Skip to main content
Browse Subject Areas

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

Effects of Hybridization and Evolutionary Constraints on Secondary Metabolites: The Genetic Architecture of Phenylpropanoids in European Populus Species

  • Celine Caseys ,

    Affiliations Unit of Ecology and Evolution, Department of Biology, University of Fribourg, Fribourg, Switzerland, Department of Botany and Biodiversity Research Centre, University of British Columbia, Vancouver, Canada

  • Christoph Stritt,

    Affiliation Unit of Ecology and Evolution, Department of Biology, University of Fribourg, Fribourg, Switzerland

  • Gaetan Glauser,

    Affiliation Neuchâtel Platform of Analytical Chemistry, Faculty of science, University of Neuchâtel, Neuchâtel, Switzerland

  • Thierry Blanchard,

    Affiliation Unit of Ecology and Evolution, Department of Biology, University of Fribourg, Fribourg, Switzerland

  • Christian Lexer

    Affiliations Unit of Ecology and Evolution, Department of Biology, University of Fribourg, Fribourg, Switzerland, Department of Botany and Biodiversity Research, University of Vienna, Vienna, Austria


The mechanisms responsible for the origin, maintenance and evolution of plant secondary metabolite diversity remain largely unknown. Decades of phenotypic studies suggest hybridization as a key player in generating chemical diversity in plants. Knowledge of the genetic architecture and selective constraints of phytochemical traits is key to understanding the effects of hybridization on plant chemical diversity and ecological interactions. Using the European Populus species P. alba (White poplar) and P. tremula (European aspen) and their hybrids as a model, we examined levels of inter- and intraspecific variation, heritabilities, phenotypic correlations, and the genetic architecture of 38 compounds of the phenylpropanoid pathway measured by liquid chromatography and mass spectrometry (UHPLC-MS). We detected 41 quantitative trait loci (QTL) for chlorogenic acids, salicinoids and flavonoids by genetic mapping in natural hybrid crosses. We show that these three branches of the phenylpropanoid pathway exhibit different geographic patterns of variation, heritabilities, and genetic architectures, and that they are affected differently by hybridization and evolutionary constraints. Flavonoid abundances present high species specificity, clear geographic structure, and strong genetic determination, contrary to salicinoids and chlorogenic acids. Salicinoids, which represent important defence compounds in Salicaceae, exhibited pronounced genetic correlations on the QTL map. Our results suggest that interspecific phytochemical differentiation is concentrated in downstream sections of the phenylpropanoid pathway. In particular, our data point to glycosyltransferase enzymes as likely targets of rapid evolution and interspecific differentiation in the ‘model forest tree’ Populus.


Plant secondary metabolites—chemical compounds not directly involved in plant growth, development, and reproduction—have a variety of roles, ranging from pigmentation to protection from ultraviolet (UV) radiation to defence against insect herbivores [14]. These key traits modulate many different types of ecological interactions and comprise a stunning diversity of active chemical compounds. The biochemical pathway targeted in the present study, the phenylpropanoid pathway, is estimated to represent twenty percent of all fixed carbon available in plant terrestrial ecosystems [5] and includes flavonoid, tannin, anthocyanin and lignin compounds.

The evolution of plant secondary metabolite diversity represents a topic of considerable interest in evolutionary biology and plant science [68]. In the current eco-evolutionary view, the origin and maintenance of plant chemical diversity is typically explained by adaptive evolution (i.e. natural selection driven by insect herbivores) and co-evolution (i.e. arms-races or frequency-dependent processes) between plant populations and interacting herbivores and pathogens [810]. However, changes in plant secondary metabolites due to hybridization can also affect specific biotic [1113] and abiotic [1, 3] interactions and can have broader community effects [14]. In this study, we focus on secondary metabolite abundances in hybridizing species as functionally important suites of traits with potential links to population divergence, speciation, and species interactions.

Hybridization between species leading to taxa with ‘porous genomes’ is often observed in recently radiated groups of animals and plants [1518]. In plants, hybridization has long been known to affect the number, diversity, and quantity of secondary metabolites [19]. Reviews of the effects of hybridization on secondary metabolism and herbivore resistance [1921] indicate that different groups of secondary metabolites are affected by hybridization in roughly the same way, although there is still a dearth of information on chemical variation within and between different hybrid generations [21]. Typically, plant hybrids synthesize traits present in both parental species for approximately 70% of chemical traits. This high degree of additivity in hybrids, observed for most chemical traits, is complemented by less frequent phenomena such as the presence of transgressive (extreme) chemical traits, including compounds that are synthesized at higher concentrations in hybrids than in the parental species, and by the synthesis of novel compounds in hybrids. In hybridizing European Populus species, these proportions of traits shared between parental species and natural hybrids (comprising different hybrid generations; [22, 23]) are similar, with 50% of phenylpropanoid traits with intermediate phenotypes in hybrids, 20% of traits with parental-like phenotypes in hybrids, and 20% of traits with transgressive phenotypes [24]. While the effects of hybridization on secondary metabolites at the phenotypic level are well characterized, the genetic causes of the frequent chemical additivity and of the rarity of chemical novelty in hybrids remain largely unknown.

Plant secondary metabolism is postulated to have emerged from primary metabolism in response to environmental pressures. It is thought that gene duplication, neofunctionalization, metabolic diversification, and convergent evolution have all contributed to its emergence and expansion during the evolution of plant lineages [6, 7, 25]. Rates of metabolic enzyme evolution in plants and animals were shown to vary with position, connectivity, and metabolic flux within metabolic pathways [26, 27]. Enzymes in the upstream portions of pathways have a strong tendency to be under stronger selective constraints, to evolve more slowly, and to be more strongly pleiotropic due to their effect sizes. For example, modifications of the first enzyme of a pathway will generally have a stronger effect on the total pathway than modifications of downstream enzymes [28, 29]. In plant secondary metabolism, such differences in evolutionary constraints were found in different plants and pathways [2831]. We apply the current knowledge and models of pathway evolution to our phenotypic data, in an attempt to further identify branch point enzymes leading to interspecific chemical differentiation within the phenylpropanoid pathway in hybridizing perennial plant species. We investigate these issues in a plant group with ongoing speciation in the face of gene flow, namely Eurasian members of the ‘model forest tree’ genus Populus.

We studied three natural hybrid zones and a common garden trial of Populus alba (White poplar) and P. tremula (European aspen), two ecologically divergent Eurasian species that form part of a species complex in section Populus of genus Populus [32]. Populus species (poplars, aspens, cottonwoods) represent textbook examples of forest foundation species with genetic variation in chemical defence traits that affect entire communities of herbivores and pathogens [14, 33]. Heritable variation for biotic interaction-related traits in Populus spp. has been found at different genetic scales, ranging from pairs of hybridizing species [14, 24, 34, 35] to local populations of the same species [36].

Using the Populus model, we ask three questions of current interest to the evolution of the plant secondary metabolism and defence: (1) Does the expression of secondary metabolites vary between species and geographic localities? (2) Do phytochemical defence traits in hybridizing species exhibit distinct genetic architectures involving major quantitative trait loci (QTL)? (3) Does evolution of interspecific chemical differentiation in hybridizing species involve chemical hubs and modification through enzymes with species-specific activity? To address these questions, we combine phenotyping of phenylpropanoids by liquid chromatography and mass spectrometry with molecular marker based quantitative genetic analysis of compound abundances. In particular, we examine patterns of variation in phenylpropanoid traits between species and across geographic localities, estimate the heritability of the same traits in natural populations and a common garden, and explore their genetic architecture by genetic mapping. Genetic architecture holds the key to understanding the genetic correlation structure among traits involved in biotic and abiotic interactions [37], an approach we complement by phenotypic correlation networks.

Materials and Methods

Plant materials and sampling design

Populus alba (White poplar) and P. tremula (European aspen) share broad and overlapping distribution ranges across Eurasia. The two species exhibit ecological divergence at multiple spatial scales, ranging from very large scales (P. alba reaching further south into Africa than its congener, P. tremula further north across the arctic circle) to small scales (P. alba found as pioneer in lowland flood plain forests, P. tremula primarily in upland forests) [38]. In geographic areas of overlap, P. alba and P. tremula form extensive homoploid (2n = 19) hybrid zones [22, 39]. Hybrids (a.k.a. P. x canescens) appear to be genomic mosaics with balanced genetic contributions from each parental species [22, 23], resulting in phenotypic mosaics of morphological and phytochemical traits [24, 40].

Two interconnected experiments formed the basis for this study: pheno- and genotyping of natural hybrid zones of the two species informed us on phytochemotypes in adult trees in different geographically separated populations that likely experienced spatially varying selection [41, 42], whereas a common garden trial facilitated comparison of phytochemotypes in a common and controlled environment. Sampling of natural populations comprised a total of 260 individuals (163 hybrids, 51 P. alba-like and 46 P. tremula-like trees; S1 Table) from three natural hybrid zones, referred to as (1) Ticino river (Italy), (2) Danube river (Austria), and (3) Tisza river (Hungary) valley hybrid zones [22]. Sampling permits for the Danube river (Austria) population were issued by the park authorities of the Donau-Auen National Park. No specific permits were required for the remaining locations. This study did not involve endangered or protected species. These central European hybrid zones represent differences in longitude ranging from 8.98 to 22.26OE from North-western Italy to eastern Hungary, but do not present large variation in latitude (between 45.28 and 48.32ON). The common garden trial was composed of 133 seedlings (S1 Table) from 15 open pollinated families with 5 to 17 individuals per family, planted in an unbalanced block design with randomization within blocks. The trial was established in 2011 in Fribourg, Switzerland, in typical P. tremula habitat from seeds collected from individual female trees in the Ticino river hybrid zone. The trial forms part of a larger, reciprocal common garden design involving multiple localities, to be genotyped and fully analysed at a later time. The common garden seedlings were characterized with microsatellites (S1 Materials; S2 Table) by estimating admixture proportions Q [23] and the correlation of paternity (Cp) for outcrossing species [43].

Identification of phenylpropanoids and measurement of inter- and intra-specific geographic variation

Phenylpropanoids were previously shown to play an important role as phytochemical defence traits in Populus [3]. Metabolomic analysis of individual samples from natural populations [24] focussed on phenylpropanoids as biologically relevant compounds. In short, secondary metabolites were extracted from ground tissue of three silica-dried leaves per tree. 60% methanol extracts were used for both condensed tannin quantification and metabolite fingerprinting by Ultra-High-Pressure-Liquid-Chromatography Quadrupole-Time-Of-Flight Mass-Spectrometry (UHPLC-QTOF-MS) (S1 Materials). Thirty-eight compounds of the phenylpropanoid pathway were identified or tentatively identified by high-resolution mass spectrometry data and chemical standards (S3 Table). The targeted secondary metabolites were quantified in a relative manner via the normalized peak areas of chromatograms. Condensed tannins were quantified by the butanol-HCl method [27].

Inter- and intra-specific components of variation in phytochemical traits were explored by Principal Component Analysis (PCA; duality diagram function, ade4 R package plotted with ellipse function of car R package). In natural hybrid zones, raw mass spectrum data of all assayed chemical compounds (including >2800 datapoints for each individual) were compared to the selected 38 phenylpropanoids to verify correspondence of our targeted dataset with the phytochemical complexity of the total leaf extracts. In addition, PCA was used to detect interspecific and intra-specific (geographic) differentiation in each of three different subgroups of the phenylpropanoid pathway: chlorogenic acids, salicinoids, and flavonoids. The repeatability of chemical patterns in hybrid zones and the common garden was checked by comparing the positions of known pure parental and hybrid plants in principal component space. The phenotypic interspecific differentiation (S3 Table) in natural hybrid zones and the common garden was assessed with the Tukey “honest significant difference” method in R. Level of significance was fixed at adjusted P-values <0.05.

Estimating the genetic component of phytochemical traits in natural populations and a common garden trial

The genetic component of phenotypic variation in the abundances of targeted phenylpropanoid metabolites was estimated by two different methods: (1) Mantel correlations between pairwise genomic similarities (derived from genomic admixture proportions Q, equivalent to hybrid indices) and phenotypic similarities in natural populations in the absence of pedigree information [24], based on the marker-inferred relatedness approach [44]; (2) formal estimation of heritability by half-sib analysis [45] in the common garden trial. Heritabilities (h2) in the common garden half-sib design were calculated from among- and within-families variance estimates [Var(s) and Var(e)] calculated with Restricted Maximum Likelihood (REML) in SPSS Statistics 20. This approach accounts for the unbalanced design resulting from variable numbers of individuals available for each family [45]. As microsatellite data indicated the presence of 31% full-siblings across the common garden families (correlation of paternity) [43], the original half-sib analysis equation was modified by multiplying the intraclass correlation by 3.

Genetic mapping of phytochemical traits in natural hybrid zones

Natural hybrid zones are considered as useful tools to study the genetic basis of trait differences between diverging populations and species [46, 47]. Therefore, we studied three natural hybrid zones [22] to assess the genetic architecture of phytochemical traits with known or suspected roles in plant defence [1, 2, 4, 35] by in situ genetic mapping (‘admixture mapping’). The mapping panel was identified based on the genetic admixture proportions Q (below) among >800 trees genotyped in our laboratory. The genotypic composition of the mapping panel was estimated with 77 DNA microsatellite markers from all 19 chromosomes of Populus [23]. Although much larger marker numbers can be obtained in these hybridizing species using genotyping-by-sequencing approaches [48], we chose to use microsatellite markers in the present study, because genetic data for them were fully available for all studied populations. Genetic data sets of this size (>40 000 data points for mapped markers from all chromosomes) are still relatively rare in studies of natural hybrid zones, and are sufficient for coarse-scale genetic mapping in recently admixed populations [49, 50].

The admixture mapping approach applied in this study is based on linear regression analysis and model selection [50]. In short, locus specific ancestries (LSA’s) were estimated for all microsatellite loci [23]. Standardized residuals from ANOVA of phenotypic data with locality as factor were calculated to account for variation among hybrid zones. Single-locus associations between phenotypes and LSA’s of each marker locus were then modelled by two generalized linear models, an additive and a dominant model, and compared using the Akaike Information Criterion (AIC). Candidate QTL with AIC≥4 were subsequently subjected to a two-step procedure of model selection (forward selection and backward elimination) to reduce false positives [50]. Percentages of Variance Explained (PVE) were estimated from linear combinations of the candidate QTL with AIC≥4 with a likelihood-ratio based pseudo-R-squared approach (r.squaredLR function from MuMIn R package).

As an indication for potential long-term selective pressures acting on these genome regions, departures from neutrality of marker loci associated with phytochemical QTL were examined in the form of interspecific heterozygote excess (an excess of heterozygotes carrying one allele of each parental species) in natural hybrids relative to the remainder of the genome [23], following the well established ‘genomic clines’ method [51]. This approach of studying departures from neutral expectations falls within a family of population genetic methods sometimes referred to as the “spy glass” approach to studying the evolutionary process, which complements the “magnifying glass” approach of studying contemporary selection with the toolbox of evolutionary ecology [52]. Interspecific heterozygote excess was extracted from genomic clines previously fitted for the same three hybrid zones [23]. Our focus was on markers with an excess of interspecific heterozygotes in the Italian hybrid zone, i.e. the population of origin of the common garden trial, and across at least two of the studied hybrid zones (Italy, Austria, and Hungary). Fisher’s exact tests were used to check whether instances of significant heterozygosity excess were more frequent for markers representing QTL for phytochemical traits than for markers from across the genome. The rationale of this test is thus similar to other QTL-based tests for selection [53, 54], with the shared caveat that the excess or dearth of particular genotype classes at the studied loci may be affected by local patterns of linkage disequilibrium (LD) along the genome.

Phenotypic correlations of targeted phytochemicals

Previous work focusing on flavonoids in Populus showed clear patterns of phenotypic correlations in full-sib hybrid families [55]. In the present study, phenotypic correlation networks from the 38 studied phytochemicals identified by UHPLC-QTOF-MS (S3 Table) were used to visualize the strength of correlations between different subgroups of phenylpropanoids, and between taxa. Correlation networks do not systematically represent compound proximity within metabolic pathways, but rather represent an overview of physiological state accounting for genetic, biochemical, regulatory and environmental variations [56, 57]. Phenotypic correlations were assessed with non-parametric Spearman’s rank correlations in SPSS to account for the non-normal distribution of 34 out of 38 targeted traits. Normality of the studied compounds was assessed by the Shapiro-Wilk normality test in R. Correlation networks were generated for Spearman correlations with r>0.5 at p<0.001 using the Fruchterman-Reingold 2D algorithm ( function from Network and gplot function from Sna R package).


Inter- and intra-specific variation of phytochemical traits

The first two PC’s of the raw mass spectrum data (Fig 1A) accounted for 15.3% of the total variance in the three studied natural hybrid zones. The major axis of variation (PC1) resided between species, whereas PC2 captured a gradient from East (Tisza river hybrid zone, Hungary) to West (Ticino river hybrid zone, Italy; Fig 1). The first two axes from PCA of 38 phenylpropanoids (Fig 1B) accounted for 47.1% of the total variance and yielded a pattern consistent with the full metabolomic data (Fig 1A). This suggested that the targeted compounds were representative of the total variation of metabolites present in the plant extracts. The first two axes of PCA of the 18 identified flavonoids accounted for 74.5% of the total variance (Fig 1C), with clear partitioning into interspecific variation along PC1 and geographical intraspecific variation along PC2. In contrast, PCA of chlorogenic acids and salicinoids (S1 Fig) indicated that these subgroups did not present substantial inter- and intraspecific differentiation. In the common garden, the first two PC axes explained 58.9% of the total variance of the 38 phenylpropanoids (S1 Fig) and indicated the repeatability of phenotypic patterns between natural populations and the common garden trial. Overall, intraspecific chemical variation in the parental species was lower than interspecific variation (S4 Table), thus resulting in substantial differentiation of phytochemical traits between P. alba and P. tremula (Fig 1).

Fig 1. Principal component analysis of secondary metabolites in hybridizing European Populus species.

Principal Component Analysis (PCA) of secondary metabolites identified by UHPLC-QTOF-MS in Populus alba (in red, square), P. tremula (in green, cross) and their hybrids, P. x canescens (in blue, crossed square). Symbols specific to each species and hues specific to hybrid zone (Ticino: Italy, Ticino river hybrid zone; Danube: Austria, Danube river hybrid zone; Tisza: Hungary, Tisza river hybrid zone) represent individuals. Ellipses represent 95% confidence intervals and coloured dots represent averages for each group. The first two principal components (PC’s) are plotted for (a) PCA of the raw mass spectrum data, i.e. all secondary metabolites detected by UHPLC-QTOF-MS, with PC1 and PC2 explaining 9.3% and 6.0% of the phenotypic variance; (b) PCA of the 38 targeted chemical traits: PC1 and PC2 explaining 30.9% and 16.2% of the phenotypic variance; (c) PCA of 18 flavonoids: PC1 and PC2 explaining 57.2% and 17.3% of the phenotypic variance, respectively.

Genetic component of the studied traits in natural populations and the common garden

Great variation in heritability across phytochemical traits was detected by both, Mantel correlations (MC) in natural populations and half-sib analysis in the common garden trial (Table 1; [24]). There was a clear tendency for greater heritability in flavonoids and chlorogenic acids compared to salicinoids. In natural populations, Mantel correlations (Fig 2A, Table 1, p<0.05) for flavonoids (mean: 0.35; range: 0.09–0.67) were larger than for chlorogenic acids (mean: 0.25; range: 0.11–0.52;) and salicinoids (mean:0.22; range: 0.1–0.35). In the common garden trial, 43% and 37% of h2 heritability estimates (Fig 2B; Table 1) were ≥0.50 for flavonoids (mean: 0.36; range: 0–1.28) and chlorogenic acids (mean: 0.46; range: 0.04–0.94) compared to 23% for salicinoids (mean: 0.34; range: 0–0.99). Low heritabilities (h2≤0.3, MC<0.2) were found for several traits of all three functional groups and were most frequent in salicinoids with low levels of substitution (salicin and salicortins) and isorhamnetin-derivatives in the flavonoid group. The h2 of one phytochemical trait exceeded 1, reflecting the experimental trade-off of measuring phenotypes at early life stages that matter most to tree establishment but are also most strongly affected by maternal effects [45].

Fig 2. Indicators of genetic determination of phytochemical traits representing different branches of the phenylpropanoid pathway.

Violin plots indicate medians (white dots), quartiles (thick black bars) and probability distributions (grey areas) of different surrogates of the degree of genetic determination or heritability in different branches of the phenylpropanoid pathway. (a) Significant Mantel correlation coefficients between phenotype and genotype in the Ticino river valley hybrid zone (Italy). (b) Estimates of heritability (h2) in hybrid seedlings from maternal families in the common garden. (c) Percentage of variance explained (PVE) by QTL for phytochemical traits detected by genetic mapping in natural hybrids.

Table 1. Thirty-eight phenylpropanoid compounds identified by UHPLC-QTOF-MS in Populus.

Genetic architecture of phytochemical traits revealed by admixture mapping in natural populations

Single-locus tests for genotype-phenotype associations revealed one to several QTL (Table 1) for 39 metabolic traits (Fig 3; S5 Table). Out of these, 41 QTL were significant in additive and 25 in dominant generalized linear models (Fig 3). The percentages of variance explained (PVE) varied among functional groups of traits (Figs 2C and 3). The PVE’s of QTL for chlorogenic acids, salicinoids, and flavonoids exhibited means and ranges of 10.5% (4–16), 19.7% (6–34), and 28.5% (3–55), respectively (Fig 2C; S5 Table). Modes of gene action inferred from standardized residual phenotypes and LSA’s indicated over-dominance for seven, under-dominance for three, P. alba-dominance for three, P. tremula-dominance for four, and additivity for the remaining QTL (Figs 3 and 4; S5 Table). Based on results from Fisher’s exact tests, all three groups of traits exhibited significantly elevated numbers of QTL with heterozygosity excess compared to markers from across the genome in the Italian hybrid zone (S6 Table), and flavonoids did so in ≥ two hybrid zones

Fig 3. Genetic architecture of phytochemical traits inferred in natural hybrid zones of European Populus species.

Seven chlorogenic acids, 13 salicinoids and 19 flavonoids (y axis) were tested for association with genetic markers (x axis) spread across all 19 chromosomes of Populus (roman numbers at the top) following an information theoretic approach [50]. For each candidate locus, additive (▼) or dominant (▲) effects are documented. Symbol sizes represent PVE’s of each candidate locus (▲<15%, ▲15–30%, ▲>30%). For details see S5 Table.

Fig 4. Examples of gene action in hybrids and effects on phenotype.

Phenotypes for each species (left panel) and phenotype-genotype associations for four exemplary QTL (right panel; for detailed information on mapped QTL see S6 Table). Boxplots in the left panel represent medians, interquartile ranges, and outliers for phytochemical traits in Populus tremula (trem), P. x canescens hybrids (middle of graphs) and P. alba (alb). Scatter plots in the right panel show standardized chemical phenotypes (y-axis) plotted against Locus Specific Ancestries (LSA’s) with 0 indicating P. tremula and 1 indicating P. alba ancestry. Fitted linear regression lines indicate gene action as additive, over-dominant and dominant for alleles of one or the other parental species [50].

Phenotypic correlations in natural populations

Correlation networks (Fig 5) largely reflected functional relationships within each group of phytochemical traits. Trait correlations, estimated as the number of significantly (p<0.001) correlated trait pairs with r>0.5 varied between trait groups and species. In hybrids from the Ticino hybrid zone (average number of correlated pairs: 4.8; Fig 5A), salicinoids (mean: 7.0; range: 1–11) were more strongly correlated than flavonoids (mean: 3.9; range: 1–10) and chlorogenic acids (mean: 3.3; range: 1–6). Correlations patterns were similarly strong in P. alba (average number of correlated pairs: 4.4; Fig 5B) and were generally weaker in P. tremula (average number of correlated pairs: 0.9; Fig 5C). Viewed by functional groups of compounds, salicinoids comprised a tightly correlated cluster of traits, and flavonoids grouped according to sugars but not according to aglycones (core molecules of flavonoids, here Quercetin, Isorhamnetin and Kaempferol) (Fig 5). A consistent correlation between salicinoids and isorhamnetin-rhamnoside (a flavonoid) was present in all taxa.

Fig 5. Correlation networks of phenolics in hybridizing European Populus species.

Metabolite correlation networks of 38 phenylpropanoids (for abbreviations see Table 1) identified by UHPLC-QTOF-MS. Chlorogenic acids are coloured in white, salicinoids in grey and flavonoids in black. Correlation networks were obtained by applying Fruchterman-Reingold 2D algorithm on Spearman correlations (r>0.5, p<0.001). Metabolite correlation networks are presented for (a) admixed individuals from the Ticino river valley hybrid zone (n = 109), using thresholds of 0.05 < genomic admixture proportion Q < 0.95 to delimit admixed trees; (b) individuals of P. alba (n = 51), (c) individuals of P. tremula (n = 46) from natural hybrid zones.


Our study on the genetic architecture and phenotypic patterns of phytochemical traits in hybridizing Populus species differs from other recent studies on plant defence and metabolomics, which have addressed related issues primarily at the within-species level [9, 10, 36]. Our focus on phytochemical variation present in hybridizing species is of particular relevance to organismal groups in which leaky species barriers and ‘porous’ genomes are widespread [15]. The effects of hybridization on the abundances of secondary metabolites in hundreds of trees from three natural hybrid zones of European Populus species in our study are in line with expectations from many other plant taxa [21, 24]: hybrids present a majority of additive traits, few parental-like traits, rare transgressive traits, and little detectable chemical novelty. However, our study goes an important step beyond accounting for the phenotypic effects of hybridization by providing insights into the genetic architecture underlying these phenotypes (Figs 3 and 4). We also examined inter- and intraspecific patterns of variation for targeted phytochemical traits across geographically separated hybrid zone localities. Moreover, our phytochemical data allowed us to single out a specific gene family (glycosyltransferases) on which hybridization may have particularly strong effects.

Inter- and intra- specific geographic variation in secondary metabolite abundances

Interspecific variation in phenylpropanoid abundances in European Populus species is sufficiently strong to potentially allow species identification with LC-MS techniques [24]. The first principal component axis of phenylpropanoid traits (Fig 1 and S1 Fig) is highly correlated with a hybrid index measured with molecular techniques such as microsatellites [23, 24] or second-generation sequencing based markers [58]. Still, intraspecific variation is an important factor, since plant defences and secondary metabolism are known to exhibit geographic variation [10, 59, 60]. Geographic structure in defence genes, metabolome traits and herbivore communities has recently been documented in Swedish populations of P. tremula [36]. However, no geographic structure was found for salicinoids in the same set of populations [61]. Here, we explored geographic patterns of phytochemical traits along a longitudinal transect of three natural hybrid zones of P. tremula and P. alba. We found no evidence of geographic structure for chlorogenic acids and salicinoids, but strong geographic structure for flavonoids in the studied species and their hybrids (P. x canescens). These differences in geographic structure between different subgroups of phenylpropanoids can plausibly be explained by (1) differences in levels of available genetic variation among phytochemical traits representing these subgroups of the pathway [10, 60] and/or (2) differential involvement of these subgroups in genotype x environment (GxE) interactions, i.e. local adaptation [62].

Our genetic mapping results show that flavonoids exhibit a stronger genetic component (e.g. larger PVE’s; Fig 2C) than salicinoids and chlorogenic acids. The apparent geographic pattern from East to West seen in these species and hybrids (Fig 1) suggests that phytochemical trait expression follows similar geographic patterns as previously seen for neutral genetic markers, presumably caused by patterns of postglacial recolonization [22, 41, 63].

Although the presence of genetic (= heritable) variation alone may explain geographic structure as observed for flavonoids in the present study (Fig 1C), local adaptation remains a suitable alternative hypothesis. Whereas salicinoids are frequently discussed as typical anti-herbivory compounds in Salicaceae, flavonoids also fulfil more general functions by conferring both biotic and abiotic protection, especially in the context of water stress (hydrology, precipitation) and UV radiation [1, 64]. In addition, the large flavonoid molecules known as condensed tannins are textbook examples for traits with ‘extended phenotypes’, thus suggesting the ecological importance of this subgroup of the phenylpropanoid pathways in Populus [65]. Reciprocal transplant experiments would allow specific tests for the potential roles of these phytochemical traits in local adaptation [62].

Genetic architecture and phenotypic correlations of phenylpropanoids

Coarse-scale genetic mapping based on genetic markers from all 19 poplar chromosomes provided a glimpse of the genetic architecture of 38 phenolics identified and quantified by UHPLC-QTOF-MS and condensed tannins (Fig 2; S4 Table). In particular, the results are consistent with the presence of ‘major genes’ controlling species differences likely arisen by divergent natural selection or a combination of selection and drift [50, 66]. We have studied genetic variation segregating between highly divergent species, which is expected to uncover more distinct genetic architectures than genetic mapping of traits at the within-species level [67]. Our results are generalizable to the extent that genome admixture between taxa with porous genomes (Fig 4) plays a role in ecology and evolution [15, 16, 66].

As in most QTL mapping studies of wild species, PVE’s in our study must be regarded as upper bounds [68]. In addition, the sample sizes employed in this study certainly allowed us to see only the “tip of the iceberg” of the QTL distribution for each trait [67]. Future studies based on greater sample sizes and marker numbers [48] will provide a more fine-grained picture of the genetic architecture of ecologically important traits in hybridizing species, especially in more highly recombinant mapping populations [58]. Within the limits of statistical power [50] afforded by the current mapping study, our results indicate high heritability (Table 1) and a relatively simple architecture of leading QTL for many of the studied phytochemical traits (Fig 2; S5 Table). QTL effect sizes (PVE’s) were on average greater for flavonoids than for salicinoids and chlorogenic acids. This trend across functional categories of chemical traits, largely consistent with our heritability estimates (Table 1), suggests that our mapping results are robust. Also, loci for two salicinoids (salicortin and HCH-salicortin) mapped to the same approximate locations in a QTL mapping pedigree of North American Populus species (Scott Woolbright, pers. comm.), which further validates our results obtained by admixture mapping. In the following paragraphs, we detail our findings for each branch of the phenylpropanoid pathway (Fig 6) regarding genetic architecture, genetic and phenotypic correlations, and the potential effects of genotypes and the environment.

Fig 6. Overview of the phenylpropanoid pathway.

Black arrows represent well-known branches of the phenylpropanoid pathways, their enzymes and numbers of genes in Populus [70, 71]. The hypothesized metabolite network for salicinoids is represented by grey arrows, following [72]. Common aglycones are represented in square boxes and moieties in hexagonal boxes. PAL, Phenylalanine ammonia lyase; C4H, cinnamate 4-hydroxylase; 4CL, 4-coumarate: CoA ligase; CHS, chalcone synthase; CHI, chalcone isomerase; F3H, flavanone 3-hydroxylase; FLS, flavonol synthase; DFR, dihydroflavonol reductase; LAR, leucoanthocyanindin reductase; HCT, hydroxycinnamoyl-CoA quinate/shikimate hydoxycinnamoyltransferase; C3H, 4-coumarate-3-hydroxylase; COMT: Caffeate O-methyltransferase; OMT: O-methyl transferase; UGT, UDP-glycosyltransferase.

Chlorogenic acids.

This group of secondary metabolites is involved in defence against herbivores, and their effect is enhanced by peroxidases [69]. Heritabilities and QTL effect sizes of these traits in our study (Fig 2) were low and there was little phenotypic variation between species (S1 Fig). Minor QTL, explaining small proportions of the phenotypic variation in each trait, were found spread across five chromosomes with three QTL positioned on chromosome VI (Fig 3). Phenotypic correlation networks suggest that chlorogenic acids are associated with intermediate compounds of the flavonoid pathway (Fig 5A and 5B), in agreement with available models of the phenylpropanoid pathway (Fig 6), whereas correlations with salicinoids were weak.


Salicinoids are ecologically important anti-herbivory compounds in the Salicaceae family [4]. Although salicinoids are described as a defined branch of the phenylpropanoid pathway [70, 72], the exact sequence of steps and molecules and their respective intermediate forms remain largely unknown. It has been proposed that salicinoids may not be synthesized within a single linear pathway but rather within a network of interconnected reactions (Fig 6; [72]). The high phenotypic correlations found in this group (Fig 5) suggest particularly strong dynamics within the pathway, with compounds functioning as substrates of one another. As [61] proposed, the dynamics within the salicinoid pathway might depend on the availability of moieties (acetyl-, Cinnamoyl- and HCH-derivatives) with common salicinoids (salicortin and tremulacin; Fig 6) most likely stocked as intermediates with various isomer forms. In this study, we provide first evidence of the coarse-scale genetic architecture of phytochemical traits in this pathway with 15 different QTL detected, six of which control the abundances of two or more salicinoid compounds (Fig 3). The abundance of many salicinoids exhibited genetic correlations (linkage and/or pleiotropy) visible as genomic clustering on the QTL map on chromosomes II and III (Fig 3). This suggests that the entire pathway may either be under the control of few regulatory genes or elements, or that salicinoid synthesis involves few enzymes, as is the case for the terpenoid pathway [61]. Chemical links with the phenylpropanoid pathway are suggested by phenotypic correlations with the abundances of chlorogenic acids, but also with the abundance of a flavonoid, Isorhamnetin-rhamnoside (Fig 5), which is mirrored by QTL co-localization on chromosome XII (Fig 3).


Flavonoids represent a large group of phytochemicals widely distributed in the plant kingdom [1]. In European Populus species, some flavonoids, especially the glucuronide and rutinoside derivatives, present high heritability and species specificity (S2 Fig, [24]). Estimates of both, flavonoid heritability and QTL effect sizes (PVE; Fig 2) were on average larger than for chlorogenic acids and salicinoids. Phenotypic correlations (Fig 5) were detectable between compounds sharing glycosylation patterns, but not between aglycones. QTL for flavonoid abundances mapped to twelve of the 19 Populus chromosomes (Fig 3) with up to two QTL co-localizations (Table 1), but without discernable genetic correlation patterns. Thus, the genetic architecture of flavonoids appears to be more complex, with many QTL of intermediate to large effects. Flavonoids were largely species-specific and exhibited geographic variation (see above), consistent with genetic control and GxE. Heritable variation for condensed tannin (= large flavonoid molecules) abundances has previously been shown to result in measurable ‘extended phenotypes’ of Populus and other tree species at the level of terrestrial, aquatic, and endophyte communities and even at the level of entire ecosystems (nutrient cycling; [14]).

Excess of heterozygosity associated with phytochemical QTL.

A growing body of literature reports the ability of herbivore and pathogen communities to drive rapid evolution in short-lived host plant species [9, 10]. Here, we explore this issue for long-lived trees representing foundation species of terrestrial ecosystems. A previous study has shown that naturally occurring hybrid populations between P. alba and P. tremula retain unexpectedly high levels of interspecific heterozygosity (individuals with hetero-specific allele combinations) at many loci in the genome [23], despite several (95% CI = 23–43) generations of recombination between the parental species [50]. Selection for increased between-species heterozygotes was put forward as a likely explanation, although [23] were cautious to pinpoint the precise mechanisms responsible. Markers identified as QTL for salicinoid and flavonoid abundances in the present study exhibited a significant enrichment for loci with such ‘heterozygote excess’ in the Italian hybrid zone. For both functional groups of traits, Fisher’s exact tests indicated that instances of ‘heterozygote excess’ among phytochemical QTL were more numerous than expected from all marker loci typed by [23]. When the test was repeated across hybrid zone localities, a significantly increased count of ‘heterozygote excess’ QTL was still observed for flavonoids, but not for salicinoids (S5 Table). The results suggest selection as a potential driver of the genotypic composition of these trees. By extension, genetic variation associated with flavonoids appears to be more consistently selected compared to salicinoids. Future work should aim to identify the genetic mechanisms responsible (i.e. heterosis or alternative hypotheses; [21]) and the precise biotic or abiotic selective agents.

Evolutionary constraints and chemical diversification of phenylpropanoids in Populus

Populus alba and P.tremula are hybridizing, ecologically divergent species with porous species boundaries [22, 23, 40, 50] that nevertheless remain differentiated in their secondary metabolism (Fig 1 and S1 Fig). We propose a prediction-based model that follows pathway evolution theory [26, 27] and accommodates experimental results [2831] to explain and further identify targets of differentiation in the phenylpropanoid pathway.

Changes in evolutionary rates with pathway position and flux control have been described in plant secondary metabolites, with selective constraints decreasing along pathways [26, 30, 73]. This model of evolution of the plant metabolism suggests differences in evolutionary constraints along the phenylpropanoid pathway that match well with the phenotypic results observed here. Chlorogenic acids and salicinoids present low levels of species differentiation (S1 Fig; S4 Table). On the contrary, some flavonoids present high levels of interspecific differentiation (Fig 1 and S2 Fig; S4 Table). These results suggest high conservation of branches in upstream portions of the pathway (i.e. chlorogenic acids and salicinoids) and point to the potential for chemical differentiation downstream in the flavonol branch of the flavonoid pathway.

As described in [24] and plotted in S2 Fig, flavonoids in the two parental species are far less differentiated in aglycones (core of molecules, square boxes in Fig 6) than in moieties (functional groups or sugars linked to the aglycone; hexagonal boxes in Fig 6). Similarities in total concentrations between species (S2 Fig) in the face of high chemical diversity (Table 1) of flavonol suggests that the flavononol aglycone pathway is conserved and that interspecific differentiation occurred by further modification of aglycones. Also note that the transgressive traits identified in the hybrids present an unusual flavonol moiety (rutinoside-pentose). This suggests that genes involved in the addition of moieties to aglycones are particularly affected by interspecific differentiation and hybridization. Two possible scenarios may explain moiety variation: (1) the chemical diversity in salicinoids and flavonoids is driven by the availability of moieties (e.g. different sugars) in different species and ecotypes; (2) the chemical diversity is driven by enzymes catalysing the transfer of moieties to aglycones and by their specificity. We argue in favour of the second scenario and propose that such enzymes are disproportionately affected by interspecific differentiation and hybridization.

Enzymes transferring sugar moieties, socalled glycosyltransferases, have been characterized in many plant species. Glycosyltransferases fulfill various functions during compound biosynthesis, storage and regulation and present high levels of genetic diversity and complexity [74, 75]. The UDP glycosyltransferase (UGTs) genes in Populus represent a large family of 236 identified genes (Fig 6), twice the number of UGT genes observed in Arabidopsis thaliana [71]. Among these genes, Flavonoid-3-O-glycosyltransferase (9 genes) and flavonol 7-O-glycosyltransferase (27 genes) present important and highly diverse gene families in Populus [71]. Glycosyltransferases have also been shown to exhibit regulatory activity on diverse enzymes of the phenylpropanoid pathway. In potato, for example, a UGT has been shown to control the anthocyanin pathway after over-expression of dihydroflavonol reductase [76]. In Populus, over-expression of a UGT with low substrate specificity targeting cinnamic and benzoic acids is known to modulate the phenylpropanoid carbon flow without interfering with the expression levels of genes of the pathway [77]. Thus, it appears likely that UGTs are responsible for the inter- and intraspecific patterns of variation observed for flavonoids in our study (Figs 1 and 2).


Phytochemical traits from different branches of the phenylpropanoid pathway differ greatly in proportions of intra- and interspecific variation, heritabilities, genetic architectures, and phenotypic correlations in hybridizing Eurasian Populus species. Flavonoids exhibit stronger differentiation between species and geographic localities, stronger G components, and larger numbers of detected QTL with intermediate to large effect sizes compared to most other studied phenylpropanoids. Inter- and intraspecific differentiation for the abundances of flavonoids can plausibly be explained by both, relaxed selective constraints and increased numbers of adaptive substitutions for sugar-transferring enzymes (glycosyltransferases), a gene family with greatly increased diversity and complexity in Populus. Our results demonstrate how different groups of secondary metabolites are differentially affected by evolutionary constraints and hybridization in plant taxa with porous genomes. Future work should combine the metabolomic approach taken here with proteomics, transcriptomics, and high-density genotyping of appropriate association mapping panels phenotyped in controlled environments. This combination of approaches should help towards closing current knowledge gaps regarding the origin and maintenance of functionally important phenotypic differences during plant population divergence and speciation.

Supporting Information

S1 Fig. Principal component analysis of chlorogenic acids and salicinoids in natural populations and a common garden trial.

Principal Component Analysis (PCA) of secondary metabolites identified by UHPLC-QTOF-MS in Populus alba (in red), P. tremula (in green) and their hybrids, P. x canescens (in blue). Ellipses represent the 95% confidence intervals and colored dots represent averages for each group. For (A) and (B), symbols specific to each species and hybrid zone (Ticino: Italy, Ticino river hybrid zone; Danube: Austria, Danube river hybrid zone; Tisza: Hungary, Tisza river hybrid zone) represent individual trees. The first two principal components (PC’s) are plotted for (A) PCA of seven chlorogenic acids explaining 71.6% of the phenotypic variation; (B) PCA of the 13 salicinoids explaining 85.1% of the phenotypic variance within this subgroup; (C) PCA of all 38 phenylpropanoid in 133 individuals from the common garden explaining 58.9% of the phenotypic variance. For PCA for flavonoids see main paper.


S2 Fig. Interspecific variation for total flavonoids, quercetin and kaempferol aglycones, rutinoside and glucuronide moieties.

Normalized peak areas used as measures of relative abundances of compounds were summed within groups of aglycones and moieties for P. alba, P. tremula and their hybrids from three natural hybrid zones. For compound diversity see S3 Table.


S1 Materials. A) High-throughput quantification of phenylpropanoids in natural populations.

B) Molecular genetic analysis of common garden seedlings.


S1 Table. Number of individuals studied in three natural hybrid zones.

Hybrid individuals and plants from each parental species (P. alba and P. tremula) were characterized with 77 microsatellite DNA markers in a previous study [23]. The common garden plants were genotyped with 16 microsatellites for the present study (see text and S2 Table for details). Assignment to each taxon was based on Bayesian genomic admixture proportions Q as described in main text.


S2 Table. Molecular marker used to characterize common garden seedlings.

Information for the 16 microsatellite marker loci used to identify parental species and hybrids and to estimate the correlation of paternity (Cp) in the common garden. These 16 microsatellites are a subset of the genome-wide marker panel used for admixture mapping in natural hybrid zones and are fully described [23]. Localization on chromosomes, allele frequency differential (delta) between the parental reference populations of the Italian hybrid zone [23], number of alleles (NA) and gene diversity (He) in the common garden trial are indicated.


S3 Table. Details of thirty-eight phenylpropanoids identified [24] by Ultra-High Pressure Liquid Chromatography coupled with Quadrupole-Time Of Flight mass spectrometry (UHPLC-QTOF-MS) in P. alba, P. tremula and their hybrids.


S4 Table. Intra- and inter-specific variation in 38 phytochemical traits identified by UHPLC-QTOF-MS and condensed tannins.

Shown are means ± standard deviations of traits quantified for three natural hybrid zones (Italy, Austria, Hungary) and a common garden trial established from seeds from the Italian hybrid zone. Numbers of individuals (n) for each species in each hybrid zone are indicated.


S5 Table. Genomic architecture of phytochemical traits inferred from admixture mapping in natural hybrid zones.


S6 Table. Excess of interspecific heterozygosity linked with phytochemical QTL.

Contingency tables presenting counts of the presence (yes) or absence (no) of an excess of interspecific heterozygote [23], for all 67 codominant genetic markers (S5 Table) studied in natural hybrid zones of P. alba and P. tremula, for markers representing putative QTL for salicinoids and flavonoids, and for all 34 markers representing putative phytochemical QTL in the present study.



We thank Dorothea Lindtke, Stefano Castiglione, Stefano Gomarasca, the University of Fribourg botanical garden team, and Thelma Barbará for help in the field, common garden, and lab, and Dorothea Lindtke for providing R-scripts from a published study. We are grateful to Loren Rieseberg, Marc Johnson and Mark Rausher for inspiring discussions, Thomas G. Whitham and Scott A. Woolbright for reading an earlier version of the manuscript and to anonymous reviewers. Thanks also go to Jean-Luc Wolfender and Bernd Schneider for providing chemical compounds standards.

Author Contributions

Conceived and designed the experiments: CC CL. Performed the experiments: CC GG CS TB. Analyzed the data: CC CS CL. Contributed reagents/materials/analysis tools: CL GG. Wrote the paper: CC CL. Revision of the manuscript: CC CL GG CS TB.


  1. 1. Treutter D. Significance of flavonoids in plant resistance: a review. Environ Chem Lett. 2006;4:147–57.
  2. 2. Philippe RN, Bohlmann J. Poplar defense against insect herbivores. Can J Bot. 2007;85:1111–26.
  3. 3. Chen F, Liu CJ, Tschaplinski TJ, Zhao N. Genomics of secondary metabolism in Populus: Interactions with biotic and abiotic environments. Critical Revs in Plant Sc. 2009;28:375–92.
  4. 4. Boeckler A, Gershenzon J, Unsicker SB. Phenolic glycosides of the Salicaceae and their role as anti-herbivore defenses. Phytochemistry. 2011;72:1497–509. pmid:21376356
  5. 5. Yu O, Jez JM. Nature's assembly line: biosynthesis of simple phenylpropanoids and polyketides. Plant J. 2008;54:750–62. pmid:18476876
  6. 6. Chae L, Kim T, Nilo-Poyanco R, Rhee SY. Genomic signatures of specialized metabolism in plants. Science. 2014;344:510–3. pmid:24786077
  7. 7. Weng JK, Philippe RN, Noel JP. The rise of chemodiversity in plants. Science. 2012;336:1667–70. pmid:22745420
  8. 8. Futuyma DJ, Agrawal AA. Macroevolution and the biological diversity of plants and herbivores. Proc Natl Acad Sci U S A. 2009;106:18054–61. pmid:19815508
  9. 9. Agrawal AA, Hastings AP, Johnson MT, Maron JL, Salminen JP. Insect herbivores drive real-time ecological and evolutionary change in plant populations. Science. 2012;338:113–6. pmid:23042894
  10. 10. Zust T, Heichinger C, Grossniklaus U, Harrington R, Kliebenstein DJ, Turnbull LA. Natural enemies drive geographic variation in plant defenses. Science. 2012;338:116–9. pmid:23042895
  11. 11. Bennett RN, Wallsgrove RM. Tansley review no. 72. Secondary metabolites in plant defence mechanisms. New phytol. 1994; 127:617–33.
  12. 12. Lattanzio V, Lattanzio VMT, Cardinali A. Role of phenolics in the resistance mechanisms of plants against fungal pathogens and insects. In: Imperato F, editor. Phytochemistry: Advances in research. Kerala: Research Signpost; 2006. pp.23–67.
  13. 13. Mithofer A, Boland W. Plant defense against herbivores: chemical aspects. Annu Rev Plant Biol. 2012;63:431–50. pmid:22404468
  14. 14. Whitham TG, Bailey JK, Schweitzer JA, Shuster SM, Bangert RK, LeRoy CJ, et al. A framework for community and ecosystem genetics: from genes to ecosystems. Nat Rev Genet. 2006;7:510–23. pmid:16778835
  15. 15. Rieseberg LH, Raymond O, Rosenthal DM, Lai Z, Livingstone K, Nakazato T, et al. Major ecological transitions in wild sunflowers facilitated by hybridization. Science. 2003;301:1211–6. pmid:12907807
  16. 16. Arnold ML. Evolution through genetic exchange. Oxford: Oxford University Press: 2007.
  17. 17. Nolte AW, Tautz D. Understanding the onset of hybrid speciation. Trends Genet. 2010;26:54–8. pmid:20044166
  18. 18. Feder JL, Egan SP, Nosil P. The genomics of speciation-with-gene-flow. Trends Genet. 2012;28:342–50. pmid:22520730
  19. 19. Rieseberg LH, Ellstrand NC, Arnold M. What can molecular and morphological markers tell us about plant hybridization? Critical Revs in Plant Sc. 1993;12:213–41.
  20. 20. Orians CM, Hochwender CG, Fritz RS, Snall T. Growth and chemical defense in willow seedlings: trade-offs are transient. Oecologia. 2010;163:283–90. pmid:20012101
  21. 21. Cheng D, Vrieling K, Klinkhamer PG. The effect of hybridization on secondary metabolites and herbivore resistance: implications for the evolution of chemical diversity in plants. Phytochem Rev. 2011;10:107–17. pmid:21475395
  22. 22. Lexer C, Joseph JA, van Loo M, Barbara T, Heinze B, Bartha D, et al. Genomic admixture analysis in European Populus spp. reveals unexpected patterns of reproductive isolation and mating. Genetics. 2010;186:699–712. pmid:20679517
  23. 23. Lindtke D, Buerkle CA, Barbara T, Heinze B, Castiglione S, Bartha D, et al. Recombinant hybrids retain heterozygosity at many loci: new insights into the genomics of reproductive isolation in Populus. Mol Ecol. 2012;21:5042–58. pmid:22989336
  24. 24. Caseys C, Glauser G, Stolting KN, Christe C, Albrectsen BR, Lexer C. Effects of interspecific recombination on functional traits in trees revealed by metabolomics and genotyping-by-resequencing. Plant Ecol Divers. 2012;5:457–71.
  25. 25. Pichersky E, Lewinsohn E. Convergent evolution in plant specialized metabolism. Annu Rev Plant Biol. 2011;62:549–66. pmid:21275647
  26. 26. Rausher MD. The evolution of genes in branched metabolic pathways. Evolution. 2013;67:34–48. pmid:23289560
  27. 27. Sellis D, Longo MD. Patterns of variation during adaptation in functionally linked loci. Evolution. 2015;69:75–89. pmid:25338665
  28. 28. Clotault J, Peltier D, Soufflet-Freslon V, Briard M, Geoffriau E. Differential selection on carotenoid biosynthesis genes as a function of gene position in the metabolic pathway: A study on the carrot and dicots. Plos One. 2012;7:e38724. pmid:22737218
  29. 29. Chu S, Wang J, Cheng H, Yang Q, Yu D. Evolutionary study of the isoflavonoid pathway based on multiple copies analysis in soybean. BMC Genet. 2014;15:76. pmid:24962214
  30. 30. Rausher MD, Miller RE, Tiffin P. Patterns of evolutionary rate variation among genes of the anthocyanin biosynthetic pathway. Mol Biol Evol. 1999;16:266–74. pmid:10028292
  31. 31. Ramsay H, Rieseberg LH, Ritland K. The correlation of evolutionary rate with pathway position in plant terpenoid biosynthesis. Mol Biol Evol. 2009;26:1045–53. pmid:19188263
  32. 32. Eckenwalder J. Systematics and evolution of Populus. In: Stettler RF BH, Heilman PE, Hinckley TM, editor. Biology of Populus and its implications for management and conservation. Ottawa: NRC Research Press; 1996. p. 542.
  33. 33. Bangert RK, Turek RJ, Rehill B, Wimp GM, Schweitzer JA, Allan GJ, et al. A genetic similarity rule determines arthropod community structure. Mol Ecol. 2006;15:1379–91. pmid:16626460
  34. 34. Whitham TG, Martinsen GD, Floate KD, Dungey HS, Potts BM, Keim P. Plant hybrid zones affect biodiversity: Tools for a genetic-based understanding of community structure. Ecology. 1999;80:416–28.
  35. 35. DeWoody J, Viger M, Lakatos F, Tuba K, Taylor G, Smulders MJM. Insight into the genetic components of community genetics: QTL mapping of insect association in a fast-growing forest tree. Plos One. 2013;8:e79925. pmid:24260320
  36. 36. Bernhardsson C, Robinson KM, Abreu IN, Jansson S, Albrectsen BR, Ingvarsson PK. Geographic structure in metabolome and herbivore community co-occurs with genetic structure in plant defence genes. Ecol Lett. 2013;16:791–8. pmid:23601188
  37. 37. Anderson JT, Mitchell-Olds T. Ecological genetics and genomics of plant defenses: Evidence and approaches. Funct Ecol. 2011;25:312–24. pmid:21532968
  38. 38. Dickmann DI, Kuzovkina J. Poplars and Willows in the world: Poplars and Willowd of the world, with empasis on sylviculturally important species. Food and Agriculture Organization of the United Nations. 2009:1–130.
  39. 39. van Loo M, Joseph JA, Heinze B, Fay MF, Lexer C. Clonality and spatial genetic structure in Populus × canescens and its sympatric backcross parent P. alba in a Central European hybrid zone. New phytol. 2008;177:506–16. pmid:18005320
  40. 40. Lexer C, Joseph J, van Loo M, Prenner G, Heinze B, Chase MW, et al. The use of digital image-based morphometrics to study the phenotypic mosaic in taxa with porous genomes. Taxon. 2009;58:349–64.
  41. 41. De Carvalho D, Ingvarsson PK, Joseph J, Suter L, Sedivy C, Macaya-Sanz D, et al. Admixture facilitates adaptation from standing variation in the European aspen (Populus tremula L.), a widespread forest tree. Mol Ecol. 2010;19:1638–50. pmid:20345678
  42. 42. Stolting KN, Paris M, Meier C, Heinze B, Castiglione S, Bartha D, et al. Genome-wide patterns of differentiation and spatially varying selection between postglacial recolonization lineages of Populus alba (Salicaceae), a widespread forest tree. New Phytol. 2015:
  43. 43. Ritland K. Extensions of models for the estimation of mating systems using n independent loci. Heredity. 2002;88:221–8. pmid:11920127
  44. 44. Ritland K, Ritland C. Inferences about quantitative inheritance based on natural population structure in the yellow monkeyflower, Mimulus guttatus. Evolution. 1996;50:1074–82.
  45. 45. Lynch M, Walsh B. Genetics and analysis of quantitative traits. Sinauer Associates Incorporated; 1998.
  46. 46. Widmer A, Lexer C, Cozzolino S. Evolution of reproductive isolation in plants. Heredity. 2009;102:31–8. pmid:18648386
  47. 47. Lexer C, Heinze B, Alia R, Rieseberg LH. Hybrid zones as a tool for identifying adaptive genetic variation in outbreeding forest trees: lessons from wild annual sunflowers (Helianthus spp.). Forest Ecology and Management. 2004;197:49–64. pmid:18677413
  48. 48. Stolting KN, Nipper R, Lindtke D, Caseys C, Waeber S, Castiglione S, et al. Genomic scan for single nucleotide polymorphisms reveals patterns of divergence and gene flow between ecologically divergent species. Mol Ecol. 2013;22:842–55. pmid:22967258
  49. 49. Rieseberg LH, Buerkle CA. Genetic mapping in hybrid zones. Am Nat. 2002;159 Suppl 3:S36–50. pmid:18707368
  50. 50. Lindtke D, Gonzalez-Martinez SC, Macaya-Sanz D, Lexer C. Admixture mapping of quantitative traits in Populus hybrid zones: power and limitations. Heredity. 2013;111:474–85. pmid:23860234
  51. 51. Gompert Z, Buerkle CA. A powerful regression-based method for admixture mapping of isolation across the genome of hybrids. Mol Ecol. 2009;18:1207–24. pmid:19243513
  52. 52. Via S. Natural selection in action during speciation. Proc Natl Acad Sci U S A. 2009;106 Suppl 1:9939–46. pmid:19528641
  53. 53. Rieseberg LH, Widmer A, Arntz AM, Burke JM. Directional selection is the primary cause of phenotypic diversification. Proc Natl Acad Sci U S A. 2002;99:12242–5. pmid:12221290
  54. 54. Karrenberg S, Lexer C, Rieseberg LH. Reconstructing the history of selection during homoploid hybrid speciation. Am Nat. 2007;169:725–37. pmid:17479459
  55. 55. Morreel K, Goeminne G, Storme V, Sterck L, Ralph J, Coppieters W, et al. Genetical metabolomics of flavonoid biosynthesis in Populus: a case study. Plant J. 2006;47:224–37. pmid:16774647
  56. 56. Camacho D, de la Fuente A, Mendes P. The origin of correlations in metabolomics data. Metabolomics. 2005;1:53–63.
  57. 57. Steuer R. Review: on the analysis and interpretation of correlations in metabolomic data. Brief Bioinform. 2006;7:151–8. pmid:16772265
  58. 58. Lindtke D, Gompert Z, Lexer C, Buerkle CA. Unexpected ancestry of Populus seedlings from a hybrid zone implies a large role for postzygotic selection in the maintenance of species. Mol Ecol. 2014;23:4316–30. pmid:24750473
  59. 59. Johnson MT, Rasmann S. The latitudinal herbivory-defence hypothesis takes a detour on the map. New Phytol. 2011;191:589–92. pmid:21770942
  60. 60. Brachi B, Meyer CG, Villoutreix R, Platt A, Morton TC, Roux F, et al. Coselected genes determine adaptive variation in herbivore resistance throughout the native range of Arabidopsis thaliana. Proc Natl Acad Sci U S A. 2015;112:4032–7. pmid:25775585
  61. 61. Keefover-Ring K, Ahnlund M, Abreu IN, Jansson S, Moritz T, Albrectsen BR. No evidence of geographical structure of salicinoid chemotypes within Populus tremula. Plos One. 2014;9:e107189. pmid:25299342
  62. 62. Kawecki TJ, Ebert D. Conceptual issues in local adaptation. Ecol Lett. 2004;7:1225–41.
  63. 63. Fussi B, Lexer C, Heinze B. Phylogeography of Populus alba (L.) and Populus tremula (L.) in Central Europe: secondary contact and hybridisation during recolonisation from disconnected refugia. Tree Genet Genomes. 2010;6:439–50.
  64. 64. Jaakola L, Hohtola A. Effect of latitude on flavonoid biosynthesis in plants. Plant Cell Environ. 2010;33:1239–47. pmid:20374534
  65. 65. Schweitzer JA, Madritch MD, Bailey JK, LeRoy CJ, Fischer DG, Rehill BJ, et al. From genes to ecosystems: The genetic basis of condensed tannins and their role in nutrient regulation in a Populus model system. Ecosystems. 2008;11:1005–20.
  66. 66. Nosil P. Ecological speciation. Oxford: Oxford University Press; 2012.
  67. 67. Rockman MV. The QTN program and the alleles that matter for evolution: all that's gold does not glitter. Evolution. 2012;66:1–17. pmid:22220860
  68. 68. Beavis WD. QTL analyses: power, precision, and accuracy. In: Paterson AH, editor. Molecular dissection of complex traits. CRC press LLC; 1998. pp. 145–162.
  69. 69. Barbehenn R, Dukatz C, Holt C, Reese A, Martiskainen O, Salminen JP, et al. Feeding on poplar leaves by caterpillars potentiates foliar peroxidase action in their guts and increases plant resistance. Oecologia. 2010;164:993–1004. pmid:20680646
  70. 70. Tsai C- JJ, El Kayal W, Harding SA. Populus, the new model system for investigating phenylpropanoid complexity. Int J Appl Sci Eng. 2006;4:221–33.
  71. 71. Yonekura-Sakakibara K, Hanada K. An evolutionary view of functional diversity in family 1 glycosyltransferases. Plant J. 2011;66:182–93. pmid:21443631
  72. 72. Babst BA, Harding SA, Tsai CJ. Biosynthesis of phenolic glycosides from phenylpropanoid and benzenoid precursors in Populus. J Chem Ecol. 2010;36:286–97. pmid:20177744
  73. 73. Ramsay H, Rieseberg LH, Ritland K. The correlation of evolutionary rate with pathway position in plant terpenoid biosynthesis. Mol Biol Evol. 2009;26:1045–53. pmid:19188263
  74. 74. Gachon CM, Langlois-Meurinne M, Saindrenan P. Plant secondary metabolism glycosyltransferases: the emerging functional analysis. Trends Plant Sci. 2005;10:542–9. pmid:16214386
  75. 75. Vogt T, Jones P. Glycosyltransferases in plant natural product synthesis: characterization of a supergene family. Trends Plant Sci. 2000;5:380–6. pmid:10973093
  76. 76. Aksamit-Stachurska A, Korobczak-Sosna A, Kulma A, Szopa J. Glycosyltransferase efficiently controls phenylpropanoid pathway. BMC Biotechnol. 2008;8:25. pmid:18321380
  77. 77. Babst BA, Chen HY, Wang HQ, Payyavula RS, Thomas TP, Harding SA, et al. Stress-responsive hydroxycinnamate glycosyltransferase modulates phenylpropanoid metabolism in Populus. J Exp Bot. 2014;65:4191–200. pmid:24803501