Mutations in Barley Row Type Genes Have Pleiotropic Effects on Shoot Branching

Cereal crop yield is determined by different yield components such as seed weight, seed number per spike and the tiller number and spikes. Negative correlations between these traits are often attributed to resource limitation. However, recent evidence suggests that the same genes or regulatory modules can regulate both inflorescence branching and tillering. It is therefore important to explore the role of genetic correlations between different yield components in small grain cereals. In this work, we studied pleiotropic effects of row type genes on seed size, seed number per spike, thousand grain weight, and tillering in barley to better understand the genetic correlations between individual yield components. Allelic mutants of nine different row type loci (36 mutants), in the original spring barley varieties Barke, Bonus and Foma and introgressed in the spring barley cultivar Bowman, were phenotyped under greenhouse and outdoor conditions. We identified two main mutant groups characterized by their relationships between seed and tillering parameters. The first group comprises all mutants with an increased number of seeds and significant change in tiller number at early development (group 1a) or reduced tillering only at full maturity (group 1b). Mutants in the second group are characterized by a reduction in seeds per spike and tiller number, thus exhibiting positive correlations between seed and tiller number. Reduced tillering at full maturity (group 1b) is likely due to resource limitations. In contrast, altered tillering at early development (groups 1a and 2) suggests that the same genes or regulatory modules affect inflorescence and shoot branching. Understanding the genetic bases of the trade-offs between these traits is important for the genetic manipulation of individual yield components.


Introduction
Barley (Hordeum vulgare) is one of the most important cereal crops used for food and feed production [1]. In cereals, grain yield can be broken down to three main components: (1) spike number per plant, (2) grain number per spike, and (3) Thousand Grain Weight (TGW) [2,3]. The barley inflorescence (spike) can appear in two main shapes: two-rowed and six-rowed ( Fig  1A). In two-rowed genotypes only the central spikelet develops and the two laterals remain sterile or staminate. In six-rowed cultivars, all three spikelets are fully fertile and develop into grains, leading to an increase in grain number per spike [4]. Over the last decades different loci that affect row type have been identified in barley. The loci have been termed six-rowed spike (vrs), hexastichon (hex-v) or intermedium spike (int), depending on the mutant screen and the phenotype [5,6]. Intermedium spike mutants typically have a phenotype in between a tworowed and a fully developed six-rowed spike [5][6][7]. The spike phenotype of six-rowed barley cultivars mainly relies on mutations in the Vrs1 (allelic with Hex-v and Int-d) and Vrs4 loci. The gene underlying the Vrs1 locus encodes a HD-ZIP transcription factor (HvHox1) which is expressed in the lateral spikelet primordia in the developing spike. Vrs1 suppresses lateral spikelet fertility resulting in the two-rowed phenotype [8]. Similar to vrs1 mutants, the vrs4 mutant shows complete fertility of central and lateral spikelets and the spikes produce additional florets. Vrs4 encodes a homolog of maize RAMOSA 2 (RA2) a gene involved in the control of branching in the female inflorescence (ear) [9,10]. An increase in seed number often results in a decrease in TGW [3,11]. In addition, the six-rowed spike phenotype has been described to negatively affect tiller number under field conditions [3,12]. Negative correlations between seed number, TGW and tillering have hampered the genetic improvement of overall yield. The opposing effects between these yield components are often attributed to limitations in amounts of assimilates available [3]. Tiller number is mainly affected by the initiation and outgrowth of axillary meristems (AM) and depends on endogenous developmental, hormonal and environmental signals [13][14][15][16]. The plant shoot apical meristem (SAM) coordinates the formation of all above ground organs such as leaves, internodes, inflorescence and AMs ( Fig  1). During vegetative development, the AM initiates several leaf primordia, which form an axillary bud. In grasses, the axillary meristems capable of giving rise to side shoots (tillers) are located only at the most basal nodes of the plant (crown) (Fig 1B) [17,18]. After transition from vegetative to reproductive development, the grass apex irreversibly differentiates into the inflorescence. Domestication of different crop species has resulted in cultivars with altered shoot architecture. For example, TEOSINTE BRANCHED 1 (TB1) gene, a TCP transcription factor and homolog of the Arabidopsis thaliana gene BRANCHED 1 (BRC1), has been identified as a major domestication related gene in maize [19]. In cultivated maize, a natural overexpressing allele was introgressed into the breeding germplasm to reduce tillering as an adaptation to agricultural cultivation [19]. A homolog of TB1, Intermedium-c (Int-c) was also targeted in barley during domestication to increased lateral spikelet development. In addition, the int-c mutant has an increased tiller number at early developmental stages [20]. Similarly, the maize gene BARREN STALK 1 (BA1), affects tiller and tassel development [21]. This exemplifies that genes involved in inflorescence branching can act on shoot branching. However, the pleiotropic effects of row type genes on shoot architecture are not straightforward: int-c and int-m are described as mutants with an increased tiller number while int-b and intl.81 (allelic to low number of tillers 1, lnt1) have a reduced tiller number [5,7,22]. It is therefore important to perform a comparative study on the pleiotropic effects of different row type genes on yield components.
To this end we recorded tiller number, seed number per spike, seed area and TGW of independent allelic mutants of vrs3/int-a, int-b, int-c, vrs1/int-d/hex-v,vrs4/int-e, int-f, int-m, lnt1/ int-l and absent lower laterals (als). The latter is included since als mutants show irregularities in floret phyllotaxy and the gene is involved in secondary AM formation [23]. Given that tiller number is a dynamic trait, affected by the environment and development, we monitored tiller number at different developmental stages under greenhouse and outdoor conditions. Our data shows that row type genes affect the correlation between shoot and spike architecture either directly most likely via regulating the initiation and/or outgrowth of tillers or indirectly due to resource limitation. We conclude that the classification based on the macroscopic phenotypes presented here helps to understand and interpret the pleiotropic effects of row type genes on major yield components.

Plant material
Seeds of different row type mutants representing the loci (vrs1/hex-v/int-d, vrs3/int-a, vrs4/inte, int-b, int-c, int-f and int-m) and their respective parental cultivars were obtained either from the Department of Agriculture of the United States of America (USDA) or from the Nordic Gene Bank (NordGen) (S1 Table). For clarity, different loci of the same gene will be designated with one name referring to the original locus with the allele name in brackets. In this study, we used the original mutants in the cultivars Bonus and Foma and Bowman backcross lines [24]. In addition, we analyzed different allelic mutations of the Vrs1 gene originating from an EMS TILLING population in cv. Barke [25]; and two backcrosses (BC 3 ) of six-rowed Vrs1 mutants cv. Barke referred to as vrs1(8408) and vrs1(11910). Seed material of the TILLING mutants was kindly provided by S. Gottwald (University of Gießen) and N. Stein (IPK Gatersleben). Leaf material was collected from four individuals per genotype and freeze-dried. The freeze-dried material was ground and genomic DNA was extracted with the Qiagen BioSprint Robot (Progam: DNAPlant) and the respective Kit, according to the manufacturer's recommendations (Qiagen, Hilden, Germany). As a control we verified the mutation in vrs1. Sequencing confirmed position and nature of the published mutations with exceptions of vrs1(hex-v.3) and vrs1(hex-v.6) (S2 Table).

Outdoor experiment
Lines harboring different allelic variants of the vrs and intermedium spike loci (vrs3/int-a, int-b, int-c, vrs1/hex-v/int-d, vrs4/int-e, int-f and int-m) were grown outdoors and compared to their respective parental backgrounds. All phenotypes recorded were, when possible, confirmed in multiple independent allelic variants of the same locus in addition to backcrossed lines in the spring barley cultivar Bowman [24] (S1 Table). For int-f, int-m and als, only one allelic variant was available. The plants were sown in 96-well trays mid-February 2014, germinated in the greenhouse and then transferred outside (MPIPZ Cologne, Germany). After six weeks (end of March), plants were transferred to 12 L pots with one plant per pot, each filled with a custommade peat and clay soil mixture (EinheitsErde 1 ED73 Osmocote, Einheitserdewerke Werkverband e.V., Sinntal-Altengronau, Germany) containing a long term fertilizer. The pots were arranged in 22 rows with a distance of 1 m between rows where each row contained 54 pots with a distance of 10 cm. The entire experiment was divided into 4 blocks with at least 4 replicate plants per genotype within each block. To avoid edge effects, the plot was surrounded by border pots containing wild type barley plants.The plot was irrigated by a sprinkling robot and treated with additional fertilizer or pesticides when necessary. Tiller number was assessed at four different stages: (1) two weeks after transplantation (50 DAS), (2) flag leaf stage, (3) anthesis stage, and (4) at full maturity. Barke vrs1 EMS TILLING lines were additionally also phenotyped at 56, 63 and 70 DAS. Flag leaf stage was recorded, when the flag leaf completely emerged and unfolded from the main shoot. At anthesis stage, at least one spike of the plant showed open florets with the anthers visibly pushed out. For vrs4.k tiller number at this stage was scored for only 10 plants instead of 16. This is due to the absence of a clear anthesis occurring in some individuals, which were therefore left out of the analysis. Total tiller number at full maturity was counted for all plants simultaneously mid of July 2014, two weeks before harvest. At this time point the number of grain bearing spikes was also recorded. At 92 DAS, pictures were taken from one representative plant per genotype. Plant height was measured (soil to base of topmost spike) at full maturity. All raw data obtained from the outdoor experiment has been added as S1 File.

Seed analysis
The grain number per spike, seed size and shape and thousand grain weight (TGW) were measured using the MARVIN Seed Analyser (GTA Sensorik, Neubrandenburg, Germany).All seeds and spikes used in the analysis of the seed parameters were derived from plants grown outdoors. With exception of int-b.6, all measurements were performed on at least 3 spikes from 4 individual plants per genotype. For int-b.6, only 2 spikes of 4 individual plants were scored due to the low tiller number. The frequency distribution of seed size was calculated and plotted against the average number of seeds per spike. The TGW was measured for all mutant and wild type lines based on the total number of seeds of the 3 spikes from 4 separate plants per genotype. In addition, the TGW for vrs3, int-b.75, int-c, vrs1, vrs4, int-m and int-f and wild type genotypes, but not for the low tillering int-b.3, int-b.6, als and lnt mutants, was measured from three different plants per genotype using 4x5 mL of seeds (240-480 seeds in total). No significant differences were observed when comparing the TGW extrapolated from the seed weight per spike with the TGW obtained from measuring 240-280 seeds per plant (S3 Table). All raw seed parameters measured are included as S2 File.

Greenhouse experiments
Variation in tiller number of mutant and wild type genotypes was also scored in a controlled greenhouse under long day (LD) conditions (16h, 22°C day; 8h, 18°C night). Tiller number at full maturity was recorded for different allelic row type mutants (vrs3/int-a, int-b, int-c, vrs1/ hex-v/int-d, vrs4/int-e, int-f and int-m). Due to spatial limitations in the greenhouse, the mutants and respective wild type genotypes were grown in separate experiments in 1.3 or 4 L pots. The mutant lines int-c.25 and -c.29 in Foma, int-c.5 in Bonus, int-f. 19 in Foma and all other row type mutant lines backcrossed to Bowman were scored for tiller number twice per week starting from 7 days after emergence (DAE) until the age of four weeks and once per week thereafter until the age of 13 weeks (n = 10 plants, grown in 1.3 L pots). The flag leaf stage was recorded as in the outdoor experiment. Pictures were taken from three representative plants per genotype at 32 DAE.
For staging the development of the shoot apical meristem (SAM) of all mutants in Bowman background and int-f in Foma background, seeds were sown in 96-cell "Mini Tray" (Einheitserde 1 ) and kept at 4°C for 3 days after which they were germinated under LD conditions (16h, 22°C day; 8h, 18°C night). The apex development of 3-5 plants was recorded at 12, 18 and 22 DAE in three independent biological replicates. The SAM was visualized using a stereo microscope (MZ FLIII, Leica) and the developmental stage was determined using the quantitative scale introduced by Waddington et al. [26]. als was not included in this analysis due to seed limitations.

Data analysis
Statistical analyses were performed using the statistical software R (http://www.r-project.org/) or SAS version 9.1 (SAS Institute 2003). Differences between wild type and mutant genotypes were determined using a student's t-test or a one-way ANOVA combined with a Tukey HSD for multiple comparison for the outdoor and greenhouse experiments separately. The mutant genotypes were classified into groups based on the relative performance of each mutant line according to the following equation: in which "mut" denotes the mutant line and "wt" wild type. Calculations were performed on the trait averages for each mutant line. The relative performance values were used to cluster the different mutants using a hierarchical cluster analysis (HCL) [27]. The HCL analysis was done using the MeV multiple experiment viewer (http://www.tm4.org/mev.html), with default settings that apply a Pearson correlation as distance measure and average linking for clustering. Significant effects of the mutant locus and environment (greenhouse or outdoors) on variation in tillering were calculated using the following generalized linear model (GLM) programmed in SAS: Were gene is the fixed effect of the mutant locus; gene(allele) the fixed effect of the allele nested within the mutant locus; environment the fixed effect of the environment (greenhouse vs. outdoor), and the interaction effects between the gene and environment and between the environment and the mutant allele nested within the mutant locus. We also tested for significant block effects in the outdoor experiment in an initial GLM model. However, since the block showed no significant effects on tillering, this factor was removed from the final model. In order to test if the mutant locus exhibited consistent effects on tillering across the mutant alleles and environments, least square means were calculated within the GLM procedure for the effect of the mutant locus on tillering under greenhouse and outdoor conditions separately.

Results
Trade-off between seed number per spike and seed size To decipher the effects of row type genes on seed setting (fertility) and seed size, we analyzed the seed number per spike, TGW and seed shape parameters (width, length and area) for different row type mutants. To compare the phenotype of the different row type mutants directly, backcrossed lines in the spring barley cultivar Bowman were used [24] in addition to the original mutants [5] and allelic series of vrs1 mutants in cv. Barke identified by EMS TILLING [25]. When available, multiple independent allelic variants of the same locus were used to test if pleiotropic effects were locus specific (S1 Table). The mutant vrs3(int-a.31) in Foma background was removed from the analysis since this mutant exhibited pre-harvest sprouting and therefore no reliable measurements on the seed size and TGW could be performed. None of the other row type mutants showed pre-harvest sprouting.
The highest increase in seed number per spike is found for six-rowed mutants: vrs1(11910), vrs1(8408), vrs1.a, vrs1(hex-v.3), vrs1(hex-v.6), vrs4.k and for vrs3(int-a.1) in Bowman background. These mutations increase the seed number per spike on average 2.5-fold. In our analysis, the majority of the row type mutants exhibit an increase in the seed number per spike and a reduction in the TGW (Table 1). This indicates a trade-off between the seed number per spike and seed size. Seed size of the two-rowed parental cultivars Barke, Bonus, Bowman or Foma show a bell-shaped distribution. When compared to these, an increase in seed number per spike results in either an overall reduction of seed size or in less homogenous sized seeds on the same spike (Fig 2A-2C, S1, S2, S3 and S4 Figs). In the vrs1.a allele about 2/3 of the seeds exhibit a size between 15-22.5 mm while 1/3 of the seeds have a size between 22.5-30 mm, resulting in a bimodal distribution (Fig 2C). In contrast to the vrs1.a allele, int-m.85 (Bowman), lnt1 (Bowman) and als1.a (Montcalm) exhibit a significant reduction of the seed number per spike and a reduction in TGW (Table 1). In these mutants, the seed number per spike and TGW thus shows a positive correlation (S5 Fig). The int-b mutants exhibit a slight reduction in seed number per spike and a significant reduction in TGW. The reduction of seed number per spike is caused by different spike phenotypes. The lateral spikelets of the vrs1(int-d.11) mutant, which harbors a mutation within the C-terminus of Vrs1, do not set seed. Therefore, the spikes of the vrs1(int-d.11) mutant remain effectively two-rowed and do not show a significant reduction of seed number per spike or TGW. The int-m.85 have very short spikes, while lnt1, int-b and als mutants display an irregular seed set [22,23]. In addition, the seed size distribution of these lines is only marginally different from the respective two-rowed wild type backgrounds ( Fig  2D, S1, S2, S3 and S4 Figs). In conclusion, the majority of the row type mutants exhibit an increased seed number per spike and an overall reduction in seed weight ( Fig 2E,

Row type genes have pleiotropic effects on tillering
To assess the effect of row type genes on tillering, the tiller number of different row type mutants was analyzed at 50 days after sowing (DAS), at the flag leaf and anthesis stage, and before harvest of dry plants (full maturity) (Fig 3, S6, S7 and S8 Figs, S4 Table, S1 File). To reduce effects on the tiller number due to spatial limitation (e.g. pot size), competition or nutrient shortage, plants were grown individually in 12 L pots under semi-controlled outdoor conditions. In addition, plants were randomized within the plot to avoid differences due to spatial effects.
All row type mutants in Bowman background show a reduced tiller number at full maturity compared to Bowman, with the exception of int-m.85 which exhibit significantly more tillers than Bowman at maturity (Fig 3). In addition, multiple allelic variants of lnt1, int-b, and the single mutant allele of int-f (int-f. 19 in Foma) and als (als1.a in Montcalm and Bowman backcross), respectively, have a low tiller number at all developmental stages, whereas int-c and vrs3 show a reduction in tillering only after the flag leaf stage (Fig 3, S6, S7 and S8 Figs, S4 Table). The vrs1.a and vrs4.k mutants in Bowman background only show a decrease in tiller number at full maturity. One exception to this is vrs1(11910) in Barke background, which exhibits a lower  tiller number over the whole development. None of the other vrs1 mutants shows a reduction in tiller number before maturity, suggesting that a background mutation rather than the vrs1 mutation itself caused the tiller phenotype in vrs1(11910). Notable differences between allelic variants and backgrounds are observed (S6, S7 and S8 Figs, S4 Table). For example, vrs1(hexv.3) and vrs1(hex-v.6), two six-rowed allelic variants of vrs1 in Bonus background, do not show a significant reduction in tiller number (Fig 4A, S6 Fig, S4 Table). In contrast, vrs1(int-d.11) and vrs1(int-d. 22), two vrs1 mutants with intermedium spike phenotypes, exhibit a reduced tiller number at full maturity ( Fig 4A, S7 Fig, S4 Table). In addition, int-b.3 and int-b.6 in Bonus background display a more severe reduction in tiller number when compared to the int-b.75 allele in the same background ( Fig 4A). Interestingly, the reduction in tiller number in int-b.3  is less severe in the Bowman backcross (Fig 4A), suggesting an effect of background variation between Bonus and Bowman or the presence of additional mutations in the original line. As flowering time can affect shoot branching by influencing the length of vegetative development [28][29][30], we also compared flowering time and apex development between the different mutants and respective wild type genotypes. The row type mutants do not show major differences in flowering time or apex development when compared to their respective wild type backgrounds (S4 Table, S9 Fig). In addition, plant height was measured at full maturity. The low tillering mutants lnt1.a, als1.a and int-b are significantly shorter compared to their respective parents. No consistent height phenotype is observed for any of the other row type mutants (S4 Table). These results indicate that row type genes mainly affect seed parameters and tiller number, but neither flowering time nor plant height. Variation within these traits is not only dependent on the allelic variants but also on the genetic background of the parental lines. To compare the effects of mutant loci across different backgrounds, the relative phenotypic performance of each mutant was calculated for all traits compared to their respective parent. Hierarchical clustering [26] of the relative performance shows that the row type mutants cluster into two main groups (Fig 4B). Mutants in group 1, which contain all alleles of vrs3, int-c, int-f, vrs4, and vrs1 (with exception of vrs1(int-d.11)) show a negative correlation between tillering and seed number per spike. Group 1 can be subdivided into two subgroups: (a) low tillering already before maturity (vrs3,int-c, int-f); and (b) low tillering only at maturity (vrs1,vrs4). Group 2 encompasses all alleles of als, lnt1, int-b, and vrs1(int-d.11). These mutants display reduced tillering and seed number per spike. Within group 2 vrs1(int-d.11) and int-b.75 form a separate cluster, as they are characterized by a less severe tillering phenotype when compared to the other mutants in group 2. The int-m.85 mutant, which forms an outgroup, is the only row type mutant with an increased tiller number at full maturity and reduced number of seeds under outdoor conditions.
To test if the mutant loci exhibit a consistent effect on tillering across environments, variation in tillering at maturity was scored outdoors and in a controlled greenhouse. Under greenhouse conditions, plants have less tillers when compared to those grown in the outdoor experiment (S10 Fig). Indeed, the analysis of variance confirmed that the environment has a significant effect on tillering (S5 Table). The effect of each mutant locus on tillering is stronger than that of the environment for the majority of genotypes. With exception of vrs1 and int-m, all mutant loci cause a consistent reduction in tiller number in the greenhouse and outdoor experiment ( Table 2). The reduction and increase of tiller number in vrs1 and int-m, respectively, is only significant under outdoor conditions ( Table 2). The effects of the mutant loci on tillering are consistent over different mutant alleles (S5 Table, S10 Fig).
Taken together, the majority of row type mutants show consistent effects on tillering across different environments, allelic variants and genetic backgrounds. Mutant loci which affect tillering only at later developmental stages, such as vrs1 and int-m, have no significant effects on tillering when plants are grown in the greenhouse. In addition, the cluster analysis classifies the row type mutants based on their macroscopic phenotype into three groups; mutants in group 1 have a negative correlation between tillering and seed number per spike already after flag leaf stage (group 1a) or only at full maturity (group 1b), while mutants in group 2 exhibit a positive correlation between these traits

Tillering phenotype of int-c changes during development
Mutants of int-c (HvTB1) have been described as high tillering [20], however in our analysis no such increase in tiller number was observed. In fact, the tiller number of int-c mutants is significantly reduced compared to the wild type parents at flag leaf stage and later stages of development (Fig 3, S6, S7 and S8 Figs, S4 Table). This observation appears to be in contrast with a study which reported an increased tiller number in int-c mutants at an early developmental stage [20]. We therefore recorded tillering of int-c over the entire life span of the plant under greenhouse conditions. Three independent mutants of int-c were selected: int-c.25 and -c.29 in Foma background and int-c.5 in Bonus and Bowman backgrounds. In addition, int-f. 19 in Foma and Bowman background was selected as another low tillering row type mutant which showed a reduction in tiller number over the whole development (Fig 5, S11 Fig). Interestingly, int-c.5 mutants in Bonus background do not exhibit an increase in tiller number while the same mutation backcrossed into Bowman background shows a significant increase in tiller number when compared to the respective parental line until 56 DAE. However, int-c.5 in Bonus background has a significant reduction in tiller number after 63 DAE (Fig 5). The parental lines Bowman, Bonus and Foma form new tillers until full maturity while all int-c mutants tested cease formation of new tillers after the flag leaf stage under both outdoor and greenhouse conditions (Fig 5, S11 Fig). As a consequence, the wild type plants have significantly more tillers at maturity when compared to flag leaf stage, while int-c mutants do not (S6 Table). No differences were observed in tiller number at full maturity between Bowman and int-c.5 under greenhouse conditions while in the outdoor experiment int-c.5 in Bowman showed a significant reduction at full maturity. When grown in the greenhouse, Bowman and int-c.5 in Bowman background reached flag leaf stage after only five weeks (32 to 36 days after emergence) while in the outdoor experiment this stage was reached after 62 DAS. The fast development under greenhouse versus outdoor conditions and between Bowman and Bonus/Foma might cause the variation observed in relative performance of int-c mutants in different backgrounds/conditions.  To evaluate if any of the other row type mutants show a consistent increase in tiller number at early development but a reduction after flag leaf stage, we analyzed the tiller number over their lifespans. For comparison, we used only row type mutants backcrossed into the Bowman parent grown under greenhouse conditions. However, none of the other row type mutants shows a similar change in tillering behavior as was seen in int-c (S12 Fig, S7 Table). Taken together, our data shows that when compared to the wild type, the tiller number of int-c is increased at early developmental stages and reduced at later stages, as the duration of tiller formation is reduced by the int-c mutation.

Discussion
Shoot and spike architecture are two important adaptive traits in barley. Our results show that an increase in the seed number per spike as observed in the intermedium and six-rowed mutants causes an overall reduction in seed size. In addition, variation in seed size is increased in the row type mutants due to the size difference between central and lateral spikelets. In the vrs1 and int-c mutants the seed area shows a bimodal distribution with a 2:1 ratio. In line with this, grains of the lateral florets are generally smaller when compared to the central florets [11]. Vrs1, Int-c and other Vrs loci have been reported to influence seed weight and shape parameters in a doubled haploid population derived from a cross between two-rowed cv. Harrington and six-rowed cv. Morex. Vrs1 was found to be the major locus affecting TGW and all seed size and shape parameters measured [11]. Similarly, in our analysis vrs1 causes the largest increase in seed number per spike and the most severe reduction in TGW when compared to the other row type mutants. The increase in the seed number per spike was likely the reason for the introduction of the vrs1 allele into the barley gene pool during domestication [8]. In addition, six-rowed cultivars generally harbour an allele of Int-c (Int-c.a) which increases lateral seed size compared to the allele present in two-rowed cultivars (Int-c.b) [11,20]. Strong variation in seed size is generally considered unfavorable for malting barley which are therefore most commonly grown in Europe as two-rowed cultivars [11,31].
While the majority of the row type mutants shows a negative correlation between seed number and TGW, int-m.85, lnt1, als and several alleles of int-b display a reduction in seed number and in seed size. A positive correlation between seed number and seed size in these mutants suggests that the causative mutations negatively affect both seed size and seed number. These findings indicate that pleiotropic effects between yield parameters are not only caused by competition of different plant organs for limited resources, but that row type genes may control development of different meristematic tissues. We therefore analyzed the effects of row type genes not only on spike, but also on shoot branching. Indeed, we observed that row type mutants are affected in tillering. As observed for the correlations between seed number and size, we identified mutants with negative and positive correlations between seed number and tiller number. Based on these correlations, we classified the row type mutants into two main groups: (1) mutants with increased seed number per spike and low tillering; and (2) mutants with a reduction in seed and tiller number. Row type mutants in group 1 show an increase in inflorescence branching (seed number per spike) while the shoot branching (tiller number) is reduced after flag leaf stage (group 1a) or only at full maturity (group 1b). A similar effect of increased inflorescence branching and decreased tillering was reported for certain alleles of rice SQUAMOSA BINDING PROTEIN-LIKE 14 (OsSPL14), which were found to be positively associated with desirable traits underlying a quantitative trait locus (QTL) termed Ideal Plant Architecture (IPL) or Wealthy Farmer's Panicle (WFP). Alleles of OsSPL14 with higher expression levels cause a reduction in shoot branching, but also stronger culms and increased panicle branching leading to higher yield. Knockdown of OsSPL14 results in more tillers and a reduction in plant height, culm diameter, panicle branches and grain number [32,33]. Furthermore, the wheat tiller inhibition (tin) mutant displays larger spikes and a higher grain number per spikelet and a low tiller number. The spike phenotype can be phenocopied by removal of tillers shortly after initiation in wild type plants [34]. This exemplifies that different plant organs can compete for the same resources, resulting in a reduction of tiller number which correlates with an increase in seed number per spike. We observed that mutants of group 1b (vrs1, vrs4) with a high seed number per spike exhibit a reduced tiller number at maturity, but not at earlier stages of development. No tillering phenotype was observed in vrs1 mutants in the greenhouse experiment or until the plant reached full maturity in the outdoor experiment suggesting that vrs1 does not directly affect the initiation and outgrowth of axillary buds. This is supported by the finding that, unlike its paralog HvHox2, which is expressed abundantly in all tissues, Vrs1 is expressed almost exclusively in developing spikes [35]. Therefore, the observed reduction in tiller number is likely caused by resource limitations. Mutants of vrs4 show a similar trend, with no reduction in tiller number until the plant reached full maturity. In line with this, Vrs4 was reported to act upstream of Vrs1 in the same pathway [9], suggesting that this signaling cascade does not directly influence shoot branching. In contrast to vrs1 and vrs4, intc mutants show an increase in tiller number at early developmental stages (this work and [20]), suggesting a direct effect of this gene on tillering. A similar increase in tiller number has been observed in loss of function mutations in maize TB1 or rice FINE CULM 1 (FC1), two homologs of int-c [20,36,37]. TB1 and FC1 directly affect tiller number by suppressing the outgrowth of lateral buds. In the fc1 mutant it has been demonstrated that the increase in tiller number is due to reduced dormancy of the buds at the prophyll node, the most basal node of the tiller [36]. Int-c in barley may affect bud outgrowth in a similar way by preventing the outgrowth of axillary buds. Interestingly, our data shows that int-c mutants cease to grow out new tillers after reaching flag leaf stage which causes the observed reduction in tiller number at late developmental stage. This indicates that the phenotype of int-c mutants differs from its homologs tb1 or fc1 which exhibit an increase in tiller number throughout all developmental stages [36,37]. None of the other row type mutants exhibit an increase in tiller number at very early developmental stages. In fact, vrs3 mutants which are also clustered in group 1a have significant reduction in tiller numbers starting from flag leaf stage on.
The als, lnt1 and int-b mutants in group 2 show an even stronger reduction in tillering throughout their development. This is in line with previous observations that lnt and als successfully initiate primary AMs, but fail to generate secondary AMs [22,23]. Despite a similar phenotype, lnt1 and int-b likely act in different pathways as double mutants do not produce any tillers [22]. As als, lnt1, and int-b exhibit a strong decrease in seed number, seed size and tiller number, the causative mutations may directly initiate the formation or outgrowth of different meristematic tissues in barley. It has been shown before that tillering genes also modify seed number. For example, the rice gene MONOCULM 1 (MOC1), a homolog of the A. thaliana boundary gene LATERAL SUPPRESSOR (LAS), affects shoot as well as inflorescence branching [38].
In line with this, we propose that row type mutations in barley directly affect tiller initiation or outgrowth as suggested by the strong effects on tillering at early developmental stages. One exception to this are mutations in vrs1 and vrs4 which cause a reduction in tillering only at maturity likely due to resource limitations.

Conclusions
Shoot architecture is crucial for crop yield in cereals. We have shown that genes involved in development of the branched inflorescence architecture of the grasses also control seed size and shoot branching. Our results indicate that correlations between shoot and spike architecture are due to a) competition between different sink organs for limited assimilates or b) the direct involvement of row type genes in the initiation and growth control of different plant organs, seeds and tillers. We thus speculate that the same regulatory genes or modules may control the development of different meristematic structures and organs in plants. Understanding how these genes are regulated and in turn control downstream targets in different plant organs is important to improve yield by modifying shoot and spike architecture. The results presented here provide a starting point for such studies.  Table. (TIF) S1 File. Raw trait data obtained for the row type mutants under outdoor and greenhouse conditions. (XLSX) S2 File. Raw trait data seed parameters measured using the MARVIN analyser. (XLSX) S1