The 22 different genotypes of B. pendula were cloned during spring 1998 from randomly selected B. pendula trees taken from a naturally regenerated B. pendula - B. pubescens Ehr forest in Punkaharju, southeastern Finland (61°48′ N, 29°18′ E), to study genetic variation in phenology, growth, reproduction and resistance-related traits among individual birch trees ²⁵ . Sampling was stratified random sampling: six spots where forest lift could be transferred were first selected around the forest, and 2–5 trees within the reach of forest lift in each spot were then randomly (by throwing a coin) selected for our study purposes. B. pendula is predominantly a sexual species, but genotypes can be cloned for study purposes or for plantations using standard tissue-culture methods ³² . Cloned B. pendula saplings were planted at the growing sites (i.e. common gardens, each approximately 0.25 ha) in June 1999 to find out the degree to which the genotype and environment affect birch traits and to test how genotypes differ in their response to the environment ²⁶ . The Kuikanniitty study site (61°47′ N, 29°21′ E) is an abandoned cultivated field and the Parikkala study site (61°36′ N, 29°36′ E) is Myrtillus type forest ³³ . Soil type was defined as fine sandy till for both sites ²⁶ . The distance between these sites was around 30 km and they were situated at approximately the same altitude (Kuikanniitty 79 m and Parikkala 93 m above sea level). Thus, the mean summer (June–August) temperatures were very similar at these sites: in 2002 mean temperatures were 17.6°C and 17.9°C and in 2003 they were 15.9°C and 15.6°C in Kuikanniitty and Parikkala, respectively. Both study sites were divided into six blocks, each of which included four saplings from each genotype. To prevent edge effects, the experimental saplings were surrounded by one row of extra saplings. From each block, one of the four saplings of a total of 22 genotypes was randomly selected for the present study in order to have six replicates per genotype.

In addition, we collected additional data from Kuikanniitty in 2003 to investigate the effect of previous insect herbivory on insect community structure and abundance, and surveyed one extra sapling from each block and genotype. These extra saplings were protected from insect herbivory in the previous growing season by regular sprayings with synthetic pyrethrin ²³ , which has no direct or side effects on the growth or chemistry of birch seedlings ³⁴ .

The insect herbivore community of each sapling was assessed by surveying the abundance of 23 (in Parikkala 2002) or 27 (in Kuikanniitty 2002–2003, and Parikkala 2003) insect taxa from diverse orders (Lepidoptera, Hymenoptera, Coleoptera, Diptera, Hemiptera; Table 1). These taxa were generally the most abundant taxa in both sites. However, species that were rare in both sites were included in the surveys as well. Species identifications were undertaken following Saalas ³⁵ species identification guide, using several web pages ( http://www.funet.fi/pub/sci/bio/life/insecta/index.html; http://www.leafmines.co.uk/index.htm; http://www.bladmineerders.nl/; http://www.nrm.se/) with the assistance of specialists. Euceraphis betulae eggs were counted from the side of twelve (2002) or eight (2003) topmost buds in April before budburst. However, in 2002 E. betulae eggs were counted from two saplings per genotype per block (sum of the eggs on the sides of 24 buds was used in the analysis), because regular sprayings with synthetic pyrethrin on the other sapling was started only after egg counts in both sites. Trichiosoma sp. pupae were counted in April/May when the timing of budburst of the same saplings was observed (Possen, submitted manuscript). The abundance of Eriocrania sp. was determined at the end of June, Deporaus betulae at the beginning of July and Heteropteran 1 (sap sucker) in August. Croesus septentrionalis larval colonies and the number of larvae in each colony were recorded along with Eriocrania and D. betulae measurements in both years. The abundance of all other insects were determined indirectly by counting damaged leaves at the beginning of September in both years, since the damage caused by most of the surveyed taxa remained identifiable for a long time after the initial damage.

In general, the insect abundance in 2002 was determined by surveying the whole sapling. The mean height of these saplings at the end of 2002 was 253 ± 4.3 cm (mean ± SE) in Kuikanniitty and 227 ± 3.8 cm in Parikkala. Because B. pendula genotypes differ in their height and diameter growth ²³ and large saplings may harbor more insects than smaller saplings, we determined the whole sapling “surface area” and used it as a covariate hereafter called “size index” in statistical analysis. Surface area was determined by photographing each sapling sideways from their southern side against a white background, converting the picture to a black and white silhouette picture in Adobe Photoshop 7.0 and determining the number of black pixels (i.e. leaf and branch area) within the picture. The number of pixels was converted to m ² using the number of pixels of a known area as a reference. The amount of pixels significantly (p < 0.001) explained over 73% of the sapling volume [Y = (3.14 * {base diameter/2} ² * height)/3] in both sites. The abundance of Phyllonorycter cavella, Phyllonorycter sp. 1, Parornix betulae and Parornix sp. was not examined on the whole sapling, but was determined as the damage (i.e. number of mines per each species) found within a period of 30 seconds. The period of time (30 sec) was chosen so that even the smallest saplings had leaves uncounted when the time was up.

Since the method of assessing herbivore abundance/resistance by time counts has been successfully used in the past ^{35, 36} we decided to use time counts to determine the abundance of almost all taxa (except E. betulae, Trichiosoma sp. and C. septentrionalis) in 2003. The same person undertook all surveys. The abundance of easily visible damage (large mines and rolls) of Eriocrania sp., D. betulae and Heteropteran 1 were determined as the number of damaged areas found within a period of 30 seconds. Epinotia tetraquetrana “knobs” in the branches of the saplings were counted within a period of 20 seconds in 2003 starting at the top of the tree. Since the “knobs” in the branches remain visible for years and we did not separate different year’s growth while surveying, the values represent the accumulation of E. tetraquetrana damage during the last few years. Therefore the same values were used in both years’ insect community analyses. The abundance of all other 20 taxa in 2003 was determined within a single time count of each sapling at the beginning of September. To examine a similar proportion of each sapling, they were divided into three size categories according to their height and number of leaves. Small saplings (average height 2.8 and 3.2 m in Parikkala and Kuikanniitty, respectively) were surveyed for 30 seconds, average sized saplings (3.5 and 3.9 m in Parikkala and Kuikanniitty, respectively) for 60 seconds and large saplings (4.5 m in Kuikanniitty, large saplings were not found in Parikkala) for 120 seconds. Surveying time was used as a covariate called size index in statistical analysis.

All multivariate analyses were performed with Primer 6 (Primer-E Ltd, United Kingdom). The full data matrix consists of the abundance of 23–27 (23 in Parikkala 2002) insect species in 264 saplings (22 genotypes, 6 blocks, 2 sites) that were surveyed in two consecutive years. All surveyed insect species were included in the statistical analysis when sites were tested separately, but those four species that were not surveyed in Parikkala 2002 were excluded also from Kuikanniitty 2002 data when sites were compared. Arthropod community composition data was analyzed using non-parametric multivariate analysis of variance (PERMANOVA), which is well suited to non-normal ecological data such as ours ^{38, 39} . Years were analyzed separately in all statistical tests, because of the changed sampling protocol between years (surveying the whole tree in 2002, using time counts in 2003). All data was fourth root transformed prior to analysis to reduce differences between common and rare species. The semimetric Bay-Curtis distance, which generally seems to provide the most meaningful measure of dissimilarity in ecological community structure ³⁹ , was used to calculate distances between each pair of observations. The resulting distance matrix was used to obtain p-values using a random subset of 4999 permutations in PERMANOVA. The permutation method was permutation of residuals under a reduced model. The statistical model was designed to test the effect of genotype, site, block (nested within site) and the interaction of genotype × site using sapling size index (sapling surface area in 2002 and surveying time in 2003, see above) as a covariate. Site was treated as a fixed factor and block and genotype as random factors in the model. In addition to these analyses, we separately tested the effect of genotype and block on insect assemblages in each site and year to calculate the proportion of variance explained by B. pendula genotype and local environment (i.e. replicated block). Additional data collected from those saplings that were protected from insect herbivory in the previous growing season in Kuikanniitty 2003, were combined with the Kuikanniitty 2003 non-treated sapling data prior to analyzing the effects of insect removal, block and genotype, and their two-way interactions with the insect assemblages with PERMANOVA. Sapling size index was used as a covariate.

To visualize the multivariate patterns among observations, non-metric multidimensional scaling (nMDS) was performed on the Bay-Curtis distances. The distance among centroids for groups of samples was determined prior to nMDS to increase clarity, e.g. when the whole data was visualized we had 88 genotype-site-year points (22 genotypes in 2 sites over 2 years) instead of 528 genotype-block-site-year points. To visualize the effect of genotype in individual site and year, we separately determined the distance among genotype centroids in each site and year and produced one nMDS plot from each of these “environments”. Additional Kuikanniitty 2003 data combined with Kuikanniitty 2003 raw data was used to visualize the effect of insect removal on insect assemblages using nMDS on the genotype centroids of those saplings that were either protected from herbivory or grown under natural herbivory.

Species richness (number of species/sapling) and total mean abundance (number of herbivores/sapling) was statistically tested by analysis of covariance using SPSS 20.0.0.1 (IBM SPSS Statistics) General Linear Models (GLM) procedure. Those four species that were not surveyed in Parikkala 2002 were excluded also from Kuikanniitty species richness and total mean abundance calculations to better enable site comparisons. Genotype and block (nested within site) were treated as random factors and site as a fixed factor in the statistical model while sapling size index was used as a covariate. Additional Kuikanniitty 2003 data combined with Kuikanniitty 2003 basic data was used to analyze the effects of insect removal, block and genotype, and their interactions with the species richness and total mean abundance. Genotype and block were treated as random factors and insect removal as a fixed factor while sapling size index was used as a covariate. Total mean abundance was log(x+1)-transformed to equalize the error variances across groups in both analyses.

B. pendula genotypes were significantly different in their insect species community composition in both study years ( Table 2). In 2002, regional scale environmental (site) variation explained more of the total variation in species composition than the genotype (13.9 and 8.0%, respectively), while in 2003 the genotype explained more of the total variation than the site (12.1 and 3.8%, respectively). Significant genotype × site interaction, which explained 8.2% of the total variation, was found only in 2003. When the sites were tested separately in both years we found that the effect of genotype was significant in Kuikanniitty 2002 and both study sites in 2003 ( Table 4, Figure 2). B. pendula genotype could account for 15.8–27.0% of the total variation in community structure, while local scale environmental (block) variation explained 5.9–7.6% of the total variation in community structure ( Table 4, Figure 2).

B. pendula genotypes also significantly differed in their total mean abundance of herbivores (mean number of herbivores/sapling): the total mean abundance of the most susceptible genotype was 5.4- and 3.2-fold compared to the total mean abundance of the most resistant genotype in Kuikanniitty and Parikkala 2002, respectively ( Table 3, Figure 3). In 2003, only the genotype × site interaction was statistically significant, which indicates that the genotype effect strongly depended on the study site. Indeed, when we tested the study sites separately, genotype effect was significant only in Parikkala (ANCOVA: Parikkala F _21,104=2.29, p=0.003; Kuikanniitty F _21,103=1.48, p=0.103). The species richness (number of insect species/sapling) was not significantly affected by the B. pendula genotype or genotype × site interactions in either year (p>0.134).

Local scale genotype × environment interaction (i.e. the interaction of genotype × replicated block) was studied in Kuikanniitty 2003. Insect species community composition was significantly affected by both genotype and genotype × block interaction ( Table 5). Genotype variation explained 10.6% and genotype × block variation 38.0% of the total variation in insect community composition, indicating that genotype effect is also strongly affected by local scale environmental variation. Total mean abundance or species richness was not affected by genotype or genotype × block interaction (p>0.097).

Previous year herbivory changed the insect community composition of B. pendula saplings ( Table 5). The genotype centroids of those saplings that were either subjected to natural herbivory or protected from it were located on the opposite sides of the two-dimensional nMDS ordination plot, although overlapping is evident ( Figure 4). Previous year herbivory did, however, explain only 4.4% of the total variation in insect community composition. Total mean abundance was affected by the previous year’s herbivory as well, but species richness was not (ANCOVA: effects of insect removal on total mean abundance F _1,5.28=34.6, p=0.002 and species richness p>0.829).

The associations between insect species across genotypes in different sites and years seemed to be based on random associations, since we found only one correlation that was significant after sequential Bonferroni correction ⁴⁰ . An unidentified gallery mine (Lepidoptera 4) and E. fuliginosella were correlated across genotypes in Kuikanniitty 2002 (Pearson’s correlation; in 2002, r = 0.88, n = 22, p < 0.0001; in 2003, r = 0.545, n = 22, p = 0.009), but not in Parikkala (p>0.199).