Background

Leaf morphology has remarkable phenotypic diversity throughout the plant kingdom, making it a favorable system in which to study the evolution of variations in form. A number of genes and networks have been described that affect initial leaf development and pattern formation (reviewed thoroughly by [1] in both simple and complex leaves [2, 3]. Similarly, mutagenesis screens have identified genes that regulate leaf blade shape, primarily width [4–6], length [6–8] and their ratios [9]. An emerging paradigm from these discoveries is that two-dimensional leaf shape can be regulated both by differential cell elongation (polarity) or differential cell proliferation favoring one dimension versus the other [10]. This well-grounded knowledge of leaf initiation, development, and shape was established largely through mutagenesis of model systems. Comparably, little is known about whether these same molecular mechanisms underlie evolutionary differences in leaf morphological variation. Few studies have directly addressed whether alternative alleles at these genes underlie variation in leaf shape within or between different plant species [11]. In light of this shortcoming and the extensive diversity in leaf form, additional studies exploiting naturally occurring variation are needed to clarify the role of previously discovered genes in evolutionary variation for leaf traits and potentially discover new genetic regulators.

Quantitative trait locus-based approaches are frequently applied to uncover loci that regulate natural phenotypic diversity. Quantitative trait loci have been identified for leaf morphological traits in several species including tomato [12], poplar [13], oak [14], maize [15], and Arabidopsis [16]. These studies clearly show several aspects of leaf morphological variation to be subject to multigenic control. Thus, loci discovered on the basis of mutagenesis may only explain a portion of the naturally occurring variation in leaf shape. However, the molecular characterization of these regulatory loci discovered by QTL analysis has been generally prohibitive because of the challenges associated with moving from QTL to gene [17]. Novel experimental approaches developed from advances in genome sequencing and transcriptome analysis have eased QTL cloning [18–20], making QTL-based techniques powerful tools to elucidate molecular mechanisms underlying naturally occurring phenotypic variation [21, 22]. One approach to uncover genes underlying QTL is based on genetical genomics [18], which combines information from genome sequence with quantitative genetic analysis of gene expression and organismal traits of interest [23]. Transcriptome analysis extends the quantitative genetics (QTL) paradigm by providing information about an intermediate step between genotype and phenotype, potentially capturing not only genotypic but also developmental and environmental sources of variation. Genetical genomics was applied early on to model systems such as yeast [24] and Arabidopsis [25], and later to forestry crops to uncover genes involved in drough response [26] and lignin biosynthesis [27]. More recently the use of genetical genomics has expanded to several plant species, to understand the genetic regulation of developmental traits [28], as well as biotic [29] and abiotic [30] stress response.

The Populus genus is a particularly favorable system in which to apply a genetical genomics approach, given its extensive genetic and phenotypic variation, the availability of several well established interspecific pedigrees, and a rapidly growing genomic toolbox founded on the genome sequence of P. trichocarpa [31]. The genus is comprised of five evolutionary sections, and leaf morphological form is widely regarded as diagnostic of evolutionary relationships at the sectional level [32]. In addition, several studies have shown that leaf morphological characters are predictive of long-term clonal performance and growth [33–35]. Therefore, a detailed study of intersectional poplar hybrids may provide an approach to identify loci associated with this important phenotypic variation.

Here we analyze an intersectional pseudo-backcross pedigree of narrow-leaf P. trichocarpa Torr. & Gray (section Tacamahaca) and wide-leaved Populus deltoides Bartr. ex Marsh (section Aigeiros) for variation in leaf morphology using a genetical genomic approach to identify potential candidate genes as important regulators of leaf shape variation. Specially, we performed QTL analyses for leaf lamina width and lenth:width ratio in multiple experiments and a transcriptome analysis in expanding leaf tissue in this pedigree. An ADP-ribosylation factor (ARF) GTPase identified as a prime candidate gene for the regulation of leaf morphology in this pedigree was further fuctionally characterized using the pharmacological agent BFA, which interferes with the ARF GTP-GDP exchange [36, 37].

Results

Identification of a major QTL for leaf blade width

We carried out a QTL analysis for variation in leaf lamina shape (measured by lamina length, width, and their ratio) in an interspecific pseudo-backcross pedigree (family 52–124) segregating between the narrow-leaf donor parent species P. trichocarpa and the broad-leaf recurrent parent P. deltoides (Fig. 1). One or more significant QTL were detected using composite interval mapping (CIM) [38, 39] using a standard threshold of the 95th percentile of 1000 permutations, for all traits measured in the population of 396 individuals grown in a greenhouse at the University of Florida (Table 1). Additional analysis focused on a QTL on linkage group (LG) X that was most significantly associated with both lamina width and length:width ratio (Fig. 2). As the direction of the effects for these QTL were opposing – i.e., the same allele that increased lamina width also decreased lamina length:width ratio, we hypothesized that a common and pleiotropic regulator of leaf shape underlies the locus. Supporting this hypothesis, a negative genetic correlation was observed between these traits (r G = −0.4628 ± 0.0653). The QTL interval, constrained by sequence-linked microsatellite markers PMGC_2855 and GCPM_2122, encompasses 625 genes and spans ~3.5 Mb of uninterrupted sequence in the genome sequence of P. trichocarpa [40]. To reduce the number of potential candidate genes in the leaf QTL interval, we increased mapping resolution by genotyping putative recombinant progeny (n = 96) for seven additional microsatellite markers identified from the genome sequence (Additional file 1). The additional microsatellites allowed us to decrease the sequence spanned by the QTL interval to ~3.0 Mb and 450 candidate genes (Additional file 2). Due to the high proportion of phenotypic variation explained and quality of the genome assembly in this QTL region, we elected to pursue the major locus on LG X for further characterization and QTL cloning.

Fig. 1
figure 1

Leaf lamina shape variation among parents and progeny of the P. trichocarpa X P. deltoides pseudobackcross pedigree family 52–124. The donor parent, P. trichocarpa, has a lamina length/width ratio of ~3.0, while the recurrent parent, P. deltoides, has a lamina ratio of ~1.0. The trait exhibits additive variation, as the hybrid parent has a ratio of ~1.5. Segregating BC1 progeny span the spectrum of phenotypic varation from the hybrid to the recurrent parent with limited transgressive segregation observed

Full size image
Table 1 Summary of QTL detected for leaf lamina shape characters in P. trichocarpa X P. deltoides X P. deltoides family 52–124, in the population growth in greenhouse in Florida. Asterisks indicate QTL detected in a subset of the same population established in a greenhouse in Michigan
Full size table
Fig. 2
figure 2

Genome-wide composite interval mapping scan for leaf lamina length/width ratio (grey line), and for the expression of PtARF1 (black line) in family 52–124

Full size image

Trait and QTL stability in different experimental settings

QTL studies often present conflicting results when repeated across environments [41]. In order to assess the stability of the QTL detected initially, the study was repeated at Michigan Technological University by measuring variation in leaf lamina traits under controlled greenhouse conditions in 2010 and at a research field site in 2012. Composite interval mapping confirmed the QTL on linkage group X to be significantly associated with length:width ratio under greenhouse conditions (Table 1). However, the correlation between leaf traits measured under greenhouse and field conditions was low and not significant, suggesting highly plastic responses of the leaf traits to the field environment. Under field conditions no significant QTL was detected for any of the leaf traits.

Genetical genomics identifies PtARF1 as a candidate gene for lamina shape regulation

After decreasing the QTL interval we elected to move forward with a genetical genomics approach that integrates quantitative genetic analysis of phenotypes and gene expression, to further reduce the pool of candidate genes. Because of the significant role of gene expression in evolutionary diversification of species [42], we hypothesized that interspecific differences in transcription of a key regulator of cell division or expansion may explain the leaf lamina variation observed in the progeny of family 52–124. Thus, we identified genes that are differentially regulated in developing leaves of family 52–124, located within the major QTL interval in LG X, and cis-regulated by the same locus. We measured genome-wide gene expression in expanding leaf tissue from one biological replicate of each of 183 segregating progeny using a microarray that included probes for all gene models annotated from the P. trichocarpa genome. The normalized transcript data was mapped as expression quantitative trait loci (eQTL). This genome-wide analysis identified 13,403 statistically significant eQTL representing 12,392 unique gene models, as described previously [43].

Since our hypothesis was that lamina shape variation was a product of differential gene regulation arising from interspecific polymorphism(s) within the leaf QTL interval, we narrowed our focus to the genes with eQTL surpassing the significance threshold in this region. Our analysis identified 161 eQTL, which we classified as cis- (n = 116) or trans- (n = 45) acting, contingent on the physical position of the gene model in the genome assembly. We classified all eQTL arising from gene models genomic scaffolds (n = 19) that were not anchored to the genome sequence as trans-acting, since the Populus genome sequence assembly in the leaf trait QTL region was predicted to be contiguous [31].

We expected that if differential transcript accumulation of a key regulatory factor(s) accounts for phenotypic variation in leaf shape, abundance of such transcripts should exhibit a significant statistical correlation with leaf lamina shape. Furthermore, as differential regulation of these transcripts influences variation in the phenotypes, they should have eQTL co-localized in the leaf QTL interval. Thus, we utilized a standard multivariate correlation to determine the relationship between the leaf lamina phenotypes and each transcript with an eQTL in the leaf trait QTL interval. Among the 161 transcripts with eQTL co-localized to the trait interval, we identified 25 correlated (|r| > 0.20) to either lamina width, length:width ratio, or both (Table 2). Among these 25 genes, only two were correlated to both measures of lamina shape at |r| > .20. Furthermore, only Potri.010G152600 was regulated in cis- by the trait QTL region (Fig. 2), indicating that polymorphisms within its regulatory region affects its transcript level (eQTL) and might be the cause for the phenotypic variation controlled by the QTL on linkage group X.

Table 2 Phenotypic-expression correlation for 25 genes with significant cis- and trans-eQTL regulated by the major lamina shape QTL locus on linkage group X
Full size table

The closest homolog of Potri.010G152600 in Arabidopsis thaliana is At2g47170, (e-value = 2e-99, 96 % amino acid sequence identity), an ADP-ribosylation factor (ARF) -type GTPase (ARF1). Thus, the predicted gene Potri.010G152600 was renamed PtARF1. PtARF1 is located in close proximity to the leaf QTL logarithm of the odds (LOD) peak positioned at 98.11 cM of linkage group X. The 2 centimorgan (cM) region (97.11 cM–99.11 cM) with the highest LOD score spans 285.1 Kbp (from 15819851–16104972 on scaffold 10) and contains 24 candiate genes with putative function or known amino acid domains, of which PtARF1 is located at position 16037970.

Disruption of ARF function leads to alteration in leaf width growth

Plant ARFs and ARF GEFs (guanine-nucleotide exchange factors for ADP-ribosylation factor GTPases) are essential for vesicular trafficking in all eukaryotic kingdoms. ARF1-GTP has higher affinity for membranes and is essential for the formation of the budding vesicle. This mechanism has been dissected extensivelly using the pharmacological agent BFA, which interferes with the ARF GTP-GDP exchange [36, 37]. In plants treated with BFA, PIN accumulates in intracellular compartments, abolishing polar localization and disrupting normal auxin flux [44, 45].

To begin addressing the role of poplar ARFs in vesicle trafficking and PIN distribution in leaf development we established cuttings of the narrow-leaf P. trichocarpa in a hydroponic solution and treated them with BFA. After 48 h in ½ Murashige and Skoog solution, plants were transferred to the same nutrient solution supplemented with 0, 5 and 10 μM of BFA. The length and width of the first fully unfolded leaf (leaf plastochron index [LPI] 2) was recorded at the time of transfer and again 24 h after cuttings were transferred to the BFA-containing hydroponic solution. Within this interval, we observed an increase in the ratio of length to width of +0.08 (standard deviation, s.d = 0.06) in the control conditions, which is commonly observed in P. trichocarpa leaves undergoing expansion. Contrastingly, there was a significant decrease in the ratio of length to width of −0.13 (s.d. = 0.11, p-value = 0.01) and −0.18 (s.d. = 0.21, p-value < 0.05) in plants treated with 5 and 10 μM of BFA, respectively. Measurements were not made for a longer period of time because development was significantly impaired in treated plants compared to controls. These results suggest that ARF-mediated vesicle trafficking and PIN localization are critical determinants of leaf morphology in Populus.

Discussion

A key goal of quantitative genetics is to move from a phenotypic QTL to the polymorphisms that regulate it. Evidence in model systems has shown that genetical genomics, which allies the quantitative analysis of phenotypes and gene expression data in segregating populations, can effectively identify regulatory genes [18–20]. Here, we considered leaf lamina shape among an interspecific hybrid progeny of Populus in a genetical genomics context. From a series of QTL for leaf lamina shape characters, we identified a major QTL implicated in lamina morphology – pleiotropically regulating both lamina width and lamina length:width ratio.

The major QTL for leaf length:width ratio of linkage group X was consistently identified in two greenhouse experiments, confirming it as a major effector for leaf lamina shape variation in a controlled physical environment and an attractive candidate for downstream characterization. However, the QTL was not identified under field conditions suggesting that expression of the trait is plastic and potentially affected by microenvironmental variations including effects of soil, water availability and temperature. Similar to our observation, in an interspecific willow progeny, QTL positions varied for growth phenology traits between controlled and field conditions reflecting the effects of different environments [46]. In a previous QTL study, we have demonstrated that a simple change in nitrogen availability completely changes the genetic control of biomass growth and allocation, as well as carbon partitioning within wood components [62]. Phenotypic plasticity, mainly due to variation in temperature, has also been reported to affect bud set traits in black poplar planted at two sites with contrasting environmental conditions [47].

Utilizing whole-genome microarrays for expression analysis of leaf tissue, we identified a small group of genes whose expression was statistically correlated with the phenotypes of interest and whose transcript abundances were cis-regulated by the phenotypic QTL interval. Considering the statistical relationships with leaf morphology traits, genetic regulation of expression, and functional annotations, an ADP-ribosylation factor GTPase – PtARF1 – was identified as a prime candidate gene governing lamina shape characters in the interspecific hybrid pedigree. Few of the other flanking genes close to the QTL maximum have been previously reported to be involved in leaf development and other processes. For example, the Myb-like DNA-binding domain controls leaf cell differentiation in Arabidopsis [48]. This gene is also responsible for longer and narrower leaves in maize [49]. However, regulation of expression (eQTL) in none of these flanking genes co-located with the major QTL.

ARF genes are implicated in vesicle trafficking – a critical process for cell polarity during development of yeast, animals and plants. The hormone auxin regulates many aspects of plant development [50, 51]. Auxin distribution in cells is mediated by PIN, which determines the direction of auxin intercellular flow and, therefore, influences plant growth and development [52]. It has been previously shown that vesicle trafficking is essential for the localization of PIN. For proper localization, PIN must cycle between the plasma membrane and the endosomal compartments [53]. Therefore, because the polar transport of auxin mediates various essential processes in plant growth, vesicle trafficking is critical for development. In Arabidopsis mutants for a known component of the vesicle trafficking machinery, the ADP-ribosylation factor GTP exchange factor (GNOM ARF-GEF), PIN localization was disrupted resulting in loss of cell alignment in the embryonic axis [54, 55]. PIN proteins are post-translationally localized to apical and basal membranes of Arabidopsis epidermal cells in an endocytic process mediated by two Rab5 GTPase homologs [53]. Similarly, the action of AtARF1 (the ortholog of PtARF1) has also been shown to modulate the kinetics of endocytosis and PIN2 localization, as well as cell polarity, in Arabidopsis roots [44]. The connection between cell polarity and auxin has been intricately studied in several model plant systems [56, 57]. Auxin has already been implicated in initial leaf formation [58, 59], lamina margin elaboration [2, 3] and leaf vasculature patterning [60]. Thus, it is likely that leaf expansion is also directly affected by auxin flux. In Populus, our evidence supports a model whereby differential expression of PtARF1 in P. trichocarpa and P. deltoides changes the dynamics of endocytosis-mediated PIN localization in leaf cells. Decreased abundance of ARF1 transcript, and hence, ARF1 protein in P. deltoides slows the process of PIN polarization through the endocytic pathway during development and expansion. Decreased PIN polarization results in increased lateral auxin flux and, subsequently, increased lamina expansion in the leaf width direction. Conversely, in P. trichocarpa, higher abundance of ARF1 transcript leads to a more abundant supply of ARF1 protein, which increases the relative rate of endocytosis and PIN polarization. More rapid PIN polarization increases auxin flux and expansion of the lamina in the leaf length direction, decreasing lateral leaf expansion.

Conclusion

Collectively, our results provide compelling evidence for the role of PtARF1 in shaping variations in leaf morphology. Similarly, we provide another possible way in which the key plant hormone auxin could shape diversity in plant morphology. We have demonstrated that natural variation for auxin response might have a clear role in plant diversity we observe in nature, and our analysis suggests that natural variation in hormone response pathways clearly warrants additional investigation as we learn more about the evolution of morphological differences among plant taxa.

Methods

Plant material and phenotyping

This study was carried out in a pseudo-backcross pedigree of P. trichocarpa × P. deltoides (clone 52–225) and P. deltoides (clone D-124), referred hereafter as family 52–124. This pedigree was previously genotyped, phenotyped and quantitatively analyzed for several growth and developmental traits measured in plants grown in a greenhouse at the University of Florida (Gainesville, Florida, USA) in the spring of 2006 [61, 62]. In the present study, leaf morphology was measured digitally using Image Pro Plus software (Media Cybernetics, Inc., Bethesda, MD) from image scans of the leaf closest to one-half the live crown height, in three biological replicates of 396 individuals. Traits measured included leaf lamina length, width, and their ratio. Leaf length was measured along the midvein from the junction of the lamina and petiole, to the distal tip of the leaf. Blade width was measured at the widest point of the lamina. An analysis of variance [62] was applied to the phenotypic measurements to generate a least-square mean genotypic value for each individual that was used for subsequent QTL analysis. Clonal repeatability and genetic correlations were calculated as described previously [62], except excluding the effects of row and column position in the experimental design.

To evaluate the stability of QTL expression in another experimental setting, a subset of 153 individuals from family 52–124 were grown in a greenhouse at Michigan Technological University (Houghton, Michigan, USA) in the spring of 2010. Cuttings were rooted with rootone (Garden Tech) and grown in one-gallon pots in sunshine mix 1 potting soil containing 70-80 % Canadian Sphagnum peat moss, perlite, dolomite limestone, Gypsum and wetting agent (Sun Grow Horticulture, Canada). Leaf morphological traits were assessed on the three largest fully expanded leaves of the leading shoot for each individual as described above. Mean values were calculated for each genotype for subsequent QTL analysis. Plants were transferred to a field site (2 m × 2 m spacing) in June 2011 at the Ford Center Research Forest, south of L’Anse, Michigan (46°38′37″N 88°28′46″W), and leaf morphological traits were scored again in summer of 2012. Correlation among leaf traits and years was calculated using Pearson’s correlation coefficients.

Genotyping and genetic mapping of progeny

A previously developed microsatellite and microarrray-based genetic map [62] that covers at least 85 % of the reference P. trichocarpa genome was used for QTL mapping of the leaf traits. Additional microsatellite markers within the primary QTL interval were identified from the 'Nisqually-1' genome sequence (Additional file 1, [40]) using MsatFinder v.2.0 software [63]. Primers were designed for these loci within the MsatFinder interface, and loci were amplified and genotyped in 96 recombinant progeny (as judged by flanking framework markers), using 1 % agarose gels (w/v) as described in a previous study [61].

Leaf trait QTL analysis

QTL for leaf traits measured at the University of Florida were initially identified in QTL Cartographer v.4.0 [65], using CIM [38, 39] with a standard threshold of the 95th percentile of 1000 permutations. QTL for leaf traits assessed at Michigan Technological University were identified using MAPQTL v.5.0 [66, 67] using the same method (CIM) and statistical threshold applied in the University of Florida experiment. The maximum LOD and genome-wide/chromosome-wide significance intervals were determined for each QTL. Putative candidate genes underlying these QTL intervals were identified using the genome sequence of P. trichocarpa available at Phytozome v. 9 [68].

Transcriptome analysis

RNA was isolated immediately after collection [64] from one leaf apical and basal to the phenotyped leaf, in one biological replicate of 183 individuals grown in a greenhouse at the University of Florida. RNA was converted to double-stranded complementary DNA, labeled, and hybridized to a customized four-plex NimbleGen microarray platform (Gene Expression Omnibus Accession# GPL7234) using probes that minimize the contribution of sequence polymorphism to estimates of gene expression. The microarray included 55,793 60-mer probes, including one for each of the previously described gene models derived from the annotation of the genome sequence of P. trichocarpa clone 'Nisqually-1', and a set of non-annotated ESTs. The RNA manipulation methods and microarray design have been previously described in detail [61]. The raw data from hybridizations were background subtracted, log2 transformed, and quantile normalized. The data is publically available (Gene Expression Omnibus Accession# GSE12623, GSE20117 and GSE20118) and was used as input for gene expression QTL analysis.

Gene expression QTL analysis

Gene expression QTLs were identified, measured for significance, and classified as cis- or trans-acting as described previously [43]. Briefly, the quantile-normalized gene expression was analyzed with CIM implemented in QTL Cartographer [65], using the genetic map of family 52–124 [61]. Expression QTLs were declared to be significant based on a global permutation threshold of 95 % and classified as cis- or trans-regulated based on them being co-localized within the same marker interval where the respective gene model was located in the 'Nisqually-1' genome sequence.

To determine the relationship between gene expression and phenotype, phenotypic values from the same biological replicate measured for gene expression were correlated with the normalized gene expression values using the multivariate Pearson correlation function of JMP 7.0 (SAS Institute, Cary, NC). Finally, trait QTL position was confirmed for the single biological replicate on which gene expression was measured, using the same approach described above.

In vivo characterization of ARF function for leaf lamina shape determination

A set of 36 cuttings (6–7 cm) were collected from the P. trichocarpa Nisqually-1 reference genotype (narrow leaf) and grown in hydroponic chambers containing ½ concentration MS solution for 48 h. Next, cuttings were transferred to the same nutrient solution supplemented with 0, 5 and 10 μM of BFA, which inhibits vesiclular trafficking. A total of 12 cuttings were used in each treatment. Length and width of the first fully unfolded leaf (leaf plastochron index [LPI]) were recorded at time 0 (when cuttings were transferred to the BFA-containing hydroponic solution) and 24 h later. Measurements were not made for a longer period of time because development was significantly impaired in treated plants compared to controls. The distribution of length to width ratio values measured in the 12 biological replicates, in plants treated with BFA (5 and 10 μM) was contrasted with values observed in the non-treated plants using a two-sample t-test, and p-values were estimated.

Availability of supporting data

The microarray data is publically available in the National Center for Biotechnology Information Gene Expression Omnibus under the accession numbers GSE12623, GSE20117 and GSE20118.