Introduction

Lolium perenne L. (perennial ryegrass) is the principle forage grass species used in temperate agriculture. High digestibility and good nutritional characteristics make it particularly suitable as a forage grass (Hannaway et al. 1999). Some important targets of perennial ryegrass breeding programmes include disease resistance, nutrient use efficiency, extending the growing season, persistency and yield. It is estimated that breeding in the past 50 years has yielded 4–5% gain in dry matter yield per decade (Wilkins and Humphreys 2003). Despite this, improving yield still remains one of the most important aims of current forage breeding programmes.

Classical breeding programmes can be assisted with the use of marker assisted breeding technology (Stuber et al. 1999). However, prior to this, the knowledge of the underlying genetic control of the target trait is required. One approach to dissect the genetic basis of complex traits is quantitative trait loci (QTL) mapping. Using molecular markers tightly linked to QTL associated with forage yield can be applied in molecular breeding programmes to assist traditional breeding. Marker-assisted selection (MAS) is one approach and relies on the establishment of a linkage relationship between a marker and a character of agronomic importance, although the success of MAS with a complex trait such as biomass presents a challenge.

QTL studies investigating biomass yield have been reported in a number of species including Arabidopsis (Lisec et al. 2008), sorghum (Lin et al. 1995), or rice (Yu et al. 1997). Some studies in ryegrass have reported QTL for biomass yield related traits of leaf width, fresh weight, and dry matter (Armstead et al. 2008; Turner et al. 2008; Yamada et al. 2004). However, to date no larger study has reported QTL with reasonable sized effects on biomass yield in perennial ryegrass. In order to perform a QTL study investigating biomass, a population with excellent segregation for biomass is required. The present study makes use of an F2 population based on two inbred grandparents where the F1 has a large biomass yield phenotype greater than both parents and therefore by definition, displays heterosis. The resulting heterotic effect subsequently segregates widely in the F2 generation. A genetic linkage map has been generated for this F2 population (Anhalt et al. 2008). The creation of F2 populations has been limited in perennial ryegrass because of the inherent difficulty in generating highly homozygous inbred lines. This was overcome in this case by an interspecific approach. This difficulty in generating homozygous inbred lines because of self-incompatibility and in-breeding depression represent the major barriers to exploiting heterosis in Lolium perenne, a technique that has yielded such enormous benefits in some of the major food crops.

The objectives of the present study were to measure biomass in an F2 mapping population, segregating widely for yield, under both field and glasshouse conditions and attempt to identify major QTL for biomass yield.

Materials and methods

Plant materials

The ‘F2 biomass’ population consisting of 360 F2 genotypes has been described elsewhere (Anhalt et al. 2008). The two parental inbred ryegrass lines, which were chosen as parents of the single F1 generation, originated each from an inter-specific cross between Festuca pratensis (meadow fescue) and L. perenne (Connolly and Wright-Turner 1984). In the pedigree of the maternal inbred parent the ryegrass cultivar ‘S24’ was included and for the paternal inbred line parent the ryegrass cultivar ‘Premo’. The initial interspecific hybrid was backcrossed for several generations to the ryegrass parent and selfed for 9 or 10 generations. The maternal parent of the F1 population was hand emasculated and was pollinated with pollen from the paternal parent. Pollinated florets were bagged in cellophane bags, individual F1 seed was harvested, and single F1 plants were raised and self-pollinated by bagging in cellophane pollination bags to generate independent F2 populations. One of these independent F2 populations was used for subsequent experiments and genetic map construction.

Experimental design

An alpha lattice design was used for both the field and greenhouse experiments. For the generation of experimental layouts, Agrobase Generation 2TM was used (Agronomix Software, Inc., Winnipeg, Manitoba, R3N 0S4, Canada). Replicated greenhouse and field experiments were designed. Each replicate comprised 45 incomplete blocks, with each block consisting of nine different genotypes and in each of the incomplete blocks either the maternal, paternal or the F1 genotype was included. The greenhouse experiment with plants in single pots was carried out from autumn 2005 to spring 2006 in three replicates under two greenhouse environments. Two out of the three replicates were placed in a greenhouse environment with the following climatic conditions: average night and day temperatures of 11 and 19°C, respectively. A third replicate was placed in a second greenhouse environment with an average night temperature of 6°C and day temperature of 19°C. The plants for the field experiment were planted in Oak Park, Carlow (Ireland) in spring 2006 with two replicates and the experiment was carried out over the two growing seasons from autumn 2006 to autumn 2007. In the field experiment, each genotype was planted in ‘mini-swards’ consisting of six clonal plants.

Collection of phenotypic data

From the greenhouse experiment three harvests were carried out in December 2005, February 2006 and April 2006. For the field experiment, four harvests were taken in August 2006, October 2006, May 2007, and August 2007. After each cut the plants were supplied with 80 kg N/ha. To reduce experimental errors during harvesting, a measuring unit was used for the cuts to achieve uniform cutting heights. The cutting height was 8 cm for the greenhouse experiment and 5 cm for the field experiment. Since the plants in the field were smaller when the first harvest was done, the cutting height in the field was lower than in the greenhouse. Plant samples were dried to constant weight in an oven at 60°C. Fresh weight (g) and dry weight (g) were measured. Dry matter (%) was calculated as (dry weight/fresh weight) × 100. The trait leaf width was measured in the greenhouse in two replicates in February 2006. Four measurements were taken on four fully expanded young leaves of each genotype. The measurements were taken in the middle of the fully expanded leaf blades using callipers. The average of these four values of each genotype was taken for further calculations.

Climatic data were collected at the Institute monitoring station in Teagasc, Oak Park, Carlow.

Data analysis

All data analyses were carried out in SAS V9.1 (SAS Institute Inc. 2004. SAS OnlineDoc® 9.1.3. Cary, NC: SAS Institute Inc.).

Data of plants, which were alive but with a too low weight to be weighed out or which were under the cutting height were analysed as zeros values.

Log transformations for dry matter in the greenhouse and field and square root transformation for fresh and dry weight in the field were computed for the non-normally distributed data. All other traits remained untransformed. Analysis of variance (ANOVA) was carried out using PROC MIXED for each of the traits and both experiments.

To calculate means for parents, F1 and F2 at each harvest, the treatment effect was modelled by a fixed effect with four levels (F1, F2 and the two parents) and a random effect for entries within F2. The random effect was fitted only for F2 using the method described in Piepho et al. (2006). When transformed data were analysed, standard errors for backtransformed means, which estimate medians on the original scale, were computed by the delta method.

Broad sense heritability for each trait was calculated on a plot basis for the greenhouse and field environment using the formula: \( H^{2} = \sigma_{g}^{2} /\sigma_{p}^{2} = \sigma_{g}^{2} /\left( {\sigma_{g}^{2} + \sigma_{e}^{2} } \right), \) where \( \sigma_{g}^{2} \) represents the genetic variance, \( \sigma_{p}^{2} \) the phenotypic variance and \( \sigma_{e}^{2} \) the environmental/error variance.

Pearson correlations were calculated for the measured traits fresh weight, dry weight, dry matter and leaf width.

QTL analysis

The sum of the yields of all harvests per plot were used for the QTL analysis in a mixed model using the model statement R : R · IBLOCK + R · IBLOCK · PLOT (Piepho et al. 2003) with the factors: plots (PLOT) within incomplete blocks and incomplete blocks (BLOCK) nested within complete replicates (R).

The F2 population described previously (Anhalt et al. 2008) was used for QTL detection. Briefly, the map consists of 75 markers (10 AFLP and 65 SSR marker) across seven linkage groups with a total map length of 592 cM and an average map density of 8 cm. Sixty-three percent of the markers in the F2 population did not fit the expected Mendelian ratios and showed significant (P < 0.05) segregation distortion. MapQTL® Version 4.0 (Van Ooijen et al. 2002) was used for QTL detection. Interval mapping (IM) followed by multiple QTL models (MQM) mapping was used to estimate the map position, LOD score, and genotypic effect of potential QTL, expressed as percentage of phenotypic variance explained. Markers with the closest positions to LOD peak of QTL identified by IM were selected as co-factors for MQM mapping. Genome wide LOD significance thresholds were determined using permutation tests with 1,000 iterations at a 5% significance level. QTL charts were displayed with MapChart Version 2.2 (Voorrips 2002).

Estimated means of the distribution of the quantitative trait associated with each genotype were used to rate the genetic effect of each QTL at the position of the highest LOD score. Only additive and dominant effects and no epistatic genetic effect can be found in an F2 breeding programme since there is only one single F1 genotype existing (Liu 1998). Estimated mean of distribution of the quantitative trait are displayed using the output data of the QTL mapping in MapQTL® Version 4.0 (Van Ooijen et al. 2002).

Results

Phenotypic data

Generally, across the greenhouse and field experiments yields were highest in the F1 generation. Across all harvests, biomass yields were much higher in the field than in the greenhouse experiment because of the mini-swards consisting of six clonal plants compared to the greenhouse experiment with single plants. Also growth was restricted in pots and the length of time between the harvests was different. Broad sense heritabilities were calculated for the three biomass traits in the greenhouse and the field experiments (Table 1).

Table 1 Broad sense heritabilities (H2) on a plot basis for the traits fresh weight, dry weight, dry matter, and leaf width in the greenhouse and the field experiment, respectively
Full size table

Biomass fresh and dry weight values (Table 2) for parental, F1 and F2 genotypes had a wide range. Maternal and paternal lines had lower values compared to the F1 and F2 line (Table 2) which could point to heterosis in the F1 and in F2 genotypes and inbreeding depression in the parental lines (Table 2). Dry matter did not show the same effect for F1 and F2 genotypes compared to the inbred lines. Parental lines had a higher range in dry matter compared to the F1 and F2 lines. Biomass yield values had a wider range in the field than the greenhouse. These high differences in values of the F1 and F2 genotypes compared to the inbred lines in the field were obtained since genotypes grew better in the field compared to their parental inbred genotypes, which could have been due to environmental influences like soil structure or water availability. The parental inbred lines performed poorly under field conditions: they either died or were shorter than the harvested height. After establishment in the greenhouse in 2005, the plants seemed to develop better in the late winter period than in the spring period, which could be influenced by higher irradiance in springtime. In 2006, more pronounced differences were found between individual harvests in the field, whereas in 2007 the genotypes had established under field conditions with the environmental conditions being more suitable for L. perenne. The rainfall distribution was notably different from 2006 to 2007, with only a third of the rainfall in June, July and August 2006 compared to 2007 (average 38 mm vs. 115 mm) and the reverse pattern during September, October and November 2006–2007 (127 mm vs. 41 mm).

Table 2 Median estimates of biomass yield and dry matter (±standard error, computed from log-transformed scale by delta method) for the three greenhouse and three field harvests averaged across the replicates: Average (Av) of fresh weight (g), dry weight (g), dry matter (%) of parental (PFa paternal, PM maternal), F1 and F2 lines
Full size table

Pearson correlation showed that all traits were significantly correlated (P < 0.0001). Fresh weight and dry weight were strong positive associated (r = 0.979), and leaf width had a moderate positive correlation with fresh weight (r = 0.385) and dry weight (r = 0.352) in the greenhouse experiment. Negative and moderate correlations were found between dry matter and fresh weight (r = −0.370), dry weight (r = −0.204), and leaf width (r = −0.246) in the greenhouse experiment. In the field experiment fresh weight and dry weight had a strong positive correlation (r = 0.996); whereas dry matter had a moderate negative correlation with fresh weight (r = −0.474) and dry weight (r = −0.411).

Identification of biomass QTL

A permutation test identified a LOD threshold between 3.2 and 3.5 for individual LGs, at which a QTL could be declared with 95% confidence. Using MQM mapping identical positions for fresh weight and dry weight QTL were consistently detected on LG 2, 3 and 7 (Table 3; Fig. 1) with one additional QTL for fresh and dry weight in the field on LG 5, which could not be found in the greenhouse experiment. The explained variance for the trait fresh weight at the QTL positions was 28% in the greenhouse and 50% for the field experiment and dry weight of 29% in the greenhouse and 53% for the field experiment.

Table 3 QTL for the greenhouse and field experiment for the traits fresh weight, dry weight, dry matter and leaf width as determined by multiple QTL mapping (MQM): Positions on linkage groups in cM, maximum LOD scores, percent of the phenotypic variation explained by the quantitative trait loci (QTL), position of QTL peak in cM on genetic map, marker selected as co-factor for MQM mapping, marker interval of QTL position
Full size table
Fig. 1
figure 1

Linkage groups (LGs) 2, 3, 4, 5 and 7 of the ‘F2 biomass’ population genetic map (Anhalt et al. 2008). Distances are given in cM. Asterisks indicate segregation distortion (* P < 0.05, ** P < 0.01, *** P < 0.001, **** P < 0.0001). LOD scans for the traits fresh weight, dry weight, dry matter, and leaf width in the greenhouse ((a) and filled shapes and the field (b) and open shapes) are shown at each QTL position. QTL were calculated by MQM mapping in MapQTL V. 4.0 and are displayed with MapChart V. 2.2

Full size image

Dry matter QTL were detected on LG 2, 3 and 7 in the field experiment (Table 3; Fig. 1). These were the only dry matter QTL detected in this study. Dry matter QTL appear to be in the same positions to other biomass QTL. Dry matter had an overall explained variance of 21% for the field experiment. Leaf width was only calculated for the greenhouse experiment and QTL were identified on LG 3 and 4 (Table 3; Fig. 1). The total variance explained by these QTL was 15%.

The QTL with the largest effect for biomass yield were identified on LG 3 and were detected using both glasshouse and field experiment data. The QTL for dry weight accounted for 15% and 33% of the phenotypic variance in the glasshouse and field experiments, respectively. When looking at the allelic contribution of parental lines at QTL positions, it was noticeable that the maternal allele (‘a’) was dominant (Table 3). This appears to be true for all major QTL identified on LG 3. However the QTL for biomass yield on LG 2 showed additive gene action, where the yield of the heterozygote (‘ab’) was greater than either of the homozygotes.

Discussion

There have been previous QTL studies in L. perenne that have looked at biomass traits (Armstead et al. 2008; Turner et al. 2008; Yamada et al. 2004). However, this manuscript represents the first large scale biomass study in perennial ryegrass reporting QTL for both fresh and dry weight. Our study took advantage of an F2 population derived from selfing an F1 plant displaying considerable heterosis. This population exhibited significant segregation for biomass yield, making it the ideal population in which to study the underlying genetic components of the trait.

All biomass yield related values in the present study were highest for fresh and dry weight in the F1 generation compared to the parental genotypes which is in accordance to the heterosis theory (Becker 1993). Decreasing heterosis values were observed over the harvest times for the field experiment, which can be explained by the much better growth over time and faster development of the F1 and F2 genotypes compared to the parental genotypes. Dry weather conditions in 2007 did not affect the performance of the F1 and F2 genotypes as much as the parental genotypes in the field, supporting the hypothesis that hybrids show increased yield stability. Similar findings were reported before in wheat hybrids (Bruns and Peterson 1998) where enhanced stability and improved agronomic traits were found in hybrids compared to pure line plants. It is likely that much of the heterosis in the F1 generation is due to the increased heterozygosity and the loss of homozygous recessive alleles that may have accumulated during inbreeding of the parental genotypes. The accumulation of homozygous recessive alleles during inbreeding is the principle reason for observed ‘inbreeding depression’.

Negative heterosis was observed for % dry matter and could be explained by increased cell sizes of the F1 and F2 plants compared to the parental lines. The cells may have been bigger and contained more water in F1 and F2 genotypes, which was removed by drying. This is supported by the results of the correlation analysis where dry matter content and leaf width were negatively correlated. Another explanation could be that in the F1 and F2 genotypes more water was stored than in the parental plants, which would explain why dry matter is increasing when leaf width is decreasing. Similar findings were reported for maize (Hoecker et al. 2006) in primary roots where the cortical cells were expanding in the hybrid plants. This would be also in agreement with another study in maize were a more rapid growth rate of leaf blades in hybrids was explained by increased cell number or enlargement of the cells (Uchimiya and Takahashi 1973). Dry matter is an important trait in plant development and ecology because it is directly connected with the growth rate and ecological behaviour of the plant.

QTL analysis was performed using phenotypic data from both a field and a glasshouse experiment. Three major QTL regions were identified on LG 2, 3 and 7 (Table 3; Fig. 1). This is the first report of QTL for fresh and dry weight on these three linkage groups in perennial ryegrass. These regions contained co-localizing QTL for both fresh weight and dry weight. In addition these QTL were identified for both the field and glasshouse experiments. Co-localizing QTL for fresh weight and dry weight are not surprising considering the significant correlation between both traits. Additional QTL for fresh weight and dry weight were detected on LG 5 in the field experiment. Overall more variation was explained in the field experiment, which could be because of more suitable environmental conditions for the genotypes. The QTL for both fresh and dry weight on LG 3 explained the largest proportion of the phenotypic variance.

When examining the allelic contribution of parental lines to the major QTL on LG 2 and LG 3 differences were observed. The QTL on LG 3 appeared to be controlled by dominant gene action with the maternal parent contributing the allele for increased biomass. The QTL on LG 3 was controlled by additive gene action and the heterozygote had higher fresh and dry weights than either homozygote. Biomass yield is a highly complex trait and QTL for fresh and dry weight represent yield related QTL but they may also represent basic physiological processes that have an impact on yield. An example of this would be maturity or heading date. In this study heading date could not be measured in parallel to biomass yield due to the cutting regime. A lack of common markers between genetic linkage maps makes it difficult to superimpose previous heading date QTL studies onto the present linkage map.

There have only been three previous QTL studies reporting on traits analysed in this study. Armstead et al. (2008) identified QTL for leaf width on LG 7 using the F2 ‘WSC’ population. Turner et al. (2008) reported on fresh weight QTL on LG 1, 4, and 5 and autumn dry matter QTL on LG 3. Yamada et al. (2004) identified fresh weight QTL on LG 5 and leaf width QTL on LG 3 in the ‘ILGI’ population. However, a lack of common markers between various linkage maps makes it difficult to predict whether or not they represent the same QTL region. A large-scale QTL study on herbage quality traits has previously been performed in the perennial ryegrass p150/112 population (Cogan et al. 2005). The traits measured were crude protein, neutral detergent fibre, water soluble carbohydrate, in vitro dry matter digestibility and estimated metabolisable energy. Interestingly, a large cluster of QTL for all traits measured was detected on LG 3. The use of common markers to link the maps together would be useful to determine the relationship between the relatively large effect QTL for biomass yield detected on LG 3 in this study and the QTL for herbage quality parameters identified by Cogan et al. (2005).

There is controversy surrounding the use of spaced plants for studies of biomass. Forage yield measured in spaced plants is considered as a poor predictor of yield (Hayward and Vivero 1984). However, there are practical limitations to performing QTL studies in swards. In this study we created mini swards in the field comprised of six clonal replicates. We also performed a glasshouse study using single plants. It is worth noting that similar results were obtained from experiments using artificial mini swards in the field and single plants in the greenhouse.

The presented study is based on a reliable dataset for biomass yield heterosis and has identified consistent QTL regions for biomass yield over environments and replications. In summary, with glasshouse data we accounted for fresh weight 28%, for dry weight 29%, and for leaf width 15% of the phenotypic variation, respectively. With the field data we accounted for 50% of the phenotypic variance for fresh weight, 53% for dry weight, and 21% for dry matter. The development of cultivars with increased biomass yield is still a target trait of breeding programmes. Some of the identified QTL are therefore very promising for the identification of candidate genes for heterotic biomass yield after future detailed fine mapping efforts.