents localized reciprocal adaptation by crossbills. In contrast, evolution in response to S. griseus has incidentally defended cones against crossbills, limiting the geographic range of the interaction between crossbills and ponderosa pine. Our results suggest that annual resource variation does not prevent competitors from shaping selection mosaics, although such fluctuations likely prevent fine-scale geographic differentiation in predators that are nomadic in response to resource variability.

Geographical variation in community context and species interactions is increasingly recognized as an important aspect of evolution. Such variation forms the basis of the geographic mosaic theory of coevolution (Thompson 2005), which posits that selection mosaics form as a result of geographic variation in the form of species interactions, that coevolution occurs in some locales (coevolutionary hot spots) but not in others (coevolutionary cold spots), and that trait remixing resulting from processes such as gene flow and extinction/recolonization dynamics continually alters patterns of variation across the landscape. Consistent with the predictions of the theory, geographic variation in the form of coevolutionary selection and in the traits shaped by interactions has been documented in a wide variety of systems, including Greya moths and the Lithophragma plants they pollinate and oviposit in (Thompson and Cunningham 2002), garter snakes (Thamnophis) preying on toxic newts (Taricha) (Brodie et al. 2002), and crossbill finches (Loxia) preying on the seeds of lodgepole pine (Pinus contorta latifolia) (Benkman et al. 2003). Empirical evidence also exists for trait remixing in coevolving interactions (Dybdahl and Lively 1996; Burdon and Thrall 1999), and a growing body of theoretical work has begun to address how variable patterns of gene flow and selection should affect geographically structured species interactions and coevolution (e.g., Nuismer et al. 1999; Gomulkiewicz et al. 2000). However, few empirical studies have explored how variation in basic ecological factors such as resource variability, the degree of resource specialization, and vagility of interacting taxa may influence the formation of geographic selection mosaics (but see Forde et al. 2004). Such studies should enhance our understanding of the general conditions contributing to the formation of geographic selection mosaics and provide an empirical basis for stimulating further theoretical development of the geographic mosaic theory of coevolution. Advances here will most likely come from studies of well-characterized systems in which the traits mediating and responding to reciprocal natural selection are well understood.

Crossbills and the conifers on which they specialize represent such a system. Crossbills have evolved a mandible crossing as an adaptation for spreading apart the scales of conifer cones and feeding on the underlying seeds (Fig. 1; Benkman and Lindholm 1991), and have diversified into an array of specialists on different conifer species (Benkman 1993, 2003). Recent studies provide evidence of coevolution between red crossbills (L. curvirostra complex) and both lodgepole pine (Benkman 1999; Benkman et al. 2001, 2003) and black spruce (Picea mariana) (Parchman and Benkman 2002), and suggest that geographic mosaics of coevolution have had an important role in the adaptive radiation of North American crossbills. For example, throughout the Rocky Mountains red squirrels (Tamiasciurus hudsonicus; hereafter Tamiasciurus) are a dominant competitor for and predator of the seeds in lodgepole pine, which has evolved defenses in response to this predation (Smith 1970; Benkman 1999; Benkman et al. 2001, 2003). In isolated forest islands in which Tamiasciurus are absent, crossbills have become resident, occur in higher abundances than in areas with Tamiasciurus, and have coevolved with lodgepole pine over the last 10,000 years in predator–prey arms races (Benkman 1999; Benkman et al. 2001, 2003). Geographic variation in these interactions has caused divergent selection on crossbills and pines, and the evolution of distinct bill and cone morphologies in coevolutionary hot spots and cold spots (Benkman 1999; Benkman et al. 2001, 2003; Edelaar and Benkman 2006). Furthermore, patterns of bill and cone trait evolution are strikingly replicated among areas of lodgepole pine and black spruce with and without Tamiasciurus (Benkman 1999; Parchman and Benkman 2002). The replicated outcome of these interactions indicates that selection mosaics arising from geographic variation in community context contribute to the evolution of crossbill diversity, and that the traits mediating and responding to selection in these interactions often evolve in a consistent and predictable manner.

Figure 1.

The distribution of ponderosa pine, including western ponderosa pine (P. p. ponderosa) and Rocky Mountain ponderosa pine (P. p. scopulorum) (representative cones drawn to scale on left and right; from Sudworth [1967] and Sudworth [1917], respectively), with the dashed line representing the boundary between the two subspecies. A drawing of an Abert's squirrel (S. aberti) in the lower right-hand corner feeding on the inner bark of a ponderosa pine twig, and a drawing of a gray squirrel (S. griseus) on the left eating a seed it removed from a cone. The illustration in the upper right-hand corner represents a crossbill laterally abducting its mandibles to spread apart cone scales to expose the underlying seeds (hatched ovals; from Benkman 1987b). The thick solid lines outline the distributions of gray squirrels (left) and Abert's squirrels (right). Study sites are abbreviated as follows: Lassen National Forest, CA (LP), Forest Hill, CA (FH), Dixie National Forest, UT (D), Fishlake National Forest, UT (F), Ashley National Forest, UT (A), South Fork, CO (SF), Nederland, CO (N), Twin Lakes, CO (TL), the Black Hills, SD (BH), Apache-Sitgraeves National Forest, AZ (AS), and Coconino National Forest, AZ (C).

Lodgepole pine and black spruce produce cone crops of similar size from year to year (coefficients of variation of annual cone production [CV]: 76% and 45%, respectively; Kelly and Sork 2002) and have serotinous or semiserotinous cones that hold seeds for extended periods (>20 years; Burns and Honkala 1990), smoothing annual resource variability. Consequently, territorial or sedentary predators can maintain high densities, specialize on cone seed, and exert strong and consistent selection on cone traits (Smith 1970; Benkman et al. 2003). This resource stability is why Tamiasciurus are such strong preemptive competitors with crossbills, and also why, in the absence of Tamiasciurus, crossbills are abundant, resident, and exert consistent selection (Benkman 1999; Benkman et al. 2003; Siepielski and Benkman 2005). In contrast, most conifers specialized on by crossbills produce variable cone crops from year to year, and release all their seeds within the first year (Benkman 1993), resulting in far greater yearly fluctuations in seed availability. Such variation in annual resource availability should limit the competitive impacts of relatively sedentary tree squirrels, as has been argued for Sciurus relative to crossbills in Europe (Summers and Proctor 1999). However, the extent to which high temporal resource variability and the presence and absence of tree squirrels affect the formation of selection mosaics for crossbill–conifer interactions is unknown.

Here we focus on two subspecies of ponderosa pine (Pinus ponderosa scopulorum and P. p. ponderosa, hereafter Rocky Mountain ponderosa pine and western ponderosa pine, respectively) that produce highly variable annual seed crops (CV: 137% and 163%, respectively; Kelly and Sork 2002) and cooccur with two predispersal seed predators with nonoverlapping distributions in the genus Sciurus, Abert's squirrel (S. aberti) and western gray squirrel (S. griseus), respectively (Fig. 1). Rocky Mountain ponderosa pine occurs in much of the Rocky Mountain and southwestern regions of the United States, whereas western ponderosa pine occurs from southern California throughout the Sierra Nevada and north into western Canada (Fig. 1). One of the largest call types of red crossbill in North America [call type 2 of Groth (1993)] has a bill specialized for removing and handling seeds from ponderosa pine cones, in particular Rocky Mountain ponderosa pine (Benkman 1993, 2003; Benkman et al. 2001). Call type 2 crossbills (hereafter ponderosa pine crossbills) are highly nomadic, engaging in eruptive movements tracking seed crops throughout western North America. Because they can track resource fluctuations (Newton 1972; Benkman 1987a), crossbills may exert more consistent selection on cone structure, although population movements and gene flow over large areas may impede local adaptation by crossbills to geographic variation in ponderosa pine. On the other hand, Sciurus are less specialized on cone seed and harvest substantially fewer cones than do Tamiasciurus (Smith and Balda 1979), and thus may be weaker competitors and selective agents than Tamiasciurus. However, Tamiasciurus are absent throughout Eurasia and in parts of the New World, so that in many areas only Sciurus are present. Thus, studies including Sciurus are necessary if we are to understand the general importance of tree squirrels in the evolution of conifers and the adaptive radiation of crossbills.

Abert's squirrels occur in Rocky Mountain ponderosa pine forests of the Southwest and southern Rocky Mountains (Fig. 1), and specialize on this pine for almost all of their resource requirements (Keith 1965; Snyder and Linhart 1994). In contrast to other tree squirrels, Abert's squirrels are highly specialized for feeding on the phloem tissue of inner bark (Murphy and Linhart 1999), and their cutting of twigs with developing cones (Fig. 1) can decrease reproductive output by up to 80% (Snyder 1993). For example, Abert's squirrels in the Coconino National Forest, AZ (C in Fig. 1) destroyed about five times more cones from inner bark feeding (mean number of immature conelets destroyed per tree = 12.3) than by harvesting mature cones to feed on seeds (mean number of cones harvested per tree = 2.3, n= 869 trees sampled in October 1988; S. W. Allred, unpubl. data). Because Abert's squirrels selectively forage on phloem in relation to resin characteristics of individual trees (Snyder 1992), this could overwhelm selection on structural cone and seed traits. For example, in the above dataset there was no correlation among trees between the number of conelets on twigs cut for inner bark feeding and the number of mature cones harvested (rs= 0.023, P= 0.49). Consequently, Abert's squirrels are unlikely to have an important selective impact on cone structure evolution, but are likely to have a competitive effect on crossbills.

Unlike Abert's squirrels, western gray squirrels (hereafter gray squirrels) are not specialized for feeding on the inner bark of ponderosa pine and do not clip twigs for feeding (Fig. 1; Steinecker and Browning 1970). Gray squirrels occur in both oak (Quercus) and conifer woodlands, and, in coniferous forests, pine seeds make up a large percentage of their diet and are a critical food source prior to the over-wintering period (Steinecker and Browning 1970). Consequently, gray squirrels may be an important selective agent on the cone structure of western ponderosa pine.

Here we test two hypotheses about how tree squirrels in the genus Sciurus may affect the formation of a geographic selection mosaic for ponderosa pine through competitive and evolutionary effects, and examine how this structures geographic variation in the interactions between crossbills and ponderosa pine. Such investigations of the mechanisms by which different species affect the formation of selection mosaics are particularly valuable because research to date indicates that geographic selection mosaics often arise because of variation in community context. We first test the hypothesis that Abert's squirrels act as a competitor with crossbills, but not as an important selective agent on the structure of ponderosa pine cones. We predict that differences in cone traits of Rocky Mountain ponderosa pine between areas with and without Abert's squirrels will be consistent with variation in selection by crossbills, but not with variation in selection by Abert's squirrels. This would be supported if cone traits that are known to evolve in response to crossbill predation, such as scale thickness (Benkman et al. 2003; Parchman et al. 2007), are elevated in areas without Abert's squirrels. Cone and seed traits of conifers generally have high heritabilities (e.g., Verheggen and Farmer 1983), so responses to selection are expected. Second, we test the hypothesis that gray squirrels have had an evolutionary effect on ponderosa pine. Because selection by tree squirrels on large-coned pines tends to favor the evolution of even larger cones (Mezquida and Benkman 2005), we predict that selection by gray squirrels will also favor larger cones. Finally, we use foraging experiments with captive crossbills to examine the consequences of geographic variation in cone structure for crossbill feeding propensity and geographically variable selection on crossbills.

Methods

PHENOTYPIC SELECTION EXERTED BY ABERT'S SQUIRRELS AND CROSSBILLS

We estimated selection exerted by Abert's squirrels on Rocky Mountain ponderosa pine by quantifying cone predation and sampling cones from 67 mature trees in the Apache-Sitgreaves National Forest on the Mogollon Rim, AZ (AS in Fig. 1) between 1 and 7 October 2004. Trees were sampled haphazardly under the constraint that cones could be reached and cut with a 9-m pruning tool. Sciurus bite off cone scales to access seeds and drop cone cores to the base of the tree (Fig. 1), so that the presence of cores reliably indicates seed predation by Sciurus (Moller 1983). Although Tamiasciurus forage on cones in a similar manner, Tamiasciurus were not observed nor are they typically found in monotypic ponderosa pine stands such as those sampled in this study. We recorded the number of fresh cone cores at the base of each tree (mean = 15.6; range: 3–94) and the number of cones remaining on the tree (mean = 163.5; range: 70–340). Counts were converted to percent unpredated cones (number of unpredated cones divided by the sum of unpredated and predated cones), and percent unpredated cones was converted to relative fitness (percent unpredated cones divided by overall mean percent, assuming tree fitness is inversely related to the percent of cones predated by Abert's squirrels [see Siepielski and Benkman (2007a) for justification of using proportion of seed crop harvested as a measure of fitness in long-lived trees]).

The following measurements were made on two unpredated cones from each tree: cone length, cone width at the widest point, cone mass, number of scales crossed by a transect along the central axis of the cone, thickness of five scales and length of three scales (measured from the scale tip to the distal end of the seed scars) from the middle third of the cone, number of full seeds, number of empty seeds (seed coat developed but lacking kernel), and mass of five seeds filled with kernel (female gametophyte and embryo). Length was measured to the nearest 0.01 mm with digital calipers, and mass was measured for oven-dried (60°C for > 48 h) cones and seeds to the nearest 0.1 mg. The seed mass to cone mass ratio was calculated as the product of mean individual seed mass multiplied by the number of full seeds divided by cone mass. The mean values of the cone traits measured for each tree were used in analyses because trees were the experimental units.

We used least squares linear regression analyses between relative tree fitness and various cone and seed traits, each standardized to zero mean and unit variance, to estimate regression coefficients and to examine which traits were under selection due to predation by Abert's squirrels (Lande and Arnold 1983). Because residuals from these analyses were nonnormally distributed, we used the regression bootstrap technique of Efron and Tibshirani (1993) to test for significance of regression coefficients and based these tests on 1000 replicates. We checked for nonlinear selection in all traits by examining regression models with quadratic terms, and by using cubic splines (Schluter 1988) to visualize the form of selection.

Selection exerted by crossbills on Rocky Mountain ponderosa pine was measured in the Black Hills, SD (BH in Fig. 1) between 21 and 24 September 2003. Forty-five mature trees were sampled as described previously. To estimate predation, seven branches were haphazardly chosen in the crown of these trees and both the number of cones predated (mean = 2.8; range: 0–16) and not predated by crossbills (mean = 15.7; range: 11–20) was recorded with the aid of binoculars. Predated cones had scales unambiguously shredded and peeled apart by crossbills, whereas those not predated were brownish-green and still closed. Percent of cones that were unpredated was then converted to relative fitness as done above. We measured the same suite of cone and seed traits as described above for two unpredated cones from each tree, and used the means for each tree in subsequent analyses. We used the same methods and types of regressions described above for estimating selection.

We used foraging trials with captive crossbills to further determine what traits may be under selection by crossbills. Captive crossbills were housed in six aviaries (3 m × 1 m × 2 m) each containing 7–10 South Hills crossbills (Benkman 1999) in the Animal Care Facility at the University of Wyoming. Using South Hills crossbills should not affect our results because they are the most morphologically similar crossbill to ponderosa pine crossbills (Benkman 1999), and because of the highly stereotypical foraging behavior of crossbills. Prior to trials, other food sources were removed, but grit, charcoal, and water were available. Four cones were secured to dowels in each aviary each morning. Two were from Rocky Mountain ponderosa pine and two were from western ponderosa pine. Cones were from trees used in analyses to characterize geographic variation in cone and seed traits (see below), so that mean values for all measured cone and seed traits could be estimated based on the tree means. We used four cones from each of 77 trees spanning the full range of geographic variation in ponderosa pine. We recorded the percentage of each cone that had been foraged on in each aviary (0%, 25%, 50%, 75%, or 100%) after 2 h. To determine which cone traits were associated with variation in foraging propensity, we used regression analyses based on the relationship between relative tree fitness (percent of cones not foraged upon for each tree divided by the overall mean percent) and the individual cone and seed traits, each standardized to zero mean and unit variance.

Foraging trials have two potential advantages over our measures of selection in the wild. First, crossbills (and squirrels) likely forage preferentially on cones both within and between trees (e.g., Mezquida and Benkman 2005). This limits our ability to detect selection in the wild because as the proportion of cones foraged on in a tree increases, the remaining cones that we can sample will increasingly tend to be representative of cones from avoided trees (Benkman et al. 2003). However, such within tree preferences should not prevent us from detecting selection in the wild, because cone variation within trees is substantially less than cone variation between trees (Smith 1968; Elliott 1974). The relatively large between-tree variation is also why limiting our sampling to two cones per tree allows us to capture the between-tree variation. Second, we can increase the range of cone variation in captivity relative to what is found in any single location where we measured selection in the wild, which should increase our ability to detect selection (e.g., Schluter 1988).

CONE EVOLUTION IN RESPONSE TO SELECTION EXERTED BY ABERT'S SQUIRRELS AND CROSSBILLS

To determine if geographic variation in cone and seed traits of Rocky Mountain ponderosa pine is consistent with evolution in response to selection exerted by either Abert's squirrels or crossbills, recently mature, closed Rocky Mountain ponderosa pine cones were sampled from mature trees at three sites in regions with Abert's squirrels (SF, N, and TL in Fig. 1) and four sites without Abert's squirrels (D, F, A, and BH in Fig. 1). Cones were sampled between 15 and 30 September 2003 from 22 to 25 randomly selected trees at each site. We measured the same cone and seed traits described above for two cones from each tree.

Two-level nested ANOVAs were used to test for differences in Rocky Mountain ponderosa pine cone traits within and between areas with sites nested within areas with and without Abert's squirrels using proc GLM in SAS version 9.1 (SAS, Cary, NC). We tested for homogeneity of variances (inspection of residual plots and Levene's tests) and normality (Shapiro-Wilkes tests), and found no cases of heteroscedasticity and only several cases of slight departures from normality. Data were log transformed to minimize departures from normality. Discriminant functions analysis was used to determine if the traits under selection by Abert's squirrels and crossbills were also the traits that most strongly distinguished cones from areas with and without Abert's squirrels. In multivariate analyses, we used the total number of seeds per cone instead of full seeds because the number of full seeds is influenced by the frequency of outcrossing (Smith et al. 1990) and there was considerable variation among sites in the proportion of full seeds per cone (range in site means: 0.47–0.72).

We used foraging experiments with captive crossbills to further determine whether defenses directed toward crossbills were enhanced in the absence of Abert's squirrels, and if selection would favor larger bills in these areas (e.g., Parchman and Benkman 2002). Crossbills were kept in indoor aviaries (1.6 m × 2.7 m × 2.2 m or larger) at the New Mexico State University Animal Care Facility. Nineteen crossbills (16 ponderosa pine and three South Hills crossbills) with bill depths (measured at the anterior end of the nares) ranging from 9.00 mm to 10.18 mm were each timed feeding on 10 Rocky Mountain ponderosa pine cones from both areas with and without Abert's squirrels (cone length and width were within one SD of the means for the respective areas). We focus on bill depth in all analyses because it is correlated with feeding performance and is a known target of selection (Benkman 1993, 2003). Experimental cones were oven-dried until the scales began to open and then reclosed with moisture the night before being used. Each bird was timed feeding on one to three cones per day, and birds were given cones in a randomized order with respect to treatment. The time required to extract five seeds was recorded beginning after the first seed was eaten and ending when the sixth seed was extracted (see Benkman 1993 for details on aviary protocols). The timer was stopped during the handling of each seed, because seed extraction time is more closely related to bill size and cone structure variation than is seed handling time (Benkman 1993).

We used a paired t-test to test for differences in seed extraction times for crossbills foraging on cones from areas with and without Abert's squirrels. Lower foraging rates (more time per seed) for areas without Abert's squirrels would indicate enhanced defenses against crossbills. We used analysis of covariance (ANCOVA) to test for a difference in the slopes of the relationships between seed extraction time and bill depth for crossbills foraging on cones from areas with and without Abert's squirrels. A steeper negative slope for areas without Abert's squirrels would indicate stronger selection for larger bills in areas without Abert's squirrels than in areas with them (Benkman 2003).

CONE EVOLUTION IN RESPONSE TO SELECTION EXERTED BY GRAY SQUIRRELS

We quantified seed predation and cone traits for 44 western ponderosa pine trees at Forest Hill, California (FH in Fig. 1) to estimate selection exerted by gray squirrels as described above for Abert's squirrels. As before, we recorded the number of fresh cone cores at the base of each tree (mean = 13; range: 0–130) and estimated the number of cones remaining on the tree (mean = 87; range: 40–220). Cones were sampled between 20 and 30 August 2005 from two areas in the Sierra Nevada (cones mature earlier here than in other areas sampled) where gray squirrels occur (LP and FH in Fig. 1), and were compared to cones from areas without gray squirrels (Rocky Mountain ponderosa pine). Two-level nested ANOVAs (sites nested within areas with and without gray squirrels) on log-transformed data were used as before to test for differences in individual cone traits between areas with and without gray squirrels (Fig. 1). Discriminant functions analysis was used to determine which cone traits most strongly distinguished cones from areas with and without gray squirrels.

CONE EVOLUTION IN RESPONSE TO SELECTION EXERTED BY GRAY SQUIRRELS: IMPLICATIONS FOR CROSSBILLS

We used data from the foraging experiments with captive crossbills and analyses described above under “Phenotypic selection exerted by Abert's squirrels and crossbills” to determine if traits that evolved in response to selection exerted by gray squirrels deter crossbills. In addition, we used a multivariate cubic spline procedure (projection pursuit approximation [Schluter and Nychka 1994]) to characterize the fitness surface for ponderosa pine in relation to the first two principal components derived from the correlation matrix of eight cone and seed traits. As before, relative fitness was determined as the percent of cones not foraged upon for each tree divided by the overall mean percent. We estimated a surface with the λ corresponding to the lowest GCV score (Schluter and Nychka 1994), where the GCV score is a metric of the prediction error associated with a particular value of λ (Schluter 1988), and then drew the surface using a Loess smoothing function in Sigma Plot (SYSTAT Inc., Point Richmond, CA).

Results

PHENOTYPIC SELECTION EXERTED BY ABERT'S SQUIRRELS AND CROSSBILLS

Selection by Abert's squirrels favored trees that produced shorter cones with fewer seeds, and a lower ratio of seed mass to cone mass (Table 1). There was no evidence of stabilizing selection on any trait.

Table 1.  Estimates of univariate selection gradients using least squares regression analyses between relative tree fitness (based on percent seed predation by Abert's squirrels) and various cone and seed traits of Rocky Mountain ponderosa pine in the Apache-Sitgraeves National Forest, Arizona (n=67 trees). One thousand bootstrap replicates were used to establish the significance of β. Significant parameter estimates are in bold.
Cone trait β±SEP
Cone length−0.032±0.0130.017
Cone width −0.035±0.0210.10
Cone width/ length  0.012±0.0230.59
Cone mass −0.034±0.0170.057
Scale number −0.017±0.0200.39
Scale thickness −0.006±0.0230.78
Scale length −0.033±0.0170.059
Number of full seeds−0.041±0.0190.036
Number of empty seeds  0.002±0.0180.92
Total number of seeds−0.046±0.0160.007
Individual seed mass −0.020±0.0140.16
Seed mass/cone mass−0.044±0.0200.035

The univariate selection gradients exerted by crossbills were similar, identical in direction, and significantly correlated between the datasets (r= 0.61, P= 0.037) from the wild and captivity (Table 2). Twice as many regressions were significant for the data from the aviaries (6 of 12 regressions) than from the wild (3 of 12 regressions), although this difference was not significant (Fisher's exact test, P= 0.20). In both datasets, crossbills foraged more often on smaller, narrower cones (Table 2), which should favor the evolution of trees with larger and wider cones. Only one trait, scale thickness, was significant in the selection analyses from the wild but not in the analyses of the aviary experiment (Table 2). Presumably selection on scale thickness was not detected in the experiments because crossbills were more likely to avoid foraging on the relatively large cones of western ponderosa pine, which also had thinner scales than cones sampled from other areas (Table 3). Individual seed mass was the only trait that experienced stabilizing selection, and this was only in selection analyses from the wild (Table 2). Our only explanation for why crossbills tended to exploit cones with either small or large seeds in the wild was that crossbills avoided trees with large cones and for unknown reasons cones with the smallest and largest seeds were also small. For example, the quadratic regression between cone mass and seed mass was significant (cone mass= 12.3 + 279.7[seed mass]– 27863.9[seed mass– 0.03]2, r2= 0.24, F1,42= 6.77, P= 0.003, quadratic term P= 0.046).

Table 2.  Estimates of univariate selection gradients using least squares regression analyses between relative tree fitness (based on percent seed predation by crossbills) and various cone and seed traits of Rocky Mountain ponderosa pine in the Black Hills, SD and in foraging trials with captive crossbills. Bootstrapping was used to establish the significance of β. Significant parameter estimates are in bold.
Cone traitSelection in the wild Foraging trials
 β±SEP β±SEP
Cone length 0.099±0.0620.110.212±0.055<0.001
Cone width0.181±0.0520.0010.220±0.054<0.001
Cone width/length−0.049±0.0550.38−0.094±0.059  0.12
Cone mass0.161±0.0560.0060.226±0.054<0.001
Scale number 0.071±0.0610.250.124±0.058  0.036
Scale thickness0.116±0.0530.033 0.081±0.059  0.17
Scale length 0.057±0.0320.0700.167±0.057  0.004
Number of full seeds−0.002±0.0560.97−0.002±0.060  0.97
Number of empty seeds 0.003±0.0610.59 0.023±0.060  0.70
Total number of seeds 0.011±0.0460.82 0.014±0.060  0.81
Individual seed mass 0.051±0.0700.460.131±0.058  0.026
Seed mass/cone mass−0.121±0.0610.052−0.111±0.058  0.062
(Seed mass)2−0.149±0.0400.001  
Table 3.  Mean cone measurements for ponderosa pine sampled from sites in areas without Sciurus, areas with Abert's squirrels, and areas with gray squirrels. The first two sets of P-values gives the significance of the differences between areas with and without gray squirrels, and of the differences among sites within these areas. The third and fourth sets of P-values give the significance of the differences between areas with Abert's squirrels and without Sciurus, and of the differences among sites within these areas. Tests are based on two-level nested ANOVAs.
MeasurementSites without SciurusSites with Abert's squirrels
Fishlake, UTDixie, UTAshley, UTBlack Hills, SDSouth Fork, COTwin Lakes, CONederland, CO
Cone length (mm)86.075.080.277.067.869.568.3
Cone width (mm)41.838.439.441.535.439.239.7
Cone mass (gm)23.1317.1122.7520.9717.1115.0816.86
Scale number19.817.317.818.415.917.615.5
Scale thickness (mm) 6.4 5.6 5.4 5.8 5.0 5.5 4.9
Scale length (mm)22.119.621.018.718.219.417.9
Number of full seeds38.842.242.658.142.031.4742.3
Number of empty seeds36.729.833.721.528.434.441.9
Total number of seeds75.572.076.379.670.566.084.2
Individual seed mass (mg)36.832.434.834.628.030.730.4
Seed mass/cone mass 0.062 0.079 0.066 0.096 0.081 0.057 0.077
Number of trees23222425242423
MeasurementSites with gray squirrelsBetween areas with and without gray squirrelsAmong sites within areas with and without gray squirrels
Forest Hill, CALassen, CAF1, 8P% of totalF8, 248P% of total
Cone length (mm)94.195.7198.9<0.000155.4 9.9<0.000111.3
Cone width (mm)45.045.9132.5<0.000146.7 7.7<0.000110.8
Cone mass (gm)31.0234.40301.7<0.000165.410.56<0.0001 9.2
Scale number16.916.1 12.8 0.0004 0.012.3<0.000130.0
Scale thickness (mm) 4.6 4.6123.0<0.000140.611.7<0.000117.3
Scale length (mm)23.723.8156.6<0.000150.4 8.5<0.000111.1
Number of full seeds25.843.0 24.6<0.000110.0 7.1<0.000116.9
Number of empty seeds27.434.1  4.2 0.042 0.0 5.3<0.000114.1
Total number of seeds53.177.2 22.3<0.0001 9.6 6.2<0.000115.1
Individual seed mass (mg)37.535.1 24.6<0.0001 9.6 7.5<0.000117.9
Seed mass/cone mass 0.032 0.045157.1<0.000153.6 5.4<0.0001 6.6
 4143      
MeasurementBetween areas with Abert's squirrels and without SciurusAmong sites within areas with Abert's squirrels and without Sciurus
F1, 6P% of totalF6, 164P% of total
Cone length (mm)54.9<0.000137.13.0 0.0121 4.9
Cone width (mm)17.5<0.0001 8.57.9<0.000120.6
Cone mass (gm)45.7<0.000129.75.3 0.000210.8
Scale number41.9<0.000125.96.7<0.000114.4
Scale thickness (mm)69.6<0.000138.26.9<0.000112.4
Scale length (mm)33.6<0.000119.48.2<0.000118.8
Number of full seeds7.56 0.007 1.26.3<0.000118.2
Number of empty seeds 9.1 0.003 0.09.6<0.000127.0
Total number of seeds 1.2 0.27 0.03.5 0.0051 9.6
Individual seed mass (mg)39.7<0.000130.12.6 0.0271 4.5
Seed mass/cone mass 1.9 0.20 0.08.2<0.000123.3

CONE EVOLUTION IN RESPONSE TO SELECTION EXERTED BY ABERT'S SQUIRRELS AND CROSSBILLS

Patterns of geographic variation in Rocky Mountain ponderosa pine cone and seed traits between areas with and without Abert's squirrels suggest that selection from crossbills, but not Abert's squirrels, has driven cone evolution. Two traits detected as being under selection by Abert's squirrels, the number of seeds per cone and the ratio of seed mass to cone mass (Table 1), were the only traits that did not differ between areas with and without Abert's squirrels (Table 3). Selection exerted by crossbills favored increases in cone length, mass, and width, and scale length and thickness (Table 2), and these traits were all larger in areas without Abert's squirrels compared to areas with Abert's squirrels (Table 3). Discriminant functions analysis correctly classified 80% of the trees from areas with Abert's squirrels and 79% of the trees from areas without Abert's squirrels. Based on the canonical discriminant functions, Rocky Mountain ponderosa pine from areas without Abert's squirrels are characterized by longer, heavier cones, with thicker scales, which were traits under selection by crossbills but, with the exception of cone length, were not under selection by Abert's squirrels.

Crossbills required more time to extract seeds from cones from areas where Abert's squirrels are absent compared to where they occur (Fig. 2; paired t-tests, t17= 11.17, P < 0.0001), implying that seeds are more defended against crossbills in the absence than presence of Abert's squirrels. The slope of the relationship between seed extraction time and bill depth (Fig. 2) for crossbills foraging on cones from areas without Sciurus (r2= 0.57, df = 18, P < 0.001) was steeper (ANCOVA: F1, 37= 4.31, P= 0.045) than from areas with Abert's squirrels (r2= 0.39, df = 18, P= 0.003), suggesting that selection for larger bills is stronger in areas where Abert's squirrels are absent than where they are present.

Figure 2.

Crossbills require more time to extract seeds from Rocky Mountain ponderosa pine cones from areas without Abert's squirrels (open circles) than from areas with them (filled circles). The solid and dashed lines represent best-fit linear regressions for each of the datasets, respectively, and each point represents the mean for a single bird feeding on seeds in 10 cones.

CONE EVOLUTION IN RESPONSE TO SELECTION EXERTED BY GRAY SQUIRRELS

Gray squirrels were more likely to forage on trees with smaller cones having a higher seed mass to cone mass ratio (Table 4). No traits were under stabilizing selection. Such selection presumably accounts for the larger, heavier cones with a lower ratio of seed mass to cone mass in areas with gray squirrels (western ponderosa pine) in comparison to areas without them (Rocky Mountain ponderosa pine) (Table 3). Discriminant functions analysis correctly classified 99% of the trees from both areas with and without gray squirrels. Based on these analyses, trees from areas with gray squirrels are most strongly characterized by larger cones with fewer seeds (i.e., a lower ratio of seed mass to cone mass), which are also the traits favored by selection from gray squirrels (Table 4).

Table 4.  Estimates of univariate selection gradients using least squares regression analyses between relative tree fitness (based on percent seed predation by gray squirrels) and various cone and seed traits of western ponderosa pine from Forest Hill, California (n= 44 trees). Bootstrapping was used to establish statistical significance of β. Significant parameter estimates are in bold.
Cone trait β±SEP
Cone length0.052±0.0250.047
Cone width0.064±0.0240.01
Cone width/length−0.007±0.0290.81
Cone mass0.055±0.0230.018
Scale number 0.002±0.0220.94
Scale thickness 0.041±0.0250.10
Scale length 0.016±0.0280.57
Number of full seeds−0.029±0.0360.43
Number of empty seeds0.051±0.0180.006
Total number of seeds 0.006±0.0290.83
Individual seed mass 0.012±0.0190.53
Seed mass/cone mass−0.060±0.0390.049

CONE EVOLUTION IN RESPONSE TO SELECTION EXERTED BY GRAY SQUIRRELS: IMPLICATIONS FOR CROSSBILLS

Crossbill foraging propensity decreased with increases in both the first principal component (accounting for 48% of the total variation and loaded positively especially by cone length [0.47], width [0.45], and mass [0.48], and scale length [0.42]) and the second principal component (accounting for 18% of the total variation and loaded positively especially by scale thickness [0.63], number of scales [0.58], and total number of seeds [0.48]) (Fig. 3). The centroids for cones sampled from each of the three regions (Fig. 3B) illustrate how trees from areas with gray squirrels are most effectively defended (relative tree fitness = 1.32), whereas areas without Sciurus are intermediately defended (relative tree fitness = 1.13), and areas with Abert's squirrels are least defended (relative tree fitness = 0.58).

Figure 3.

(A) The first two principal components from an analysis using eight cone and seed traits for ponderosa pine from areas with Abert's squirrels, with gray squirrels, and without Sciurus. (B) Fitness surface from projection pursuit approximation for relative fitness in relation to the propensity of crossbills to avoid foraging on cones in relation to the first two principal components of the cone and seed traits as in A. The circles represent the centroids for areas with S. aberti, with S. griseus, and without Sciurus with the shades of the symbols matching those used above in A.

Discussion

Our results indicate that the presence and absence of Sciurus has given rise to a geographic selection mosaic for ponderosa pine through both competitive and evolutionary effects, and that this should lead to geographically variable selection on crossbills. In particular, the two Sciurus species affected the selection mosaic in different ways, with Abert's squirrels acting mostly as preemptive competitors and gray squirrels acting as important selective agents. Below we discuss variation in cone and seed characteristics of ponderosa pine in light of the ecological and  evolutionary consequences of predation by Sciurus, and the evolutionary consequences of this and resource variability for geographic variation in the interactions between crossbills and ponderosa pine.

ABERT'S SQUIRRELS AS A PREEMPTIVE COMPETITOR, AND ROCKY MOUNTAIN PONDEROSA PINE EVOLUTION IN RESPONSE TO SELECTION EXERTED BY CROSSBILLS

Although Abert's squirrels feed on ponderosa pine seeds when available (Keith 1965; Allred et al. 1994), the heavy defoliation of ponderosa pine by Abert's squirrels and the resulting selection exerted based on resin characteristics (Snyder 1992) appears to overwhelm selection exerted directly on cone structure. This conclusion is supported by the lack of geographic differentiation between areas with and without Abert's squirrels in cone traits experiencing selection from Abert's squirrels. Their twig-clipping for inner bark feeding (Fig. 1) and their feeding on cone seed, however, depress seed availability (Snyder 1993; Allred et al. 1994) and should depress both the abundance of crossbills and their selective impact on cone structure.

Our data indicate that selective foraging by crossbills has had a greater impact on Rocky Mountain ponderosa pine cone evolution in the absence of Abert's squirrels where defenses against crossbills have evolved. For example, as a measure of increased seed defense, captive crossbills had significantly longer seed extraction times when feeding on cones from areas without Abert's squirrels than on cones from areas with Abert's squirrels (Fig. 2). Furthermore, crossbills in both the wild and captivity avoided foraging on larger ponderosa pine cones with thicker scales (Table 2, Fig. 3B), which are traits that characterize cones from areas without Abert's squirrels (Table 3). Interestingly, the average shift in cone structure from areas with Abert's squirrels to areas without Abert's squirrels (Fig. 3B) is in the direction one would predict if cones were simply increasing defenses directed toward crossbills. We have also detected selection exerted by crossbills on a similar suite of cone traits in studies of other pines and found patterns of geographic differentiation consistent with the inferred form of selection (Benkman et al. 2003; Parchman et al. 2007). In particular, we have identified scale thickness as a key, geographically variable trait. Thicker scales provide resistance to crossbills spreading cone scales apart (Fig. 1), and we have consistently documented increased scale thickness as evolving in response to crossbill predation (Parchman and Benkman 2002; Benkman et al. 2003; Parchman et al. 2007).

WESTERN PONDEROSA PINE EVOLUTION IN RESPONSE TO SELECTION EXERTED BY GRAY SQUIRRELS, AND IMPLICATIONS FOR CROSSBILLS

Variation in the cone and seed traits of ponderosa pine between areas with and without gray squirrels suggests that these squirrels have been an important selective agent on cone structure. Selection exerted by gray squirrels favored trees with large cones and a low ratio of seed mass to cone mass (Table 4), and these are the traits that characterize cones in which gray squirrels are present (Table 3). Similar decreases in the seed mass to cone mass ratio have been documented for other conifers in areas in which tree squirrels are important selective agents, including lodgepole pine (Benkman et al. 2001), black spruce (Parchman and Benkman 2002), limber pine (P. flexilis), and whitebark pine (P. albicaulis) (Siepielski and Benkman 2007b). The evolution of larger cones in areas in which tree squirrels are important seed predators and selective agents (i.e., gray squirrels) is consistent with patterns seen in other studies of relatively large-coned conifers, including Aleppo pine (Pinus halepensis) (Mezquida and Benkman 2005) and several serotinous-coned pines in California (Linhart 1978). The tendency for squirrels to prefer and select against trees with smaller cones of these relatively large-coned species would occur if handling time increases disproportionately with increases in cone size (a positive second derivative for handling time with increases in cone size). Such an increase (exponential) in handling time occurs for seed-eating birds with increases in seed size (Benkman and Pulliam 1988), and we suspect a similar relationship occurs for tree squirrels foraging on conifer cones.

Although both Abert's and gray squirrels preferentially foraged on trees having cones with more seeds and a higher seed mass to cone mass ratio, the direction of preferences for size-related traits differed between these predators (Tables 1 and 4). This pattern is consistent with results in Mezquida and Benkman (2005), where the European red squirrel (S. vulgaris) in Spain preferred smaller Allepo pine cones when trees had relatively large cones, but preferred larger cones when trees had relatively small cones. Similarly, Molinari et al. (2006) report that in Italy European red squirrels prefer trees with larger cones of the relatively small-coned Scots pine (P. sylvestris). When data for both Abert's and gray squirrels are pooled, significant quadratic relationships exist between standardized relative tree fitness in relation to seed predation by Sciurus and size-related traits (cone length: relative fitness= 0.488 – 0.008[cone length]+ 0.001[cone length– 82.39]2, r2= 0.06, F2, 108= 3.51, P= 0.03, quadratic term P= .009; cone mass: relative fitness= 0.140 – 0.013[cone mass]+ 0.003[cone mass– 25.212]2, r2= 0.05, F2, 108= 3.10, P= 0.049, quadratic term P= .014). Interestingly, the cone length at which fitness is minimized and selection would be disruptive is similar for ponderosa pine (approximately 82 mm) and Aleppo pine (70 mm [Mezquida and Benkman 2005]). This, however, does not explain why Rocky Mountain ponderosa pine cones initially tended to be small and western ponderosa pine cones initially tended to be large. One explanation is that cone size is positively correlated with seed size (r= 0.91, P < 0.0001 for the correlation between cone length and seed mass for 18 North American species in the subgenus Pinus with seed masses less than 100 mg [data from McCune 1988]) and seeds were initially larger in western ponderosa pine (Table 3: western ponderosa pine seeds averaged about 11% heavier than Rocky Mountain ponderosa pine seeds; western ponderosa pine seeds were 38% heavier in a larger sample summarized in Wells [1964]). In general, seed size variation is related to abiotic conditions during and soon after germination (Westoby et al. 1996), but how this is related to seed size variation in ponderosa pine is unknown. For example, no significant correlation existed between annual precipitation and seed mass (r= 0.319, df = 7, P= 0.209) or cone mass (r= 0.521, df = 7, P= 0.093) across the sites in our study (precipitation data from USDA NRCS; http://www.ncgc.nrcs.usda.gov/).

Cone traits that have evolved in response to predation from gray squirrels incidentally result in these cones being avoided by crossbills. For example, the shift in average cone structure from areas without gray squirrels to areas with them (Fig. 3B) results in a decrease in seed accessibility for crossbills; however, the direction of cone change is not what one would predict if cones were evolving to deter only crossbills. Indeed, crossbills in North America rarely feed on cones as large as those of western ponderosa pine, as the massive size of these cones makes seed removal difficult for crossbills (Benkman 1993). We infer that crossbills have had little selective impact on western ponderosa pine evolution because they tend to be uncommon in these forests (Craig, W. Benkman, pers. obs.), and the cone trait that consistently increases in response to selection exerted by crossbills is scale thickness (Benkman et al. 2001, 2003; Parchman and Benkman 2002; Parchman et al. 2007), which is relatively thin in western ponderosa pine (Table 3).

Alternative explanations for geographical variation in the cone and seed traits of ponderosa pine appear less likely to account for the variation we document here. Other than crossbills and tree squirrels, insects are the most prevalent predispersal seed predators (Smith and Balda 1979) and may exert selection on cone and seed traits (Siepielski and Benkman 2004). However, insects that attack ponderosa pine cones exist throughout the range of ponderosa pine (Hedlin et al. 1980), so that evolution in response to these taxa should not confound our results.

THE GEOGRAPHIC SELECTION MOSAIC FOR PONDEROSA PINE AND CROSSBILLS

As in previous studies (Benkman et al. 2001; Parchman and Benkman 2002; Mezquida and Benkman 2005), our results suggest that crossbills have a stronger impact on the evolution of cone traits in areas in which tree squirrels are absent than in areas in which tree squirrels are present. Preemptive competition from Abert's squirrels limits but does not eliminate the selective impact of crossbills, so that seed defenses directed at crossbills are greater in the absence than presence of Abert's squirrels (note relatively thick scales in areas with Abert's squirrels compared to areas with gray squirrels in Table 3). The resulting geographic variation in cone structure in Rocky Mountain ponderosa pine causes geographically variable selection on crossbills, so that crossbills require more time to extract seeds from cones in which Abert's squirrels are absent than present and larger bills are favored by selection in areas without Abert's squirrels (Fig. 2). The result is that reciprocal selection between crossbills and ponderosa pine will be stronger in areas lacking Abert's squirrels.

Although reciprocal selection between ponderosa pine and crossbills will vary between areas with and without Abert's squirrels, crossbill nomadism resulting from high variability in the availability of ponderosa pine seed may cause enough gene flow to impede local adaptation of crossbills to geographic variation in the cone structure of Rocky Mountain ponderosa pine. This suggestion is consistent with Groth's (1993) finding of no detectable morphological differentiation of ponderosa pine crossbills sampled from throughout western North America and with recent genetic analyses (Parchman et al. 2006). In previous studies of crossbills specialized on conifers representing temporally stable resources, divergent selection on crossbills between coevolutionary hot spots and cold spots has fueled phenotypic diversification, and perhaps speciation (Benkman 1999; Benkman et al. 2001, 2003; Smith and Benkman 2007). One reason for this divergence is that resource stability allows crossbills to form resident, locally adapted populations continually exerting and responding to reciprocal selection in the absence of Tamiasciurus. In contrast, gene flow resulting from nomadic movements of crossbills in response to cone crop fluctuations should result in ponderosa pine crossbills being somewhat maladapted for feeding on ponderosa pine in parts of its range and suggests that in certain regions this interaction may be characterized by a trait mismatch. Nomadism should also limit local escalatory coevolutionary adaptations resulting from predator–prey arms races (Benkman et al. 2003), and likely reduces the strength of divergent selection on crossbills and ponderosa pine across the selection mosaic.

Gray squirrels have a selective impact on western ponderosa pine causing cones to become so large that the feeding propensity of crossbills is greatly reduced (Fig. 3B). Indeed, crossbills are less common in ponderosa pine forests in the Sierra Nevada than in the Rocky Mountain regions (Craig W. Benkman, pers. obs.). This, along with data from previous studies (Benkman 1993; Benkman et al. 2001), suggests why ponderosa pine crossbills are adapted for feeding on the cones of Rocky Mountain ponderosa pine. Thus, by favoring the evolution of such large cone size, gray squirrels may have influenced the selection mosaic by reducing the geographic range of the interaction between crossbills and ponderosa pine. Similarly, selection from European red squirrels on the mainland of Spain may have limited the geographic extent of the interaction between crossbills and Aleppo pine in the Mediterranean region (Mezquida and Benkman 2005). These findings indicate that the ability of crossbills to specialize on relatively large-coned conifers may be limited where tree squirrels are important selective agents causing the evolution of even larger cone size. Gray squirrels also forage on the extremely large-coned pines in California (e.g., P. coulteri and P. sabiniana; Craig W. Benkman, pers. obs.), and selective foraging by gray squirrels (and the pines' extremely large seeds) may have also contributed to the evolution of massive cone size in these pines.

Conclusions

Although Abert's squirrels depress seed abundance and thereby reduce the selective impact exerted by crossbills on ponderosa pine, selection by gray squirrels causes the evolution of larger cone size preventing crossbills from using much ponderosa pine seed in areas that gray squirrels occur. The result is that ponderosa pine crossbills have evolved to specialize on Rocky Mountain ponderosa pine (i.e., where gray squirrels are absent) and interact most strongly with ponderosa pine in the complete absence of squirrels. However, the high interannual variation in seed production by ponderosa pine forces crossbills to be nomadic and regularly move between areas with and without Abert's squirrels. Nevertheless, ponderosa pine crossbills and Rocky Mountain ponderosa pine have likely coevolved. Here, the extent of seed defenses varies depending on the presence of Abert's squirrels, and crossbills in turn have adapted to cones across the range of Rocky Mountain ponderosa pine. Consequently, increasing resource variability prevents strong selection mosaics and local differentiation of crossbill populations between areas with and without squirrels that we find in conifers with more consistent seed availability (e.g., Benkman 1999; Benkman et al. 2001, 2003; Parchman and Benkman 2002), but does not prevent coevolution between crossbills and conifers (e.g., Parchman et al. 2007). We suspect that, in general, plants and more sedentary predispersal seed predators (e.g., insects and squirrels) will experience selection mosaics and show geographic variation regardless of variation in annual seed production as long as the distributions of important seed predators are not completely coincident. However, with increasing annual variation in seed crops, species capable of tracking seed crops that vary asynchronously in space (e.g., crossbills and some parrots) are unlikely to adapt to local plant populations. This is consistent with the expectation that high levels of gene flow and population mixing constrain local adaptation (e.g., Hendry et al. 2001).

Finally, our study provides further evidence that geographic variation in community context often gives rise to geographic selection mosaics. However, the actual mechanisms by which variation in community context affect geographic selection mosaics are likely to be diverse and idiosyncratic. This diversity of mechanisms suggests both that geographic selection mosaics are likely a common feature of species interactions and that a close consideration of natural history will often be necessary to understand the mechanisms by which geographic selection mosaics arise.

Associate Editor: J. Shykoff

ACKNOWLEDGMENTS

National Science Foundation grants (DEB-0212271 and DEB-0344503) to CWB, and an Environmental Protection Agency GRO Fellowship and a Sigma Xi GIAR grant to TLP supported this work. The manuscript was improved by comments by A. Badyaev, J. Pauli, A. Siepielski, and anonymous reviewers. We thank W. S. Allred for generously supplying data and K. Larsen for help with measuring cones. E. Rahel drew the squirrels in Figure 1.

Ancillary