•It has been suggested that autumn-migrating insects drive the evolution of autumn leaf colours. However, evidence of genetic variation in autumn leaf colours in natural tree populations and the link between the genetic variation and herbivore abundances has been lacking.
•Here, we measured the size of the whole aphid community and the development of green–yellow leaf colours in six replicate trees of 19 silver birch (Betula pendula) genotypes at the beginning, in the middle and at the end of autumn colouration. We also calculated the difference between green leaf and leaf litter nitrogen (N) and estimated the changes in phloem sap N loading.
•Autumn leaf colouration had significant genetic variation. During the last survey, genotypes that expressed the strongest leaf reflectance 2–4 wk earlier had an abundance of egg-laying Euceraphis betulae females. Surprisingly, the aphid community size during the first surveys explained N loss by the litter of different birch genotypes.
•Our results are the first evidence at the tree intrapopulation genotypic level that autumn-migrating pests have the potential to drive the evolution of autumn leaf colours. They also stress the importance of recognizing the role of late-season tree–insect interactions in the evolution of herbivory resistance.
One of the most striking senescence-related phenomena in temperate and boreal forests is that of autumn leaf colouration of deciduous trees. Trees produce the yellow leaf pigments, carotenoids, to facilitate photosynthesis throughout the season, while red pigments, mostly composed of anthocyanins, advance physiological leaf performance in the autumn (Lee, 2002; Ougham et al., 2005). The timing and intensity of autumn leaf colouration are related to day length and temperature sum (Hänninen et al., 1990; Andersson et al., 2004), as well as to plant characteristics, such as defence against herbivores, sexual reproduction and level of nutrition (Schaberg et al., 2003; Sinkkonen, 2006a,b, 2008; Karageorgou et al., 2008). There is, however, tremendous variation in autumn leaf colouration between tree species and between geographical areas of comparable climate, which seems to suggest that plant ecophysiology may not be the sole explanation for the production and visibility of autumnal leaf pigments (Vollenweider & Günthardt-Goerg, 2005; Lev-Yadun & Gould, 2007; Archetti et al., 2009; Lev-Yadun & Holopainen, 2009; Lev-Yadun et al., 2012). This has encouraged the development of herbivory-related hypotheses: some assume coevolution between trees and autumn-migrating aphids, while others suggest that autumn leaf colours provide camouflage against the aphids, or undermine the aphid’s camouflage when they occupy the leaves, thus exposing them to predators (Hamilton & Brown, 2001; Lev-Yadun et al., 2004; Holopainen, 2008; Archetti, 2009a). However, empirical studies considering both the genetic variation in autumn leaf colouration and the link between leaf colour and aphid abundance are lacking for wild tree populations, and the only data set published so far consists of geographically separated apple (Malus pumila) varieties that for generations have been under both conscious and unconscious artificial selection as fruit and ornamental trees (Archetti, 2009b). This lack of data has prevented the separation of phenotypic and heritable, genotypic variation in autumn leaf colouration and aphid load, which is a necessary prerequisite for testing evolutionary hypotheses.
Silver birch (Betula pendula) is one of the main forest tree species in Eurasia. It has yellow autumn leaf colouration and together with its specialist aphid, Euceraphis betulae, it forms one of the most widespread tree–insect interactions in Western Eurasia (Heie, 1982; Holopainen et al., 2009). Winged E. betulae males and the winged parthenogenetic females migrate and search for new hosts predominantly during the first half of the period of silver birch autumn leaf colouration. Before leaf fall, the winged parthenogenetic females give birth to a generation of nonmigrating sexual wingless females (oviparae). This leads to an unequal sex ratio with two to three times more sexual females than males on the same tree (Dixon, 2005). Wingless sexual females continue laying over-wintering eggs until they are eradicated by severe frosts (Heie, 1982). Such aphid occupancy is likely to expose the trees to significant resource losses: precious nutrients, especially nitrogen (N) and phosphorus, are relocated from the senescing leaves at the time that the wingless females feed on the phloem contents, while in the spring a similar transport of nutrients to developing leaves occurs, and these nutrients are partly lost when the aphid’s offspring emerge and tap the phloem (White, 1993; Holopainen & Peltonen, 2002). Phloem-feeding insects are known to significantly affect tree performance (Dixon, 1971, 2005; Zvereva et al., 2010), so trees that through favourable colouration can reduce aphid arrival at the time of aphid migration should consequently benefit (Archetti, 2000; Hamilton & Brown, 2001; Archetti et al., 2009).
Here we examined whether silver birch autumn leaf colouration has significant intrapopulation genetic variation. We measured the leaf colour (i.e. leaf reflectance of light of different wavelengths) of cloned, field-grown progeny of a local silver birch population in an experiment consisting of three surveys conducted during one autumn. We focused on a wavelength interval of 530–535 nm in the middle of the green light waveband, which was chosen because the reflectance of a yellow surface is always greater than the reflectance of a similar green surface; this allows measurement of the green–yellow gradient of birch autumn leaf colouration using a single reflectance value. Based on earlier studies reporting significant genetic variation in silver birch populations (Järvinen et al., 2003; Rusanen et al., 2003; Silfver et al., 2009), we hypothesized that silver birch autumn leaf colouration will also have a significant genetic component. As both asexual and sexual generations of E. betulae aphids have been shown to favour yellow over green birch leaves when searching for new host trees (Holopainen et al., 2009), we predicted that the genetic variation in silver birch autumn leaf colouration will be associated with variation in E. betulae abundance on individual silver birch trees. To test this, we counted E. betulae individuals on randomly selected branches of each tree in concert with leaf colour measurements. The reason why aphids prefer yellow over green is not entirely clear. Döring & Chittka (2007) suggested that preference of yellow is a physiologically unavoidable side-effect of green preference as the reflectance of a yellow object is higher than the reflectance of a similar green object in the green light waveband. By contrast, the nutrient retranslocation hypothesis proposes that migrating aphids use yellow as a signal of available N exported from senescing leaves (Holopainen & Peltonen, 2002). To test whether the aphids can benefit by selecting specific birch genotypes, we contrasted phloem sap N availability and E. betulae abundance over three different phases of autumn leaf colouration. We assumed that a positive correlation would be in keeping with the nutrient retranslocation hypothesis (Holopainen & Peltonen, 2002); while no correlation or a negative correlation would support the side-effect hypothesis (Döring & Chittka, 2007). Finally, to examine whether autumnal aphids can reciprocally affect the N economy of their host trees, we tested whether the amount of N lost by different silver birch genotypes in leaf litter can be explained by the abundance of aphids residing on these genotypes during autumn leaf colouration.
Materials and Methods
The silver birch (Betula pendula Roth.) trees used in this study were cloned from their parental trees using a micropropagation technique in 1998. Since then the trees have been growing in an abandoned agricultural field (Kuikanniitty in Punkaharju, south-eastern Finland; 61°48′N, 29°18′E) as a common garden experiment consisting of six replicate blocks, each block including individuals of each of the 19 genotypes (Laitinen et al., 2004, 2005; Silfver et al., 2009). For this study, each tree was surveyed three times during the autumn of 2008: the survey dates were 8–10 September, 22–23 September and 5–6 October. These dates corresponded with the onset of leaf colour change, mid-autumn leaf colours and the onset of leaf fall, respectively. To estimate the leaf colour change and the development of the aphid population, a randomly chosen south-facing branch (c. 50 cm long, at 3–5 m height) was collected from each tree during each survey. The branch was rapidly enclosed in a plastic bag to prevent aphid escape, excised and taken to the laboratory. The bagged branches were frozen (− 20°C), and all aphids in the frozen samples were later counted and identified. Euceraphis betulae (Koch) individuals were sorted into winged males, winged viviparous females, wingless oviparous females and nymphs (in total, 1906 aphids were identified of which 84% were E. betulae). Other species of aphids (e.g. Holopainen et al., 2009) were classified as either adults or nymphs. The number of leaves and their dry mass were measured for each branch, but, because variation in these variables (mean ± SE for leaf number 18.4 ± 0.6 and for leaf dry mass 1.23 ± 0.04 g) did not explain the variation in aphid abundance on the branches, only the aphid data are presented per branch.
For leaf colour measurements, five leaves were randomly picked from each frozen branch and the colour of each leaf, that is, the spectral irradiance of the reflected light (referred to hereafter as reflectance), was measured using a spectroradiometer for the waveband 400–700 nm with 5-nm intervals (LI-1800 Portable Spectroradiometer, LI-COR, Lincoln, Nebraska), calibrated according to the manufacturer’s instructions. Before the study, tests were conducted to show that freezing did not affect the reflectance pattern of either green or yellow leaves. The measurements focused on the 530–535 nm wavelength interval, where the reflectance is most sensitive to the green–yellow gradient and can express changes in both autumn leaf colouration (higher values indicate yellower leaves) and leaf chlorophyll concentration (higher values indicate less chlorophyll; see Gitelson & Merzlyak, 1994). Accordingly, the reflectance difference between two field surveys was used as an indication of chlorophyll breakdown and phloem sap N loading between the surveys. The reflectance values obtained between 400 and 700 nm were further inserted into an aphid colour preference model developed by Döring et al. (2009). In this model, the leaf reflectance is balanced by the visual capability of aphids to recognize different wavebands and by their response to different wavelengths (Döring et al., 2009). Using the model, an attractiveness index was calculated for each tree in each survey. Finally, to estimate the total N transfer from the senescing leaves of each tree in the autumn the N concentrations in green leaves and leaf litter were measured. Green leaves (30–40 fully grown leaves) were randomly collected from a south-facing branch of each tree at a height of 2–4 m at the end of June 2008. The senescent leaves were collected by enclosing two south-facing branches of each tree in white polyethylene mesh bags before leaf abscission; the fallen leaves were then randomly picked from the bags. N concentrations were analysed using a Leco-CNS-2000 analyser (LECO Corp., St. Joseph, MI, USA).
The data were analysed using the spss 15.0.1. statistical package (SPSS, Chicago, Illinois, USA). The effects of the silver birch genotype on (1) leaf reflectance at 530–535 nm (using the mean of the five measured leaves); (2) the leaf attractiveness index; (3) differences in the reflectance between surveys, first vs second, and second vs third (used as an estimate of phloem N availability to aphids between the respective surveys); and (4) the difference in N concentration of green leaves vs leaf litter (used as an estimate of total N transfer from the senescing leaves) were tested using univariate analyses of variance, with the genotype and the replicate block as random factors. The responses of E. betulae life-cycle stages to leaf reflectance, leaf attractiveness and reflectance differences were tested using a regression analysis, while the association between the numbers of flying E. betulae (winged males and females) sampled for the first and second surveys and the subsequent phloem N availability was tested using Pearson’s correlation analysis (all analyses performed using genotypic means). Finally, the effect of the aphid community on leaf litter N concentration was analysed using a multiple linear regression and the backward procedure of the stepwise variable selection. All E. betulae life-cycle stages (except for first survey wingless females and second and third survey winged females which were too scarce) and the nymphs of other species (the adults of other species were excluded as they were too scarce) were included in the initial stepwise model together with green leaf N concentration. The assumptions of the statistical analyses were met, except for the leaf reflectance data, where one to three outliers occurred per survey. Exclusion of these outliers did not, however, affect the outcome of the statistical tests.
Progress of autumn leaf colouration and N availability among birch genotypes
Leaf reflectance at 530–535 nm differed significantly among silver birch genotypes at the beginning (first survey) and middle (second survey), but not at the end (third survey) of autumn leaf senescence (Fig. 1a, Table 1). The genotype means of the leaf reflectance data measured in the first and second surveys correlated positively (r =0.70, P =0.001), while there was no correlation between measurements made in the second and third surveys (r =0.07, P =0.766) (Fig. 1a). Unlike leaf reflectance, leaf attractiveness index also differed significantly between the genotypes in the last survey (Fig. 1b, Table 1). However, while the correlation between the genotype means of leaf attractiveness in the first and second surveys was positive (r =0.85, P <0.001), the correlation between means in the second and third surveys was negative (r = − 0.50, P =0.025) (Fig. 1b). Leaf reflectance and attractiveness correlated positively in each survey, but this correlation was much weaker for the third survey (r =0.55, P =0.011) than the first (r =0.92, P <0.001) and second (r =0.97, P <0.001) (Fig. 1a,b).
Table 1. F-statistics of ANOVA of the effects of silver birch (Betula pendula) genotype and replicate block on: leaf reflectance (measured at 530–535 nm) and attractiveness (calculated according to Döring et al., 2009) at the beginning (first survey, mostly green leaves), middle (second survey, greenish and yellow leaves) and end (third survey, yellow and dark yellow leaves) of leaf senescence; between-survey differences in leaf reflectance (used to estimate the rate of chlorophyll breakdown and phloem sap nitrogen (N) loading between surveys); and green leaf vs leaf litter difference in N concentration (used to estimate the total N export from the senescing leaves during the autumn)
First survey leaf reflectance
Second survey leaf reflectance
Third survey leaf reflectance
First survey leaf attractiveness
Second survey leaf attractiveness
Third survey leaf attractiveness
First vs second survey difference in reflectance
Second vs third survey difference in reflectance
Green leaf vs leaf litter difference in N%
The parameters that were used to estimate the rate of chlorophyll breakdown and phloem sap N loading (i.e. the between-survey differences in leaf reflectance) and the total N export from the senescing leaves (i.e. the green leaf vs leaf litter difference in N concentration) differed significantly among the silver birch genotypes (Fig. 1c,d, Table 1). The green leaf vs leaf litter difference in N concentration correlated strongly with green leaf N concentration (r =0.91, P <0.001) at the tree genotype level (Fig. 1d), but neither of these variables had a significant genotypic correlation with leaf reflectance or attractiveness in any survey (|r| < 0.34 and P >0.1 in each case) (Fig. 1a,b,d). As an indication of a positive link between leaf attractiveness and N availability, the first survey leaf reflectance and attractiveness had positive genotypic correlations with the first vs second survey difference in reflectance (r =0.45, P =0.047 and r =0.59, P =0.007, respectively) (Fig. 1c). However, the second survey reflectance and attractiveness values had strong negative genotypic correlations with the second vs third survey difference (r = − 0.84, P <0.001 and r = − 0.81, P <0.001, respectively) (Fig. 1a–c), and the genotype means of the two between-survey differences in leaf reflectance correlated negatively (r = − 0.72, P <0.001) (Fig. 1c).
Development of the aphid community
The aphid community was dominated by E. betulae: 74%, 90% and 87% of all aphids belonged to E. betulae in the first, second and third surveys, respectively. The abundance of the winged females decreased and the abundance of wingless females and nymphs increased along with the advance of autumnal leaf colouration (Table 2). The abundance of winged males did not have a clear trend with leaf senescence (Table 2).
Table 2. Total number of individuals of different Euceraphis betulae life-cycle stages found in the silver birch (Betula pendula) trees in the first, second and third field surveys
Winged viviparous females
Wingless oviparous females
The genetic variation in leaf reflectance and attractiveness recorded in the first and second surveys significantly explained the variation in the abundance of wingless E. betulae females recorded in the third survey: those silver birch genotypes that expressed the highest leaf reflectance and attractiveness at the beginning and middle of the leaf senescence process had the highest abundance of wingless females at the end of senescence (Fig. 2a–d). The abundance of wingless E. betulae females was also significantly explained by attractiveness in the third survey, but this association was negative (Fig. 2e). The abundances of E. betulae nymphs, males or winged females were not significantly explained by the genotypic variation in leaf reflectance or leaf attractiveness data of any survey (data not shown).
When testing whether phloem sap N availability could partly explain the E. betulae response to leaf reflectance, the second vs third survey reflectance difference was found to be negatively related to the abundance of wingless E. betulae females in the third survey (Fig. 2f). In the correlation analyses that tested the potential benefit of aphids inhabiting certain genotypes, the abundance of winged E. betulae males and females in the first and second surveys did not correlate with the estimates of subsequent phloem sap availability (|r| < 0.32 and P >0.10 in each case; data not shown). The numbers of males, however, correlated positively with the numbers of wingless females in the third survey at the tree phenotype level (r =0.37, n =58, P =0.004; the data set excludes trees where neither males nor females were found) (Fig. 3). Lastly, when testing the ability of the aphid community to affect the silver birch N economy, the abundances of E. betulae males and other species nymphs in the first and second surveys were found to significantly explain the variation in the amount of N lost in leaf litter (estimated using litter N concentration) at the tree genotype level (F4,18 = 3.81, R2 = 0.52, P =0.027).
Genetically controlled autumn leaf colouration and the aphid load in silver birch
Our results show that the green–yellow reflectance of silver birch leaves differs significantly among silver birch genotypes in the beginning and middle of the period of autumn leaf colouration. As far as we know, this is the first evidence that intrapopulation variation in autumn leaf colouration can have a significant genetic component. We also show that the genetic variation in silver birch leaf reflectance is positively correlated with the variation in leaf attractiveness for aphids. This suggests that the variation in leaf reflectance can affect aphid host selection in the autumn, and, supporting this, we found that the level of aphid infestation of silver birch trees was significantly explained by their genetically controlled autumn leaf colouration. In particular, we found that the abundance of wingless E. betulae females, which lay the over-wintering eggs, was highest at the end of leaf senescence in the silver birch genotypes that had the highest leaf green–yellow reflectance during the first phase of the senescence period. This suggests that those B. pendula trees that have leaves of lowest reflectance (i.e. the least yellow leaves) in the early autumn should gain over others in the spring when the new aphid generation emerges from the eggs. Significantly, however, it appears that the rank of silver birch genotypes is not stable through the entire period of autumn leaf colouration: in the latter phase of senescence, genetic differences in leaf reflectance disappear and there is a reversal in the rank of genotypes in terms of leaf attractiveness. This is an important finding and suggests that the aphid load of a particular birch genotype may depend on how the period of aphid host searching relates to the progress of autumn leaf colouration in different autumns. If the host searching period varies in different years, and unlike in our study does not always match with the early autumn leaf colouration period, the aphid load may not consistently be higher on certain genotypes and the coevolution between the two species might be attenuated. Likewise, year-to-year variation in weather conditions may have affected the coevolution. Short autumns with early frost should lead to tighter coevolution than long mild autumns that offer more temporal variation in genotypic differences. Noteworthily, if autumns are typically short and winters arrive early, aphid avoidance could efficiently reduce yellow leaf reflectance in early autumn, while aphid egg-laying could remain a weak selective agent in late autumn when there is not much time for aphid activity.
In contrast to what we expected, we did not find a significant relationship between the numbers of winged females, that is, the flying parents of the wingless females, and autumn leaf colour. This is probably a consequence of the low number of winged E. betulae females in our data set, as the preference of winged females for yellow over green leaves has been demonstrated earlier (Holopainen et al., 2009). The reason for wingless females being positively related to the green–yellow reflectance in our study is therefore probably a result of their winged parents preferring trees of high reflectance during host selection. Taken together, these findings indicate that E. betulae has the potential to drive the evolution of autumn leaf colouration in B. pendula populations. To conclusively demonstrate this, one should also provide evidence of the effect of aphid abundance on tree fitness (Archetti et al., 2009), but there are no such data available for silver birch. However, aphids are known to reduce tree growth and reproduction (Dixon, 1971; Kilpeläinen et al., 2009; Zvereva et al., 2010), and, combined with our data, such evidence indicates that natural selection should ultimately reduce yellow leaf colouration, or, as leaf yellowing is imposed by chlorophyll degradation, the evolution could be towards changing the timing or the duration of autumn leaf colouration in silver birch (e.g. Sinkkonen, 2006b; Archetti, 2009a).
To the human eye, hues of red are the most striking autumn leaf colours, but, unlike humans, aphids lack red photoreceptors and prefer both green and yellow over red (Döring et al., 2009). Red autumn leaf colours, which are caused by autumnal anthocyanin production, can therefore act as an antiherbivory adaptation. It was recently suggested that such adaptation was lost in most North European tree species during glaciations when insect populations became extinct, but tree populations survived in various refuges (Lev-Yadun & Holopainen, 2009). This would explain why tree species, like silver birch, in North Europe have yellow rather than red autumn leaf colours even though yellow attracts aphids. Our study provides novel findings that support this hypothesis: as intense yellow leaf colour exposes silver birch trees to aphid infestation and leaf reflectance has a genetic background, any mechanism that can camouflage the attractiveness of yellow leaves, such as red pigment synthesis in the autumn, should be favoured by natural selection. This suggests that herbivores can act as selective agents in the evolution of autumn leaf colours even when the active colour pigments, such as the carotenoids in silver birch, are not synthesized in the autumn but earlier in the growing season.
Our results do not support the hypothesis that yellow autumn leaf colours have evolved as a warning signal for aphids (Archetti, 2000; Hamilton & Brown, 2001), but suggest that the distinction between yellow and red autumn leaves in later hypotheses (e.g. Lev-Yadun & Gould, 2007; Archetti et al., 2009; Lev-Yadun & Holopainen, 2009) is indeed relevant. Although we found that leaf attractiveness in late autumn was negatively associated with the abundance of wingless E. betulae females (Fig. 2e), we did not find any winged females in our last survey and therefore host tree selection at this stage of leaf senescence had no potential to influence antiherbivory colouration. This suggests that the negative association between leaf attractiveness and aphid abundance was rather a legacy of the positive association that originated earlier in leaf senescence and host selection processes. In any case, this finding also stresses the importance of considering the temporal variation in the aphid–leaf colour association during the process of leaf senescence. Interestingly, we found that the abundance of E. betulae males was not related to leaf reflectance, not even at the tree phenotype level (r =0.259, P =0.116, n =38, with trees involved that contained males), but was instead significantly correlated with the abundance of wingless females. This indicates that not all autumn-migrating aphids automatically concentrate on trees with the most attractive leaves, but also use other cues for their orientation, such as female pheromones in the case of male aphids (Holopainen et al., 2009).
The population density of E. betulae was low in our study: the average number of adults varied from 3 to 11 and the number of nymphs from 14 to 28 per 100 leaves in the three successive surveys. In the same year, in another locality, Holopainen et al. (2009) found an average of 56 adults and 63 nymphs per 100 yellow senescing leaves in their mid-summer survey of a population of flood-stressed 2-yr-old silver birch saplings. The reasons for the low densities in our study may be manifold, ranging from the low position of the inspected branches within tree canopies to the locality of the birch population in the landscape and the weather and season of the survey. Nevertheless, the difference in the abundance of wingless E. betulae females was over 20-fold between the genotypes of the lowest and highest aphid load. This shows that genotypic variation in the autumn leaf colours can lead to a noteworthy genotypic variation in aphid load within a local birch population.
Aphids and tree N dynamics during autumn leaf colouration
Our results show that at the time of the first survey leaf reflectance was positively associated with the reflectance difference between the first and second surveys (an indication of chlorophyll breakdown and phloem sap N loading between the surveys), which suggests that the most attractive trees at the onset of leaf colour change provided the highest sap N loading during early autumn leaf colouration. This seems to be consistent with the nutrient retranslocation hypothesis (Holopainen & Peltonen, 2002), which proposes that migrating aphids can use yellow as a signal of available N. However, the positive association between leaf attractiveness and subsequent sap N loading was reversed at mid-autumn leaf colouration and the most attractive trees in the second survey had the lowest sap N loading between the second and third surveys. This finding seems to support the side-effect hypothesis (Döring & Chittka, 2007) rather than the nutrient retranslocation hypothesis: choosing the yellowest trees does not seem to benefit the aphids (in terms of subsequent phloem N availability) in the middle of autumn leaf colouration. This shows, once again, how the genetically controlled attributes of leaf senescence and autumn leaf colouration can involve significant temporal variation. The remarkably low numbers of wingless E. betulae females in the trees of our genotype ‘30’, despite the highest reflectance and attractiveness of these trees at the time of the second survey (Fig. 2b,d), might be an indication that aphids are adapted to recognize this variation: that is, once the reflectance of a tree increases above a certain threshold value, aphids could start avoiding the tree because of a minimal subsequent N availability in such a tree. Among our genotypes, ‘30’ was the only genotype for which the survey 1 vs 2 difference in leaf reflectance was significantly higher than the survey 2 vs 3 difference (Fig. 1c), and, when contrasting the second survey reflectance and wingless E. betulae female abundance, ‘30’ emerges as a statistical outlier (a regression without ‘30’ would give F =22.5, P <0.001 and R2 = 0.58; Fig. 2b). Although this piece of evidence that aphids avoid trees above a threshold reflectance level is not unequivocal, it supports earlier observations of aphids avoiding leaves that will soon be shed (Glinwood & Pettersson, 2000) as well as the hypothesis that aphids can use leaf reflectance as an indication of upcoming leaf fall (Lev-Yadun & Gould, 2007). Finally, it seems to suggest that the possibility of coevolution between aphids and trees having yellow autumn leaf colours should not be completely ignored (e.g. Sinkkonen, 2006a; Archetti et al., 2009).
We further tested whether leaf attractiveness could be used as a sign of total autumnal N transfer from the senescing leaves. Our silver birch genotypes differed significantly in summer leaf N concentration and the variation caused a matching variation in the autumnal N transfer (measured as a difference between green leaf and leaf litter N concentrations). However, this genetic variation did not correlate with leaf reflectance or attractiveness at any stage of leaf senescence and therefore cannot be a significant factor in the context of colour-induced aphid host selection. Lastly, we tested if the responses of E. betulae life-cycle stages to leaf reflectance could be explained by N availability, but did not find evidence for this. The abundance of E. betulae wingless females in the third survey was associated with the reflectance difference between the second and third surveys, but, as this association was negative, it is not a likely reason for the observed abundance in the third survey. However, it has to be noted that, as aphids utilize soluble amino acids in leaf phloem, but not other forms of leaf N (White, 2009), and can recognize amino acid concentration in phloem sap (Nowak & Komor, 2010), our estimates of aphid N availability in the autumnal trees can only be considered approximate.
We found an indication that the autumnal aphid community could also reciprocally affect the silver birch N economy: the abundance of E. betulae males and the nymphs of the other aphid species found on silver birch genotypes in the first and second surveys explained a significant part of the variation in leaf litter N loss among the genotypes. These aphid groups do not appear to respond to leaf reflectance (this study and Holopainen et al., 2009) and we cannot suggest a mechanism to explain their effect on nutrient resorption efficiency, except that the effects seem to develop during early and mid leaf senescence when chlorophyll breakdown and N transfer are most active. Autumnal nutrient resorption has a major effect on tree growth and fitness (May & Killingbeck, 1992) and B. pendula spring leaf growth, for example, can exclusively rely on internal cycling for 2 wk after bud burst (Millard et al., 1998). As the association between the aphid abundance and N loss further seems to take place at the tree genotype level, this may cause additional selection pressure on those silver birch attributes, such as volatile emissions (Holopainen et al., 2010; Powell et al., 2006), that are not related to leaf colouration but may affect aphid load in trees in the autumn.
Plant–insect interactions have always been of considerable interest in ecology, but studies of late-season host–insect dynamics have been lacking, with the exception of migrating cereal pests (Klueken et al., 2009). The birch population that we used in our study has been shown to express significant genotypic variation in herbivore resistance during several leaf growing seasons (Silfver et al., 2009). Our results show that such intrapopulation genotypic variation in leaf characteristics and herbivore load are not restricted to the main leaf growing season, but also prevail during leaf senescence.
With respect to the coevolution hypothesis of autumn colouration (e.g. Archetti et al., 2009), this means that coevolution of autumn-migrating insects, like aphids, and tree autumn leaf colouration, is also possible when the colours attract (yellow) rather than discourage (red) the insects. Likewise, our results indicate that such coevolution may arise with aphid species that do not change their host species in the autumn, as long as they search for an optimal host while nutrient transport from leaves to woody tissues is ongoing. Taken together, although our study needs to be repeated with other tree species and genera, our findings indicate that, besides being one potential explanation for the myriad autumn leaf hues of trees in the temperate and boreal climate, the behaviour of autumn migrating insects may have a significant role in understanding tree–insect interactions in general.
Thomas F. Döring helped with the analyses of attractiveness and gave constructive comments on an earlier version of the manuscript. Hanni Sikanen helped in the field and Matti Salovaara in the laboratory. We would also like to thank Dr James Blande for editing the language and Prof. Simcha Lev-Yadun and an anonymous reviewer for constructive comments on an earlier version of the manuscript. The study was funded by the Academy of Finland (decision #1122444 for U.P., E.S. and J.M. and decision #133322 for J.K.H.).