PpeTAC1 promotes the horizontal growth of branches in peach trees and is a member of a functionally conserved gene family found in diverse plants species


For correspondence (e-mail chris.dardick@ars.usda.gov).


Trees are capable of tremendous architectural plasticity, allowing them to maximize their light exposure under highly competitive environments. One key component of tree architecture is the branch angle, yet little is known about the molecular basis for the spatial patterning of branches in trees. Here, we report the identification of a candidate gene for the br mutation in Prunus persica (peach) associated with vertically oriented growth of branches, referred to as ‘pillar’ or ‘broomy’. Ppa010082, annotated as hypothetical protein in the peach genome sequence, was identified as a candidate gene for br using a next generation sequence-based mapping approach. Sequence similarity searches identified rice TAC1 (tiller angle control 1) as a putative ortholog, and we thus named it PpeTAC1. In monocots, TAC1 is known to lead to less compact growth by increasing the tiller angle. In Arabidopsis, an attac1 mutant showed more vertical branch growth angles, suggesting that the gene functions universally to promote the horizontal growth of branches. TAC1 genes belong to a gene family (here named IGT for a shared conserved motif) found in all plant genomes, consisting of two clades: one containing TAC1-like genes; the other containing LAZY1, which contains an EAR motif, and promotes vertical shoot growth in Oryza sativa (rice) and Arabidopsis through influencing polar auxin transport. The data suggest that IGT genes are ancient, and play conserved roles in determining shoot growth angles in plants. Understanding how IGT genes modulate branch angles will provide insights into how different architectural growth habits evolved in terrestrial plants.


Over the last 40 years, work on tree architecture has intensified as it represents a critical aspect of the management of landscapes, plantations and forests, as well as for the aesthetics of our environment. Tree architecture is heavily influenced by the spatial patterning of branches (timing and number of branches formed, spacing, degree of branching, branch angle and orientation of branches), which is ultimately the result of the interaction of numerous genetic, developmental and environmental factors (Halle et al., 1978; Barthélémy and Caraglio, 2007; Broeckx et al., 2012). We focus here on one prominent feature, the branch angle, which makes a major contribution to the overall tree architecture (Tomlinson, 1978).

In general, branch angles vary among tree genotypes, but even within a single genotype they are not uniform and are subject to extensive variations within any given tree (Broeckx et al., 2012). Most familiar tree canopy shapes display a phenomenon whereby branches closer to the apical meristem grow more vertically, whereas those lower in the canopy tend to be more horizontally oriented. Decapitation of the apical stem will lead to subsequent vertical growth of the nearest axillary shoot(s), a well-studied phenomenon associated with apical control (Cline, 1997; Wilson, 2000). The opposite happens to branches of so-called weeping or pendulate types, found in numerous tree species (Govaerts et al., 2011). Shoots in weeping trees emerge and grow vertically at the beginning, whereas later the direction of their growth is reversed, resulting in a downward orientation of the shoots (DeVries, 1904). Collectively, these findings indicate that the growth pattern of branches shows a high degree of plasticity, and is coordinated by both developmental programs and environmental conditions in order to minimize interference with other shoots, and to optimize their access to light (Broeckx et al., 2012).

To simplify management, optimizing tree architecture to maximize fruit or biomass productivity under changing environmental conditions is a chief goal of numerous tree-crop industries. Well-established studies in forestry and horticulture include pruning management, hormone treatments, fertilizer applications and the effects of rootstock–scion interactions (Costes et al., 2006; Scorza et al., 2006; Miller and Scorza, 2010; Schneider et al., 2012), as well as the modeling of plant architecture (Allen et al., 2005; Tang et al., 2011). However, the molecular basis for different tree-growth habits is poorly understood, despite its importance in practical applications. Architectural tree types suited for high-density production systems offer great opportunities to dramatically increase yields in tree-based agricultural systems (Jacob, 2010; Schneider et al., 2012). In this regard, fastigiated tree forms (also called columnar, pillar or broomy) have been used for their narrow branch angles and reduced canopy diameters (Scorza et al., 1989; Kelsey and Brown, 1992). In Malus domestica (apple), it is estimated that orchard yields can be increased by up to 300% with columnar trees (Jacob, 2010). Prunus persica (peach) mutant varieties with pronounced vertical growth of branches, resulting in a fastigiated tree shape, have been successfully used for breeding cultivars suited for high-density production systems (Scorza et al., 2006). This trait, called ‘broomy’ (br), and later designated ‘pillar’, was shown to be incompletely dominant, as heterozygous individuals have intermediate branch angles, referred to as ‘upright’ (Scorza et al., 1989, 2002; Tworkoski and Scorza, 2001) (Figure 1). The most prominent feature of pillar trees is that the branches tend to grow vertically, regardless of their canopy position. These trees exhibit additional phenotypes, including alterations in branch growth dynamics and stem diameter (Tworkoski et al., 2006). There are two different sources of the br trait available in peach breeding: (i) the ‘Italian Pillar’ and (ii) the ‘New Jersey Pillar’. The latter was introgressed from a Japanese ornamental peach variety (Yamazaki et al., 1987). The br locus in peach (‘Italian’ as well as ‘New Jersey’ origin) was mapped to linkage group 2 of the Prunus general genetic map using microsatellite markers (Sosinski et al., 2000; Dirlewanger et al., 2004; Sajer et al., 2012).

Figure 1.

Typical architecture of 2-year-old pillar, upright and standard peach trees from an F2 segregating population. Trees were photographed during the spring bloom of 2010.

Traits controlling tree shape have also been mapped in other Rosaceae species. Two quantitative trait loci (QTLs) were identified in Prunus armeniaca (apricot), which were shown to competitively contribute to fastigiated versus drooping phenotypes (Socquet-Juglard et al., 2012). These were located on distinctly different linkage groups on the Prunus general map than the peach br trait, and thus may be different genes. In apple, a fastigiated phenotype was previously described as columnar, which occurred as a mutation at the Co locus in a spontaneous sport of the variety ‘MacIntosh’ that has a spur phenotype (Fisher, 1970). However, this trait is dominantly inherited in apple, and although it has narrow branch angles, other features such as shortened branches make it distinguishable from peach br. Furthermore, map-based cloning approaches located the co gene on chromosome 10 (Tian et al., 2005; Zhu et al., 2007; Kenis et al., 2008; Moriya et al., 2009), which does not show any synteny with peach linkage group 2 (Illa et al., 2011), suggesting that apple co is in fact distinct from the peach pillar trait. However, fine mapping in apple narrowed the locus down to a 193-kb interval (Bai et al., 2012), and in combination with the transcriptome analysis (Zhang et al., 2012) might soon reveal the causal polymorphism for the columnar trait in apple.

Although the control of branching, including axillary meristem initiation and the outgrowth of axillary buds, has been intensively studied at the molecular level (Bennett and Leyser, 2006), the molecular mechanism determining the branch angle has not yet been elucidated. In the current study we sought to identify the gene(s) responsible for the pronounced vertical growth habit of branches in peach tree varieties with a pillar tree shape. In the current report, we present the identification of a previously unknown gene family that plays a role in plant architecture as well as the development of a whole genome sequencing method applicable for universal map-based cloning approaches.


Sequence-based mapping of br using pnomes

To identify the polymorphism responsible for the pillar trait, we devised a strategy for simultaneous genetic mapping and candidate gene identification using next-generation sequencing of pooled genomes, dubbed here ‘pnomes’. The pnomes strategy is based on sequencing a population(s) of segregating individuals pooled by a specific trait(s). In theory, the linkage of individual polymorphisms to a trait of interest should be measurable by calculating the abundance of each polymorphism within a given pnome assembled against a reference genome. Tightly linked polymorphisms should occur at high frequency in the pnome containing the trait, whereas those same polymorphisms should be rare or absent in the pnome lacking the trait, and vice versa. Consequently, when plotted by nucleotide position, the data should produce a bell-shaped curve delineating the location of the trait. A schematic describing the method used here is shown (Figure 2). To test the efficacy of the pnomes strategy we sampled a peach population segregating for the pillar trait. Details regarding the population, phenotyping and trait segregation are provided in Table S1. DNA was extracted from 27 standard trees and 56 pillar individuals from an F2 segregating population, and combined into two pools (standard and pillar) for pnome sequencing. The two DNA pools were sequenced via Illumina 100-bp paired-end reads to an estimated coverage of 2× and 1.6× (relative to the number of genomes in each pnome), respectively. Next, the pnomes were separately assembled against the peach genome (The International Peach Genome Initiative, 2013; sequence available at http://www.rosaceae.org/species/prunus_persica/genome_v1.0) and subjected to pnome-wide single nucleotide polymorphism (SNP) and deletion insertion polymorphism (DIP) searches using clc genomics workbench (CLC Bio, http://www.clcbio.com). Approximately 300 000 SNPs and 36 000 DIPs were identified from both pnomes, and filtered to identify linked polymorphisms. After filtering, a total of 487 SNPs and 23 DIPs remained, and all were located on scaffold 2 (Table S2). The resulting average SNP/DIP frequencies were plotted by reference nucleotide position to reveal the physical location of the br gene responsible for the pillar trait (Figure 3). The collective results showed a bell-shaped curve with a peak near the distal end of scaffold 2 (Figure 3a).

Figure 2.

Schematic of the pnome workflow used in this study. Details including software parameters and filtering steps are presented in 'Experimental Procedures'.

Figure 3.

Pnome mapping of the pillar trait in peach.

(a) SNP/DIP pnome frequency map for the peach pillar trait. The x-axis represents the scaffold position, whereas the y-axis shows the average SNP frequency. Data prior to SNP filtering (top) and after filtering (bottom) are shown. SNPs/DIPs unique to pillar (i.e. not present in the peach reference genome of the cultivar ‘Lovell’ or the standard pnome of ‘True Gold’) are shown in blue, whereas those unique to standard (i.e. not found in peach reference genome ‘Lovell’ or the ‘Italian Pillar’ pnome) are in red. The dashed line represents a trend line. SNP clusters are indicated by brackets and numbered. A 2-Mb region (blue bar) was identified with the highest SNP frequencies.

(b) Genetic linkage map showing the HRM marker positions (blue) on scaffold 2 and the resulting calculated cM distances (red). The region delineated from the frequency map is indicated by the shaded bar. SNPs within the mapped region are indicated by circles, whereas DIPs are indicated with triangles.

We further analyzed the more distant linkage clusters, some of which were located on the proximal end of scaffold 2. We compared the filtered SNP/DIP data set with the unfiltered data, and found that the SNP clusters furthest from the mapped location represent only a small proportion of the total SNPs present at those loci (Figure 3a; Table S2). Consequently, these distant clusters appear to be artifacts that occur as a consequence of variations in SNP density.

In the region showing the highest level of linkage, polymorphisms were rare, with only 15 SNPs/DIPs identified within a 2-Mb interval. The defined position was consistent with previous pillar mapping studies that had positioned the trait on the distal half of scaffold 2 (Chaparro et al., 1994; Sosinski et al., 2000; Sajer et al., 2012).

To confirm and further narrow the interval, seven high-resolution melting (HRM) SNP markers, spanning the region from 17.4 to 23.1 Mb, were designed from the pnome polymorphism data and were then tested on all 83 F2 individuals (56 pillar and 27 standard; Table S3). The results confirmed the accuracy of the allele frequency graph as the identification of recombinant individuals narrowed the causative polymorphism between positions 19.349 and 20.128 Mb (Figure 3b; Table S3). Within the mapped interval, only two pillar SNPs remained, located at positions 19.526 and 19.566 Mb, neither of which fell within or near annotated genes. Based on the results, we hypothesized that the causative polymorphism could be a larger structural anomaly not revealed by the SNP or DIP searches. To assess this, we used the clc genomics workbench structural variation detection tool to identify potential insertions, deletions or rearrangements within a 5-Mb segment spanning the mapped interval (17.000–22.000 Mb). A total of 95 putative structural polymorphisms were identified and filtered using the same method described for SNPs/DIPs. The vast majority of identified structural variations were found to be assembly artifacts arising from homopolymer or short repetitive regions. After filtering, five putative insertion events remained: three enriched in the standard pnome and two within the pillar pnome. Only one of the events specific to the pillar pnome fell within the mapped interval. It consisted of a putative insertion element that could be ascertained by the presence of unaligned flanking sequences on both ends of stacked reads (Figure S1). This insertion event had the highest pillar pnome frequency of all polymorphisms identified.

BLAST analysis of the flanking, unaligned overhang sequences indicated the insertion is a previously uncharacterized, non-coding, 1410-bp repetitive element, with a copy number >90, found throughout the peach genome (Table S4). The putative insertion was located at position 19 659 067 bp and fell within the third exon of the predicted gene Ppa010082, annotated as encoding an unknown protein. The insertion site in Ppa010082 was marked by a short nucleotide repeat (GAT)7 within exon 3 that encodes a contiguous stretch of aspartic acid residues. Marker P19.659, which was designed to flank the insertion, confirmed that the element was present in all 56 pillar individuals, and in none of the 27 standards. This marker, along with HRM markers at positions P19.652 and P20.128, were tested on an additional 157 pillar individuals derived from several segregating populations with similar pedigrees to confirm the location. No recombinants were found for either the P19.652 or P19.659 markers. In contrast, three recombinants were identified for the P20.128 marker. Collectively, the pnome and marker mapping data excluded all but two SNP polymorphisms, neither of which fell within or near gene sequences, indicating that the insertion event within Ppa010082 was highly likely to be the causative polymorphism for the pillar trait.

Sequence analysis of Ppa010082

Ppa010082 was found to encode a predicted protein of 302 amino acids that was later confirmed by RT-PCR, followed by sequencing. When translated, the insertion element in pillar results in a premature stop codon at amino acid position 102. Translation initiation from the 3′ end of the insertion element leads to stop codons in all three reading frames prior to the resumption of the Ppa010082 coding sequence. BLAST analysis of Ppa010082 indicated that it is present in most, if not all plant species, and occurs most often as a single or low-copy gene. Weak similarity (23% amino acid identity) was found with monocot genes from Oryza sativa (rice) and Zea mays (maize), previously characterized as the cereal-specific gene tiller angle control 1 (TAC1). OsTAC1 and ZmTAC1 were found to be components of QTLs associated with tiller growth angle. SNPs within these genes disrupted splicing and gene expression, respectively (Yu et al., 2007; Ku et al., 2011), resulting in a more vertically oriented stature. Because of the similarities between TAC1 and Ppa010082 we named this peach ortholog PpeTAC1.

PpeTAC1 gene expression

To determine the expression pattern of PpeTAC1, we performed qPCR studies using a set of tissue samples collected from both vegetative and reproductive tissues of the standard double haploid cultivar (‘True Gold’) at various stages of growth and development (Figure 4a). Significant expression was observed in axillary tissues, including vegetative buds, branch nodes, apical meristems, young fruit and flower buds. The high expression level of PpeTAC1 in flower buds is consistent with the nearly horizontal growth of the flower pedicels in standard trees, compared with the narrow growth angles in the pillar mutants, where the flower pedicels grow nearly parallel with the stems (Figure 4b). Interestingly, PpeTAC1 shows a relatively high level of expression in attachment sites of actively growing branches, where its role in the patterning of vertical versus horizontal growth may be required. In contrast, very low relative expression levels of PpeTAC1 were observed in mature or dormant tissues, suggesting it is specific to actively growing tissues. Collectively, these data indicate that PpeTAC1 is specifically expressed within or near actively growing vegetative and reproductive tissues.

Figure 4.

Tissue-specific expression of PpeTAC1.

(a) Expression of PpeTAC1 in anatomical tissues collected from standard trees cv. ‘True Gold’. Tissue types are labeled. The y-axis represents relative expression values derived from qPCR results after normalization and standardization. Expression was highest in axillary flower and vegetative buds, as well as axillary branch attachment sites.

(b) Image of flower pedicel angles from standard (top left) and pillar (bottom left) trees. Quantitative differences in axillary floral bud emergence angles from pillar, upright and standard trees are shown on the right. Error bars represent the standard deviations of three biological replicates derived from three independent trees.

(c) qPCR results for PpeTAC1 expression in branch attachment sites from pillar, upright and standard trees. No transcript was detected in either ‘Italian Pillar’ or ‘New Jersey Pillar’ samples. Error bars represent standard deviations of three biological replicates.

(d) Mutations in PpeTAC1 are associated with the pillar phenotype; the PpeTAC1 gene structure and the relative positions of putative disruptions are shown.

To test whether PpeTAC1 gene expression is altered in pillar cultivars, transcript levels were measured via qPCR. RNAs were extracted from branch attachment sites collected from 1-year-old field-grown shoots of pillar, upright and standard stature trees. PpeTAC1 transcript could not be detected in either ‘Italian Pillar’ or ‘New Jersey Pillar’. Similarly, transcript levels in the heterozygous upright trees were reduced relative to standard controls (Figure 4c).

The lack of expression in ‘New Jersey Pillar’ prompted us to assess whether this cultivar possessed the same insertion element found in ‘Italian Pillar’. Previous mapping studies had positioned the ‘New Jersey Pillar’ trait to the same region of scaffold 2 (Sosinski et al., 2000). A 3-kb genomic fragment of PpeTAC1 was PCR amplified and sequenced from ‘New Jersey Pillar’. Surprisingly, the insertion element present in ‘Italian Pillar’ was not found. Instead, PpeTAC1 in ‘New Jersey Pillar’ contained four SNPs within the third and fourth introns, and a fifth SNP in the 3′ untranslated region (3′-UTR; Figure 4d). None of the SNPs showed obvious deleterious impacts; however, one of the SNPs was located at position +11 of the intron-3 donor splice site. (Figure S2).

Attac1 mutant exhibits a vertical growth habit

The discovery that TAC1 influences the angle of axillary shoot growth in rice, maize and peach prompted us to assume that a similar function should be present in the model plant Arabidopsis thaliana. We were able to identify a single TAC1 homolog in the Arabidopsis genome, At2g46640, which we named AtTAC1. A putative attac1 T-DNA mutant line from The Arabidopsis Resource Center (CS825872; ABRC, http://abrc.osu.edu) that carried a T-DNA tag within intron 4 of AtTAC1 was obtained and confirmed via PCR. A homozygous line was identified from T1 progeny and further characterized. This mutant lacked any detectable transcript of AtTAC1, thereby confirming the knock-out of the gene (Figure 5). With regard to growth habit, Arabidopsis plants first grow as a rosette, but then after bolting produce axillary shoots from inflorescence bolts. Measuring lateral axillary branch angles from both mutant and wild-type inflorescences showed that in the attac1 mutant the angle of lateral axillary shoot growth was significantly narrower (P < 0.0001) than in the wild-type, resulting in a more vertical growth habit (Figure 5a). The architecture of attac1 mutant plants was also assessed upon repeated bolt removal and re-growth. After three rounds, the resulting bolt clusters in the mutant were more densely packed and had a significantly smaller diameter (Figure 5b).

Figure 5.

Phenotypic characterization of the attac1 mutant in Arabidopsis thaliana.

(a) Inflorescence branch angles are altered in the attac1 mutant; the Columbia control is shown next to the attac1 T-DNA mutant. The box plot (right) is derived from a Student's t-test analysis, and shows significantly different mean axillary branch emergence angles (from the horizontal axis) between attac1 and control plants (P < 0.0001). Dots represent individual measurements.

(b) Image showing architecture of bolt clusters after three cycles of inflorescence removal and re-growth. The graph (top) shows mean number of bolts per plant in Col-0 versus attac1. The bottom graph shows the mean circumference of bolt cluster in mm.

(c) Graph showing mean branch angles (y-axis) in Col-0 (white) and attac1 (black) transformed with 35S::AtTAC1 or empty vector. Values shown were derived from branch angle measurements taken from T0 lines (≥ 115). Letters above each bar indicate significance group, derived from pairwise Student's t-tests.

(d) Average AtTAC1 expression values of transformed plants shown in (c). The y-axis represents the average relative qPCR expression value derived from a minimum of 14 T0 lines. The standard deviation is shown. No TAC1 expression could be detected in attac1 mutant lines.

Complementation experiments were performed using a 35S::AtTAC1 construct. In addition, plants were transformed with empty vector as a control. Details regarding the sequences are provided in Figure S3. 35S::AtTAC1 complemented the mutant phenotype relative to Col-0 plants transformed with the same construct (Figure 5c); however, the mean branch angles were slightly lower than in Col-0 plants transformed with the empty vector, suggesting that the overexpression observed in 35S::AtTAC1 plants may result in some inhibitory effect (Figure 5d).

Evolution of TAC1-like genes

Given the potentially conserved function of TAC1 in both monocots and dicots, we performed iterative BLAST searches to determine the phylogeny of TAC1 genes. Multiple alignments of diverse TAC1-like proteins revealed the presence of four conserved domains (Figure S4; Data S1). We named this gene family IGT because of the universally conserved motif (GϕL(A/T)IGT) present in domain II of all sequences identified. This motif was subsequently used to search additional eukaryotic genomes. Phylogenetic analyses of a representative set of sequences showed the presence of a second class of TAC1-like genes that formed a distinct clade (Figure 6). This second clade included LAZY1, which was previously shown to have the opposite effect as TAC1 on tiller growth angle in rice and branch angle in Arabidopsis (Li et al., 2007; Yoshihara and Iino, 2007; Yoshihara et al., 2013). LAZY1 was shown to positively influence gravitropism by negatively regulating polar auxin transport. LAZY1 mutants show altered auxin distribution in axillary shoots, which is associated with horizontal axillary shoot growth (Li et al., 2007; Yoshihara and Iino, 2007; Yoshihara et al., 2013).

Figure 6.

Phylogenetic tree of various IGT family members identified in diverse plant species. The tree was constructed using the UPGMA algorithm from a manually refined amino acid multiple alignment, generated using clc genomics workbench (CLC Bio). TAC1 and LAZY1 clades are indicated on the right. Plant classifications are color coded according to the legend. Plant common names and sequence accession IDs are listed.

IGT homologs were also found within the genome of the primitive moss Physcometrilla patens, as well as the fern (spikemoss) genome Selaginella moellendorffii, both of which lack true axillary shoots. In contrast, no hits were identified from the aquatic algal genome Chlamydomonas reinhardtii. All Physcometrilla and Selaginella homologs share the domain-V, C-terminal motif, WIKTD, which is highly conserved among the LAZY clade, suggesting primitive plants lack a TAC1 homolog (Figure S5). The intron–exon structure of IGT genes is highly conserved, marked by the first two codons (nucleotide sequence ATGAAG) occurring at the end of the first exon (Figure S6). This conserved gene structure suggests that despite their overall relatively low sequence similarities, all IGT genes have a common evolutionary origin.

Spatial expression of TAC1 and LAZY1

Yoshihara et al. (2013) previously showed that AtLAZY1 is predominately expressed in the upper part of the inflorescence vasculature, including the apical meristem. Here, we compared the spatial expression patterns of TAC1 and LAZY1 to determine whether or not they are co-expressed. The primary shoot and all lateral shoots of 5-week-old mature Arabidopsis plants and 12-week-old peach seedlings were dissected into terminal, central and basal sections, and then subjected to qPCR analyses (Figure 7; Table S6). Results showed that TAC1 and LAZY1 expression patterns are similar in both peach and Arabidopsis, as expression was predominately in the apical shoots and the upper sections of the main stem, as well as in the upper laterals. In contrast, lower lateral shoots and the basal sections of the main trunk showed little or no expression. The results for LAZY1 are consistent with those previously reported by Yoshihara et al. (2013). Although the overall expression patterns between TAC1 and LAZY1 were similar, TAC1 expression was more prevalent near the apical meristems, whereas LAZY1 expression tended to be higher in the central stem region. Collectively, these data suggest that TAC1 and LAZY1 expression are coordinately regulated.

Figure 7.

Expression of TAC1 and LAZY1 in dissected 4-week-old wild-type Arabidopsis (Col-0) plants (bottom) and 4-month-old peach saplings (cv. True Gold) (top). Short black lines indicate cut points where sections of each individual plant were dissected and flash frozen. RNA could not be recovered from a small number of samples, which are highlighted with dashed lines. The Arabidopsis image is a redrawn representation of an individual plant, whereas the peach image is derived from an actual picture of the plant that was dissected. Expression values obtained for each segment via qPCR are color coded according to an expression scale (bottom). The experiment was repeated on a total of three individual plants for each species, with similar results. The actual values are presented in Table S6.


Pnomes mapping

Identifying genes responsible for specific traits in non-model systems remains a significant challenge, particularly in tree species where long juvenility periods and limited orchard space impede genetic studies. Here, we present a method using pnomes that overcomes some of these limitations by streamlining the process of mapping and candidate gene identification. This method is similar to the concept described by Lister et al. (2009) and demonstrated by Schneeberger et al. (2009), who were able to identify a causative point mutation in a segregating EMS-mutagenized A. thaliana population. The advantage of this technique is that, in principal, a trait can be simultaneously mapped and the underlying causative polymorphism identified in a very short period of time (Schneeberger et al., 2009). In the case of the peach pillar trait, this approach was successful as we were able to map and identify a single candidate gene (PpeTAC1) in less than 4 months, starting from leaf material. The polymorphism frequency data derived from two relatively small populations (27 standard and 56 pillar individuals) was sufficient to map the trait to within 2 Mb. The near-complete identification of polymorphisms within the mapped region enabled the rapid SNP marker development and subsequent fine mapping of the locus. Both the pnome allele frequency data and the molecular marker data were in very close agreement. Here, relatively small numbers of individuals were used to generate the pnomes. With the rapid advances of next-generation sequencing, both in regard to sequencing volume and read length, it should soon become possible to sample much larger populations and identify relatively narrow physical mapping intervals. Likewise, this technique could, in theory, be applied to multilocus traits or QTLs.

TAC1 is required for horizontal shoot growth in peach and Arabidopsis

Several lines of evidence support the identified insertion in the PpeTAC1 gene as the causative mutation responsible for the peach pillar phenotype. First, only three polymorphisms could be identified within the mapped region: two were SNPs that did not fall within or near predicted genes, leaving the insertion event identified within PpeTAC1 as the mostly likely candidate. Second, sequencing of PpeTAC1 from a distinct peach cultivar, ‘New Jersey Pillar’, revealed that it lacked the insertion event, but instead had four SNPs that fell within introns 3 and 4 and a fifth SNP in the 3′-UTR. In both ‘Italian Pillar’ and ‘New Jersey Pillar’, PpeTAC1 expression could not be detected, confirming that it is knocked-out in both cases but arises from independent mutations. In the case of ‘Italian Pillar’, the loss of expression is likely to stem from the presence of the insertion element within exon 3. The SNP(s) responsible for loss of expression in ‘New Jersey Pillar’ is less clear, although it should be noted that an SNP in the 3′-UTR of rice TAC1 was previously shown to lead to reduced TAC1 expression levels (Yu et al., 2007). Third, a T-DNA knock-out mutant line of the homologous AtTAC1 gene in Arabidopsis thaliana showed a similar phenotype, as the lateral shoots of inflorescence bolts had narrower branch angles. A similar result was previously reported for rice and maize, where mutations in TAC1 were associated with more erect architecture (Yu et al., 2007; Ku et al., 2011). Lastly, the overexpression of AtTAC1 under a 35S promoter partially complemented the Attac1 mutant phenotype. Similar results were reported for LAZY1, where overexpression in rice or Arabidopsis led to only partial complementation (Yoshihara and Iino, 2007; Yoshihara et al., 2013). Our finding that overexpression in wild-type Col-0 plants led to a decrease in branch angle suggests that overexpression is slightly inhibitory. Collectively, these data indicate that TAC1 is indeed associated with the determination of branch angle in dicots.

The finding that TAC1 determines the axillary shoot growth angle in grasses, herbaceous plants and trees suggests that the mechanisms controlling axillary shoot growth angle are, at least in part, universal. How TAC1 contributes to shoot growth trajectory is unknown. Given the similarity of TAC1 to LAZY1 and their overlapping expression patterns, it seems appropriate to hypothesize that TAC1 may regulate LAZY1 either directly or indirectly. The silencing of LAZY1 leads to the horizontal growth of Arabidopsis inflorescence lateral shoots and rice tillers, and is associated with weakened gravitropic responses, probably as a consequence of impaired polar auxin transport (Li et al., 2007; Yoshihara and Iino, 2007; Yoshihara et al., 2013). In comparison, pillar peach trees were previously shown to have higher levels of auxin in axillary shoots, consistent with an effect on polar auxin transport (Tworkoski et al., 2006). Moreover, the pillar trait was previously found to be epistatic to the weeping trait in peach, demonstrating that the reversal of agravitropism in weeping requires PpeTAC1 (Werner and Chaparro, 2005). Comparison of TAC1 and LAZY1 expression in peach and Arabidopsis revealed significant overlap in their expression patterns, particularly within the central and terminal stem sections of the apical shoot and upper lateral shoots. These patterns deviated, as TAC1 expression was predominant in apical meristems and shoot tips, whereas LAZY1 expression was more pronounced in the central stem sections adjacent to the shoot tip. Collectively, the data suggest that TAC1 and LAZY1 coordinate lateral shoot growth angle, although at this time we can only speculate on the nature of this presumed interaction.

TAC1 and LAZY1 are members of a previously unknown gene family

The phylogenetic analyses described here suggest LAZY1 pre-dated TAC1, hence TAC1 is likely to have evolved from LAZY1 at a point that coincides with the evolutionary advancement of axillary shoots. All angiosperms and gymnosperms contain at least one TAC1 and one LAZY1 homolog, whereas primitive plants such as mosses and ferns contain LAZY1 but appear to lack TAC1. The major structural difference between these two clades of IGT genes is the conserved C-terminal domain V that is present only in LAZY1. Domain V contains an EAR-like motif (LxLxL), which may confer the ability to repress auxin response genes to LAZY1 (Tiwari et al., 2004; Yoshihara et al., 2013). We predict that, concurrent with or shortly after the inception of axillary shoots in primitive plants, TAC1 arose as a C-terminal truncation of LAZY1 and adopted a role in promoting the horizontal growth of axillary shoots, possibly through the inhibition of LAZY1-induced agravitropism.

Although it is tempting to speculate that TAC1 and LAZY1 influence axillary shoot growth angle via altered gravitropic responses, such a model is insufficient to account for all aspects of tree branch angle control. Primary lateral shoot growth angles can be explained in relation to the gravity vector, but the growth of higher order axillary shoots cannot. Rather than being driven by gravitational set points, the growth angle of higher order shoots is set to the growth angle of the parent branch, sometimes resulting in downward growth trajectories. Moreover, the horizontal growth habit of primary axillary shoots is subject to their respective distance from the apical meristem: a phenomenon associated with a poorly understood process known as apical control (Wilson, 2000). Apical control is distinguishable from apical dominance, and influences numerous aspects of branch growth in addition to branch angle, such as branch diameter, growth rate and upward-bending mobility. Pillar peach trees show many altered growth characteristics, including changes in lateral branch growth rates and stem diameters that are consistent with potential alterations in apical control. Moreover, expression data suggest that TAC1 and LAZY1 are predominately expressed in the upper shoots, and are low or absent in lower shoots, suggesting that their mode of action is developmentally regulated. Taken together, these data suggest that a more complex model, that takes into account developmental signaling programs and apical control, will be needed to fully explain how IGT genes influence axillary shoot growth patterns in trees. Understanding how TAC1 and LAZY1 cooperate to influence plant spatial patterning should shed light on our understanding of plant development and how modern plant architectures first developed.

Experimental Procedures

Peach tree populations

For pnome mapping, an F2 peach (Prunus persica) population of 250 individuals segregating for the pillar trait was used. This was derived from a selfed F1 individual that was a progeny of an F0 cross between the pillar cultivar ‘Crimson Rocket’ and the double-haploid variety ‘True Gold’ that has a standard architecture (Table S1). The trees were phenotyped and leaf tissue was collected from selected 10-year-old trees. Fine mapping was performed using trees from several different segregating populations, in which the pillar trait was derived from ‘Italian Pillar’. From these, 157 pillar trees were chosen based on their having a clear pillar phenotype. Standard and upright individuals were not used for fine mapping because of the inability to accurately phenotype them.

DNA extraction and quantification

Genomic DNA was extracted from liquid nitrogen-treated ground leaf samples with the EZNA™ High Performance (HP) DNA Kit (Omega Bio-Tek Inc., http://www.omegabiotek.com). Modifications to the Frozen Specimens protocol were the addition of 2% Polyvinylpyrrolidone-40 (PVP-40) (w/v) to Buffer CPL and the optional addition of 2-mercaptoethanol. The fluorescent dye PicoGreen was used to obtain specific double-stranded DNA measurements using the Quant-iT PicoGreen kit (Invitrogen, http://www.invitrogen.com/site/us/en/home/brands/Molecular-Probes.html). The resulting DNA samples were combined in equal molar ratios to generate the pillar and standard pools for pnome sequencing.

Pnome mapping

The two DNA pools were subjected to Illumina GAII 100-bp paired-end sequencing: 174 and 218 million reads were obtained for the standard and pillar pnomes, giving an estimated coverage of 2× and 1.6×, respectively. Next, the pillar and standard reads were separately assembled against the peach genome (The International Peach Genome Initiative, 2013; sequence available at http://www.rosaceae.org/peach/genome) using clc genomics workbench (CLC Bio, http://www.clcbio.com). Default parameters were used and non-specific reads were excluded. SNPs and DIPs were identified from each pnome assembly using the respective CLC tools. Default parameters for SNP/DIP significance were changed as follows: minimum coverage and paired coverage = 20; maximum coverage = 500; minimum variant count required = 20; maximum expected variations = 2. Frequency values for each SNP/DIP are reported by clc based on the number of high-quality reads in which each polymorphism occurs relative to the total number of high-quality reads that span that nucleotide position. To compare the pillar versus standard SNP/DIP data sets, they were filtered as follows. First, SNPs/DIPs with the same variant allele (relative to the reference) occurring at a frequency >75% in both the pillar and standard pnomes were removed. These largely represent non-segregating polymorphisms that only occur relative to the Lovell genome. Second, SNP/DIPs with a variant frequency below 75% were removed from each data set to eliminate unlinked polymorphisms. Finally, the remaining SNPs and DIPs were manually verified by inspection of the corresponding sequence assemblies to eliminate artifacts arising from assembly differences.

To create an allele frequency map, the frequencies of SNPs/DIPs in the opposing pnomes (in which they are low or absent) were obtained using low-stringency SNP/DIP searches (with the same parameters as above, except that the minimum variant count was set to 1). Next, we calculated the inverse frequencies (1/x) for these values. The data sets were filtered again based on the frequency scores of the inverse pnome (removed if less than 75%). This frequency was averaged with the primary pnome frequency to get a single frequency value. This additional step captures segregation data for all SNPs/DIPs in both pnomes (high abundance in one and low abundance in the other), providing a more robust frequency estimate. The resulting average SNP/DIP frequencies were plotted by reference nucleotide position to generate the allele frequency map.

Structural variation analysis was performed using clc genomics workbench. A 5-Mb region (17–22 Mb) from both the pillar and standard pnome assemblies was extracted and used for the input. Default parameters were used and interchromosomal variation was excluded. The results were compared and filtered as described above.

High-resolution melt marker assay

Primers were designed from the pnome sequence to have an annealing temperature of 60°C. These are presented in Table S5. The HRM technique was performed in a single run on a LightCycler 480® (Roche Applied Science, http://www.roche-applied-science.com) in a reaction mix containing 2.5 ng of genomic DNA, 2 nm of each primer and 1 mm MgCl2 in the LightCycler 480 High Resolution Melting Master, with PCR-grade water adjusted to a total volume of 10 μl. The reaction conditions included an activation step at 95°C for 10 min, followed by 50 cycles of 95°C for 15 sec, 60°C for 15 sec and 72°C for 15 sec. HRM was carried out over the range from 65 to 95°C, rising at 1°C per second, with 25 acquisitions per degree. Individuals were scored based on their melting curve profiles relative to parental homozygous and heterozygous controls.

Quantitative real-time PCR (qPCR)

Peach tissue samples for whole-tree anatomy studies were collected from 10-year-old trees of ‘True Gold’, a double haploid variety. Tree seedling and Arabidopsis dissections were performed by cutting each shoot into sections from the top down. Sections were immediately immersed in liquid nitrogen. RNA extraction and qPCR was performed as previously described by Dardick et al. (2010). Briefly, each reaction was run in triplicate using 50 ng of RNA in a 15-μl reaction volume, using the Superscript III Platinum SYBR Green qRT-PCR Kit (Invitrogen, http://www.invitrogen.com). The reactions were performed on a 7900 DNA sequence detector (Applied Biosystems, http://www.appliedbiosystems.com). Quantification was performed using a relative curve derived from a serially diluted standard RNA run in parallel. A primer set designed to amplify 26S ribosomal RNA was run on all samples and used to normalize the data. A dissociation curve was run to verify that a single desired amplified product was obtained from each reaction. Primers used for peach and Arabidopsis TAC1 and LAZY1 genes are presented in Table S5.

Branch and flower bud growth angle measurements

Arabidopsis Attac1 and Col-0 controls were grown in 10 cm pots (one plant per pot) under fluorescent light (130 μmol m−2 sec−1) in an environmental growth chamber set at 21°C and 50% humidity. Approximately 2–3-week-old inflorescence bolts were removed and pressed. Pressed inflorescence bolts were photographed and axillary branch angles were measured using imagej (Rasband, 2008; Abramoff et al., 2004). A total of 87 angles were measured from 24 plants each, and subjected to statistical analyses. This experiment was repeated three times with similar results. The circumference of inflorescence bolt clusters was measured after three rounds of bolt removal using a tape measure. For flower bud growth angles, five branches were collected prior to anthesis from three trees each of pillar, upright and standard field-grown trees sampled from an F2 segregating population. Branches were imaged and growth angle measurements were taken as described above.

Construction of plasmids and transgenic plants

The sequences for pAtTAC1::AtTAC1 and 35S::AtTAC1 were commercially synthesized with flanking EcoRI and BamHI restriction sites, and cloned into the commercial vectors pJ201 and pJ344, respectively (DNA 2.0, http://www.dna20.com). Sequences are shown in Figure S3. Resulting clones were digested with EcoRI and BamHI, and ligated into a pBIN-based vector that had been previously modified to enable cloning behind a 35S promoter. The resulting 35S::AtTAC1 and pAtTAC1::AtTAC1 constructs, along with the empty vector, were transformed into Agrobacterium tumefaciens GV3101 by electroporation. Arabidopsis thaliana transformation was performed using the floral-dip method as described by Mara et al. (2010). Transformed seedlings were selected by growth on MS plates with 50 μg ml−1 kanamycin, and then transferred to 4-inch pots with Metromix 360® soil (Sun-Gro Horticulture, Inc., http://www.sungro.com). Plants were grown under short days (12-h light), at 22°C, with approximately 100 μmol light. A minimum of n = 47 was used for all branch angle measurements.


We would like to thank our technical support staff Elizabeth Lutton, Linda Dunn and Mark Demuth (USDA-ARS, Appalachian Fruit Research Station, Kearneysville, WV, USA) for their assistance. Also we give special thanks to Tom Tworkoski (USDA-ARS, Appalachian Fruit Research Station, Kearneysville, WV, USA) for his input and Mike Wang at the David H. Murdock Research Institute (Kannapolis, NC, USA) for his guidance with Illumina Sequencing applications. Pollen from ‘Italian Pillar’ was kindly provided by A. Liverani, Istituto Sperimentale per la Frutticoltura, Forli. The authors have no conflicts of interest. This project is supported by the Agriculture and Food Research Initiative Competitive Grants Program, grant no. WVAW-2011-04220, from the USDA National Institute of Food and Agriculture.