Evolutionary Ecology

, Volume 28, Issue 5, pp 811–828

Growth and genotype × environment interactions in Betula pendula: can tree genetic variation be maintained by small-scale forest ground heterogeneity?


    • Department of Environmental SciencesUniversity of Helsinki
  • Ulla Paaso
    • Department of Environmental SciencesUniversity of Helsinki
  • Tarja Silfver
    • Department of Environmental SciencesUniversity of Helsinki
  • Mira Autelo
    • Department of Environmental SciencesUniversity of Helsinki
  • Katariina Koikkalainen
    • Department of Environmental SciencesUniversity of Helsinki
  • Seppo Ruotsalainen
    • The Finnish Forest Research InstitutePunkaharju Unit
  • Matti Rousi
    • The Finnish Forest Research InstituteVantaa Research Unit
Original Paper

DOI: 10.1007/s10682-014-9708-9

Cite this article as:
Mikola, J., Paaso, U., Silfver, T. et al. Evol Ecol (2014) 28: 811. doi:10.1007/s10682-014-9708-9


Forest ground heterogeneity can affect interactions among tree species and control the assembly of local forest communities. Less is known of the effects of spatial heterogeneity on the maintenance of tree genetic variation through small-scale genotype × environment (G × E) interactions. We measured growth variation within and among 17 Betula pendula genotypes, planted in a clear-cut forest site through the summers 2009–2011. We assessed the spatial heterogeneity at two scales: among forest stands having history of the same or different tree species combinations (treated as replicate blocks), and along a gradient of within-block forest density, revealed by the stump density. To add a temporal perspective, we distinguished between old (cut 50 years earlier) and new stumps (cut one year earlier). The broad-sense heritabilities for growth were 0.093–0.055 and the coefficients of genotypic variation 0.37–0.21 in 2009–2011. The growth difference among the genotypes was 3.5–5.5 fold, significant in all years, and the rank of genotype means correlated positively between the years. The most favourable block had 106 % higher growth than the least favourable block and the amount of total variation explained by block increased from 0.4 % in 2009 to 6.9 % in 2011. Genotype × block interaction was marginally significant in 2009, but not later. Similarly, the response of growth to old stump density in sapling vicinity varied among the genotypes in 2009, but not later. In 2010 and 2011, the mean growth increased by 50–91 % along the old stump density gradient. Our results suggest that despite creating significant variation in sapling growth the small-scale forest ground heterogeneity, which reflects the recent forest history, may not significantly contribute to the maintenance of genetic variation in B. pendula populations.


Forest historyGenetic variationPioneer tree speciesSilver birchSpatial variationTree populationTree stump


Spatial heterogeneity, or patchiness, has a significant role in the assembly and functioning of terrestrial ecosystems. In forests, much of this heterogeneity is created, sustained and experienced by trees in the different phases of their life cycle. The variation in canopy cover creates spatial variation in light, which has a significant influence on the growth and competitive ability of understory plants (Reich et al. 1998; Brokaw and Busing 2000; Tanaka et al. 2008; Imaji and Seiwa 2010). Single trees, either of different or same species, can also create small-scale patchiness in soil pH, C to N ratio and nutrient concentrations (Zinke 1962; Boettcher and Kalisz 1990; Häkkinen et al. 2010; Weber and Bardgett 2011), and this can further generate spatially structured communities of microbes responsible for soil organic matter breakdown and nutrient liberation (Pennanen et al. 1999; Saetre and Bååth 2000; Ushio et al. 2010). Due to the commonness of such heterogeneity, differences among tree species in their responses to the heterogeneity, and the significance of these differences in structuring forest communities, have been comprehensively tested (e.g. Poorter 1999; Beckage and Clark 2003; Gómez-Aparicio et al. 2005). In contrast, the role of such heterogeneity in structuring tree populations, and in particular maintaining their genetic variation, has received little interest.

Genetic variation is needed for populations to be able to evolve, but is also important in the ecological framework. For instance, genetic variation can increase the productivity of plant populations and enhance their resistance and recovery when exposed to climatic and herbivore perturbations (Hughes and Stachowicz 2004; Reusch et al. 2005; Crutsinger et al. 2006). Understanding the forces that can maintain genetic variation in populations subjected to natural selection is thus of uttermost importance for ecological and evolutionary theory. For environmental heterogeneity to be important for the preservation of intrapopulation genetic variation, the fitness rank of genotypes has to vary among environments, i.e. to show a genotype × environment (G × E) interaction. G × E interactions represent genetic variation in reaction norms, or variation in phenotypic plasticity, and when the reaction norms cross, selection favours different genotypes in different environments. The scale of environmental heterogeneity is, however, critical and in order to preserve the genetic variation in a local population, G × E interactions need to take place within the dispersal limit of the population (Stratton 1994). The small-scale heterogeneity of canopy cover and soil attributes in forest stands is within the dispersal limit of tree populations and therefore, can be assumed to have a potential to contribute to the conservation of genetic variation in tree populations through the G × E interactions. On the other hand, when the environmental variation occurs at small spatial scales, individuals from any generation meet various environments and selection can erode the G × E interaction by removing the genetic variation in reaction norms (Rodríguez 2012). In such case, only a generalist genotype with a high mean and low variance in fitness across the environments would remain (Stratton and Bennington 1998). G × E interactions are commonly found for forest trees when the performance of genotypes is compared among remote or contrasting growing sites (Hodge and White 1992; Haapanen 1996; Yu and Pulkkinen 2003; Silfver et al. 2009), but little is known of the commonness of G × E interactions within forest stands, where the genotypes are exposed to the environmental variation of their local habitats.

Betula pendula Roth is one of the most common tree species in Europe, covering most parts of the continent (Atkinson 1992), and is particularly abundant in the northern and eastern parts (Hynynen et al. 2010). It is a light-demanding, fast-growing pioneer species, which thrives in fertile soils of adequate moisture and rapidly colonizes open forest patches after a forest fire or clear-cutting (Atkinson 1992; Fischer et al. 2002). Local B. pendula populations have high genetic variation in many traits, ranging from phenology, growth and herbivore resistance (Rousi and Pusenius 2005; Silfver et al. 2009) to leaf autumn colouration and leaf litter decomposition rate (Silfver et al. 2007; Sinkkonen et al. 2012). A common belief is that this variation is maintained by cross-pollination, supported by the abundant production and efficient dispersal of the pollen (Hynynen et al. 2010). The pollen production is much greater in Betula than in other European tree genera (Geburek et al. 2012) and the light pollen can travel long distances along the atmospheric pathways (Siljamo et al. 2008). Accepting that this process governs the local genetic variation implicitly assumes the existence of long-distance G × E interactions. These are also known to exist: when B. pendula saplings were planted in abandoned agricultural fields and less fertile forest sites, significant G × E interactions were found for growth and secondary chemistry (Laitinen et al. 2005; Silfver et al. 2009). These comparisons, however, exaggerate the environmental variation that can be found in any single forest site and may not give an accurate picture of local G × E interactions.

To test whether significant G × E interactions can also emerge locally, we followed the growth of 17 B. pendula genotypes in a single forest site for three growing seasons. We assessed the heterogeneity of the site at two spatial scales typically created by tree communities and encountered by their seedlings: i.e. (1) among the different tree stands of the forest, represented by the six replicate blocks in our study, and (2) along a gradient of former within-stand forest density, revealed by stump density within the blocks. Two of our replicate blocks were placed on a former B. pendula stand, while the other four were on a former mixed stand of Pinus sylvestris L. and B. pendula. Variation between these blocks represents the environmental variation that can be created in a mixed forest site by stands of different tree species. For instance, it has been shown that soil nutrient mineralization is higher under B. pendula than P. sylvestris forest (Kanerva and Smolander 2007; Kanerva et al. 2008). The stump density, in turn, displays the heterogeneity that is created within forest stands by varying tree density, and to add a temporal perspective, we distinguished between old (cut 50 years earlier) and new stumps (cut one year earlier). Due to greater litter production in dense forest patches, high density of new stumps should correlate with high soil organic matter content (Paluch and Gruba 2010) and be related to fast plant growth due to the nutrients liberated from the litter layer (Xiong and Nilsson 1999; Sayer 2006). High density of old stumps could instead be related to plant growth through several mechanisms. The relationship could be negative if the fungi, which decay the old wood material, sequestrate significant amounts of nutrients from the soil (Clinton et al. 2009), but also positive if the decomposition of wood material is in a phase that releases nutrients or if the decaying wood retains water and offers more stable moisture conditions.

The target of our study is to test whether the structure of tree communities can generate environmental heterogeneity that is able to maintain the genetic variation of their tree populations. In particular, we test a hypothesis that local forest ground variation, originating from the variation in the former tree community structure and stand density, can affect tree seedling growth and lead to local G × E interactions. If supported, such hypothesis would suggest that small-scale environmental variation within forest stands can contribute to the conservation of genetic variation in tree populations.

Materials and methods

Site description

The field site is situated 140 m above the sea level near the Haapastensyrjä Unit of the Finnish Forest Research Institute (FFRI) in Loppi, south Finland (60°36′N, 24°24′E). In the region, the long-term (1981–2010) mean daily temperatures of the warmest (July) and coldest (February) month are 17 and −7 °C, respectively, and the mean annual precipitation is 660 mm (Finnish Meteorological Institute, 2012). The thermal growing season (mean daily temperature >5 °C) starts at the end of April and ends in the middle of October (Finnish Meteorological Institute, 2012). The soil is post-glacial sorted fine sand and is topped by a humus layer of a few centimeters thick. To characterize the soil, 24 samples were taken for pH, loss on ignition (at 550 °C), total C and N (analyzed using LECO CNS-2000 Analyzer; LECO Corporation, USA) and water-extractable inorganic N and P (analyzed using Lachat QuikChem 8000; Zallweger Analytics, Inc., Lachat Instruments Division, USA) in different parts of the study area in the middle of the 2010 growing season. On average, the upper 0–5 cm layer of soil had 15 % loss on ignition, total C and N concentrations of 6 and 0.3 %, respectively, a C to N-ratio of 20.4 and a pH of 5.0. The pH of the lower 5–10 cm mineral layer was 5.2. Concentrations of water-extractable inorganic N and P were on average 0.3 and 1.1 µg g−1 dry soil in the upper soil layer and <0.1 and 0.1 µg g−1 dry soil in the lower soil layer.

The field site had originally been covered by an old coniferous forest, which was harvested in 1961. Most, if not all, of the old stumps recorded in our study originate from this harvest. The forest, which grew in the site before the present experiment, was planted after a controlled burning in 1962 and contained stands of different mixture of B. pendula and P. sylvestris with sporadic Picea abies (L.) Karst. In the 2008 clear-cut, made for this study, two experimental areas, 100 m apart, were established into these stands. The larger area (0.42 ha) includes the replicate blocks 1–4 and was covered by a mixed P. sylvestris (75 %) and B. pendula (25 %) forest. The smaller area (0.22 ha) includes the blocks 5 and 6 and had a B. pendula monoculture. In both experimental areas, the blocks were founded by dividing the available area to quadrangular parts of equal size and shape. The blocking captures the different tree species compositions of earlier forests in the two areas, but also the minor slopes of the ground and the distance from the nearby forests in the areas. The areas were harvested in the winter 2008, the logs and logging residues were immediately removed and minimal damage was inflicted on the ground layer vegetation and soil surface.

The ground layer vegetation was surveyed in 2011, 3 years after the clear-cut (24 plots examined, each 1 m × 1 m). The ferns were represented by Pteridium aquilinum (L.) Kuhn with a mean cover of 11 %, while Pleurozium schreberi (Willd. ex Brid.) Mitt. (39 %) dominated the mosses. Calamagrostis arundinacea (L.) Roth (43 %) was the dominant graminoid, but Deschampsia flexuosa (L.) Trin. (13 %) and Luzula pilosa (L.) Willd. (3 %) were also found in most plots. The dwarf shrubs Vaccinium myrtillus L. (2.7 %) and Vaccinium vitis-idea L. (4.5 %) did not cover much area, but were found in most plots. The herbs were the most species rich group, but only Melampyrum pratense L. (mean cover 1.4 %), Trientalis europaea L. (1.8 %) and Maianthemum bifolium (L.) F. W. Schmidt (0.8 %) were found in more than half of the examined plots. The naturally sprouting woody plants were continuously weeded in the study area and their cover was minimal.

Plant material

The mother trees of those B. pendula genotypes that were used in this study were selected in 1997 from a 0.9-ha B. pendulaBetula pubescens Ehrh. forest stand in Punkaharju, south-east Finland (61°48′ N, 29°18′ E). The forest had naturally regenerated after 1979 logging and the selected 30 trees grow in six separate blocks, located 10–60 m apart. The coefficient of variation (CV) of mean tree growth among the blocks was 0.06–0.26 (mean 0.15) in 1997–2004. In this study, we use this variation as a rough estimate of the magnitude of spatial variation of tree growth in the site where the genetic variation of the study population evolved, and compare it to the variation in the experimental site, where the G × E interactions were tested.

To produce plant material for the present study, twigs were collected from mother trees and their vegetatively cloned nine-year old progeny, also growing in Punkaharju, in late 2007. Plantlets of 19 genotypes (2, 3, 8, 12, 14, 15, 16, 17, 18, 19, 20, 22, 23, 24, 25, 26, 27a, 28 and 30) were then propagated from twig buds using a micropropagation technique at the Haapastensyrjä Unit of FFRI in early 2008. Once the plantlets had rooted, they were transferred to a nursery and to prepare them for field planting, grown in the nursery through the summer 2008. In the nursery, individuals of each genotype were divided to two or more multi-cell containers, except for three genotypes, which had only one container. To avoid the confounding effects of spatially varying light and temperature on genotype performance, the containers were randomly placed in the nursery. The produced saplings were then overwintered in a cold room and planted in the spring 2009. In the experimental site, each of the six replicate blocks was divided into 132 planting plots of 2 m × 2 m area and five to nine saplings of each genotype were randomly allocated and planted into these plots in each block. Two of the genotypes (16 and 24) grew poorly in the nursery and most of their individuals died in the field. These genotypes are excluded from this study. To diminish the edge effects, two rows of B. pendula pot seedlings were planted to surround the planting area of the cloned individuals in each block.

Measurements and treatments

The height of the saplings was measured in the late autumn of 2008, in the early spring of 2010 and 2011, and in the late autumn of 2011. Using these measurements, the annual height increment of the saplings was calculated for years 2008–2011 (the first growing season in the nursery, the following three in the field). Saplings, whose leader shoots were damaged (i.e. the shoot was dark brown, black, dry or buds were not swollen), were not excluded from height measurements since B. pendula is able to regenerate from basal buds by sprouting (Atkinson 1992). Instead, the condition of the leader shoot (damaged or not damaged) was recorded at each survey. The height increment of damaged saplings (33, 15, 15 % of all saplings in 2009–2011, respectively) was significantly reduced and was on average negative in 2010 (−2.4 cm) and slightly positive in 2009 (1.9 cm) and 2011 (0.7 cm).

The number and diameter of new (originating of the 2008 logging) and old stumps (originating of the 1961 logging) were measured for each 2 m × 2 m planting plot. Total density (cm2 m−2) of each group of stumps was then calculated for 3 × 3 plots surrounding each sapling (equalling to 36 m2) and used as an approximation of the density of the former forest in the vicinity of the planted sapling. The density of new tree stumps was on average higher (mean 34 cm2 m−2) and varied less (min–max 0–85 cm2 m−2, SD = 16) than the density of old stumps (mean 18 cm2 m−2, min–max 0–148 cm2 m−2, SD = 27). The densities of both stumps varied among the replicate blocks, but were not correlated at the planting plot level (“Appendix 1”).

To elucidate the mechanisms that could link sapling growth to forest ground heterogeneity and explain the potential G × E interactions, we measured soil organic matter content and grass cover (an estimate of inter-specific competition) in the vicinity of the saplings, as well as leaf N (a proxy of soil N availability) and leaf water (a proxy of soil moisture accessibility) concentrations of the saplings. The organic matter content of the top 0–3 cm layer of soil was estimated by collecting soil cores (diameter 3 cm) at 19 randomly chosen plots in each block in 2010. The cores were dried in 70 °C for 3 d and the loss on ignition was determined at 550 °C. Grass cover (including all graminoid species) was visually estimated in 34 randomly selected planting plots in each block and each plot was classified into one of the seven coverage classes: 0, 1–10, 11–20, 21–40, 41–60, 61–80 or 81–100 % of the plot area covered by grasses. For measuring leaf N and water concentrations, three to six random leaves were collected from a random subset of saplings in each block in the middle of August in 2010 (n = 114) and 2011 (n = 204). In 2010, the leaf material was ground to powder using a ball mill and analysed using a LECO CNS-2000 Analyzer (LECO Corporation, USA), while in 2011 the leaf material was ground using liquid nitrogen and a mortar and analysed at Iso Analytical Ltd., UK.

For the purpose of another study, the insect herbivory was reduced in one randomly selected sapling of each genotype in each block in 2010 and 2011 by spraying the saplings once a week with a 0.1 % solution of Decis EC25 deltamethrin insecticide (Bayer CropScience, Germany). As the reduction of herbivory has a significant effect on sapling growth (Silfver et al. 2013), we include the insecticide treatment in the statistical analyses of this study to control the effect. The effects of herbivory will otherwise be beyond the objectives of this study.

Statistical analyses

The broad-sense heritabilities (H2) of sapling growth were calculated on individual plant basis according to the Eq. 1, where σG2, σB2, σG×B2 and σE2 are variance components for genotypes, blocks, genotype × block interaction and (within-block) environment, respectively. The variance components were calculated using the SPSS GLM Variance components procedure. Because the aim of the study was to estimate the block-scale environmental variation, the block was, contrary to the common usage in forest tree breeding, treated in the calculation model as a random factor (the difference in H2 between the models where the block was treated as fixed and random was, however, marginal). In 2010 and 2011, the insecticide treatment was included in the variance component model as a fixed factor.
$$H^{2} = {\raise0.7ex\hbox{${\sigma_{G}^{2} }$} \!\mathord{\left/ {\vphantom {{\sigma_{G}^{2} } {\left( {\sigma_{G}^{2} + \sigma_{B}^{2} + \sigma_{G \times B}^{2} + \sigma_{E}^{2} } \right)}}}\right.\kern-0pt} \!\lower0.7ex\hbox{${\left( {\sigma_{G}^{2} + \sigma_{B}^{2} + \sigma_{G \times B}^{2} + \sigma_{E}^{2} } \right)}$}}$$
Coefficients of genotypic variation (CVG) were calculated according to the Eq. 2, where \(\bar{x}\) is the phenotypic mean. Whether the growth rank of the genotypes remained the same through the years was tested using Spearman’s rank correlation coefficients of the genotype means of the four consecutive years.
$$CV_{G} = {\raise0.7ex\hbox{${\sqrt {\sigma_{G}^{2} } }$} \!\mathord{\left/ {\vphantom {{\sqrt {\sigma_{G}^{2} } } {\bar{x}}}}\right.\kern-0pt} \!\lower0.7ex\hbox{${\bar{x}}$}}$$

The statistical significance of the effects of genotype, block, density of old and new stumps and the genotype × block and genotype × stump interactions on sapling growth were tested using mixed ANCOVA models (see Table 1 for the full structure of the models) and the PASW statistical package (Release 18.0.0). The genotype and block were treated as random factors, the insecticide treatment as a fixed factor and the stump densities were used as covariates. Since height increment, which we used as a measure of annual sapling growth, is significantly affected by the leading shoot damage and the insecticide treatment, both were included in the models as additional two-level fixed factors. The experimental area was not included in the models, because the replicate block can better capture the different aspects of the environmental variation than the area and can also explain the variation between the areas. The effects of block and stump variation on leaf N content, leaf water content and soil organic matter content were tested using equal ANCOVA models (see the model structure in Table 3), whereas the effects on planting plot grass cover, an ordinal response variable, were analysed using a generalized linear model with a multinomial probability distribution and a cumulative logit link function. Since the mean density of new and old stumps varied significantly among the blocks (“Appendix 1”), stump densities were mean centred by block (i.e. each block mean was set to zero) before statistical analyses. Such focus on within-block variation guarantees that stump effects are not confounded by the among-block variation of various environmental parameters that possibly covary with stump densities. The homogeneity of error variances was tested using the Levene’s test, and since this assumption was violated in the 2008 nursery data, the genotype effect was tested using the Kruskal–Wallis nonparametric test.


Height growth differed among the genotypes in the nursery in 2008 (Kruskal–Wallis H16 = 82.5, P < 0.001) and the difference remained significant through the years in the field (Table 1; Figs. 1, 2). However, while the rank of genotype means correlated positively among the years in the field (Spearman’s ρ = 0.53, P = 0.029, n = 17 for the 2009–2010 comparison and ρ = 0.52, P = 0.034 for the 2010–2011 comparison; Fig. 2), the genotype rank in the nursery correlated negatively with the 2009 field rank (ρ = −0.59, P = 0.013, n = 17), leading to markedly crossing reaction norms (Fig. 1). The broad-sense heritability for growth decreased from 0.093 to 0.055 and the coefficient of genotypic variation from 0.37 to 0.21 during the three growing seasons (Table 2). The difference between the fastest and slowest growing genotype was 1.4-, 5.5-, 5.5- and 3.5-fold in 2008–2011, respectively (Figs. 1, 2).
Table 1

F and P-statistics of mixed ANCOVA models of the effects of genotype (a random factor; 17 genotypes, n = 32–56 for each), replicate block (a random factor; six blocks), density of new and old tree stumps (covariates) and their interactions on B. pendula sapling growth in the field in the growing seasons 2009–2011





































New stump density











Old stump density




















Leading shoot damage (LSD)











Genotype × block











Genotype × new stump density











Genotype × old stump density











The insecticide treatment (a fixed factor in 2010 and 2011 models) and leading shoot damage (a fixed factor; no damage vs. damage during the target growing season) were included in the models as supplementary explaining factors (see “Materials and methods” for details)

dfT degrees of freedom for treatment, dfE degrees of freedom for error

Table 2

Variance components (σ2), broad-sense heritability (H2), phenotypic mean (\(\bar{x}\)) and the coefficient of genotypic variation (CVG) for sapling growth (height increment) in three growing seasons

































G genotype, B block, G × B genotype × block interaction, E error, or residual

Fig. 1

Reaction norm plot for 17 B. pendula genotypes growing in the nursery in 2008 and in the field in 2009. For clarity, the worse performing half of the genotypes (the genotypes 2, 3, 8, 15, 17, 25, 27a, 28 and 30 growing less than 43 cm during the three field seasons) is shown in black and the better performing half (the genotypes 12, 14, 18, 19, 20, 22, 23 and 26 growing more than 47 cm) in grey; n = 32–55 for each reaction norm

Fig. 2

Reaction norm plots for 17 B. pendula genotypes growing in six replicate blocks in the field in 2009–2011 (the grey and black lines are as in Fig. 2); n = 5–9 for each reaction norm

Sapling growth varied significantly among the replicate blocks (Table 1) and the amount of total variation explained by the block increased from 0.4 % in 2009 to 3.1 and 6.9 % in 2010 and 2011, respectively (Table 2). The coefficient of variation (CV) of block mean growth (an estimate of the magnitude of spatial variation in sapling growth in the experimental site) was 0.15, 0.20 and 0.25 in 2009–2011. The genotype × block interaction was marginally significant in 2009, but was not found in the following years despite partly crossing reaction norms (Table 1; Fig. 2), and the amount of growth variation explained by the interaction decreased from 2.7 % in 2009 to 0 and 0.3 % in 2010 and 2011 (Table 2). The magnitude of growth variation of a genotype among the blocks (measured as a CV of block mean growth) was inversely related to its mean growth (Pearson’s r = −0.61, P = 0.009; r = −0.78, P < 0.001; and r = −0.55, P = 0.021, n = 17 in 2009–2011, respectively), i.e. the fastest growing genotypes varied least among the blocks.

Sapling growth was affected by the density of old stumps, but not the density of new stumps (Table 1). In 2009, a significant genotype × stump interaction was found for the old stumps (Table 1, “Appendix 2”), whereas in 2010 and 2011, the mean growth of the saplings increased by 91 and 50 %, respectively, along the old stump density gradient (Table 1; Fig. 3). The proportion of total variation in sapling growth, which was explained by the variation in the old stump density, was 1.6 % in 2010 and 0.8 % in 2011 (Fig. 3).
Table 3

Effects of replicate block (six blocks) and the density of new and old tree stumps on soil organic matter (OM) content, B. pendula leaf N and water concentrations (analyzed using ANCOVA with F and P statistics) and planting plot grass cover (analyzed using a generalized linear model with Wald Chi2 and P statistics)




New stump density

Old stump density







Leaf N %








Leaf water %








Soil OM content








Grass cover








N = the number of observations included in the analysis

Fig. 3

Sapling growth along the old stump density gradient in 2010 and 2011. The lines and equations represent linear regressions

Leaf N and water concentrations (proxies of soil N and moisture availability), soil organic matter content (0–3 cm layer) and the grass cover of the planting plots differed among the blocks, but none was significantly affected by the density of new or old stumps (Table 3; Figs. 4, 5). Leaf N concentration correlated positively with sapling growth (r = 0.59, P < 0.001, n = 199) and leaf water concentration (r = 0.46, P < 0.001, n = 199) in 2011. The highest level of grass cover (80–100 % of area covered) was mainly found in the blocks 5 and 6 (Fig. 5).
Fig. 4

Leaf N content (mean + SE; n = 50–58, data combined for years 2010 and 2011), leaf water content (mean + SE; n = 31–36) and the organic matter (OM) content of the top 3-cm soil layer (mean + SE; n = 19) in the six replicate blocks

Fig. 5

Frequency of planting plots of different grass cover (cover classes 1–7 represent grass areal cover of 0, 1–10, 11–20, 21–40, 41–60, 61–80 and 81–100 %, respectively) in the six replicate blocks (n = 34 for each block)


We tested a hypothesis that small-scale forest ground variation can affect tree seedling growth and lead to local G × E interactions. Our results show that the small-scale heterogeneity of boreal forest ground can create significant variation in B. pendula growth. Growth varied among replicate blocks, but also in response to the spatial variation created by varying tree densities of earlier tree rotations. There was also significant genotypic variation in sapling growth and local G × E interactions emerged, both across the blocks and along the old stump density. However, these interactions were weak and they were found in the first field season only. This suggests that after the early stages of tree seedling establishment, forest ground variation may not cause G × E interactions that would be strong enough to contribute to the maintenance of genetic variation in B. pendula populations subjected to natural selection.

Lack of local G × E interactions could be due to lack of intrapopulation genetic variation in the measured trait. This was not the case in our study, however, since all analyses showed that growth had a significant genetic component. In a series of six field trials in southern Sweden, the broad-sense heritability for B. pendula growth varied between 0.07 and 0.56 (Stener and Jansson 2005). In our own, older field trial with cloned 9–12 years old trees, planted in an agricultural field and representing the same genotypes that we used in the present study, the broad-sense heritability of growth was 0.08–0.17 in four consecutive years (unpublished data). The values recorded in the present study, 0.055–0.093, are at the lower end of these estimates, but since the heritability is a measure of the proportion of total phenotypic variation explained by the genetic variation, it depends a lot on the environmental variation of the test site. That there was significant genetic variation for growth among our saplings, despite relatively low heritability, is shown by the coefficients of genotypic variation, which were higher in our present study (0.21–0.37) than in our older experiment (0.10–0.19, unpublished data) or in other B. pendula populations (0.046–0.133) (Stener and Hedenberg 2003; Stener and Jansson 2005). Coefficients of genotypic variation are more relevant estimates of evolutionary significant genotypic variation than heritabilities (Houle 1992; Hansen et al. 2011), but it is worth noting that the coefficients decreased with increasing age in our study and the other estimates for B. pendula are based on older, around 10 years old trees (Stener and Hedenberg 2003; Stener and Jansson 2005; own unpublished data). The decreasing genotypic variation could be due to selective mortality, and thus a sign of significant selection in the early life, but this was not the case since 97.5 % of the planted saplings were still alive in 2011. Another potential explanation is the developmental plasticity of the plants: during their growth, the saplings may gradually express a phenotype better suited to the field conditions and this can lead to declining genotypic variation with increasing age. The decreasing relative difference between the fastest and slowest growing genotype with increasing age, which we observed in the field, could also have similar consequences. Whatever the reason, it is likely that the declining genotypic variation in growth reduces the probability of finding G × E interactions in older saplings.

Lack of small-scale G × E interactions could also be due to lack of significant environmental variation at the spatial scale of the measurement. In our study, the heterogeneity effects (42–106 % higher mean growth in the best than worst block in 2009–2011 and 50–91 % increase in growth along the old stump density gradient) were smaller in magnitude than the genetic effects (3.5–5.5-fold difference in mean growth between the worst and best genotype). However, the block-to-block variation increased in time and in 2011, it explained more (6.9 %) of the total phenotypic variation than the genotypic variation (5.5 %). These values suggest that there was significant environmental variation, in terms of growth, at the scale of our G × E interaction tests. It also appears that the growth variation among the replicate blocks was of similar magnitude to the growth variation among the blocks in the original growth site, suggesting that the spatial variation in our site is comparable to the variation in the site where the population and its genetic structure evolved. On the other hand, 86 % of phenotypic growth variation in 2011 could not be explained by genotype, block, their interaction or old stump density, and was thus due to within-block, stump-independent environmental variation. For herbaceous plants, G × E interactions may occur at very small spatial scales (Stratton 1994; Stratton and Bennington 1998), which raises the question whether the most significant G × E interactions could also in the forest ground occur at smaller than block scale. The genotype × old stump density interaction that we found in 2009 is an example of such smaller-scale interaction, and the fact that we observed this and the genotype × block interaction in 2009, but not later, suggests that we focused on meaningful spatial scales, but the G × E interactions were restricted to early sapling establishment only.

There are several examples of G × E interactions among field sites, also for B. pendula (Laitinen et al. 2002, 2005; Silfver et al. 2009), but few studies have been able to point out the factor that governs these interactions. A comparison of Pinus elliottii Engelm. progeny tests, however, suggests that G × E interactions emerge when the productivity between two sites differs “more than on average” (Hodge and White 1992). In our study, the comparison of plant growth between the nursery and the field site and the block-to-block comparisons in the field support this conclusion. Those genotypes that grew best in the productive nursery conditions in 2008 (mean sapling growth 26 cm) grew least in the harsher field conditions in 2009 (mean growth 9 cm) and a clear G × E interaction emerged. In the field site, the blocks also differed in productivity since sapling growth, leaf N % and leaf water concentration varied among the blocks and were positively related in 2011. However, while the difference in productivity between the nursery and the field was 2.9-fold, the difference between the best and worst block was 1.4-, 1.6- and 2.1-fold in 2009–2011, respectively, and the G × E interactions remained weak. This suggests that the spatial variation of productivity in local field sites may not reach the level that would lead to G × E interactions. It is good to note, though, that in experimental trials the transfer of plants from the nursery to field conditions and planting in the field causes stress for the plants and this may influence their performance and response to environmental conditions in the beginning of field growth. Such initial stress might affect our nursery vs. field site comparison, whereas our main finding—significant genetic and spatial variation of growth, but a lack of G × E interactions—which arises from the second and third year observations should not be significantly affected.

The two main attributes of forest community structure are the tree species composition and the forest density. For the seedlings of a pioneer tree species, the effects of these attributes are mediated through the heterogeneity of soil and vegetation parameters, such as soil organic matter content, moisture, nutrient availability and grass cover. In our experimental site, the blocks 5 and 6 were established in the place of a former B. pendula stand and since soil nutrient mineralization is typically higher under B. pendula than P. sylvestris (Kanerva and Smolander 2007; Kanerva et al. 2008), we expected that growth would be most vigorous in these blocks. In line with our expectations and earlier studies (Olsson et al. 2012), the soil had 50 % less organic material in these blocks (Fig. 4), which is an indication of faster decomposition, but unexpectedly the saplings grew on average better and had higher leaf N % in the blocks 1–4, earlier dominated by P. sylvestris. Although the reasons for this may be manifold and difficult to disentangle, the very dense grass cover in the blocks 5 and 6 (Fig. 5) is one likely reason for poor growth as B. pendula is sensitive to competition with herbaceous vegetation (Atkinson 1992; Ferm et al. 1994; Hynynen et al. 2010). We also predicted that the growth of B. pendula saplings would respond positively to higher densities of new stumps because a thicker litter layer, originating from a dense tree stand, could provide more nutrients for growth (Xiong and Nilsson 1999; Sayer 2006). We did not find evidence of this, however, which may be explained by the insignificant effect of new stump density on soil organic matter content. In the case of the old stumps, we assumed that the growth could respond to the stump density either positively or negatively depending on which of the potential mechanisms was dominant. We found that sapling growth was positively associated with increasing old stump density, and the estimated average increase of growth along the stump density gradient was high, 91 and 50 %, in the two last years of measurements. This demonstrates how forest density can have substantial, unforeseen effects on tree seedling growth still half a century after the forest cut. The mechanism that could explain this association, however, remained open in our study. We did not find evidence of increasing fertility with increasing old stump density as the stump density and leaf N concentrations were not related. Better moisture conditions, due to the decaying wood retaining water, or less dense grass cover and competition in the plots of high old stump density were neither supported by our data.

Lack of significant G × E interactions at the later stage of sapling growth in our study suggests that all genotypes responded to forest ground variation in the same way, i.e. there was no genetic variation in the growth reaction norms. This could be because selection has eroded local G × E interactions (sensu Rodríguez 2012) in the population, where the genotypes were originally collected, but also due to the variation in our experimental site not reaching the level, where G × E interactions emerge. We also found that those genotypes that grew best varied least due to spatial heterogeneity, indicating that fast growing genotypes are better in tolerating less favourable environments. This is reminiscent of studies that have found no trade-offs between B. pendula growth and other plant traits, such as herbivore resistance (Tikkanen et al. 2003) and seed production (Rousi et al. 2011), and suggests that selection should decrease the magnitude of variation in sapling growth along the local gradients of environmental heterogeneity. When studying the growth responses of 15 rain-forest tree species to a light gradient, Poorter (1999) found that fast growing species had highest growth rate in both shade and light. Similarly, in an Appalachians mixed-oak forest, Beckage and Clark (2003) found that species survival ranks varied little across the microenvironments of varying light and nutrient availability. In our study, this pattern was discovered at the genotype level: the growth rank of B. pendula genotypes did not differ among the field blocks of high and low productivity. Together all these findings suggest that the environmental heterogeneity may not significantly contribute to the maintenance of either species or genetic diversity in local forest stands.

To conclude, our results suggest that selection within boreal forest sites produces generalist genotypes rather than maintains the genetic variation of local B. pendula populations. This is due to the G × E interactions disappearing at later sapling growth, due to the rank of genotypes in the annual growth remaining the same through the years and due to the size of adult trees being positively associated with their seed production (Rousi et al. 2011). In particular, it appears that the small-scale forest ground heterogeneity, which reflects the recent forest history, may not significantly contribute to the maintenance of genetic variation in B. pendula populations. In such case, the genetic variation in growth should vanish through natural selection. Since B. pendula populations, nevertheless, have high intrapopulation genetic variation in growth, processes that were not addressed in this study are needed to explain the variation. The long-distance G × E interactions (Laitinen et al. 2002, 2005; Silfver et al. 2009), which mix the genetic material favoured by selection in remote forest sites, are one likely explanation due to the long-distance dispersal of Betula pollen (Siljamo et al. 2008). Another potential explanation, which cannot be addressed using our vegetatively propagated plant material, is that significant local G × E interactions occur, but at the moment of seed germination and early seedling growth.


We thank Seija Vanhakoski for the micropropagation work, Pentti Kananen and Markku Ahlquist for field site preparation and the staff in FFRI Haapastensyrjä Unit for participating in planting and site maintenance. We are also grateful to Anni-Mari Pulkkinen for the vegetation survey, Marianne Lehtonen, Viivi Toivio and Santeri Savolainen for laboratory assistance and the anonymous reviewers for many instructive comments on the earlier version of the paper. The study was funded by the Academy of Finland (decision #1122444).

Copyright information

© Springer International Publishing Switzerland 2014