Spatial Location Dominates Over Host Plant Genotype in Structuring an Herbivore Community

Ecology, 91(9), 2010, pp. 2660–2672 Ó 2010 by the Ecological Society of America

Spatial location dominates over host plant genotype in structuring an herbivore community

1,4 1 2 1,3 AYCO J. M. TACK, OTSO OVASKAINEN, PERTTI PULKKINEN, AND TOMAS ROSLIN 1Metapopulation Research Group, Department of Biosciences, University of Helsinki, P.O. Box 65, Viikinkaari 1, Helsinki FI-00014 Finland 2Finnish Forest Research Institute, Haapastensyrja¨ Tree Breeding Station, Haapastensyrja¨ntie 34, La¨ylia¨inen FI-12600 Finland 3Department of Applied Biology, University of Helsinki, P.O. Box 27, Latokartanonkaari 5, Helsinki FI-00014 Finland

Abstract. Recent work has shown a potential role for both host plant genotype and spatial context in structuring insect communities. In this study, we use three separate data sets on herbivorous insects on oak (Quercus robur) to estimate the relative effects of host plant genotype (G), location (E), and the G 3 E interaction on herbivore community structure: a common garden experiment replicated at the landscape scale (;5km2); two common gardens separated at the regional scale (;10 000 km2); and survey data on wild trees in various spatial settings. Our experiments and survey reveal that, at the landscape scale, the insect community is strongly affected by the spatial setting, with 32% of the variation in species richness explained by spatial connectivity. In contrast, G and G 3 E play minor roles in structuring the insect community. Results remained similar when extending the spatial scale of the study from the more local (landscape) level to the regional level. We conclude that in our study system, spatial processes play a major role in structuring these insect communities at both the landscape and regional scales, whereas host plant genotype seems of secondary importance. Key words: community genetics; Finland; genotype–environment interaction; intraspeciﬁc genetic variation; metacommunity; plant–insect interactions; plant traits; Quercus robur; spatial context; spatial scale.

INTRODUCTION Recent work on metacommunities has shown the Since the recent emergence of ‘‘community genetics’’ importance of the spatial distribution of the host plant (Price 1983, Antonovics 1992, Neuhauser et al. 2003, on the structure of insect communities (Hanski 1999, Whitham et al. 2003), a multitude of studies has Holyoak et al. 2005), but the spatial context has seldom convincingly shown that genotypic variation of the host been considered in studies of community genetics. plant can influence the associated insect community Within the latter field, the few studies that have (Boecklen and Spellenberg 1990, Maddox and Root incorporated a spatial setting have either selected the 1990, Aguilar and Boecklen 1992, Dungey et al. 2000, most distinct genotypes (Fritz and Price 1988, Graham Wimp et al. 2005, Tovar-Sańchez and Oyama 2006b, et al. 2001) or the most divergent habitats present in the Whitham et al. 2006). In addition, relatively high area (Graham et al. 2001, Johnson and Agrawal 2005). heritabilities for community descriptors have been Given such systematic sampling, these studies do not reported: for example, host plant genotype explained allow for generalizations regarding the relative strength up to 41% of the variation in the diversity of an insect of host plant genotype and location in structuring the community on the evening primrose (Johnson and insect community; such comparisons should instead rely Agrawal 2005) and 50–63% of the community compo- on unbiased variance estimates as derived from a sition on a single cottonwood species (Shuster et al. random sampling of the population (Littell et al. 2006). However, the presence of a significant effect of 2006). Moreover, the relative strength of genotypic host plant genotype in a common garden does not and location-related effects may also change with the necessarily imply that host plant genotype would more spatial scale of the study (Wiens 1989, Levin 1992). It generally be an important factor in structuring commu- has been argued that environmental variation increases nities in nature (Johnson and Stinchcombe 2007). For faster than genotypic variation with increasing spatial this, the relative roles of host plant genotype and of scale (Johnson and Agrawal 2005, Johnson and Stinch- other ecological factors in structuring the insect com- combe 2007), but these predictions remain largely munity should be addressed by multifactorial designs. untested. Based on these considerations, a key question for community ecology is how much the factors genotype Manuscript received 9 June 2009; revised 16 November 2009; (G), location (E), and their interaction contribute to accepted 21 December 2009; final version received 19 January 2009. Corresponding Editor: R. J. Marquis. structure insect communities at multiple spatial scales. 4 E-mail: ayco.tack@helsinki.fi In this study, we address this question in the context of 2660 September 2010 INSECT COMMUNITIES SHAPED BY SPACE 2661

TABLE 1. Species guilds, taxonomy, and numerical code for insect species that feed exclusively on the oak Quercus robur in southern Finland.

Guild Order: Family Genus and species Species number L Hymenoptera: Tenthredinidae Profenusa pygmaea 1 L Lepidoptera: Coleophoridae Coleophora flavipennella/C. kuehnella 2 L Lepidoptera: Eriocraniidae Dyseriocrania subpurpurella 3 L Lepidoptera: Gracillariidae Caloptilia alchimiella 4 L Lepidoptera: Gracillariidae Phyllonorycter harrisella/P. quercifoliella 5 L Lepidoptera: Nepticulidae Ectoedemia albifasciella 6 L Lepidoptera: Nepticulidae Stigmella svenssoni/S. ruficapitella/ 7 S. roborella/S. samiatella L Lepidoptera: Tischeriidae Tischeria ekebladella 8 G Diptera: Cecidomyiidae Macrodiplosis dryobia 9 G Homoptera: Triozidae Trioza remota 10 G Hymenoptera: Cynipidae Andricus callidoma 11 G Hymenoptera: Cynipidae Andricus curvator 12 G Hymenoptera: Cynipidae Andricus inflator 13 G Hymenoptera: Cynipidae Cynips divisa 14 G Hymenoptera: Cynipidae Neuroterus anthracinus 15 G Hymenoptera: Cynipidae Neuroterus numismalis 16 G Hymenoptera: Cynipidae Neuroterus quercusbaccarum 17 O Homoptera: Asterolecaniidae Asterolecanium variolosum 18 O Lepidoptera: Bucculatricidae Bucculatrix ulmella 19 O Lepidoptera: Heliozelidae Heliozela sericiella 20 O Lepidoptera: Tortricidae Ancylis mitterbacheriana 21 M Erysiphales: Erysiphaceae Erysiphe alphitoides 22 F Lepidoptera/Hymenoptera/Coleoptera multiple spp. 23 Notes: In Finland, all these species feed exclusively on the oak Quercus robur, but adults of Trioza remota hibernate on conifers (Ossiannilsson 1992). The guilds are abbreviated as follows: L, leaf miner; G, galler; O, other; M, mildew; and F, free-feeding. The guild ‘‘other’’ contains four species with deviant feeding modes: a scale insect (Asterolecanium variolosum); a species combining petiole galling at early larval stages with leaf mining in the final instar (Heliozela sericiella); a species combining leaf mining in early instars with free-feeding in later instars (Bucculatrix ulmella); and a leaf folder (Ancylis mitterbacheriana). an herbivorous insect community on oak (Quercus and sustains more than 20 specialist leaf mining, gall- robur). We first conduct a reciprocal transplant common inducing, and leaf-folding species (Table 1, Plate 1). These garden experiment to assess the relative effect of guilds offer ideal targets for studies of insect abundance genotype and location at the landscape scale. We then and community structure, as they can be identified and use observational data of the insect community on wild counted even after the larva has died or the adult has trees within the same landscape to pinpoint the role of emerged. In addition, the powdery mildew Erysiphe spatial context. Subsequently, we use an experiment with alphitoides (syn. Microsphaera alphitoides) occurs at high two common gardens at the regional scale to test densities in some parts of the landscape (Roslin et al. whether the relative importance of genotype, location, 2007), offering scope for comparing patterns observed in and their interaction persists when changing the spatial insectswiththatofthisdistincttaxon. scale. More specifically, we address the following Landscape scale questions: (1) What is the relative effect of genotype, To quantify the effect of host plant genotype on insect location, and their interaction on the insect community community structure in a spatial setting, we initiated a and its component species? (2) What is the broad-sense reciprocal transplant common garden experiment using heritability of the insect community, if considered as a cloned trees in a natural landscape. Details regarding the trait of the host tree? (3) Do the relative strengths of exact experimental design are given in Appendix A. In genotype, location, and their interaction persist over short, branch tips (n ¼ 100 tips/tree) were collected from multiple spatial scales? (4) Can the effects of genotype 10 haphazardly selected large trees in different parts of and location on the insect community be attributed to the island Wattkast, an island located in the southwest- covariates such as spatial connectivity or host tree ern archipelago of Finland (Fig. 1). The sampling took chemistry? (5) Do insect species co-occur more or less place in the spring of 2004. The resulting shoots were often than expected by chance on certain genotypes or at subsequently grafted onto randomly selected rootstocks certain locations? grown from acorns. The grafts were grown for three MATERIALS AND METHODS summers in a common environment to minimize maternal effects and to increase in size to ;1.5 m. On Study system 23 April 2007, well before leaf flush, the successful grafts The pedunculate oak (Quercus robur) harbors a large (total n ¼ 172) were transported back to Wattkast as community of insects (Southwood 1961). In Finland, the planted in 50-L plant pots and placed in reciprocal pedunculate oak is the only representative of its genus common gardens (n ¼ 25–31 grafts per garden). In order 2662 AYCO J. M. TACK ET AL. Ecology, Vol. 91, No. 9

FIG. 1. Locations of common gardens, host plant provenances, and surveyed trees. (A) A map of the regional scale (southwestern Finland) where host plant genotypes were collected from six populations (circles) and planted in two common gardens (stars). The arrow marks the location of the island Wattkast. The line of the natural northern oak (Quercus robur) limit is adopted from Vakkari et al. (2006). (B) A close-up of the island Wattkast, in which gray labeled squares represent the locations of the six reciprocal common gardens and white squares represent the locations of the four additional genotypes used in the experiment. To assess the effect of spatial connectivity on local community structure, we surveyed 89 wild oak trees (small gray symbols). Small black circles show the locations of all oak trees on the island (n ¼ 1868). to increase the number of replicates per genotype per To characterize insect abundance and community common garden, we restricted the design to six locations composition, we visited the experimental trees three (Fig. 1B). Within each common garden, grafts of times (1–4 June, 10–13 July, and 9–13 September 2007) different genotypes were randomly mixed and placed and recorded the abundance of focal herbivore taxa on in a regular grid at distances of ;1 m from one another the full foliage. Multiple recordings of the same leaf (Appendix A). mine, gall, or leaf fold were prevented by marking September 2010 INSECT COMMUNITIES SHAPED BY SPACE 2663 respective leaves with permanent ink. In September, we and their spatial context ranged from individuals also recorded the presence/absence of mildew and growing in dense oak stands to isolated oaks (Fig. 1B). damage by free-feeding insects on 10 randomly selected In September 2008, we recorded the occurrence (pres- leaves per tree. Hence, for the endophagous insects and ence/absence) of the focal species by examining all leaves leaf folders, we used the counts per tree as the response of each tree. In addition, we recorded the abundance of variable, and for mildew and free-feeder damage, we each species by counting the individuals on 20 randomly used the fraction of damaged leaves on each tree as the selected shoots per tree. response variable. For some analyses, we grouped the We characterized the spatial connectivity of each of responses scored as being caused by ﬁve different the selected trees with a connectivity metric modiﬁed

‘‘guilds’’: leaf miners, gallers, ‘‘other’’ (a compound from Hanski (1999): Si ¼ Rj6¼i Nj exp(adij). Here i is the group of species being neither clear-cut gallers nor focal tree, j ranges over all the trees in the island miners; Table 1), mildew, and damage by free-feeding (selected or not), dij is the distance between trees i and j herbivores (Table 1). in meters, and Nj is the estimated number of leaves on Host plant attributes were analyzed by focusing on tree j. The parameter a was set to the value 1/a ¼ 250 m, three plant traits identified as important for herbivorous reflecting the short average dispersal of insect individ- insects in previous work: leafing phenology (Crawley uals (cf. Gripenberg et al. 2008). Nj was estimated using and Akhteruzzaman 1988, Hunter 1992, Mopper 2005), the formula log(number of leaves) ¼ 0.92 þ 2.55 3 nitrogen and carbon content (West 1985, Cornelissen log(GBH), where GBH stands for girth at breast height and Stiling 2008), and phenolic compounds (Feeny 1970, (see Gripenberg et al. 2008: Appendix S3) and reflects Roslin and Salminen 2008). On 30 May 2007, we the relative size of each oak tree as a habitat to specialist classified the leafing phenology of each tree on a scale plant-feeding insects (Gripenberg et al. 2008). ranging from 0 to 6, according to the median While the use of a single value for the dispersal developmental phase of the leaves. This scale was parameter a across all taxa may seem crude, we note modified from Crawley and Akhteruzzaman (1988), that the relative ranking of the oak trees in terms of with 0 implying that buds are still completely closed; 1, spatial connectivity is only weakly affected by the value green is visible between the brown bud scales; 2, the bud of a (cf. Hanski 1999): Si values for individual trees has elongated and is predominantly green; 3, the leaves maintain a Spearman rank correlation value of no less protrude beyond the tip of the bud; 4, leaves start to than rS ¼ 0.95 (n ¼ 1868) across 10-fold variation in a (1/ separate, but no shoot is visible yet; 5, the shoot is a ¼ 100 and 1/a ¼ 1000). Hence, the connectivity metric clearly visible, but leaves are not separated yet; 6, leaves combines estimates of habitat size and of inter-tree are fully expanded and adopt their mature, dark green distances into a robust measure of expected immigration coloration. of insects from all trees in the landscape. Connectivity In order to measure chemical contents of the oak values were also calculated for each of the six common leaves, we collected five randomly selected leaves from gardens. each graft on 6 June 2007. On collection, leaves were placed in a cooler and subsequently air-dried in Regional scale aluminium trays, sheltered from sunlight. The leaves In order to assess whether the relative impact of host were left to dry until they reached a constant mass and plant genotype varies with spatial scale, we took were subsequently frozen. Whereas air-drying might not advantage of a tree-breeding experiment established by be the optimal drying mode, only small differences in the Finnish Forest Research Institute. In autumn 1996, hydrolyzable tannin and flavonoid glycoside contents acorns were collected from six mother trees within each have been observed among air-dried and freeze-dried of six populations (n ¼ 36 mother trees) spread over the leaves in both birch (Salminen 2003; J.-P. Salminen, distributional range of oaks in Finland (Fig. 1A). These unpublished data) and oak (J.-P. Salminen, unpublished acorns were sown in a nursery in the spring of 1997 and data), and this method will thus yield accurate results were transplanted in the spring of 1998 to common (cf. Salminen et al. 2004). Samples were analyzed for the gardens at two sites in southern Finland (Fig. 1A). At concentration of individual phenolic compounds using both sites oaks occurred naturally. Both common methods described in Salminen et al. (2004). Nitrogen gardens followed a randomized complete block design and carbon concentrations were determined using a (Littell et al. 2006), where each block contained a single vario MAX CHN analyzer (Elementar Analysensys- offspring per mother tree, planted at distances of ;4.5 teme, Hanau, Germany). m. Offspring of a single mother were assumed to be half- sibs, as they may have been sired by pollen from Spatial connectivity different fathers. Species presence/absence was recorded To examine the potential role of spatial connectivity on each tree from 2 to 12 September 2008 (n ¼ 1072). on insect community structure, we selected 89 trees (ranging in size from 1 to 3 m). These trees were Statistical analyses naturally distributed across the island in which we We used generalized linear mixed models to partition conducted the reciprocal common garden experiment, the variance in various aspects of the community 2664 AYCO J. M. TACK ET AL. Ecology, Vol. 91, No. 9

PLATE 1. Species guilds in oak-associated communities: (top left) galler Neuroterus quercusbaccarum; (top right) mildew Erysiphe alphitoides; (bottom left) leaf miner Dyseriocrania subpurpurella; and (bottom right) free-feeder Favonius quercus. For most of these taxa, abundances vary more among sites within a landscape than among host plant genotypes. Photo credits: A. J. M. Tack. structure and plant traits to the effects of genotype, variation (45.9% and 22.6%, respectively), they were location, and their interaction (all random effects). selected for subsequent analyses. Models were fitted to the data using a Bayesian Markov Regional scale.—As response variables at the regional chain Monte Carlo (MCMC) approach implemented in level, we used species richness and species occurrence. the package MCMCglmm in R (Hadfield 2009). Both were modeled as functions of location, population, Landscape scale.—As response variables at the mother tree (nested within population), block (nested landscape level, we used community descriptors (species within location), and first-order interactions among richness and Shannon diversity index), species-specific location and population, block and population, and abundances, guild abundances (i.e., sums of counts mother tree and location (all random effects). For across species within a given guild), and plant traits. As species richness we assumed a Poisson distribution with rare species will add little information regarding the a log-link function and for species occurrences a relative contribution of host plant genotype and binomial distribution with a logit link. The effect of location, we omitted them from analyses of species- mother tree was estimated using the animal model specific counts. Specifically, we chose to remove any (Lynch and Walsh 1998), corresponding to the half-sib species represented by fewer than 30 individuals, since breeding design. This allowed us to compare heritability this subset corresponded to 1% of the total material. For estimates between the experiments at the landscape scale species richness and species abundances, we assumed a (with clones) and the regional scale (with half-sibs). Poisson distribution with a log-link function; for mildew Variance components and heritability estimates.—For and herbivory (which were scored at the leaf-level as both the landscape and regional scale analyses, we presence/absence), we assumed a binomial distribution partitioned the total variance by dividing the individual with a logit-link function; and for Shannon diversity and variance components by the total variance. Among for plant traits, we assumed a normal distribution. We these, the broad-sense heritability is the relative variance 2 used multivariate models to quantify correlations among associated with genetic variation, H ¼ VG/(VG þ VL þ guilds and among species within guilds. To reduce the VL3G þ VR), where VG is the variance estimate for the number of plant traits, we conducted a principal genetic variation, VL the variance among locations, VLG components analysis on the 10 phenolic compounds thevarianceestimatefortheinteractionbetween detected (castalagin, castavaloninic acid, casuarictin, genotype and location, and VR the residual variation cocciferin D2, flavonoid glycosides, monogalloylglucose, (Shuster et al. 2006). pedunculagin, stachyurin/casuarinin, vescalagin, and a Prior distributions.—Following the default choices single, unidentified ellagitannin compound). As the first implemented in the MCMCglmm package, we assumed two components accounted for the main part of the a normal distribution, N(0, r2) with r2 ¼ 102 prior for September 2010 INSECT COMMUNITIES SHAPED BY SPACE 2665

FIG. 2. Examples of the effect of host plant genotype, location, and their interaction on the insect community. (A, B) Species richness at the landscape scale and the regional scale, respectively. (C, D) Examples of variation in the abundance of two selected leaf miner species (Phyllonorycter spp. and Tischeria ekebladella) at the landscape scale. Lines connect linear predictors for means of each genotype across each location, as derived from the posterior estimates for each combination of location and host plant genotype. Tree-speciﬁc species richness and counts are plotted within respective graphs as gray circles. To prevent overlap, the data points were slightly jittered in a horizontal and vertical direction. Individual common gardens are labeled as in Fig. 1 and arranged in order of increasing connectivity (see Fig. 5 for exact connectivity values). Note the log scales on all y-axes.

2 the overall mean, and the inverse-Wishart prior IW(V, plained by the pseudo-R ¼ 1–(Dres/Dnull), where Dres is t) for the variance components. Following the recom- the residual deviance left in the full model and Dnull is mendation by Hadfield (2009), we set the degrees of the deviance under the null hypothesis (Hagle and freedom t equal to the dimension of the matrix. The Mitchell 1992). Species with fewer than five occurrences inverse scale matrix V was set to V ¼ sI, where I is the were omitted from the individual analyses. identity matrix and s is a scalar, for which we used the value s ¼ 0.01. As it is well known that the estimates for RESULTS the variance components can be sensitive to the choice Variation in insect community structure of the prior (Gelman 2006), we also computed the at the landscape scale posterior densities assuming the values s ¼ 0.0001 and s ¼ 1. Each model was run for a minimum of 500 000 Considering community structure, species richness iterations, using a minimum burn-in of 50 000 iterations. was clearly affected by the location of the tree within the Spatial connectivity.—We examined the effect of landscape, whereas host plant genotype and G 3 E spatial connectivity on community structure by fitting interaction accounted for little, if any, variation (Figs. the linear regressions log(species richness þ 0.5) ; 2A and 3A). Much variation in the community structure log(spatial connectivity), and Shannon diversity ; remained unexplained, and a large fraction of this log(spatial connectivity). The amount of variation unexplained variation was accomodated by the inherent explained was described by R2 values. Occurrence data Poisson error. Estimates of the relative effects of host were modeled by species-specific logistic regressions, plant genotype and location were for most cases with incidence as a function of log(spatial connectivity). insensitive to the choice of the prior (see Appendix B Here, we quantified the proportion of deviance ex- for results obtained for alternative priors). 2666 AYCO J. M. TACK ET AL. Ecology, Vol. 91, No. 9

FIG. 3. Percentage of variation attributed to genotype, location, and their interaction. Results are shown for several response variables at the (A–D) landscape scale and (E) regional scale. Vertical bars reﬂect 95% highest posterior density intervals, with the median shown as a circle. For each response variable, either location or genotype is colored black if that factor is larger (with at least 95% posterior probability) than the other factor. Guild abbreviations are as in Table 1; R, species richness; D, species diversity. Species numbers along x-axis in panels (C) and (E) are identiﬁed in Table 1.

At the level of individual guilds, each guild responded abundance of each of the ﬁve leaf miner species was differently: leaf miners, free-feeding herbivores, and explained by spatial location, while among the gallers, mildew were mainly affected by location, while much of only the asexual generation of Neuroterus quercusbacca- the variation in the distribution of gall-inducing insects rum (Appendix C) was signiﬁcantly affected by the and of ‘‘other’’ species with a mixed life history remained location of the tree (Fig. 3C). The leaf folder Ancylis unexplained (Fig. 3B). The same pattern emerged for the mitterbacheriana showed a strong response to spatial individual species: the majority of variation in the location, whereas variation in the distribution of Helio- September 2010 INSECT COMMUNITIES SHAPED BY SPACE 2667

FIG. 4. Correlations in abundances (A) among guilds and (B–D) among species within guilds. The correlations are shown separately for the four variance components included in the model, i.e., across locations, host plant genotypes, combinations of genotypes and locations, and across residuals. Estimates of the median correlation coefficients are derived from multivariate models of data at the landscape scale. An asterisk denotes that a given correlation coefficient is positive (or negative) with at least 0.95 posterior probability. Abbreviations are as in Table 1. zela sericiella and Bucculatrix ulmella remained largely Correlations among guilds were detected across loca- unexplained (Appendix C). Overall, the most striking tions, where we found a positive association between pattern observed across individual species was their mildew and damage by free-feeding insects (Fig. 4A). relatively weak response to both host plant genotype Among species within guilds, patterns were variable: and the interaction between genotype and location. As a while leaf miner species per se were all strongly affected result, the broad-sense heritabilities were estimated to be by location (Fig. 3C), their responses were not uniform, ,0.1 for all species and community descriptors, except resulting in both positive (Fig. 2C, D) and negative for Andricus inflator (for which H2 was 45%). correlations among individual species pairs (Fig. 4B). In principle, guilds and species within guilds could Furthermore, individual members of the gall-inducing correlate in their abundances either across locations, guild and of the mixed guild ‘‘other’’ did not show any across host plant genotypes, across combinations of host detectable correlations. Overall, across host plant plant genotypes and locations, or across residuals. genotypes and G 3 E combinations, we detected no 2668 AYCO J. M. TACK ET AL. Ecology, Vol. 91, No. 9

corresponding to the natural distribution of oak in Finland; see Fig. 1A), results remained strikingly similar: much of the variation in the occurrence of the species was explained by the location of the common garden (Figs. 2B and 3E), and heritabilities were generally low (all H2 , 0.1). Notably, a small-scale effect of location, i.e., variation among blocks (;0.2 ha) within common gardens (;5 ha), accounted for a high percentage of the variation (Fig. 3E).

DISCUSSION Few previous studies have tried to disentangle variation in community structure caused by host plant genotype, location, and their interaction at realistic spatial scales. Our study indicates that landscape context is a major force in structuring insect communities. In FIG. 5. The effect of spatial connectivity on species particular, we found that the spatial connectivity of the richness. Plotted are tree-specific observations for each of 89 trees distributed throughout the landscape. The fitted line is host tree explained a large fraction of the variation in the based on the estimates from the regression log(species richness insect community. By comparison, host plant genotype þ 0.5) ; log(spatial connectivity) . The five species that showed and the G 3 E interaction had minor effects at both the no significant relationship between occurrence and spatial landscape and the regional scales. connectivity (P . 0.1) are not depicted. Connectivity values of the six common gardens established at the landscape level are The small amount of variation attributed to host plant shown as vertical arrows on the x-axis (see Fig. 1 for site genotype and the associated low heritabilities appear to locations). Species numbers are identified in Table 1. be in striking contrast with results reported from other study systems. In a recent review of community genetics significant correlations either within or among guilds, by Whitham et al. (2006), all eight study systems while some negative and positive correlations were examined revealed a significant effect of host plant found among locations and among residuals (Fig. 4). genotype on the insect community. We propose that two factors could contribute to the apparent particularity of Variation in host plant attributes at the landscape scale our study system. While insect community structure varied little among First, most experiments published so far have been host plant genotypes, plant traits revealed a different conducted as carefully controlled experiments, which pattern: here, variation was mainly attributed to the have eliminated much of the variation associated with genotype of the host plant (Fig. 3D), with phenology being spatial location. In these studies, host plant genotypes particularly strongly associated with host plant genotype have often been collected from a large geographical area (H2 ¼ 48%). The exception was the second principal to maximize genotypic variation, whereas the experi- components (PC) axis of the phenolic compounds, which ment itself has been conducted in a single common was mainly attributed to location. Little variation in the garden to minimize environmental variation (e.g., Fritz host plant traits was attributed to the interaction between and Price 1988, Dungey et al. 2000, Ito and Ozaki 2005, host plant genotype and location (Fig. 3D). Wimp et al. 2005, Shuster et al. 2006, Crutsinger et al. 2009). Hence, while these studies have convincingly Spatial connectivity shown the potential for host plant genotype to structure Spatial connectivity explained 32% of the variation in the insect community, the finer contribution of genotype species richness and 21% of the variation in species vs. location observed may not be generalizable to other diversity on the island Wattkast, with well-connected settings. On the contrary, the presence of a significant trees sustaining higher species richness and diversity block effect in many of these studies (Maddox and Root than isolated trees (Fig. 5). The majority of individual 1987, Whitham et al. 1994, Fritz et al. 1998, Hoch- species showed the same positive relationship between wender and Fritz 2004, Ito and Ozaki 2005, Johnson the spatial connectivity of a tree and the presence of the and Agrawal 2005, 2007, Wimp et al. 2005, Barbour et species (P , 0.1 for 14 out of 19 species; Fig. 6). In these al. 2009) suggests that the role of environmental species, spatial connectivity explained between 4% and variation may actually swamp the effect of host plant 32% of the variation. genotype at spatial scales deviating from the confines of a specific common garden. Indeed, single-species studies Variation in insect community structure have shown that abundances are strongly affected by the at the regional scale environment as soon as distances among common When extending the spatial scale from the landscape gardens become larger than several tens of meters (Fritz scale (;5km2) to the regional scale (;10 000 km2, 1990, Quiring and Butterworth 1994, Stiling 1994, September 2010 INSECT COMMUNITIES SHAPED BY SPACE 2669

FIG. 6. The effect of spatial connectivity on species occurrence. Plotted are tree-specific observations for each of 89 trees distributed throughout the landscape: 0 ¼ absence; 1 ¼ presence, where data points for each species are offset for illustrative purposes. The fitted lines are based on estimates from species-specific models: logit(occurrence) ; log(spatial connectivity). The five species that showed no significant relationship between occurrence and spatial connectivity (P . 0.1) are not depicted. Species numbers are identified in Table 1.

Stiling and Rossi 1995, 1996, Rossi and Stiling 1998, Importantly, our results seemed independent of the Stiling and Bowdish 2000, Kittelson 2004). spatial scale: the relative effect sizes of host plant The second factor potentially inflating the effect of genotype and location were similar when comparing genotype in previous studies is that they tend to focus on common gardens at the landscape scale (5 km2) and the plant taxa with particularly high levels of intraspecific regional scale (10 000 km2). Despite substantial levels of genotypic variation, including a large number of hybrid genetic variation both among and within oak popula- species (Boecklen and Spellenberg 1990, Aguilar and tions in southern Finland (Mattila et al. 1994, Lahtinen Boecklen 1992, Fritz et al. 1994, Dungey et al. 2000, et al. 1997, Vakkari et al. 2006), we did not detect any Fritz et al. 2003, Hochwender and Fritz 2004, Drew and effect of host plant population, nor of the specific Roderick 2005, Wimp et al. 2005, Tovar-Sańchez and mother tree within a population. While in itself this Oyama 2006a, Barbour et al. 2009). In these groups, the finding adds further support to the notion that host host plant genotype may then be especially liable to plant genotype plays a minor role in structuring the make a difference, and hence the results may not be insect community, it contrasts with two other studies applicable to a broader range of communities. addressing patterns at multiple spatial scales. Johnson 2670 AYCO J. M. TACK ET AL. Ecology, Vol. 91, No. 9 and Agrawal (2005) and Bangert et al. (2006, 2008) microevolutionary processes and spatial location (Tack reported that while the effect of host plant genotype was and Roslin 2010). clear at small spatial scales (within common gardens), it The genetic correlations that we estimated did not was partly (but not fully) swamped by environmental differ significantly from zero, and different species were variation with increasing spatial scale. Nonetheless, this consequently no more likely to reach high abundances apparent discrepancy may perhaps be attributed to a on the same genotypes than on different ones. While this methodological difference: in the previous studies, the result is not surprising given the low heritability maximum amount of genotypic variation was present estimates, it is in striking contrast to previous studies already at the smallest spatial scale. In our study, we on arthropods, which have frequently detected positive simultaneously increased the spatial scale of the associations among herbivores across host plant geno- environment and that of host plant provenances (Fig. types (Fritz 1990, Maddox and Root 1990, Roche and 1), finding no major amount of variation explained by Fritz 1997, Leimu and Koricheva 2006, Johnson and genotype at any spatial scale. Agrawal 2007). In contrast to the lack of correlations But why do communities differ among locations? across plant genotypes, individual leaf miner taxa Differences in community structure among locations covaried both positively and negatively across locations. may be attributed to either differences in host plant This suggests that some species pairs co-occur more quality (due to abiotic conditions) or to spatial often than predicted by chance on certain host plants in processes. Of these hypotheses, we found little support the landscape, increasing the potential for negative and for location-specific differences in host plant traits: the positive interactions. Regarding negative interactions, examined plant traits were only weakly affected by the importance of direct and plant-mediated competi- location, in strong contrast with the response of the tion in leaf miner communities has recently been herbivore community (Fig. 3). Overall, the discrepancy questioned (Tack et al. 2009). Nonetheless, indirect between patterns of variation in host plant traits and in interactions caused by parasitoid overlap may still lead herbivore response suggests that the plant traits to apparent competition among certain leaf miner measured here fail to account for variation in insect species (Rott and Godfray 2000, Morris et al. 2004, abundances among locations. By contrast, the second Hirao and Murakami 2008). Regarding positive inter- hypothesis gained more support: spatial connectivity actions, recent work has shown the potential for explained 32% of the variation in species richness in our processes such as (apparent) mutualism among herbi- observational data, with the majority of species being vores (Van Zandt and Agrawal 2004, Roslin and Roland less likely to occur on the isolated host trees than on the 2005), the incidence of which deserves closer scrutiny in well-connected trees (Figs. 5, 6). Indeed, this pattern was work to come. also reflected in the common gardens, where trees in the In conclusion, this study reveals that, in our study garden with the lowest connectivity also harbored the system, both community structure and species distribu- fewest insects (Fig. 2A, C, D). tions are strongly affected by the spatial setting of the As defined by us, spatial connectivity is roughly host tree. This pattern is consistent across multiple reflective of potential immigration from other sites. How species and across two spatial scales, and it results in immigration contributes to structure local communities, spatial variation in the intensity with which species co- and how it will modulate the importance of host plant occur and thus potentially interact with one another. genotype, is illustrated by another analysis. Exploring More generally, we then suggest that spatial processes how plant-feeding insects adapt to their individual host may be more important in structuring insect communi- plants, Tack and Roslin (2010) have shown that within ties than hitherto assumed within the emerging field of our study system, the strength of local adaptation in six community genetics and that in real landscapes, spatial different insect species varies inversely with the fraction impacts might relegate host plant genotype to a of immigrants in the insect population. When immigra- secondary role. We hope that this claim will be tion is low, insect populations adapt to their host tree substantiated, or refuted, by further work in multiple individual, and when immigration is high, local popu- study systems. To achieve this goal, we recommend that lations remain in a maladapted state. Importantly, the future studies be based on establishing multiple common current analysis does not resolve the contrast between gardens in explicit spatial settings, on the random the ‘‘local’’ and the ‘‘foreign’’ genotype, the very focus of sampling of host plant genotypes and environments, Tack and Roslin (2010). Instead, the signature of local and on matching the spatial scale of the experiment with adaptation is captured in the genotype 3 location the spatial scale of host plant origins. interaction. While this variance component is generally small as compared to variation among sites per se (Fig. ACKNOWLEDGMENTS 3), differences in insect performance on individual trees We are grateful to the Haapastensyrja¨Tree Breeding Station are still strong enough to elicit a local evolutionary and in particular Tommi Salmela, Piritta Lohela, and Raimo Jaatinen for grafting, transporting, and maintaining the grafts. response. Hence, while small overall, genotypic effects Many thanks go to the fieldworkers who surveyed the trees for may still be biologically relevant in the current system the presence of species (in particular to Elena Blasco Martıń), and suffice to create a systematic relation between and to Sofia Gripenberg for her comments on the manuscript. September 2010 INSECT COMMUNITIES SHAPED BY SPACE 2671

We further thank Juha-Pekka Salminen and Piia Saarinen for Graham, J. H., E. D. McArthur, and D. C. Freeman. 2001. analyzing the phenolics. This study was supported by the Narrow hybrid zone between two subspecies of big sagebrush Academy of Finland (grant numbers 124242 and 21347 to O. (Artemisia tridentata: Asteraceae). XII. Galls on sagebrush in Ovaskainen and grants 111704 and 213457 to T. Roslin). a reciprocal transplant garden. Oecologia 126:239–246. Gripenberg, S., O. Ovaskainen, E. Morrie¨n, and T. Roslin. LITERATURE CITED 2008. Spatial population structure of a specialist leaf-mining Aguilar, J. M., and W. J. Boecklen. 1992. Patterns of herbivory moth. Journal of Animal Ecology 77:757–767. in the Quercus grisea 3 Quercus gambelii species complex. Hadfield, J. D. 2009. MCMC methods for multi-response Oikos 64:498–504. generalised linear mixed models: the MCMCglmm R Antonovics, J. 1992. Towards community genetics. Pages 426– package. Version 1.03. hhttp://cran.r-project.org/web/ 449 in R. S. Fritz and E. L. Simms, editors. Plant resistance packages/MCMCglmm/index.htmli to herbivores and pathogens: ecology, evolution and genetics. Hagle, T. M., and G. E. Mitchell. 1992. Goodness-of-fit University of Chicago Press, Chicago, Illinois, USA. measures for probit and logit. American Journal of Political Bangert, R. K., G. J. Allan, R. J. Turek, G. M. Wimp, N. Science 36:762–784. Meneses, G. D. Martinsen, P. Keim, and T. G. Whitham. Hanski, I. 1999. Metapopulation ecology. Oxford University 2006. From genes to geography: a genetic similarity rule for Press, Oxford, UK. arthropod community structure at multiple geographic Hirao, T., and M. Murakami. 2008. Quantitative food webs of scales. Molecular Ecology 15:4215–4228. lepidopteran leafminers and their parasitoids in a Japanese Bangert, R. K., E. V. Lonsdorf, G. M. Wimp, S. M. Shuster, D. deciduous forest. Ecological Research 23:159–168. Fischer, J. A. Schweitzer, G. J. Allan, J. K. Bailey, and T. G. Hochwender, C. G., and R. S. Fritz. 2004. Plant genetic Whitham. 2008. Genetic structure of a foundation species: differences influence herbivore community structure: evi- scaling community phenotypes from the individual to the dence from a hybrid willow system. Oecologia 138:547–557. Holyoak, M., M. A. Leibold, and R. D. Holt, editors. 2005. region. Heredity 100:121–131. Metacommunities: spatial dynamics and ecological commu- Barbour, R. C., J. M. O’Reilly-Wapstra, D. W. De Little, G. J. nities. University of Chicago Press, Chicago, Illinois, USA. Jordan, D. A. Steane, J. R. Humphreys, J. K. Bailey, T. G. Hunter, M. D. 1992. A variable insect–plant interaction: the Whitham, and B. M. Potts. 2009. A geographic mosaic of relationship between tree budburst phenology and popula- genetic variation within a foundation tree species and its tion levels of insect herbivores among trees. Ecological community-level consequences. Ecology 90:1762–1772. Entomology 16:91–95. Boecklen, W. J., and R. Spellenberg. 1990. Structure of Ito, M., and K. Ozaki. 2005. Response of a gall wasp herbivore communities in two oak (Quercus spp.) hybrid community to genetic variation in the host plant Quercus zones. Oecologia 85:92–100. crispula: a test using half-sib families. Acta Oecologica 27:17– Cornelissen, T., and P. Stiling. 2008. Clumped distribution of 24. oak leaf miners between and within plants. Basic and Applied Johnson, M. T. J., and A. A. Agrawal. 2005. Plant genotype Ecology 9:67–77. and environment interact to shape a diverse arthropod Crawley, M. J., and M. Akhteruzzaman. 1988. Individual community on evening primrose (Oenothera biennis). Ecology variation in the phenology of oak trees and its consequences 86:874–885. for herbivorous insects. Functional Ecology 2:409–415. Johnson, M. T. J., and A. A. Agrawal. 2007. Covariation and Crutsinger, G. M., M. W. Cadotte, and N. J. Sanders. 2009. composition of arthropod species across plant genotypes of Plant genetics shapes inquiline community structure across evening primrose, Oenothera biennis. Oikos 116:941–956. spatial scales. Ecology Letters 12:285–292. Johnson, M. T. J., and J. R. Stinchcombe. 2007. An emerging Drew, A. E., and G. K. Roderick. 2005. Insect biodiversity on synthesis between community ecology and evolutionary plant hybrids within the Hawaiian silversword alliance biology. Trends in Ecology and Evolution 22:250–257. (Asteraceae: Heliantheae-Madiinae). Environmental Ento- Kittelson, P. M. 2004. Sources of variation in insect density on mology 34:1095–1108. Lupinus arboreus Sims: effects of environment, source Dungey, H. S., B. M. Potts, T. G. Whitham, and H.-F. Li. population and plant genotype. American Midland Natural- 2000. Plant genetics affects arthropod community richness ist 152:323–335. and composition: evidence from a synthetic eucalypt hybrid Lahtinen, M.-L., P. Pulkkinen, and M.-L. Helander. 1997. population. Evolution 54:1938–1946. Potential gene flow by pollen between English oak (Quercus Feeny, P. 1970. Seasonal changes in oak leaf tannins and robur L.) stands in Finland. Forestry Studies 28:46–50. nutrients as a cause of spring feeding by winter moth Leimu, R., and J. Koricheva. 2006. A meta-analysis of genetic caterpillars. Ecology 51:565–581. correlations between plant resistances to multiple enemies. Fritz, R. S. 1990. Effects of genetic and environmental variation American Naturalist 168:E15–E37. on resistance of willow to sawflies. Oecologia 82:325–332. Levin, S. A. 1992. The problem of pattern and scale in ecology. Fritz, R. S., C. G. Hochwender, S. J. Brunsfeld, and B. M. Ecology 73:1943–1967. Roche. 2003. Genetic architecture of susceptibility to Littell, R. C., G. A. Milliken, W. W. Stroup, R. D. Wolfinger, herbivores in hybrid willows. Journal of Evolutionary and O. Schabenberger. 2006. SAS for mixed models. Second Biology 16:1115–1126. edition. SAS Institute, Cary, North Carolina, USA. Fritz, R. S., C. M. Nichols-Orians, and S. J. Brunsfeld. 1994. Lynch, M., and B. Walsh. 1998. Genetics and analysis of Interspecific hybridization of plants and resistance to quantitative traits. Sinauer, Sunderland, Massachusetts, herbivores: hypotheses, genetics, and variable responses in a USA. diverse herbivore community. Oecologia 97:106–117. Maddox, G. D., and R. B. Root. 1987. Resistance to 16 diverse Fritz, R. S., and P. W. Price. 1988. Genetic variation among species of herbivorous insects within a population of plants and insect community structure: willows and sawflies. goldenrod, Solidago altissima: genetic variation and herita- Ecology 69:845–856. bility. Oecologia 72:8–14. Fritz, R. S., B. M. Roche, and S. J. Brunsfeld. 1998. Genetic Maddox, G. D., and R. B. Root. 1990. Structure of the variation in resistance of hybrid willows to herbivores. Oikos encounter between goldenrod (Solidago altissima) and its 83:117–128. diverse insect fauna. Ecology 71:2115–2124. Gelman, A. 2006. Prior distributions for variance parameters in Mattila, A., A. Pakkanen, P. Vakkari, and J. Raisio. 1994. hierarchical models (comment on article by Browne and Genetic variation in English oak (Quercus robur) in Finland. Draper). Bayesian Analysis 1:515–534. Silva Fennica 28:251–256. 2672 AYCO J. M. TACK ET AL. Ecology, Vol. 91, No. 9

Mopper, S. 2005. Phenology: how time creates spatial structure Southwood, T. R. E. 1961. The number of species of insect in endophagous insect populations. Annales Zoologici associated with various trees. Journal of Animal Ecology 30: Fennici 42:327–333. 1–8. Morris, R. J., O. T. Lewis, and H. C. J. Godfray. 2004. Stiling, P. 1994. Coastal insect herbivore populations are Experimental evidence for apparent competition in a tropical strongly influenced by environmental variation. Ecological forest food web. Nature 428:310–313. Entomology 19:39–44. Neuhauser, C., D. A. Andow, G. E. Heimpel, G. May, R. G. Stiling, P., and T. I. Bowdish. 2000. Direct and indirect effects Shaw, and S. Wagenius. 2003. Community genetics: expand- of plant clone and local environment on herbivore abun- ing the synthesis of ecology and genetics. Ecology 84:545– dance. Ecology 81:281–285. 558. Stiling, P., and A. M. Rossi. 1995. Coastal insect herbivore Ossiannilsson, F. 1992. The Psylloidea (Homoptera) of communities are affected more by local environmental Fennoscandia and Denmark. E. J. Brill, Leiden, The conditions than by plant genotype. Ecological Entomology Netherlands. 20:184–190. Price, P. W. 1983. Hypotheses on organization and evolution in Stiling, P., and A. M. Rossi. 1996. Complex effects of genotype herbivorous insect communities. Pages 559–598 in R. F. and environment on insect herbivores and their enemies. Denno and M. S. McClure, editors. Variable plants and Ecology 77:2212–2218. herbivores in natural and managed systems. Academic Press, Tack, A. J. M., O. Ovaskainen, P. J. Harrison, and T. Roslin. New York, New York, USA. 2009. Competition as a structuring force in leaf miner Quiring, D. T., and E. W. Butterworth. 1994. Genotype and communities. Oikos 118:809–818. environment interact to influence acceptability and suitability Tack, A. J. M., and T. Roslin. 2010. Overrun by the neighbors: of white spruce for a specialist herbivore, Zeriaphera landscape context affects strength and sign of local adapta- canadensis. Ecological Entomology 19:230–238. tion. Ecology 91, in press. Roche, B. M., and R. S. Fritz. 1997. Genetics of resistance of Tovar-Sańchez, E., and K. Oyama. 2006a. Community Quercus Salix sericea to a diverse community of herbivores. Evolution structure of canopy arthropods associated to crassifolia 3 Quercus crassipes complex. Oikos 112:370–381. 51:1490–1498. Tovar-Sańchez, E., and K. Oyama. 2006b. Effect of hybridiza- Roslin, T., A.-L. Laine, and S. Gripenberg. 2007. Spatial tion of the Quercus crassifolia 3 Quercus crassipes complex population structure in an obligate plant pathogen colonizing on the community structure of endophagous insects. Oeco- oak Quercus robur. Functional Ecology 21:1168–1177. logia 147:702–713. Roslin, T., and J. Roland. 2005. Competitive effects of the Vakkari, P., A. Blom, M. Rusanen, J. Raisio, and H. Toivonen. forest tent caterpillar on the gallers and leaf-miners of 2006. Genetic variability of fragmented stands of pedunculate trembling aspen. Ecoscience 12:172–182. oak (Quercus robur) in Finland. Genetica 127:231–241. Roslin, T., and J.-P. Salminen. 2008. Specialization pays off: Van Zandt, P. A., and A. A. Agrawal. 2004. Specificity of contrasting effects of two types of tannins on oak specialist induced plant responses to specialist herbivores of the and generalist moth species. Oikos 117:1560–1568. common milkweed Asclepias syriaca. Oikos 104:401–409. Rossi, A. M., and P. Stiling. 1998. The interactions of plant West, C. 1985. Factors underlying the late seasonal appearance clone and abiotic factors on a gall-making midge. Oecologia of the lepidopterous leaf-mining guild on oak. Ecological 116:170–176. Entomology 10:111–120. Rott, A. S., and H. C. J. Godfray. 2000. The structure of a Whitham, T. G., P. A. Morrow, and B. M. Potts. 1994. Plant leafminer–parasitoid community. Journal of Animal Ecology hybrid zones as centers of biodiversity: the herbivore 69:274–289. community of two endemic Tasmanian eucalypts. Oecologia Salminen, J.-P. 2003. Effects of sample drying and storage, and 97:481–490. choice of extraction solvent and analysis method on the yield Whitham, T. G., et al. 2003. Community and ecosystem of birch leaf hydrolyzable tannins. Journal of Chemical genetics: a consequence of the extended phenotype. Ecology Ecology 29:1289–1305. 84:559–573. Salminen, J.-P., T. Roslin, M. Karonen, J. Sinkkonen, K. Whitham, T. G., et al. 2006. A framework for community and Pihlaja, and P. Pulkkinen. 2004. Seasonal variation in the ecosystem genetics: from genes to ecosystems. Nature content of hydrolyzable tannins, flavonoid glycosides, and Reviews Genetics 7:510–523. proanthocyanidins in oak leaves. Journal of Chemical Wiens, J. A. 1989. Spatial scaling in ecology. Functional Ecology 30:1693–1711. Ecology 3:385–397. Shuster, S. M., E. V. Lonsdorf, G. M. Wimp, J. K. Bailey, and Wimp, G. M., G. D. Martinsen, K. D. Floate, R. K. Bangert, T. G. Whitham. 2006. Community heritability measures the and T. G. Whitham. 2005. Plant genetic determinants of evolutionary consequences of indirect genetic effects on arthropod community structure and diversity. Evolution 59: community structure. Evolution 60:991–1003. 61–69.

APPENDIX A Details of the experimental design (Ecological Archives E091-190-A1).

APPENDIX B Effect of prior choice on the relative variance estimates (Ecological Archives E091-190-A2).

APPENDIX C Percentage of variance attributed to genotype, location, and their interaction for members of the guilds ‘‘gallers’’ and ‘‘others’’ (Ecological Archives E091-190-A3). 1 Appendix A. Detailed experimental design.

2 In the spring of 2004, branch tips were collected from ten large trees in different

3 parts of the island of Wattkast. These shoots were subsequently grafted onto

4 randomly selected root stocks grown from acorns collected across several sites. Any

5 foliage and branches sprouting from the root stock were successively pruned,

6 resulting in a tree crown composed only of the grafted genotype. The grafts were

7 grown for three summers, until large enough (ca 1.5 meter) to be used in the field.

8 On 23 April 2007, well before leaf flush, the grafted treelets (n=172) were

9 transported back to Wattkast and placed in common gardens, located under the

10 trees from which the grafted treelets originated. Five trees died during the early

11 spring, resulting in slight variation in the number of genotypes among gardens. Each

12 garden still consisted of a similar number of 25-31 treelets (cf. Table A1). Within

13 gardens, the trees were randomly placed in a regular grid at distances of ca. 1 m

14 from each other, with treelets of different genotype distributed at random positions

15 (Fig. A1).

16 Due to previous mortality between 2004 and 2007, there was some variation

17 in the exact number of treelets alive per genotype in 2007. Hence, to increase the

18 number of replicates per genotype and common garden, we departed from a fully

19 reciprocal design by establishing common gardens around only six of the ten mother

20 trees. For this purpose, we selected the six trees of which the largest numbers of

21 grafted treelets were available. To meet the rationale of another study (Tack and

22 Roslin, in press), in each common garden we also placed a slightly higher number of

23 treelets representing the “local” genotype (i.e. the genotype of the larger tree under

24 which the garden was positioned) than would follow from distributing treelets

25 randomly among gardens. This resulted in a mean of 7.7 replicates of the local

1 genotype (range n=4-11 trees) and a mean of 20.2 trees of other genotypes

2 combined (range n=18-23 trees; for exact numbers, cf. table A1).

1 Table A1. Number of treelets per genotype and common garden, as used in the

2 analyses (n=167). Here, gardens and oak genotypes are named as in Fig. 1B of the

3 paper proper. Unlabeled columns represent genotypes of mother trees next to which

4 no common gardens were constructed, and which hence lack individual letter codes

5 (cf. the white squares in Fig. 1B).

Genotype A B C D E F TOTAL Garden A 11 2 3 2 4 2 1 1 1 2 29 B 4 4 3 2 4 2 3 2 1 2 27 C 4 1 11 2 4 2 2 2 1 2 31 D 4 2 0 7 4 2 1 2 1 2 25 E 4 2 3 2 9 2 2 1 1 2 28 F 4 2 3 2 4 4 3 2 1 2 27 TOTAL 31 13 23 17 29 14 12 10 6 12 167

Genotype E D C F B A

}1 m

}

1 m

1 Fig. A1. The design of an individual common garden. This graph shows the

2 distribution of different genotypes within Garden E – for the labeling of individual

3 genotypes, cf. Fig. 1B in the paper proper. Note that Genotype E (white)

4 corresponds to the genotype of the larger tree standing next to this common garden.

5 Genotypes lacking letter codes represent mother trees next to which no common

6 garden was constructed (cf. white squares in Fig. 1B).

Appendix B. Effect of prior choice on the relative variance estimates

To examine the sensitivity of the results to the choice of the prior, we changed the default value s=0.01 (see Materials and methods) to either s=0.0001 or to s=1. In the multivariate models, the value s=0.0001 led to numerical problems (ill-conditioned G/R structure), so we could use this value only for the univariate models. The ranking of relative variance components is relatively insensitive to the prior (compare Fig. 3 with Figs. B1 and B2 below), but the differences among the different variance components become more pronounced with small values of s. The reason for this can be seen from Figure B3: the data are informative about particular variance components being large (e.g. the residual and location components in Figure B3), but there is little statistical power to tell that the other variance components are necessarily small.

Figure B1, corresponding to Fig. 3 but for s=0.0001. The circles give the median estimate

(error bars the 95% highest posterior density interval) for the percentage of variance explained by the different variance components. The cases for which no results are shown correspond to multivariate models for which the estimation failed (ill-conditioned G/R structure) for this choice of the prior.

Figure B2, corresponding to Fig. 3 but for s=1. The circles give the median estimate (error bars the 95% highest posterior density interval) for the percentage of variance explained by the different variance components.

Figure B3, Violin graphs corresponding to models with A) s=0.0001, B) s=0.01 and C) s=1, showing the prior distribution (Inverse Wishart prior - IW) and the posterior distributions of the variance estimates for the Shannon diversity at the landscape scale. Y-axis is in the log-scale.

Appendix C (Figure C1). Percentage of variance attributed to genotype, location and their interaction for members of the guilds A) Gallers and B) Others. Vertical bars reflect 95% highest posterior density intervals, with the median shown as a circle. For each response variable, either location or genotype is colored black if that factor is larger (with at least 95% posterior probability) than the other factor. Species

17 (Neuroterus quercusbaccarum) is treated as two taxa, as the asexual and sexual generation differ substantially in terms of both biology and average abundances.

Species abbreviations as in table 1.