Genome-wide association implicates numerous genes underlying ecological trait variation in natural populations of Populus trichocarpa


  • Athena D. McKown,

    Corresponding author
    1. Department of Forest and Conservation Sciences, Faculty of Forestry, University of British Columbia, Forest Sciences Centre, Vancouver, BC, Canada
    • Author for correspondence:

      Athena D. McKown

      Tel: +1 604 822 6023


    Search for more papers by this author
    • These authors contributed equally to this work.
  • Jaroslav Klápště,

    1. Department of Forest and Conservation Sciences, Faculty of Forestry, University of British Columbia, Forest Sciences Centre, Vancouver, BC, Canada
    2. Department of Dendrology and Forest Tree Breeding, Faculty of Forestry and Wood Sciences, Czech University of Life Sciences, Prague, Czech Republic
    Search for more papers by this author
    • These authors contributed equally to this work.
  • Robert D. Guy,

    1. Department of Forest and Conservation Sciences, Faculty of Forestry, University of British Columbia, Forest Sciences Centre, Vancouver, BC, Canada
    Search for more papers by this author
  • Armando Geraldes,

    1. Department of Botany, University of British Columbia, Vancouver, BC, Canada
    Search for more papers by this author
  • Ilga Porth,

    1. Department of Forest and Conservation Sciences, Faculty of Forestry, University of British Columbia, Forest Sciences Centre, Vancouver, BC, Canada
    2. Department of Wood Science, Faculty of Forestry, University of British Columbia, Forest Sciences Centre, Vancouver, BC, Canada
    Search for more papers by this author
  • Jan Hannemann,

    1. Department of Biology and Centre for Forest Biology, University of Victoria, Victoria, BC, Canada
    Search for more papers by this author
  • Michael Friedmann,

    1. Department of Botany, University of British Columbia, Vancouver, BC, Canada
    Search for more papers by this author
  • Wellington Muchero,

    1. BioSciences Division, Oak Ridge National Laboratory, Oak Ridge, TN, USA
    Search for more papers by this author
  • Gerald A. Tuskan,

    1. BioSciences Division, Oak Ridge National Laboratory, Oak Ridge, TN, USA
    Search for more papers by this author
  • Jürgen Ehlting,

    1. Department of Biology and Centre for Forest Biology, University of Victoria, Victoria, BC, Canada
    Search for more papers by this author
  • Quentin C. B. Cronk,

    1. Department of Botany, University of British Columbia, Vancouver, BC, Canada
    Search for more papers by this author
  • Yousry A. El-Kassaby,

    1. Department of Forest and Conservation Sciences, Faculty of Forestry, University of British Columbia, Forest Sciences Centre, Vancouver, BC, Canada
    Search for more papers by this author
  • Shawn D. Mansfield,

    1. Department of Wood Science, Faculty of Forestry, University of British Columbia, Forest Sciences Centre, Vancouver, BC, Canada
    Search for more papers by this author
  • Carl J. Douglas

    1. Department of Botany, University of British Columbia, Vancouver, BC, Canada
    Search for more papers by this author


  • In order to uncover the genetic basis of phenotypic trait variation, we used 448 unrelated wild accessions of black cottonwood (Populus trichocarpa) from much of its range in western North America. Extensive data from large-scale trait phenotyping (with spatial and temporal replications within a common garden) and genotyping (with a 34 K Populus single nucleotide polymorphism (SNP) array) of all accessions were used for gene discovery in a genome-wide association study (GWAS).
  • We performed GWAS with 40 biomass, ecophysiology and phenology traits and 29 355 filtered SNPs representing 3518 genes. The association analyses were carried out using a Unified Mixed Model accounting for population structure effects among accessions.
  • We uncovered 410 significant SNPs using a Bonferroni-corrected threshold (< 1.7 × 10−6). Markers were found across 19 chromosomes, explained 1–13% of trait variation, and implicated 275 unique genes in trait associations. Phenology had the largest number of associated genes (240 genes), followed by biomass (53 genes) and ecophysiology traits (25 genes).
  • The GWAS results propose numerous loci for further investigation. Many traits had significant associations with multiple genes, underscoring their genetic complexity. Genes were also identified with multiple trait associations within and/or across trait categories. In some cases, traits were genetically correlated while in others they were not.


The genetic basis of phenotypic variability is the fundamental underpinning of evolutionary biology and key in understanding factors that define speciation, biogeographical distributions and fitness under natural conditions (Stapley et al., 2010; Savolainen et al., 2013). Achieving such understanding is becoming more attainable as the ability to cast a wider net for gene discovery in traits of interest emerges. In plant biology, the integration of extensive genetic and phenotypic data is finding application in development and improvement of crop species, but is also extending our understanding of the genetics underlying traits of evolutionary and ecological importance (Ingvarsson et al., 2008; Eckert et al., 2009, 2010, 2012; Fournier-Level et al., 2011; Parchman et al., 2012; Olson et al., 2013). Genome-wide association studies (GWAS) can be powerful for identifying putative causal genes, or suites of genes, underlying phenotypic variation, particularly in traits with complex genetic architecture (Vandenkoornhuyse et al., 2010; Ingvarsson & Street, 2011; Savolainen et al., 2013; Sork et al., 2013). Where traits are complex (i.e. involving a number of genes or gene networks), GWAS using high genome coverage of single nucleotide polymorphisms (SNP) markers has been very effective for identifying the genetic architecture underlying variability in these traits (Eckert et al., 2012; Parchman et al., 2012; Riedelsheimer et al., 2012; Morris et al., 2013; Porth et al., 2013a). GWAS can also uncover loci with potential pleiotropic effects that may be important to natural variation within species and their capacity for adaptation (Mackay et al., 2009; Stapley et al., 2010; Porth et al., 2014).

Defining the roles of genotypic and phenotypic variability in adaptation across a landscape are key to understanding the evolution and adaptability of species (Sork et al., 2013). Within tree species, phenotypic variability is influenced by wide geographic distributions and numerous traits are considered to be under polygenic control (Savolainen et al., 2007; Ingvarsson & Street, 2011; Cooke et al., 2012; Sork et al., 2013). High genetic complexity is reported for many adaptive traits in trees, such as cold hardiness, bud break, bud set, cone serotiny, disease resistance and growth (Ruttink et al., 2007; Holliday et al., 2008, 2010; Ingvarsson et al., 2008; Eckert et al., 2009; Ibáñez et al., 2010; Ma et al., 2010; Rohde et al., 2010; Keller et al., 2012; Parchman et al., 2012; La Mantia et al., 2013; Olson et al., 2013). By comparison, the underlying genetic variability for numerous physiological traits considered important in range-wide adaptation of tree species, such as nutrient uptake, leaf anatomy, photosynthetic rate and water-use efficiency (cf. Soolanayakanahally et al., 2009; Chamaillard et al., 2011; Keller et al., 2011; McKown et al., 2014), is only beginning to be explored (González-Martínez et al., 2008; Cumbie et al., 2011).

In this study, we focused on the genetics underlying phenotypic trait variation in black cottonwood (Populus trichocarpa), a species of high ecological, scientific and economic value (Cronk, 2005; Tuskan et al., 2006). Like many poplars, P. trichocarpa trees are outbreeding, fast growing and often function as pioneers and/or constitute major canopy-forming components of riparian forest ecosystems (Farrar, 1995; Braatne et al., 1996). The species is common throughout the Pacific Northwest of North America and has high natural phenotypic variation relating to its geographical distribution spanning environmental and climatic gradients (Gornall & Guy, 2007; McKown et al., 2014). Trait variation within P. trichocarpa relates primarily to its latitudinal distribution and gradients in photoperiodic regime (daylength) and/or temperature across its natural range (McKown et al., 2014). Furthermore, heritability is generally highest in traits that co-vary strongly with these ecological and geographical gradients.

Extensive genomic tools available for P. trichocarpa (Tuskan et al., 2006; Geraldes et al., 2013) and high intraspecific variability in traits (McKown et al., 2014) support using the GWAS approach to provide significant insights into the genetic architecture of ecologically important phenotypic variation (Eckert et al., 2010, 2012; Parchman et al., 2012; Morris et al., 2013; Porth et al., 2013a; La Mantia et al., 2013; Olson et al., 2013). Nevertheless, GWAS is challenging to implement using natural populations across a landscape (Ingvarsson & Street, 2011; Neale & Kremer, 2011; Sork et al., 2013). As genetic structure reflects the effects of family relatedness, demography and adaptive history, model-fitting in GWAS as a corrective measure is necessary to balance the risk of false-positives with that of false-negatives (Balding, 2006; Ingvarsson & Street, 2011; Sork et al., 2013). However, attempts to minimize the loss of some associations where relationships exist between loci, demography and geography should be made by assessing corrective measures on a trait-by-trait basis (La Mantia et al., 2013; Porth et al., 2013a,b).

Using accessions originating from wild populations of P. trichocarpa, we investigated the genetic basis of intraspecific variation in 40 biomass, ecophysiology and phenology traits in an association genetics framework. We employed GWAS, integrating extensive biological information on quantitative variation in these traits assayed within a common garden over multiple years (McKown et al., 2013, 2014) and SNP genotype data from the same trees obtained using an Illumina iSelect Infinium 34K Populus SNP genotyping array developed for P. trichocarpa (Geraldes et al., 2013). We predicted that certain traits considered genetically complex, such as growth or bud set, might retrieve multiple associations underscoring the genetic complexity of the trait. Additionally, we expected that genes underlying trait variation would associate repeatedly with the same trait when phenotyped over multiple years. Finally, we expected that the same loci would associate with multiple traits where traits are genetically correlated. Based on the results from our GWAS, we propose numerous key loci for further testing in trait variation, highlighting these as important in the evolution and ecology of P. trichocarpa.

Materials and Methods

We performed a GWAS with 448 unrelated individuals using clonal means for 40 traits and 29 355 filtered SNPs (detailed later). Data in the association analysis are publicly available at the University of Victoria PhenoDB website (URL: and within the Supporting Information included within this publication (Table S1A,B).

Phenotypic trait measurements

Tree materials were obtained from wild genotypes of Populus trichocarpa Torr. & A. Gray originally collected by British Columbia Ministry of Forests, Lands and Natural Resource Operations (FLNRO) spanning the northern two thirds of the species’ range (44–60°N, 121–138°W) (Xie et al., 2009; McKown et al., 2013). Phenotyping of individual accessions in the Totem Field common garden, University of British Columbia, was replicated in space (4–20 clonal ramets of similar age and condition) and in time (repeated measurements across years) to confirm the patterns observed in phenotypic traits. This extensive phenotyping effort of all accessions for phenology events, growth and biomass accumulation, photosynthetic gas exchange, leaf traits and stable isotopes has been previously described (McKown et al., 2013, 2014; Tables 1, S1A). Before GWAS, all trait data were checked for normality using a regression model approach. We note that bud set data were analyzed either including all data or removing premature bud set dates (bud set1) occurring before the solstice (21 June, day 186) due either to photoperiodic mismatch or other stressors (cf. Soolanayakanahally et al., 2013).

Table 1. Phenotypic traits within three categories (biomass, ecophysiology, phenology) measured in Populus trichocarpa accessions indicating number of years measured and total number of significant single nucleotide polymorphisms (SNPs)/genes uncovered using genome-wide association study (GWAS) (< 1.7 × 10−6)
  1. a

    Bud set dates occurring before the summer solstice (day 186) removed.

Biomass traitsActive growth rate (cm d−1)2009–20101/1
Bole fresh mass density (kg m−3)20120
Bole fresh mass (kg)20124/3
Branches (total number)200927/17
Height:diameter (H:D; cm:cm)2009–20112/2
Height (cm)2008–201114/12
Height gain (cm)2008–201111/10
Log height growth rate (log cm d−1)20095/4
Log volume growth rate (log cm3 d−1)20093/3
Volume (cm3)2009–201113/9
Volume gain (cm3)2009–201114/12
Whole-tree mass (kg)20124/3
Ecophysiology traitsCarbon to nitrogen ratio (C:N; g g−1)20091/1
Chlorophyll content – spring (Chlspring; CCI)20090
Chlorophyll content – summer (Chlsummer; CCI)2009, 201110/7
Instantaneous water-use efficiency (WUE; μmol CO2 mmol−1 H2O)20090
Leaf carbon isotope discrimination (Δleaf; ‰)20090
Leaf mass per unit area – spring (LMAspring; mg mm−2)2010–201113/7
Leaf mass per unit area – summer (LMAsummer; mg mm−2)2009–20110
Leaf nitrogen content per unit area (Narea; mg mm−2)20090
Leaf nitrogen content per unit dry mass (Nmass; g g−1)20096/5
Leaf shape (length:width)20096/5
Leaves per bud (total number)2011–20121/1
Photosynthetic rate per unit area (Amax; μmol CO2 m−2 s−1)20090
Photosynthetic rate per unit dry mass (Amax/mass; μmol CO2 g−1 s−1)20092/2
Photosynthetic nitrogen-use efficiency (NUE; μmol CO2 g−1 N s−1)20090
Stable carbon isotope ratios (δ13Cwood; ‰)20120
Stable nitrogen isotope ratios (δ15N; ‰)20090
Stomatal conductance (gs; mol H2O m−2 s−1)20090
Phenology traitsBud break (Julian date)2010–20118/2
Bud set (Julian date)2008–2010149/104
Bud seta (Julian date)2009–2010203/145
Canopy duration (d)2009–201011/6
Growth period (d)2009–201068/51
Height growth cessation (HGC; Julian date)200947/34
Leaf drop (Julian date)2008–2010180/130
Leaf flush (Julian date)2010–20129/4
Leaf lifespan (days)20106/6
Post-bud set period (PBS; d)2009–201056/40
25% total canopy leaf yellowing (Julian date)20101/1
50% total canopy leaf yellowing (Julian date)20100
75% total canopy leaf yellowing (Julian date)20103/2
100% total canopy leaf yellowing (Julian date)201033/23

SNP genotyping

A total of 448 unrelated, phenotyped P. trichocarpa accessions (with > 0.03% genetic distance) were successfully genotyped with a 34K Populus Illumina Infinium® SNP genotyping array designed for P. trichocarpa (Table S1B). Full details of SNP discovery/selection, array development, performance and data filtering criteria are given in Geraldes et al. (2011, 2013). Candidate gene selection for the chip resulted in the inclusion of 34 131 SNP markers within 3543 genes and intergenic regions (± 2 kb up- or downstream from the longest transcript) across the genome. Genotyping was carried out as described by Geraldes et al. (2013) and array hybridizations performed at Oak Ridge National Laboratory (ORNL, TN). Genotype calls were filtered with GenomeStudio v2010.3 (

Only SNPs with GenTrain score ≥ 0.5 and genotypes with GenCall score ≥ 0.15 were exported, criteria maximizing genotype call accuracy while minimizing missing data (Geraldes et al., 2013). We further excluded SNPs with minor allele frequency < 0.05 and call rate < 0.9. Following this filtering process, we used 29 355 SNPs representing 3518 genes for associations. Each significant trait-associated SNP identified by GWAS was visually inspected for quality using the corresponding clustering plot (GenomeStudio v2010.3). The ‘Nisqually-1’ genome sequence P. trichocarpa v2.2 SNP positions and gene models described in Geraldes et al. (2013) were translated into v3.0 positions by aligning sequences flanking the SNP with the latest Populus reference genome assembly on Phytozome 9.1 (

Population structure analysis

We evaluated the effects of genetic structure within our population using the Unified Mixed Model framework (Balding, 2006; Yu et al., 2006) and compared log likelihood values between models with the Bayesian Information Criterion (BIC) (Yu et al., 2006). We assessed a number of options for population structure fit on a trait-by-trait basis. We constructed family relatedness using a kinship (K) model and population structure using a principal component analysis (P) model or a clustering matrix (Q) model. We also calculated combinations of structures (P + K, Q + K), and a ‘simple’ model (i.e. simple linear regression without any additional correction).

SNPs used for population and kinship estimates were further filtered for Hardy-Weinberg Equilibrium using the ‘Chisq’ function in the R package ‘HardyWeinberg’ (Graffelman & Morales, 2008) and for linkage disequilibrium (LD) at r2 < 0.2 (Wang et al., 2009). Following these filtering criteria, 8749 SNPs (distributed throughout the genome) were used to fit all model analyses. The K model was calculated following Loiselle et al. (1995) and the relationship matrix was estimated by first multiplying the kinship matrix by two, then setting diagonal elements as one and negative off-diagonal elements as zero (Yu et al., 2006). The ‘nearPD’ function implemented in the R package ‘Matrix’ (Higham, 2002) was used to obtain the positive definite relationship matrix required in the mixed model framework. The P model was done using the ‘prcomp’ function implemented in the base R package (R Core Development Team, 2011) and significant principal components (PC) were selected according to the broken-stick rule (Jackson, 1993) implemented in the R package ‘vegan’. Within our population, only PC1 was significant. The parametric clustering model-based inference (Q matrix) was performed using the R package ‘popgen’ (Marchini, 2013) which implements both the uncorrelated allele frequency model of STRUCTURE (Pritchard et al., 2000) when using the function ‘ps’, and the correlated allele frequency model (Falush et al., 2003) by using the function ‘ps’ and ‘popdiv’ in conjunction. The number of populations tested ranged from = 1 to = 10 populations. Both the burn-in period and the number of sampling iterations after the burn-in period were set to 60 000, and thinning was set at the default (1). For each scenario (K), 20 runs were performed to obtain both mean and standard deviation for the log likelihood value to construct a delta coefficient for the most probable number of populations (Evanno et al., 2005). While the uncorrelated allele frequency model did not detect any population structure (i.e. no peak appeared indicating the best fit was reached in the scenario considering = 1, results confirmed with GENELAND; Guillot et al., 2005), the correlated allele frequency model detected = 5. We used the = 5 cluster results from the correlated allele frequency for the Q matrix in our GWAS.

We evaluated the model fit on a trait-by-trait basis using the Bayesian Information Criterion (BIC) where the lowest BIC value indicates the best model fit. Among all studied traits, BIC selected the simple, P or Q models depending on the trait (Table S2). In no case was the kinship (K) component within the K, P + K or Q + K models considered the best fit for the data structure. This lack of importance of the K component confirmed the absence of familial relatedness within the study population (see also La Mantia et al., 2013; Porth et al., 2013a). By comparison, QQ plots (i.e. the ranking of observed P-values from smallest to highest against the expected values) showed that inclusion of the K component generated a uniform distribution of P-values (simply reflecting the tested null hypothesis that no marker is a causal variant) and indicated a substantial decrease in the power to detect true positives (Figs S1–S3). We consider this result likely to be related to the presence of linkage between the actual true positives and other SNPs due to dense SNP coverage, rather than to the confounding effect of population structure in our sample set (Pearson & Manolio, 2008). In such a case, the QQ plot may fail to identify the real source of deviation from the null hypothesis and thus risks exaggerating confounding factors resulting in an excess of false-negatives.

Association genetics

We used the GLM procedure implemented in TASSEL (Bradbury et al., 2007) to perform the association analysis as follows:

display math(Eqn 1)

(y, vector of measurements; μ, overall population mean; S and X, index matrices assigning fixed effects for both SNP genotype and population to the measurements, respectively; α and β, vectors of fixed effects for both SNP genotype and population, respectively; e, residual effect). Following the GWAS, we used Bonferroni multiple testing correction (α/29 355) rather than the false discovery rate (FDR) correction owing to nonindependence of the tests where test statistics were correlated due to LD between SNPs used in the array (cf. Schwartzman & Lin, 2011). We considered SNP–trait associations significant at α = 0.05/29 355 where < 1.7 × 10−6 and report these. As subsidiary signal, we also included trait associations at α = 0.1/29 355 where < 3.4 × 10−6 in the Supporting Information if the SNP in question was already considered significant by association to another trait at the lower cut-off of < 1.7 × 10−6. Composite pairwise LD between all significant trait-associated SNPs was calculated based on genotype correlations (Weir et al., 2004).

Cumulative R2 of significant SNPs

In order to address the total phenotypic variance accounted for by all trait-associated SNPs on a trait-by-trait basis, we calculated a ‘cumulative R2’ metric. These values were obtained by the difference in R2 between full and reduced models (Ingvarsson et al., 2008). The full model comprises all significant SNPs detected by GWAS for the trait in question and population structure (as selected by BIC, see ‘Population structure analysis’ above) while the reduced model contains only population structure. Analysis was performed using the ‘glm’ function and R2 values were extracted using the ‘RsquareAdj’ function implemented in the R package ‘vegan’ (Peres-Neto et al., 2006). We then repeated this test using < 3.4 × 10−6 to include our subsidiary SNP association information (see above).

Genetic correlations between phenotypic traits

In order to confirm that trait correlation was not solely responsible for detection of potential functional pleiotropy, we assessed the pairwise genetic correlations of all traits to identify a common genetic basis for independent variation (Porth et al., 2013b). These genetic correlations are ‘broad-sense’ (i.e. using phenotype trait data from all clonal replicates) and based on clonal best linear unbiased predictions (BLUPs) using PC1 from the PCA for structure correction (McKown et al., 2014). The broad-sense genetic correlation matrix was performed using the ‘cor’ function in the ‘stats’ R package and Pearson product-moment correlations were estimated following:

display math(Eqn 2)

(Covgxgy, covariance between clonal BLUPs of traits x and y; Vargx, variance in clonal BLUPs for trait x; Vargy,variance in clonal BLUPs for trait y). The clonal breeding values were obtained from linear mixed model results presented in McKown et al. (2014).

Tests for Gene Ontology enrichment

All genes uncovered by GWAS were tested for Gene Ontology (GO) enrichment using ‘function’ and ‘process’ categorizations against the available genes from the SNP array (i.e. genes included in GWAS following SNP filtering). We tested all genes, and subgroupings of genes based on individual trait categories or groupings of categories. Significant GO terms were determined with GOTermFinder software ( using FDR correction for multiple comparisons (Boyle et al., 2004).


SNP discovery through GWAS

The GWAS using 29K SNPs uncovered a total of 1118 significant SNP–trait associations (involving 410 unique SNPs) across the three studied trait categories (i.e. biomass, ecophysiology and phenology). Most traits required population structure correction (either P model or Q matrix; 65 out of 71 tests), decided on a trait-by-trait basis using BIC model selection (Table S2). Significant trait-associated SNP markers were found across all 19 chromosomes with the highest numbers of significant SNPs (n ≥ 35) on chromosomes 2, 6 and 9 (Fig. 1, Table S3). The number of trait-associated SNPs/chromosome was significantly different from the number of SNPs/chromosome on the array (after filtering) (χ2 test, = 0.0075) and trait-associated SNP distribution across chromosomes did not correlate strongly with the density of SNPs/chromosome on the array (using 500 kb windows along each chromosome; r2 = 0.13). Most trait-associated SNP markers were located in noncoding regions (78%) while a smaller number of SNP markers were within coding regions (nonsynonymous = 10%, synonymous = 12%) (Tables S3, S4). This largely reflected the relative distribution of the SNPs used and no enrichment based on position within gene region was found (χ2 test, not significant).

Figure 1.

Genomic distribution of single nucleotide polymorphisms (SNPs) on the 34K Populus genotyping array and significant trait-associated SNPs uncovered using genome-wide association study (GWAS) across 19 chromosomes in P. trichocarpa. SNP density on the array per 500 kB windows on each chromosome is illustrated by a heat map (outermost ring). All SNPs retrieved by GWAS are indicated in black (second ring). These are further distinguished by trait category where SNPs related to phenology traits are marked in red (third ring), biomass traits in yellow (fourth ring) and ecophysiology traits in blue (fifth, inner ring). Image courtesy of N. Farzaneh.

In total, 275 genes were identified with at least one significant trait-associated SNP (Tables S4, S5). Where multiple trait-associated SNPs within a gene were retrieved, a range in LD values between such SNPs was observed (r2 = 0–1.0; Table S6). This variability in LD within genes is likely due to the high variability in recombination rate throughout the genome (Slavov et al., 2012). Nevertheless, on average, LD within genes was high (r2 = 0.73; Table S6). Among the 18 genes with low or no LD between trait-associated SNPs (r2 = 0–0.3), 12 had multiple associations within the same trait category while six had associations across trait categories (Tables S5, S6). Some trait-associated SNPs located in different genes but within the same genomic regions also showed moderate to complete linkage (r2 = 0.35–1.0; Table S6). Among these, two genomic regions had multiple associations within the same trait category while four had associations across trait categories.

The 410 significant SNPs within 275 genes were associated with 30 of the 40 assayed biomass, ecophysiology and phenology traits (Table 1). Total numbers of identified SNP–trait associations varied, depending on the trait, and SNP markers explained between 1.2 and 13.2% of the phenotypic variation, depending on the association (average r2 = 0.037; Table S5). The phenology category retrieved the largest number of SNP–trait associations whereas both the biomass and ecophysiology categories had far fewer associations (Tables 1, S5). SNP–trait associations at P < 1.7 × 10−6 identified 53 genes associated with biomass (20 genes were solely associated with biomass traits), 25 genes associated with ecophysiology (15 genes solely with ecophysiology traits), and 240 genes associated with phenology (200 genes solely with phenology traits) (Fig. 2, Table S5). Correspondingly, the cumulative proportion of phenotypic variance explained by significant SNPs (cumulative R2) was highest within the phenology category and lower in both biomass and ecophysiology categories (Figs 3, S4).

Figure 2.

Diagram depicting 275 unique genes identified through genome-wide association study (GWAS). Numbers of associations are arranged in circles by trait category (biomass, ecophysiology, phenology), with circle size representing relative proportion of significant genes and circle overlaps representing numbers of genes associated with more than one trait category. See Tables 3-6 and Supporting Information Table S5 for detailed information on significant SNP-trait associations and gene identities.

Figure 3.

Quantile distribution of proportions of phenotypic variance explained by significant single nucleotide polymorphism (SNP)–trait associations for each trait (cumulative R2) within each trait category (biomass, ecophysiology, phenology). Cumulative R2 values for individual traits are shown in Fig. S4.

Among phenology traits, bud set, growth period, height growth cessation, post-bud set period, 100% leaf yellowing and leaf drop had the greatest number of associations. Within the biomass category, branch numbers, height/height gain and volume/volume gain yielded the most SNP associations. The highest numbers of associations among ecophysiology traits included leaf mass per area of preformed leaves (LMAspring), nitrogen per unit mass (Nmass), summer chlorophyll content (Chlsummer) and leaf shape. Many genes were repeatedly associated with the same trait where year-to-year data existed and/or with multiple traits within trait categories, particularly among phenology traits (see ‘Genes with effects on phenology’ below). GWAS further identified genes with significant associations across two trait categories (42 out of 275) and three genes with associations across all trait categories.

The genes uncovered by GWAS were largely transcription factors/regulators, transferases, kinases, transporters, hydrolases and other/unknown gene functions (based on the Arabidopsis homologs) (Tables 2, S5). These genes were tested for enrichment of Gene Ontology (GO) terms using all results, phenology-related, biomass-related, ecophysiology-related and multiple category-related (Table S7). Significant enrichment was only found considering genes associated in the biomass-related group (auxin binding (GO:0010011), hormone binding (GO:0042562)) and genes with associations across trait categories (substrate-specific channel activity (GO:0022838), nitrate transmembrane transporter activity (GO:0015112), channel activity (GO:0015267), passive transmembrane transporter activity (GO:0022803)). Other high-ranking GO terms included response to red/far red light, binding (e.g. DNA, hormone, kinase, protein) and circadian rhythm but were not significantly enriched after multiple testing correction.

Table 2. General functional classifications of genes identified by genome-wide association study (GWAS) with significant single nucleotide polymorphism (SNP) markers associated to growth, ecophysiology and phenology traits
Putative functionaNumberb
  1. a

    Functional gene prediction based on Geraldes et al. (2013).

  2. b

    Number of genes with SNPs associated with trait variation. Full details are given in Table S5.

  3. c

    Includes genes with associations across two trait categories.

  4. d

    Includes genes with associations across three trait categories.

Binding - otherc8
Cell division1
Cell wall metabolism7
DNA repair1
Ion binding8
Ion transporterc8
Protein bindingd8
RNA binding2
Transcription factor/regulatorc,d62
Zinc fingerc5

Genes underlying phenotypic variation

From the large number of SNP–trait associations, we highlight examples of specific loci providing the Arabidopsis homologue annotation, location information (i.e. chromosome/SNP/feature), allelic variation among accessions and the underlying phenotypic variability (Table S8). We focused on genes associated: to biomass or ecophysiology traits; to phenology traits; and with multiple traits within and/or across trait categories. Full SNP results, marker r2 and LD values are available in Tables S4–S6.

Genes with effects on biomass or ecophysiology

A small number of genes (35 out of 275) exhibited significant associations only with variation in biomass or ecophysiology traits (Fig. 2, Tables 3, 4, S5). These encompassed a range of functions, such as transcription factors, kinases, phytochrome, transporters and binding elements. In many cases, genes were retrieved either by year-to-year data from the same trait and/or from multiple traits in the same category. Within the biomass-related category, Potri.010G250500 (protein binding EXO70G1; EXOCYST SUBUNIT EXO70 FAMILY PROTEIN G1) was associated with active growth rate, height (2009–2011), and bole and whole-tree mass (Table 3). Effects of the SNP (10_22286918; intergenic) linked the common allele with substantially greater biomass overall. Without any apparent geographic pattern, accessions homozygous for the common allele were 39% taller (each year) and had greater bole and whole tree mass (76% and 89%, respectively) compared to the minor homozygotes, with heterozygous accessions being intermediate compared to both homozygotes (Table S8). Among the genes uncovered within the ecophysiology category, leaf N content (Nmass) and the correlated C : N ratio were associated with Potri.010G221600 (EMB1144; EMBRYO DEFECTIVE 1144 chorismate synthase) (Table 4). Accessions homozygous for the minor allele (SNP 10_20651512; 3′UTR) had 14% lower Nmass and 1.2× greater C : N ratio compared to the other accessions, but no difference was observed comparing heterozygotes with homozygotes of the major allele (Table S8). Another gene, Potri.011G024000 (SPK1; SPIKE1), was associated with maximum photosynthetic rate per unit mass (Amax/mass) (Table 4). Allelic effects of the SNP (11_2007822; intron) linked the minor homozygotes with 22% higher photosynthesis than the major homozygotes, and heterozygotes with having intermediate trait values (Table S8).

Table 3. Genes identified by genome-wide association study (GWAS) with single nucleotide polymorphism (SNP) markers associated to biomass traits
Gene modela Traitb AT homologAnnotated descriptiona
  1. a

    Poplar gene models are annotated to v3 of the genome. See Table S5 for full gene details, associated SNPs, and complete annotation description.

  2. b

    See Table 1 for trait explanations and units.

  3. H : D, height : diameter.

Potri.001G256100Volume gainAT3G21070 NADK1 (NAD KINASE 1)
Potri.001G323100Height gainAT3G26810 AFB2 (AUXIN SIGNALING F-BOX 2)
Potri.002G005800Volume gainAT1G76420 CUC3 (CUP SHAPED COTYLEDON3)
Potri.002G052100Height gainAT4G02780 GA1 (GIBBERELLIC ACID REQUIRING 1)
Potri.002G111900Log volume growth rateAT1G50010 TUA2 (TUBULIN ALPHA-2 CHAIN)
Potri.003G059400Active growth rateAT1G15490Hydrolase, alpha/beta fold family protein
Potri.003G139300Volume, Volume gainAT1G64380AP2 domain-containing transcription factor
Potri.003G195300Height gainAT3G54390Transcription factor GT-2
Potri.005G142300Log height growth rateAT2G23300Leucine-rich repeat transmembrane protein kinase
Potri.006G150400BranchesAT2G19580 TET2 (TETRASPANIN2)
Potri.010G019000Log height growth rateAT3G06350 MEE32 (MATERNAL EFFECT EMBRYO ARREST 32)
Potri.010G250500Bole mass, Height, Whole-tree massAT4G31540 EXO70G1 (EXOCYST SUBUNIT EXO70 FAMILY PROTEIN G1)
Potri.014G141400Log height growth rateAT4G18880 HSFA4A (HEAT SHOCK TRANSCRIPTION FACTOR A4A)
Potri.015G127200Volume gainAT4G25240 SKS1 (SKU5 SIMILAR 1)
Potri.016G000300H : DAT2G44190 EDE1 (ENDOSPERM DEFECTIVE 1)
Potri.016G128300Log volume growth rateAT2G38470 WRKY33 (WRKY DNA-BINDING PROTEIN 33)
Potri.018G076400Log height growth rateAT3G24450Copper-binding family protein
Table 4. Genes identified by genome-wide association study (GWAS) with single nucleotide polymorphism (SNP) markers associated to ecophysiology traits
Gene modela Traitb AT homologAnnotated descriptiona
  1. a

    Poplar gene models are annotated to v3 of the genome. See Table S5 for full gene details, associated SNPs, and complete annotation description.

  2. b

    See Table 1 for trait explanations and units.

  3. Amax/mass, assimilation rate per unit mass; C:N, carbon:nitrogen ratio; Chl, chlorophyll; LMA, leaf mass per area; Nmass, nitrogen per unit mass.

Potri.005G072700LMAspring AT4G31700 RPS6 (RIBOSOMAL PROTEIN S6)
Potri.005G073000LMAspring AT5G65270 RABA4A (RAB GTPASE HOMOLOG A4A)
Potri.006G097300Chlsummer AT2G38090MYB family transcription factor
Potri.006G116900Chlsummer AT5G03760 CSLA9 (CELLULOSE SYNTHASE LIKE A9)
Potri.008G105200Leaves per budAT2G18790 PHYB (PHYTOCHROME B)
Potri.009G110500LMAspring AT2G16050Thioredoxin-related/zinc ion binding
Potri.010G121500Leaf shapeAT1G25380Mitochondrial FAD carrier protein
Potri.010G221600C : N, Nmass AT1G48850 EMB1144 (EMBRYO DEFECTIVE 1144)
Potri.011G024000Amax/mass AT4G16340 SPK1 (SPIKE1)
Potri.011G107900Leaf shapeAT2G34250Protein transport protein SEC61 subunit alpha
Potri.013G032500Leaf shapeAT3G47590Esterase/lipase/thioesterase family protein
Potri.014G103600LMAspring AT2G46710RAC GTPase activating protein, putative
Potri.014G116800LMAspring AT2G47180 GOLS1 (GALACTINOL SYNTHASE 1)
Potri.015G009100Chlsummer AT4G27740Yippee putative zinc-binding protein
Potri.018G019900Leaf shapeAT5G10930 CIPK5 (CBL-INTERACTING PROTEIN KINASE 5)

Genes with effects on phenology

The majority of genes uncovered by GWAS had SNPs associated with phenology traits (Fig. 2, Tables 5, S5). Genes ranged in function, including cytochromes, hydrolases, ion binding/transport, transcription factors/regulators and transferases. Many encoded proteins putatively related to light perception, photoperiod and/or circadian rhythm, or were phytohormone-related/response proteins (involving auxin, cytokinin, gibberellin, abscisic acid, and ethylene). Numerous genes were repeatedly associated with phenology traits across different years (Tables 5, 6, S5). Among 89 genes associated with phenology traits measured over multiple years, 80 genes were found to be associated with the same trait in at least 2 yr. In addition, we found 13 genes with associations to the same trait in multiple years just below our stringent cutoff criteria (< 3.4 × 10−6). Similarly, GWAS uncovered 185 genes with associations to 2–7 different phenology traits, and an additional 15 genes had multiple phenology trait associations detected just below our stringent cutoff criteria (< 3.4 × 10−6). Analyses using all bud set dates available for the population vs removing premature bud set dates occurring before the solstice (i.e. bud set1) largely resulted in the same SNP–trait associations; however, a handful of genes were found only using bud set1.

Table 5. Selected genes identified by genome-wide association study (GWAS) with significant single nucleotide polymorphism (SNP) markers associated to phenology traits across multiple years and/or multiple phenology traitsb
Gene modelb Traitc AT homologAnnotated descriptionb
  1. a

    Indicates association only retrieved with bud set dates following the summer solstice (occurrences before day 186 removed).

  2. b

    Poplar gene models are annotated to v3 of the genome. See Table S5 for full association results with all phenology traits, gene details, associated SNPs, and complete annotation description.

  3. c

    See Table 1 for trait explanations and units.

  4. HGC, Height growth cessation; PBS, post-bud set period.

Potri.001G000600Bud setAT1G55570 SKS12 (SKU5 SIMILAR 12)
Potri.001G110800Bud seta, Leaf dropAT4G25480 DREB1A (DEHYDRATION RESPONSE ELEMENT B1A)
Potri.001G190800Leaf dropAT2G19770 PRF3 (PROFILIN3)
Potri.001G252600Bud set, Leaf dropAT5G58620zinc finger (CCCH-type) family protein
Potri.001G327100Bud seta, Canopy duration, Growth period, Leaf lifespanAT3G27010 TCP20 (TEOSINTE BRANCHED 1, CYCLOIDEA, PCF (TCP)-DOMAIN FAMILY PROTEIN 20)
Potri.001G375500Bud set, Growth period, Leaf drop, PBSAT1G53210sodium/calcium exchanger family protein
Potri.002G013400Bud set, Growth period, PBS, Leaf dropAT5G42250alcohol dehydrogenase, putative
Potri.002G055400Bud seta, Leaf dropAT3G59060 PIL6 (PHYTOCHROME INTERACTING FACTOR 3-LIKE 6)
Potri.002G074400Canopy duration, Growth periodAT1G43890 RAB18 (RAB GTPASE HOMOLOG B18)
Potri.002G099800Leaf drop, PBSAT1G78300 GRF2 (GENERAL REGULATORY FACTOR 2)
Potri.002G184300Bud set, Growth periodAT1G02305cathepsin B-like cysteine protease
Potri.002G242500Bud set, Growth period, Leaf drop, PBSAT2G32720 CB5-B (CYTOCHROME B5 ISOFORM B)
Potri.002G242700Bud set, PBSAT5G48740leucine-rich repeat family protein
Potri.003G050100Leaf dropAT1G52150 ATHB-15
Potri.003G126900Bud set, HGC, Leaf dropAT4G23100 GSH1 (GLUTAMATE-CYSTEINE LIGASE)
Potri.003G128100Bud set, Leaf dropAT4G233402OG-Fe(II) oxygenase family protein
Potri.003G131700Bud seta, Leaf dropAT4G23500glycoside hydrolase family 28 protein
Potri.003G173000Bud set, Leaf dropNAunknown function
Potri.004G002700Bud seta, Leaf dropAT2G32950 COP1 (CONSTITUTIVE PHOTOMORPHOGENIC 1)
Potri.004G013400Bud set, Growth period, PBS, Leaf dropAT1G11790 ADT1 (AROGENATE DEHYDRATASE 1)
Potri.004G116100Bud set, Leaf dropAT3G02150 PTF1 (PLASTID TRANSCRIPTION FACTOR 1)
Potri.004G168600100% Leaf yellowing, Leaf dropAT4G38770 PRP4 (PROLINE-RICH PROTEIN 4)
Potri.004G174400Bud seta, Canopy duration, Growth period, PBSAT4G38620 MYB4 (MYB DOMAIN PROTEIN 4)
Potri.005G086400Bud seta, Leaf dropAT4G39410 WRKY13 (WRKY DNA-BINDING PROTEIN 13)
Potri.005G111600Leaf dropAT2G17840 ERD7 (EARLY-RESPONSIVE TO DEHYDRATION 7)
Potri.005G138400Bud set, Leaf flush, Leaf dropAT5G67030 ABA1 (ABA DEFICIENT 1)
Potri.005G140200Bud set, HGC, Leaf dropAT2G23380 CLF (CURLY LEAF)
Potri.005G156500Bud set, HGC, PBSNAUnknown function
Potri.005G166100Bud seta, Leaf dropAT5G65170VQ motif-containing protein
Potri.005G170500Bud set, Leaf dropAT1G77920 TGA7
Potri.006G008300Bud setNAProtease inhibitor
Potri.006G039000Bud set, Canopy duration, Growth period, 100% Leaf yellowing, Leaf dropAT5G06950 AHBP-1B; CAMP-RESPONSE ELEMENT BINDING PROTEIN-RELATED
Potri.006G054500Bud set, Growth period, Leaf dropAT3G57600 DREB2F (DEHYDRATION RESPONSIVE ELEMENT BINDING PROTEIN 2F)
Potri.006G057700Bud set, Growth period, PBSAT3G12160 RABA4D (RAB GTPASE HOMOLOG A4D)
Potri.006G209200Bud seta, Leaf dropAT5G22380 ANAC090 (NAC DOMAIN CONTAINING PROTEIN 90)
Potri.006G241600Bud set, Leaf drop, PBSAT5G11520 ASP3 (ASPARTATE AMINOTRANSFERASE 3)
Potri.006G249900Bud set, HGC, Leaf dropAT2G25600 SPIK (SHAKER POLLEN INWARD K+ CHANNEL)
Potri.006G263000Bud setAT2G37585glycosyltransferase family 14 protein
Potri.006G264500100% Leaf yellowing, Leaf dropAT5G10840endomembrane protein 70
Potri.006G264600Bud set, 100% Leaf yellowing, Leaf dropAT2G25060plastocyanin-like domain-containing protein
Potri.007G076500Bud seta, Leaf dropAT4G39350 CESA2 (CELLULOSE SYNTHASE A2)
Potri.008G086800Bud seta, Leaf dropAT1G26820 RNS3 (RIBONUCLEASE 3)
Potri.008G140700Bud set, Growth period, PBSAT2G01980 NHX7 (NA+/H+ ANTIPORTER 7)
Potri.008G161900Bud set, Leaf dropAT5G43650basic helix-loop-helix (bHLH) family protein
Potri.008G162800Bud set, Growth period, HGC, 100% Leaf yellowing, Leaf drop, PBSAT3G23090 TPX2 (TARGETING PROTEIN FOR XKLP2)
Potri.008G195500Bud seta, Leaf dropAT3G07630 ADT2 (AROGENATE DEHYDRATASE 2)
Potri.009G006500100% Leaf yellowing, Leaf dropAT2G28110 FRA8 (FRAGILE FIBER 8)
Potri.009G011000Bud set, 100% Leaf yellowing, Leaf dropAT2G28315DUF707, protein of unknown function
Potri.009G014500Leaf dropAT5G60690 REV (REVOLUTA)
Potri.009G017400Bud set, Growth period, 100% Leaf yellowing, Leaf drop, PBSAT2G35940 BLH1 (BEL1-LIKE HOMEODOMAIN 1)
Potri.009G021800Bud seta, Leaf dropAT2G26930 CDPMEK (4-(CYTIDINE 5\’-PHOSPHO)-2-C-METHYL-D-ERITHRITOL KINASE)
Potri.009G035000Bud set, Leaf dropAT3G46640 PCL1 (PHYTOCLOCK 1)
Potri.009G099800Bud set, HGC, Leaf dropAT4G34050 CAFFEOYL COENZYME A O-METHYLTRANSFERASE 1
Potri.009G106000Bud set, HGC, Leaf dropAT2G15780plastocyanin-like domain-containing protein
Potri.010G077000Bud set, Growth period, Leaf dropAT5G43650basic helix-loop-helix (bHLH) family protein
Potri.010G093900Bud set, Growth period, Leaf dropAT1G14310haloacid dehalogenase-like hydrolase family protein
Potri.010G179300Bud set, Growth period, HGC, Leaf drop, PBSAT5G16250unknown protein
Potri.010G187600Growth period, Leaf dropAT3G55990 TBL28 (TRICHOME BIREFRINGENCE-LIKE 28)
Potri.010G212900Bud set, Leaf dropAT3G55260 HEXO1 (BETA-HEXOSAMINIDASE 1)
Potri.010G215200Bud set, Growth period, HGC, Leaf dropAT5G02810 PRR7 (PSEUDO-RESPONSE REGULATOR 7)
Potri.011G094400Bud set, Growth periodAT5G55180glycosyl hydrolase family 17 protein
Potri.011G140300Growth period, HGCAT1G17200integral membrane family protein
Potri.011G153300Bud set, Growth periodAT2G46770 ANAC043 (NAC DOMAIN CONTAINING PROTEIN 43)
Potri.012G014500Bud set, Growth period, Leaf dropAT3G49220pectinesterase
Potri.012G088200Leaf dropAT5G03340 CDC48 (CELL DIVISION PROTEIN 48)
Potri.012G13240075%,100% Leaf yellowingAT5G51810 GA20OX2 (GIBBERELLIN 20 OXIDASE 2)
Potri.013G013100Bud set, Growth period, HGC, Leaf drop, PBSAT5G27920F-box family protein
Potri.013G062400Leaf dropNADehydrin
Potri.014G047000Bud seta, Leaf dropAT2G44840 ERF7 (ETHYLENE-RESPONSIVE ELEMENT BINDING FACTOR 7)
Potri.014G087600Bud set, PBSAT5G41390 PLAC8 family
Potri.014G129400Leaf dropAT3G62820pectin methylesterase inhibitor family protein
Potri.014G160000Bud set, PBSAT1G04980 PDIL2-2
Potri.015G008300Bud set, Growth period, PBSAT1G55580 LAS (LATERAL SUPPRESSOR)
Potri.015G013700Bud seta, Leaf dropAT3G49220pectinesterase family protein
Potri.015G078600Bud seta, PBSAT5G63000uncharacterized conserved protein
Potri.015G105000Bud set, Growth periodAT5G23720 PHS1 (PROPYZAMIDE-HYPERSENSITIVE 1)
Potri.015G125500Bud set, Growth period, Leaf dropAT5G23260 TT16 (TRANSPARENT TESTA16)
Potri.015G129100Bud set, PBSAT4G22680 MYB85 (MYB DOMAIN PROTEIN 85)
Potri.015G136400Bud set, 100% Leaf yellowing, Leaf dropAT5G51990 DREB1D (DEHYDRATION-RESPONSIVE ELEMENT-BINDING PROTEIN 1D)
Potri.016G000100Bud set, 100% Leaf yellowing, Leaf dropAT1G80260 EMB1427 (EMBRYO DEFECTIVE 1427)
Potri.016G000200Bud set, Leaf dropAT1G79610 NHX6 (NA+/H+ ANTIPORTER 6)
Potri.016G134600Leaf dropAT3G51630 WNK5 (WITH NO LYSINE (K) KINASE 5)
Potri.017G042200Bud set, Leaf dropAT3G21175 ZML1 (ZIM-LIKE 1)
Potri.017G086200Bud set, Growth period, 75%,100% Leaf yellowing, Leaf drop, Leaf lifespan, PBSAT5G61430 ANAC100 (NAC DOMAIN CONTAINING PROTEIN 100)
Potri.017G090800Bud set, Leaf dropAT5G15470 GAUT14 (GALACTURONOSYLTRANSFERASE 14)
Potri.018G033600Bud set, Growth period, HGC, Leaf drop, PBSAT1G15550 GA3OX1 (GIBBERELLIN 3-OXIDASE 1)
Potri.018G090100Bud set, Growth period, Leaf dropAT2G36460fructose-bisphosphate aldolase, putative
Potri.019G076800Bud setAT1G71692 AGL12 (AGAMOUS-LIKE 12)
Table 6. Genes with significant single nucleotide polymorphism (SNP) markers associated with traits from 2 to 3 categories (phenology, biomass, ecophysiology)
Gene modelb Phenologyc Biomassc Ecophys.c AT homologAnnotated descriptionb
  1. a

    Association only retrieved with bud set dates following the summer solstice (occurrences before day 186 removed).

  2. b

    Poplar gene models are annotated to v3 of the genome. See Table S5 for full gene details, associated SNPs, and complete annotation description.

  3. c

    See Table 1 for trait explanations and units.

  4. Chl, chlorophyll; Ecophys, ecophysiology; HGC, height growth cessation; LMA, leaf mass per area; Nmass, nitrogen per unit mass; PBS, post-bud set period.

Potri.001G057400Bud set, Growth period, HGC, Leaf flush, PBSHeight gain AT1G27320 HK3 (HISTIDINE KINASE 3)
Potri.001G093800Bud seta Branches AT4G11090Unknown protein
Potri.001G320800Bud set, Leaf drop, PBSBranches AT5G60490 FLA12 (FASCICLIN-LIKE ARABINOGALACTAN-PROTEIN 12)
Potri.002G002000Bud set, 100% Leaf yellowing, Leaf dropBole mass, Whole-tree mass AT1G21050DUF617, protein of unknown function
Potri.002G165900Bud set, Growth period, HGC, Leaf drop, PBSBranches AT2G46225 ABIL1 (ABI-1-LIKE 1)
Potri.002G206400Bud set, Growth period, 100% Leaf yellowing, Leaf drop, PBSHeight, Volume gain AT2G47750 GH3.9 (PUTATIVE INDOLE-3-ACETIC ACID-AMIDO SYNTHETASE GH3.9)
Potri.002G257900Bud set, Leaf dropBranches, Volume AT5G44030 CESA4 (CELLULOSE SYNTHASE A4)
Potri.003G128600Bud set, Growth periodVolume, Volume gain AT1G01620 PIP1C (PLASMA MEMBRANE INTRINSIC PROTEIN 1C)
Potri.003G143600Bud set, Growth period, HGC, Leaf drop, PBSHeight, Height gain AT5G28540 BIP1/HSP70 PROTEIN
Potri.003G152700Bud seta, Leaf dropBranches NAUnknown function
Potri.003G214200Bud set, Growth period, HGC, 100% Leaf yellowingBranches AT5G13000 GSL12 (GLUCAN SYNTHASE-LIKE 12)
Potri.004G089800Bud set, Leaf drop, PBSBranches, Height gain, Volume, Volume gain AT2G01570 RGA1 (REPRESSOR OF GA1-3 1)
Potri.004G174500Bud set, PBSVolume AT4G35000 APX3 (ASCORBATE PEROXIDASE 3)
Potri.004G230500Bud set, Growth period, Leaf dropBranches, Volume, Volume gain AT1G10320DUF3594; PHD Zn-finger protein
Potri.005G141200Bud setBole mass, Height, Height gain, Whole-tree mass AT5G67200leucine-rich repeat transmembrane protein kinase
Potri.005G199600Bud seta, Leaf dropBranches AT1G71790F-actin capping protein beta subunit family protein
Potri.006G038600Bud set, Growth period, PBSHeight gain, Volume gain AT2G41200unknown protein
Potri.006G068400Bud set, Growth period, Leaf dropBranches, Height gain AT5G35410 SOS2 (SALT OVERLY SENSITIVE 2); CBL-INTERACTING PROTEIN KINASE 24
Potri.006G158400Bud set, HGC, Leaf dropBranches AT1G03390transferase activity
Potri.006G275500Bud break LMAspring AT5G10630 EF-1-alpha (ELONGATION FACTOR 1-alpha)
Potri.007G010700Bud set, Leaf dropVolume, Volume gain AT5G10470Kinesin (KAR3 subfamily)
Potri.008G038900Bud seta Leaf shapeAT3G54810zinc finger (GATA type) family protein
Potri.009G008500Bud set, Growth period, HGC, 100% Leaf yellowing, Leaf drop, PBSHeight AT5G60770 NRT2.4 (NITRATE TRANSPORTER 2:4)
Potri.009G008600Bud set, Growth period, HGC, 100% Leaf yellowing, Leaf drop, PBSHeight AT1G08090 NRT2:1 (NITRATE TRANSPORTER 2:1)
Potri.009G034500Bud set, PBSHeight gain AT2G29130 LAC2 (LACCASE 2)
Potri.009G136600Bud set, Growth period, HGC, Leaf drop, PBSVolume AT4G35100 PIP3 (PLASMA MEMBRANE INTRINSIC PROTEIN 3)
Potri.010G165700Bud setBranches AT3G01140 MYB106 (MYB DOMAIN PROTEIN 106)
Potri.010G184000Bud set, Growth period, HGC, 100% Leaf yellowing, Leaf drop, Leaf lifespan, PBSBranches, Volume gain AT2G40320 TBL33 (TRICHOME BIREFRINGENCE-LIKE 33)
Potri.010G250600Bud set, HGC, 100% Leaf yellowing, Leaf drop Amax/mass, Nmass AT1G51630 MSR2 (MANNAN SYNTHESIS RELATED 2)
Potri.010G254400Bud set, HGC, Leaf drop Nmass AT3G54540 GCN4 (GENERAL CONTROL NON-REPRESSIBLE 4)
Potri.013G021700Bud set, HGC, Leaf drop, PBSBranches, Volume, Volume gain AT4G14950 VMP1 (VACUOLE MEMBRANE PROTEIN 1)
Potri.014G102700Bud break, Canopy duration, Leaf flush LMAspring AT3G61880 CYP78A9 (CYTOCHROME P450 78A9)
Potri.014G109800Bud set, Growth periodLog volume growth rate AT1G02305cathepsin B-like cysteine protease
Potri.014G113700Bud set, Growth period, PBSHeight AT4G01840 KCO5 (CA2+ ACTIVATED OUTWARD RECTIFYING K+ CHANNEL 5)
Potri.015G002300Bud set, Leaf dropHeightChlsummer AT5G24470 PRR5 (PSEUDO-RESPONSE REGULATOR 5)
Potri.015G002600Bud set, HGC, Leaf drop, PBSHeightChlsummer AT5G24520 TTG1 (TRANSPARENT TESTA GLABRA 1)
Potri.015G004100Bud set, HGC, 100% Leaf yellowing, Leaf drop, PBS Chlsummer, Nmass AT3G49530 ANAC062 (ARABIDOPSIS NAC DOMAIN CONTAINING PROTEIN 62)
Potri.015G009300Bud set, Growth period, 100% Leaf yellowing, Leaf drop, PBS Chlsummer AT4G24060Dof-type zinc finger domain-containing protein
Potri.017G040800Bud set, Growth period, 100% Leaf yellowing, Leaf dropHeightNmass AT4G15210 BAM5 (BETA-AMYLASE 5)
Potri.017G079600Bud seta, Leaf dropBranches AT1G74690 IQD13 (IQ-DOMAIN 13)

Relating to light perception, Potri.010G215200 (PRR7; PSEUDO-RESPONSE REGULATOR 7 transcription regulator), was associated with fall phenology events of bud set (2008–2010), growth period, height growth cessation and leaf drop (Table 5). Allelic effects of the SNP (10_202495; coding sequence, nonsynonymous) linked the minor homozygote accessions with earlier height growth cessation, bud set and leaf drop (32, 36 and 23 d, respectively) and correspondingly shorter growth period (51 d) compared to the major homozygotes, with the heterozygous state intermediate to both homozygotes (Table S8). Among the phytohormone-related genes, Potri.018G033600 (GA3OX1; GIBBERELLIN 3-OXIDASE 1) was linked with multiple phenology traits. A single SNP (18_2683640; intergenic) was associated with bud set (2008–2010), growth period (2009–2010), height growth cessation, post-bud set period and leaf drop (Table 5). The minor homozygotes showed later height growth cessation, bud set and leaf drop (25, 31, and 31 d, respectively) resulting in a longer growth period (44 d) and shorter post-bud set period (28 d) compared to the major homozygotes with the heterozygous state intermediate to both homozygotes (Table S8). Transcription factor Potri.009G017400 (BLH1; BEL1-LIKE HOMEODOMAIN 1) was associated with bud set (2009–2010), growth period, post-bud set period, leaf yellowing and leaf drop (2008,10) (Table 5). Both significant SNP markers (09_2874013; coding sequence, synonymous/09_2874898; intron) are in high pairwise LD (r2 = 0.95). The double minor homozygote accessions had earlier bud set, canopy yellowing and leaf drop (22, 17, and 12 d, respectively), and subsequently shorter growth period (26 d) and longer post-bud set period (18 d) compared to the double major homozygote accessions (Table S8). The common heterozygote had equivalent trait values to the double major homozygote, while the less common heterozygote (6 trees total) showed trait values similar to the double minor homozygote.

Individual genes with effects across trait categories

GWAS identified 40 genes with SNPs associated with variation in two trait categories, and three genes with SNPs associated across all trait categories (Fig. 2, Tables 6, S5). All genes with multiple trait category effects had associations to phenology events, particularly bud set and leaf drop. One example, Potri.001G057400 (HK3; HISTIDINE KINASE 3 cytokinin receptor) was associated with leaf flush, bud set (2009–2010), growth period, height growth cessation, post-bud set period and height gain (Table 6). Allelic effects of the SNP (01_4368872; intron) linked the minor homozygotes accessions with earlier leaf flushing (6 d), later height growth cessation and bud set (18, 24 d, respectively), longer growth period (36 d), shorter post-bud set period (22 d), and correspondingly greater height gain (30%) compared to the major homozygote accessions (Table S8). The heterozygous state also showed earlier leaf flushing (3 d) but other traits were equivalent to the major homozygotes.

Some genes had extensive complexity in both genetic variation and the resulting phenotype. Potri.014G102700 (CYP78A9; CYTOCHROME P450 78A9) had numerous SNPs associated across spring traits, including phenology events bud break (2010–2011), leaf flush (2010–2012) and the ecophysiology trait LMAspring (2010–2011) (Table 6). The six significant SNPs (14_8045578/14_8045889/14_8046287; intergenic, upstream; 14_8047714; coding sequence, synonymous, 14_8048878/14_8049068; intergenic, downstream) are in moderate to high pairwise LD (average r2 = 0.47, range = 0.19–0.99) (Table S6). Genetic variation was highly complex and different combinations of the six SNPs resulted in 36 genetic variants (haplotypes), all with varying phenotypes (not shown). Individual SNPs had differing effects on phenotypic traits among accessions homozygous for the major or minor allele (depending on the SNP) and did not appear to show phenotypic change in the same direction. Each SNP resulted in variable bud break (6–10 d), leaf flush (2–8 d), canopy duration (11–18 d) and LMAspring (9–13%) with heterozygote accessions intermediate to both homozygotes (Table S8).

Multiple, linked genes with effects across trait categories

In some cases, GWAS uncovered significant SNPs with multiple trait associations in high pairwise LD to SNPs in other genes (Fig. 4, Table S9). These formed ‘gene blocks’ comprising adjacent genes or genes within up to 17 kb on the individual chromosome. Blocks involved 1–6 trait-associated SNPs per gene primarily located within intronic, flanking regions and coding sequence. Pairwise LD between SNPs from different genes ranged from moderate to complete linkage (r2 = 0.45−1.0) (Table S9). Not unexpectedly, genes in linkage often had similar phenotypic effects but dissimilar annotated functions. One cluster c. 8 kb in length included Potri.010G250600 (MSR2; MANNAN SYNTHESIS RELATED 2) and Potri.010G254400 (GCN4; GENERAL CONTROL NON-REPRESSIBLE 4 transporter) with associations to four phenology traits (bud set (2009–2010), height growth cessation, 100% leaf yellowing and leaf drop (2008,10)) and three ecophysiology traits (Amax/mass, Nmass and Chlsummer) (Fig. 4a, Tables 6, S9). Allelic effects of the significant SNPs from Potri.010G250600 (10_22291570; 5′UTR/10_22295252; 3′UTR) and Potri.010G254400 (10_22492661; 5′UTR) showed similar phenotypic change (i.e. in the same direction) when the SNPs were analyzed independently (Table S8). The minor homozygote accessions had earlier phenology events and greater leaf N content/chlorophyll/photosynthetic rates compared to the major homozygote accessions while the heterozygous accessions ranged in phenotypic effect from equivalency to either homozygote to divergent phenotypes. Combined, the two genes had 10 genetic variants (haplotypes) with different allelic combinations of the three SNPs and varying phenotypes observed (not shown).

Figure 4.

Linkage of single nucleotide polymorphism (SNP) markers in gene blocks with associations to different traits. (a) Potri.010G250600 (MSR2; MANNAN SYNTHESIS RELATED 2) and Potri.010G254400 (GCN4; GENERAL CONTROL NON-REPRESSIBLE 4) have multiple associations with seven traits across two trait categories. Genes are not immediately adjacent; total physical length is 8 kb. (b) Potri.015G002300 (PRR5; PSEUDO-RESPONSE REGULATOR 5), Potri.015G002600 (TTG1; TRANSPARENT TESTA GLABRA 1) and Potri.015G004100 (ANAC062; NAC-DOMAIN PROTEIN 62) have multiple associations with nine traits across three trait categories. Genes are not immediately adjacent; total physical length is 17 kb. Gene scaling and SNP locations are accurate within genes but distances between gene markers are not indicated. Gene regions are identified by coding (dark blue), intron (solid line), 3′UTR/5′UTR (light blue), and noncoding (lines extending beyond UTR regions). Hatch marks on noncoding regions indicate extensive segments of intergenic regions that could not be illustrated within the figure.

In other clusters, phenotypic effects varied depending on the SNP, but in total, associations spanned all three categories across the linked genes. One region c. 17 kb in length included a putative light-response gene Potri.015G002300 (PRR5; PSEUDO-RESPONSE REGULATOR 5), Potri.015G002600 (TTG1; TRANSPARENT TESTA GLABRA 1, protein binding) and Potri.015G004100 (ANAC062; NAC-DOMAIN PROTEIN 62 transcription factor) (Fig. 4b, Tables 6, S9). SNP alleles from Potri.015G002300 (15_141448; coding sequence, nonsynonymous/15_141921; coding sequence, synonymous/15_142205 coding sequence, nonsynonymous), Potri.015G002600 (15_162241; intron/15_163004; coding sequence, synonymous) and Potri.015G004100 (15_277979; coding sequence, nonsynonymous) affected varying combinations of phenology traits (bud set (2008–2010), growth period, height growth cessation, 100% leaf yellowing and leaf drop (2008,10), post-bud set period], two ecophysiology traits (Chlsummer (2009,11), Nmass) and one biomass trait (tree height). The underlying genetic variation was relatively complex and different combinations of the six SNPs resulted in 13 genetic variants (haplotypes) with diverse effects on the phenotypes (not shown).

Gene with potential pleiotropic effects on unrelated traits

In some cases, SNPs were associated with multiple traits that were genetically uncorrelated. The genes Potri.001G057400 (HK3; HISTIDINE KINASE 3) and Potri.005G138400 (ABA1; ABA DEFICIENT 1) each had single SNPs associated with numerous phenology traits including leaf flush, which was not correlated with any of the other associated phenology traits (see ‘Genes with effects on biomass or ecophysiology’ and ‘Genes with effects on phenology’ above; Tables 5, 6, S10). In other instances, GWAS uncovered separate SNPs within the same gene associated with different traits or suites of traits. Potri.008G038900 (encoding a homolog of Arabidopsis zinc finger (GATA type) family protein) had different SNPs associated with either leaf shape or bud set while Potri.014G109800 (encoding a homolog of Arabidopsis cathepsin B-like cysteine protease) had different SNPs associated with either log volume growth rate or multiple phenology traits (Table 6). Both cases lacked trait correlation, as neither leaf shape nor log volume growth rate is genetically correlated to any phenology trait (Table S10). Many single genes or gene clusters had SNPs with associations to Amax/mass, Nmass and Chlsummer, which are themselves are correlated, but not to any phenology or biomass trait (Tables 6, S10). For example, Potri.015G009300 (encoding a homolog of the Arabidopsis Dof-type zinc finger domain-containing protein) had SNPs associated with phenology traits and the ecophysiology trait Chlsummer while Potri.017G040800 (BAM5; BETA-AMYLASE 5) had SNPs associated with biomass and phenology traits and the ecophysiology trait Nmass. In addition, the two genes blocks previously described (see earlier) also included uncorrelated ecophysiology trait associations (Potri.010G250600/Potri.010G254400 and Potri.015G002300/Potri.015G002600/Potri.015G004100).


In this study, GWAS combining extensive genomic and phenotypic information from natural populations of P. trichocarpa uncovered numerous loci underlying variation in biomass, ecophysiology and phenology traits based on: a large collection of individuals spanning much of the natural species range; detailed, replicated trait phenotyping studies; and the largest genome-wide dataset of genetic polymorphisms in P. trichocarpa to date.

Genes underlying biomass and ecophysiology

Certain genes implicated by GWAS in determining rates of growth, whole-plant biomass and ecologically related physiological traits in P. trichocarpa may have some relationship to the associated phenotype while other associations implicate differing involvement or functionality for P. trichocarpa genes compared to their annotated Arabidopsis gene homologues that were used for poplar gene annotations (solely based on sequence homologies). For instance, Potri.010G250500 (EXO70G1; EXOCYST SUBUNIT EXO70 FAMILY PROTEIN G1) was associated with a major effect on biomass variation in the intercorrelated, complex traits of height and tree mass (Table 3). Notably, Potri.010G250500 is the upstream, neighboring gene to Potri.010G250600 (highlighted as potentially pleiotropic and linked to another high-effect gene Potri.010G254400; Fig. 4a). While Potri.010G250500 is unlinked to this gene block, it had substantial effects on tree biomass and may be related and/or affected by the potentially pleiotropic action of this genomic region. In other species, the specific function of EXO70G1 is unknown but EXO70 proteins are thought to be involved in auxin efflux carrier recycling contributing to polar auxin transport (Drdová et al., 2013). Arabidopsis mutants in a related exocyst component (EXO70A1) show reduced fertility and altered cellular development/organogenesis (Synek et al., 2006). Additional associations implicate potentially novel functionality in P. trichocarpa related to phenotypic variation. For instance, Potri.010G019000 (MEE32; MATERNAL EFFECT EMBRYO ARREST 32) was associated with log height growth in P. trichocarpa (Table 3) and has only been previously linked with tension wood growth in P. tremula (Andersson-Gunnerås et al., 2006). Another example linked Potri.008G105200 (PHYB; PHYTOCHROME B) with the number of preformed leaves in terminal buds but not phenology (see later).

Despite high intraspecific variation among accessions of P. trichocarpa in ecophysiology and biomass/growth-related traits, we found fewer associations relative to phenology and lower total phenotypic variance accounted for by trait-associated SNPs (explained by cumulative R2) (Figs 2, 3, S4). This may be due to lack of sufficient genomic coverage (i.e. SNPs not on the genotyping array) and would be ameliorated by using a broader sampling of genetic variation. Another possibility might relate to the effects of rare alleles which are hard to detect using GWAS (Ingvarsson & Street, 2011). A third possibility is loss of associations where relationships between SNP loci and geography exist (Balding, 2006) and as identified by PCA for the present study population (McKown et al., 2014). Finally, the heritability values of many biomass and ecophysiology traits are low to moderate suggesting a high local environment-response component (McKown et al., 2014; Fig. S5), and thus, detecting underlying genetic variation in these traits may be inherently difficult using GWAS.

Genes underlying phenology

The greatest number of genetic associations in P. trichocarpa involved phenology and also provided the highest cumulative R2 values found in any studied trait category (Figs 2, 3, S4) supporting high genetic complexity in such traits. The large number of genes involved in phenology is not necessarily unwarranted. In previous studies, numerous genes have also been found that control the bud activity–dormancy cycle in Populus (Ruttink et al., 2007; Jackson, 2009; Ma et al., 2010; Rohde et al., 2010, 2011; Olson et al., 2013) and distantly-related Salix (Ghelardini et al., 2014). Within this study, most SNPs provided a small contribution to the overall trait, suggesting that the evolution of variation in phenological traits involves numerous loci with small effects (cf. Rockman, 2012). This complex genetic architecture for many phenology traits in P. trichocarpa reflects the activity–dormancy cycle of the meristem. The whole-plant switch from active growth to quiescence is intricate and triggered by a number of signals, including daylength, temperature and environmental stressors (Cooke et al., 2012).

In our study, the genetic complexity of phenology traits was observed in the broad span of putative functions in associated genes, particularly late summer and fall phenology traits of bud set and leaf drop (Tables 5, 6, S5). Loci implicated by GWAS underlying phenology trait variation included multiple genes related to environmental response in Arabidopsis, such as light perception, hormone signaling, heat shock stress, cold response, water relations and drought stress. Others were related to different types of signaling in Arabidopsis, such as calmodulins/calcium, ion transport, phosphatases and kinases. We note that gene numbers and cumulative R2 identified for phenology traits did not necessarily relate to trait heritability (Figs S4, S5, Table S1). For instance, both bud break and leaf flush (H2 = 0.88, 0.85, respectively) yielded only five genes (one with associations to both traits), while bud set and leaf drop (H2 = 0.74, 0.60, respectively) yielded 222 genes (80 with associations to both traits). Correspondingly, cumulative R2 was much higher in bud set and leaf drop compared to bud break and leaf flush (Fig. S4), despite the similarly high heritability values (Fig. S5).

The timing of individual phenology events within our population is generally correlated across years but actual dates shifted annually depending on the timing of seasonal environmental cues (McKown et al., 2013). Strong genetic correlations between different phenology events exist and traits tend to be highly intercorrelated within a ‘season’ but not across seasons (i.e. spring vs late summer/fall; Table S10). Nevertheless, these correlations are not necessarily predictive and intraspecific phenotypic variation in phenology can be somewhat modified from year to year depending on environmental conditions (e.g. the timing of bud set and leaf drop is not fixed). Thus, retrieving repeated associations between SNPs and phenology traits measured across different years supports the biological relevance of these genes. Genes, such as Potri.009G017400 (BLH1; BEL1-LIKE HOMEODOMAIN 1), Potri.010G215200 (PRR7; PSEUDO-RESPONSE REGULATOR 7) and Potri.018G033600 (GA3OX1; GIBBERELLIN 3-OXIDASE 1), were each associated with multiple late summer/fall phenology traits across numerous years and have some precedent for understanding phenology timing. In Arabidopsis, BLH1 regulates the high irradiance response of PHYTOCHROME A (PHYA) (Staneloni et al., 2009) and modulates signaling by abscisic acid during development (Kim et al., 2013). BLH1 is also linked to the initiation of bud formation in P. tremula × P. alba (Ruttink et al., 2007) and is related to late summer Melampsora susceptibility in P. trichocarpa (La Mantia et al., 2013). PRR7 is a core clock gene in circadian rhythm determination within Arabidopsis through transcription–translation feedback loops (Haydon et al., 2013) and may participate in a similar role in Populus. Likewise, GA3OX1 is implicated in photoperiodic perception (Song et al., 2013) and seed dormancy (Footitt et al., 2013) in Arabidopsis, and gibberellins also have well-established roles in the transition to dormancy in Populus (Ruttink et al., 2007).

Other genes identified have a characterized function in Arabidopsis, but are novel loci for understanding phenotypic variation in P. trichocarpa. Potri.014G102700 (CYP78A9; CYTOCHROME P450 78A9) was repeatedly associated with spring phenology events while Potri.002G242500 (CB5-B; CYTOCHROME B5 ISOFORM B) was repeatedly found with late summer/fall phenology events. Although CB5-B is not known to relate to phenology, CYP78A genes in Arabidopsis are generally related to plant size, fertility, and the timing of bud opening and organ abscission (Sotelo-Silveira et al., 2013). Potri.014G102700 (CYP78A9) also showed high genetic complexity with variable effects (among related traits), and may be an example of ‘conditional neutrality’ or ‘antagonistic pleiotropy’ where different alleles might be favorable depending on the environment (Savolainen et al., 2013). Some genes associated with phenology highlighted links to nutrient availability. Two genes in high linkage, Potri.009G008500 (NRT2.4; NITRATE TRANSPORTER 2:4) and Potri.009G008600 (NRT2:1; NITRATE TRANSPORTER 2:1), were associated with late summer/fall phenology events across all years (Tables 6, S5, S9) suggesting that nitrate transporters or nitrogen availability/allocation might affect the regulation of these events. Nitrate transporters have been implicated in nitrogen sensing and auxin signal transduction, and NRT2.4 is highly expressed in numerous aboveground tissues in Populus, including the meristem (Bai et al., 2013). However, neither gene has been previously invoked in phenology in any species.

Many light-associated genes previously implicated in Populus phenology were found in our association study while others were not, despite inclusion on the SNP array. This has also been reported in the sister-species P. balsamifera, where significant phenology-related SNPs did not necessarily correspond with SNPs uncovered in other association studies for Populus (Olson et al., 2013). In addition to previously discussed genes, our GWAS uncovered COP1 (CONSTITUTIVE PHOTOMORPHOGENIC 1), FAR1 (FAR-RED IMPAIRED RESPONSE 1), PCL1 (PHYTOCLOCK 1), PIL6 (PHYTOCHROME INTERACTING FACTOR 3-LIKE 6) and PRR5 (PSEUDO-RESPONSE REGULATOR 5) (Table S5). Yet, notable genes were not among the associations, including PHYA, PHYB, CCA1 (CIRCADIAN CLOCK-ASSOCIATED1), FRI (FRIGIDA), GI (GIGANTEA), LATE ELONGATED HYPOCOTYL (LHY), and TOC1 (TIMING OF CHLOROPHYLL a/b BINDING PROTEIN/PRR1) (Ruttink et al., 2007; Ingvarsson et al., 2008; Ma et al., 2010; Rohde et al., 2011; Cooke et al., 2012; Fabbrini et al., 2012; Keller et al., 2012; Olson et al., 2013). SNPs from CCA1, LHY, TOC1/PRR1 were retrieved by phenology traits using the simple model (not shown); thus, it is possible that their signal was diminished by correcting for population structure in the mixed model, as these loci have known relationships with geography (McKown et al., 2014; A. Geraldes, unpublished). Nevertheless, it also suggests that variation within these genes does not underlie intraspecific variation of such traits in P. trichocarpa, as observed in the closely related P. balsamifera (Olson et al., 2013) and more distantly related P. nigra (Rohde et al., 2011).

Genes with associations across trait categories and potential functional pleiotropy

One of the significant findings of this study were the numerous genes with multiple significant associations to different traits, including associations across trait categories (Fig. 2, Tables 3-6), and blocks of linked genes with shared genotype–phenotype associations (Fig. 4). We consider these to be indications of pleiotropy in a broad sense (cf. Mackay et al., 2009). The repeated occurrence of ecophysiology traits associated with pleiotropic loci was notable (Table 6), particularly as these had little or no correlative relationship to biomass and/or phenology traits. Other examples of potentially pleiotropic loci have also been uncovered in P. trichocarpa (Porth et al., 2014), including a set of genes affecting phenology, wood fiber properties and disease resistance (I. Porth & J. Klápště, unpublished). In this case, phenology traits and fiber properties are functionally uncorrelated traits and the evolution of pleiotropy suggests that the developmental integration of these different traits might have led to their genetic integration (evidenced as trait co-selection; cf. Cheverud, 1996).

The pleiotropic loci in this study provide novel candidates underlying phenotypic variation in P. trichocarpa and suggest the presence of genomic regions with importance for environmental response in P. trichocarpa. The gene block with Potri.010G250600 (MSR2; MANNAN SYNTHESIS RELATED 2) and Potri.010G254400 (GCN4; GENERAL CONTROL NON-REPRESSIBLE 4) is potentially pleiotropic in P. trichocarpa (Fig. 4a) but the individual genes are not known to be functionally related or pleiotropic within other plant species. In Arabidopsis, MSR2 is localized to the Golgi apparatus, and has been implicated in mannan biosynthesis in a number of tissues, including developing vascular tissue, leaves, stems and flowers (Wang et al., 2013). The Arabidopsis transporter GCN4 is a putative ATP-binding transporter family protein but is not fully characterized in any plant species. Within P. trichocarpa, Potri.010G254400 (GCN4) is also associated with rates of Melampsora infection (La Mantia et al., 2013) and may play a role in disease resistance/susceptibility.

Another block (Fig. 4b) with Potri.015G002300 (PRR5; PSEUDO-RESPONSE REGULATOR 5), Potri.015G002600 (TTG1; TRANSPARENT TESTA GLABRA 1) and Potri.015G004100 (ANAC062; NAC-DOMAIN PROTEIN 62) suggests genes related to environmental sensing and stress response may have pleiotropic activity. PRR5 is highly upregulated with the onset of short days in P. tremula × Palba (Ruttink et al., 2007). It has also has been implicated in growth cessation and bud set in association studies of P. tremula × Palba (Ruttink et al., 2007) and P. tremula (Ma et al., 2010), and is associated with cell wall crystallinity in P. trichocarpa (Porth et al., 2013a). In Arabidopsis, PRR5 plays a role in directly regulating circadian clock genes (Nakamichi et al., 2012). Other direct targets of this regulator include transcription factors involved in flowering, hypocotyl extension and cold-stress responses, suggesting that PRR5 has light-mediated effects on many physiological processes. Both TTG1 and ANAC062 are transcription factors associated with stress responses. TTG1 affects many plant processes, including flavonoid biosynthesis, response to abscisic acid and root growth in relation to water stress in Arabidopsis (Nguyen et al., 2013). Likewise, ANAC062 is a membrane-associated stress response transcription factor in Arabidopsis and involved in abscisic acid response, cold stress and salinity tolerance (Seo & Park, 2010).

The potential pleiotropic loci detected by GWAS in this study spanned a number of functions and may have effects by acting upstream of signaling pathways that affect multiple traits (such as hormone signaling) or by directly targeting multiple genes for regulation. Within gene blocks with pleiotropic effects, such genomic regions may contain individual genes involved in signaling whose direct targets are in linkage, linked genes with similar functionality, or may represent genes with adaptive influence resulting in linkage through selective forces (Yeaman, 2013).


Employing the GWAS approach to scan the P. trichocarpa genome for significant allelic variation underlying important biomass, ecophysiology and phenology traits, we identified numerous individual genes and genomic regions where allelic variation was associated with intraspecific trait variation. The large number of SNP–trait associations highlights the polygenic nature of phenology traits in particular (Fig. 2). It is unlikely, however, that all contributing SNPs or genes are acting equally. Some may be large-effect quantitative trait nucleotides (QTNs) (Rockman, 2012; Martin & Orgogozo, 2013). The complexity of genetic trait architecture also encompasses nonadditive genetic effects such as epistasis (Hansen, 2013) and gene × environment interactions, which might modify the resulting gene effect (Hill, 2010). We noted that many allelic frequencies often accompanied phenotypic change in the same direction, suggestive of directional epistasis (Hansen, 2013) or constitute ‘hotspots’ where particular genes repeatedly are elements of phenotypic variation in similar traits (Martin & Orgogozo, 2013). Yet, we need to be cautious about the discrepancies between functional vs statistical epistasis (i.e. relative independence from population variation, cf. Hansen, 2013). The employed linear model in GWAS assumes only additive effects and may be partially fitting epistasis, which we cannot clearly dissect, and thus can exaggerate the ‘additive’ effect of the detected causative variants. Further work is required to differentiate between phenotypic variation related to epistasis and large-effect QTNs that constitute dispersed adaptive modifications, and more numerous, smaller-effect allelic variations. In the case of the latter, these may be employed to ‘fine tune’ a phenotype (Martin & Orgogozo, 2013) and/or encompass smaller trait changes required for local adaptation (Savolainen et al., 2007).

Our association results suggest a number of markers with potential ecological effects in P. trichocarpa. Many genes identified by GWAS are considered to affect growth and development and/or to respond to signaling and environmental stressors. Numerous loci, including potentially pleiotropic loci, have also been retrieved in parallel FST outlier studies indicating adaptive potential (A. Geraldes, unpublished; I. Porth & J. Klápště, unpublished). The extensive results from SNP–trait associations within this study highlight multiple avenues for further work, such as investigating functional roles of the genes implicated, genetic pleiotropy between genetically correlated and uncorrelated traits, relationships of genes with geography and local adaptation, and operative roles of important SNP variants in noncoding regions. Conclusively, this study presents an essential platform for future detailed exploration aimed at understanding species-wide ecology and evolution, particularly where numerous genetic mechanisms are invoked.


We thank L. E. Gunter, M. S. Azam, E. Drewes, N. Farzaneh, L. Liao, E. Moreno, L. Muenter and L. Quamme for data monitoring, collection and image presentation. We also thank anonymous reviewers for their suggestions and revisions in improving the manuscript. This work was supported by the Genome British Columbia Applied Genomics Innovation Program (Project 103BIO) and Genome Canada Large-Scale Applied Research Project (Project 168BIO) funds to R.D.G., J.E., Q.C.B.C., Y.A.E-K., S.D.M. and C.J.D. and by funds within the BioEnergy Science Center, a US Department of Energy Bioenergy Research Facility under contract DE–AC05–00OR22725.