Introduction

Sorghum bicolor is a diploid species (2n = 20) that is closely related to sugarcane, a polyploid species. Cultivated sorghum can be subdivided into grain, fiber, forage, and sugar (sweet) types. Each of these groups represents a broad genetic diversity as has been shown, e.g., for grain types (Assar et al. 2005). Sweet sorghum stores a high amount of sugar in its stem, and produces plenty of juice. Sweet sorghum stalks have already been processed to syrup, sugar, and molasses for a long time. Recently, sorghum breeders worldwide are aiming at improving various characteristics of the crop that can result in commercialization. Although for many years the main trait of interest in sorghum has been grain yield, recently interest has also shifted from grain to stalk to manipulate biofuel-related traits: fiber, sugar, juice, and biomass. As in grain yield, biofuel-related traits cannot be manipulated directly, because they are quantitative, polygenically inherited traits. Therefore, to successfully improve these complex traits, they need to be dissected into smaller morphological, physiological, and genetical components, which are then analyzed and evaluated.

Sorghum like maize and sugarcane is a C4 plant characterized with high photosynthetic efficiency. Sugar yield is the major breeding objective in sweet sorghum. Aitken et al. (2006) defined sugar yield as a function of cane yield and recoverable sucrose content from the harvested cane. Milligan et al. (1990) reported that genetic correlations of sugar traits are a good strategy of selection in sugarcane. After screening a population in 2 years at five locations, they found that the stalk number shows a high correlation with cane yield (r = 0.77) and concluded that selection for sucrose should emphasize cane yield and attention may be given to stalk number. A high correlation between stalk weight and stalk diameter (r = 0.70) was also observed, and brix was highly correlated to sucrose. In sugarcane, stalk diameter and stalk number are highly heritable with more than 80% reported (Aitken et al. 2008), and a number of QTLs influencing these traits have been identified (Ming et al. 2002). In sorghum, Murray et al. (2008) reported a weak but significant correlation between sugar in dry stem and stem diameter (r = 0.11) and tillers (r = −0.09). Total sugar yield had no significant correlation with diameter but a significant correlation with tillers (r = 0.16), while juice sucrose was not correlated with tillers number but had a negative, weak but significant correlation with stem diameter (r = −0.06). These weak correlations in sorghum may limit indirect selection for sugar traits or sucrose using morphological traits.

Ritter et al. (2008) reported QTLs for sugar-related traits in a population of 184 recombinant inbred lines derived from grain × sweet sorghum. They used 247 AFLP, as well as sorghum and sugarcane SSR markers. Using composite interval mapping at a LOD score of 3.00, they detected 29 QTLs distributed in five different linkage groups for sucrose content (11 QTLs), sucrose yield (8 QTLs), and sugar yield (10 QTLs). Natoli et al. (2002) reported several QTLs for sugar-related traits in a population of 129 F2 progenies derived from unrelated sweet sorghum parents. Yun-Long et al. (2006) detected several QTLs in 207 F2 individuals derived from a cross between grain and sweet sorghum. Natoli et al. (2002) detected QTLs on chromosomes SBI-01, SBI-03, and SBI-05, Yun-Long et al. (2006) found QTLs on SBI-04, SBI-10, Ritter et al. (2008) and Murray et al. (2008) detected QTLs for sugar components on SBI-01, SBI-03, SBI-05, SBI-06, SBI-07, SBI-10, but with an additional one on SBI-02 in Murray et al. (2008). The distribution of these QTLs among chromosome and the less phenotypic variation explained suggest that sugar traits are controlled by more loci with complicated interactions.

To further understand the genetics of these traits, further studies should be done in populations with different genetic background and aim at partitioning these complicated interactions through epistasis, QTL × environmental interaction, and pleiotropic effect. The objectives of the present study were to (1) identify additive QTLs linked to sugar-related traits in sorghum, (2) detect possible additive × additive interaction among these QTLs, and to (3) identify QTLs showing pleiotropic effect(s).

Materials and methods

Plant material

A population of 188 recombinant inbred lines derived from a cross between the inbreds SS79 (sweet sorghum) and Macia-SA denoted as M71 (grain sorghum), developed in the Agricultural Research Council of South Africa (ARC) by Dr. Willy Wenzel, was used for this study. The female parent SS79 was collected from a traditional farmer in Limpopo province of South Africa. It is approximately 300 cm tall with long internodes, has thin but sweet juicy stalks. It has open and loose panicles with light brown seeds. The male parent, M71, originated from ICRISAT-Bulawayo in Zimbabwe bred under a Sorghum and Millet improvement program. It is an early maturing and high grain yielding, with white grains. It has short internodes and is approximately 140 cm tall. The stems are juicy, but the juice is insipid. After the introduction of SS79 and M71 at the ARC, a single ear was selected from each variety and selfed until homozygosity was reached. The mapping population of our study was developed using conventional single seed descent method until F5:6 generation (W. Wenzel, personal communication).

Experimental design and phenotypic evaluation

A total of four experiments were planted at 04 May 2007 and 14 May 2008 at two different locations: Gross Gerau (two field trials), and Rauischholzhausen (two pot trials). The characteristics of the locations are presented in Table 1. All experiments were planted in a randomized complete block design (RCBD) with three replications. At Gross Gerau (GG), a one-row plot of 5 m row length, and a three-row plot of 3 m row length were planted in 2007 and 2008, respectively. In both years, the inter-row space was 0.3 m, and seeds were sown at 0.03 m depth and at 0.15 m interval. In the pot experiment of Rauischholzhausen (RH), two seeds were sown per pot with a diameter of 0.20 m at 0.02 m depth. Best conditions for plant growth were implemented in both locations: In Gross Gerau, the plants were thinned to 100,000 plants per ha. Basal application of nitrogen (N) in the form urea was applied at a rate of 25 kgN/ha followed by irrigation. Regular monitoring of pest was done. Spider mites were identified and sprayed with proper insecticide in the RH 2009 pot experiment.

Table 1 Biophysical information during field experiments in Gross Gerau (GG) and Rauischholzhausen (RH)
Full size table

The trials were harvested manually by cutting the plant at the base by a scissor in both the two locations. In the two field experiments, three middle plants were selected at random for all data collection and later the average of the three plants was used. In the pot trials both plants per pot were harvested and averaged. Some lines were photoperiod sensitive, which resulted in differences in flowering dates of up to 2 months. To harvest the plots at the same physiological stage, each line was harvested at 40 days after anthesis (DAA), which estimated the hard dough stage (Zhao et al. 2009). At this stage, sugar accumulation has been reported to reach the optimum. Because the plants developed tillers at an early stage in both locations, the main stem and its tiller were treated as a single plant during harvest. Sampling was done in the following sequence: (1) Flowering date was recorded when 50% of the plot had flowered. (2) The length of the plant from the ground to the panicle tip was measured to estimate plant height. (3) Stem diameter was measured 20 cm above ground and the sum of tillers per plant was counted to estimate number of tillers per plant. (4) Panicles were cut at the flag leaf, weighed to estimate fresh panicle weight per plant. (5) The juice weight was estimated as the difference between fresh stripped stalk mass and dry stripped stalk mass. The fresh stripped stalks were milled and then dried in an oven at 105°C for dry mill. This difference gives a similar juice weight as that collected by roller miller in the same material (W. Wenzel, personal communication). (7) The fresh main stalk was pressed, and 2–3 droplets of juice were collected on a sucrose-sensitive refractometer to measure the Brix and another juice was squeezed to fill 2 ml tubes for future glucose and sucrose analysis.

Glucose and sucrose analysis

Carbohydrate quantification was carried only in the main stem. The samples were frozen at −20°C until the end of harvest. The three plants per replication were pooled to obtain an average sample extract. The pooled juice was diluted to 1:600 μl ml−1 ratios and then filtered. Analyses were carried out according to the manufacturer’s recommendation using the enzymatic bioanalysis kit produced by Boehringer Mannheim/r-biopharm. 96-well microtitre plates were used for the analysis. Reagents were adjusted to make 200 μl total reaction volume. Measurements were taken at 340 nm wavelength using the Tecan, sunrise spectrophotometer (Tecan Trading AG, Switzerland). The final glucose and sucrose concentration was calculated based on the linear regression as reported by Henderson et al. (1998) and Viola and Davies (1992), and verified using the supplied manufacturer’s formula. Sugar content was calculated according to Ritter et al. (2008). The concentration of fructose was replaced by concentration of glucose because fructose was not quantified in our study. Glucose and fructose can give similar content in sorghum (Godwin Ian, personal communication) and sugarcane (Robertson et al. 1996).

Phenotypic data analysis

Descriptive statistics and analysis of variance was done using SAS® 9.1 software. Proc GLM procedure was used to test differences among the lines in each and across environments. Broad-sense heritability on RILs-mean basis was estimated for all traits as reported by Littell et al. (2006) and Fehr (1987)

$$ h_{\text{B}}^{2} \, = \, {{\sigma_{\text{G}}^{2} } \mathord{\left/ {\vphantom {{\sigma_{\text{G}}^{2} } {\left( {\sigma_{\text{G}}^{2} + \sigma_{\text{GY}}^{2} /y + \, \sigma_{\text{GL}}^{2} /l + \sigma_{\text{GYL}}^{2} /yl + \, \sigma^{2} E/ryl} \right)}}} \right. \kern-\nulldelimiterspace} {\left( {\sigma_{\text{G}}^{2} + \sigma_{\text{GY}}^{2} /y + \, \sigma_{\text{GL}}^{2} /l + \sigma_{\text{GYL}}^{2} /yl + \, \sigma^{2} E/ryl} \right)}} $$

where \( \sigma_{\text{G}}^{2} \) is genetic variance, \( \sigma_{\text{GL}}^{2} \) is the variance of genetic × location interaction, \( \sigma_{\text{GY}}^{2} \) is the variance of genetic × year interaction, \( \sigma_{\text{GYL}}^{2} \) is the variance of genetic × year × location interaction, σ²E is the error variance, and r is the replication.

Coefficients of determination (R 2) were obtained from the linear regression of individual yield in different environments on the mean yield of all the genotypes in each environment (Leon and Becker 1988):

$$ R^{2} = 1 - {{s_{d}^{2} } \mathord{\left/ {\vphantom {{s_{d}^{2} } {s_{x}^{2} }}} \right. \kern-\nulldelimiterspace} {s_{x}^{2} }} $$

where \( {\text{s}}_{d}^{2} \) is the deviation mean squares from regression and \( S_{x}^{2} \) is the environmental variance.

Genotyping

The RIL population was planted in small pots in the greenhouse at Giessen in three replications with two pots per replication. Fresh leaf samples were collected 20 days after planting from two replications and bulked. Tissues were placed in Eppendorf tubes, frozen immediately in liquid nitrogen, and then stored in −80°C freezer. DNA extraction was performed according to Doyle and Doyle (1990). A total of 200 SSR primer pairs reported by Bhattramakki et al. (2000), Dean et al. (1999), Kong et al. (2000), and Schloss et al. (2002) were used to screen parents of the mapping population. A few EST-SSRs published by Srinivas et al. (2008, 2009a) were also included. The reaction protocol and thermal cycler touch-down protocol were performed according to Hasan et al. (2006) and Xu et al. (2005). Most of the SSR primers were labeled with M13 tailing as described by Berg and Olaisen (1994). The fluorescently labeled universal M13 primer 5′-AGGGTTTTCCCAGTCACGACGTT-3′ was added to the PCR reaction, and the forward primer of each SSR has been appended with the sequence 5′-TTTCCCAGTCACGACGTT-3′. After the first cycle of PCR, fragments were successively amplified with the labeled primer (Hasan et al. 2008; Rygulla et al. 2007).

AFLP markers were amplified according to Vos et al. (1995) using Gibco AFLP core reagents. A total of 150 ng (6 μl × 25 ng) genomic DNA was restricted using 2 μl EcoRI/MseI in 5 μl × 5× reaction buffer and 12 μl AFLP-grade water. The solution was incubated for 2 h in 37°C followed by 15 min in 70°C in Perkin Elmer 9700 thermal cycler. After incubation, a ligation master mix consisting of 24 μl adapter ligation solution and 1 μl T4 DNA ligase was added to the solution, which was again incubated for 2 h at 20°C. Ligation was followed by a pre-amplification step (+0) using the 5 μl of 1:10 diluted ligation with non-selective primers in order to minimize background noise at electrophoresis and to receive more distinct bands. The +0 sequence is complementary to the adapters and to the enzyme cutting-sequence. The following +1 pre-amplification was carried out using 5 μl of 1:10 dilutions of +0 pre-amplification prior to amplification (+3) using selective primer combinations. The sequence of the primers +0 EcoR1, +0 Mse1, +1 EcoR1, and +1 Mse1 were 5′-GAC TGC GTA CCA ATT C-3′, 5′GAT GAG TCC TGA GTA A-3′, 5′-GAC TGC GTA CCA ATT CA-3′, and 5′GAT GAG TCC TGA GTA AC-3, respectively, as reported by Uptmoor et al. (2003).

For electrophoresis, a total of 10 μl fusion dyes was added to the PCR products; the samples were denatured at 94°C for 5 min, and then 2 μ1 samples was loaded in 8% polyacrylamide gel. Amplification products were separated and visualized using the LI-COR model 4200 (MWG Biotech, Ebersberg, Germany) automated DNA sequencing system. This system can detect amplification products that were labeled with two different dyes: IRD-700 nm and IRD-800 nm.

Linkage map construction and QTL analysis

A total of 215 scored markers and 188 RILs were used to construct linkage maps using the Joinmap® 4.0 software. The tentative linkage groups were grouped using the LOD score of 3.0, and the Kosambi mapping function was used to calculate map distances in cM from the recombination frequencies. The linkage maps were compared to the sorghum maps published by Bhattramakki et al. (2000) and Mace et al. (2009), and assigned to chromosomes according to a recent sorghum chromosome nomenclature (Kim et al. 2005).

QTLs were detected from the combined environmental means using composite interval mapping with the linear regression approach implemented in software PLABQTL version 1.2 (Utz and Melchinger 1994). Cofactors were selected on basis of stepwise regression using Akaike’s information criterion. A LOD score of 3.0 was used for detection of QTLs. A QTL was considered suggestive if it had a LOD score between 2.5 and 3.0, but had significantly influenced the phenotype in a certain environment. When two peaks were detected for a single trait on the same chromosome, they were considered as two QTLs only if the distance between the QTLs was greater than 20 cM, otherwise only the higher peak was considered for better estimation of the QTL position (Parth et al. 2008; Ungerer et al. 2002).

Digenic epistasis (aa) was estimated from the detected set of QTLs in a stepwise regression based on f-to-enter value. Only additive × additive epistasis was estimated because no reliable dominance effect is expected in the lines after F5 lines of self pollinating crops. For QTL × environment analysis, the detected QTLs undergo a simultaneous fit for each environment, and the results are presented in an ANOVA. The difference between the fits of the data from individual environments and the means across environments give the mean squares of QTL × environment. The significance of QTL × environment has been tested using an F test with Bonferroni adjustment.

Results

Trait performance

The phenotypic performance of parents and recombinant inbred lines are presented in Table 2. The parents performed as expected for sugar-related traits, where SS79 was taller than M71 in all locations. Besides, SS79 was late flowering, had high sugar content, and low fresh panicle weight. Trait means of the recombinant inbred lines were intermediate, although transgressive segregation was observed.

Table 2 Mean values, standard deviations, and range for sugar components and sugar-related agronomic traits in recombinant inbred lines across environments
Full size table

The effect of the genotypes was significant (P > 0.0001) for all the traits studied except for estimated juice weight (Table 3). Although the estimated juice weight showed segregation among the RILs, both parents are known to have juicy stems. Environments were significantly different in all traits and high genotype by environment interaction was observed in all traits except for estimated stem juice weight. Most of the lines were photoperiod sensitive resulting in differences in harvest dates; therefore, harvest dates were used to estimate the same physiological state of the lines.

Table 3 Mean squares, heritabilities, repeatabilities, and coefficient of variation for sugar components and sugar-related agronomic traits in recombinant inbred lines across environments
Full size table

The heritability of the traits ranged from 0.18 to 0.81 (Table 3). Estimated juice weight showed a lower heritability (\( h_{\text{B}}^{2} \) = 0.18), indicating the impact of the environment in this trait. The moderate–high heritability in these sugar-related traits shows that breeding for these traits is possible. Repeatability of the traits ranged from 0.39 to 0.97 (Table 3). As in heritability, estimated juice weight showed a lower repeatability (R 2 = 0.39). As repeatability is used to estimate predictable performance of the genotypes, genotypes showed moderate–high stable performance across environments.

Correlation estimates

Pearson correlation coefficients (r) of sugar-related traits are shown in Table 4. Flowering date correlated significantly with all the traits. The strongest positive correlation was found between flowering date and estimated stem juice (r = 0.445), followed by stem diameter (r = 0.369) and Brix (r = 0.251), and the closest negative correlation was observed between flowering date and fresh panicle weight (r = −0.626). Fresh panicle weight has negatively and significantly correlated with all traits, except for a positive, non-significant correlation to number of tillers and a moderate significant positive correlation to sugar content. Number of tillers was only significantly but weakly correlated to estimated juice content (r = 0.195).

Table 4 Pearson’s correlation coefficient of sugar-related traits
Full size table

A relatively close positive correlation was found between Brix and sucrose (r = 0.606), even though a higher correlation estimate was expected. Glucose correlated negatively and significantly to sucrose (r = −0.334) but positively to Brix (r = 0.122). The estimated sugar content correlated with all traits except estimated juice weight. The strongest but correlation was observed between sugar content and sucrose (r = 0.988); estimated correlations to glucose (r = 0.556), Brix (r = 0.386), and stem diameter (r = 0.320) were moderate.

Linkage map construction and QTL detection

The genetic map was constructed from 188 lines and consisted of 102 AFLP, 49 SSR, and 6 EST-SSR markers, where 59 markers were excluded because of missing values or divergence from the expected segregation ratio using a chi-square test with one degree of freedom at α = 0.05. Eleven linkage groups were formed with an average distance of 6.55 cM between markers. The longest distance between markers was 20 cM and the shortest between markers was 1 cM.

Figure 1 presents the results of the identified significant QTLs either in each environment or from the mean of the combined environments or both. Supplementary Table S1 shows the details of the identified additive QTLs. The QTLs that significantly affected the traits using combined environmental means or mean of each single environment are reported here as significant. Environments 01, 02, 03, and 04 throughout the chapter represent the field experiments as presented in Table 1.

Fig. 1
figure 1

Genetic linkage map of the M71 × SS79 RILs consisting of 102 AFLP, 49 SSR, and 6 EST-SSR. Markers starting with E are AFLPs; X and Sb are SSRs; Stg and Dsb are EST-SSRs. All QTL shown were detected by composite interval mapping (CIM), using the combined environments (four) for each trait, and also found significant in one or many environments, respectively. The top and bottom borders of the QTL boxes represent support intervals of the detected QTLs associated with the respective trait on the right side of the box. The peak of each QTL is shown by the position of the arrow inside the box, and the direction indicates the allele contribution by parent M71 (pointing left) or by parent SS79 (pointing right)

Full size image

Flowering date

CIM detected four QTLs and one putative QTL on SBI-03, SBI-04a, SBI-06, SBI-07, and SBI-08 associated with flowering dates (LOD score ≥2.5). The QTLs on SBI-03, SBI-04a, and SBI-06 were found significant in various environments. The phenotypic variance explained by these significant QTLs was 33.4%, and their additive effects were 3.90, 0.62 and −2.72 days on SBI-03, SBI-04a, and SBI-06, respectively.

Plant height

Seventeen QTLs were detected in nine chromosomes associated with plant height (LOD score ≥3.0). Nine significant QTLs were identified on SBI-01, SBI-02, SBI-04a, SBI-4b, SBI-06, SBI-07, SBI-08, SBI-09, and SBI-10. The phenotypic variation explained by the detected QTLs ranged from 7.5 to 26.1%, whereas the additive effects ranged from −18.79 to −0.20 cm for SS79 and 3.05–9.48 cm for M71 parent. Most of the QTLs (61%) had negative additive effects suggesting that the parent SS79 had a higher contribution of tall alleles.

Stem diameter

Thirteen QTLs and one putative QTL influencing stem diameter were distributed over eight chromosomes. Nine out of 13 QTLs were significant across the environments. The QTL on SBI-03 at position 102 cM seems to be the major QTL because it was significant in three environments and explaining 13.3% of phenotypic variation. The additive effect of this QTL was −0.042 cm suggesting that the QTL allele derived from the female parent. Besides, the QTL on SBI-07 at position 30 cM also seem to be a major QTL because of its high additive effect (0.069 cm), high LOD score of 5.03 and because it explains 11.6% of the phenotypic variation.

Number of tillers per plant

Six significant QTLs were detected (LOD ≥ 2.5) that were associated with number of tillers per plant. A QTL on SBI-06 at position 64 was found significant in all the environments and was detected at LOD score of 2.90 using the overall mean of the data set. In total, 14 QTLs were detected on nine chromosomes explaining a phenotypic variation ranging from 6.7 to 15.7% and their additive effect ranging from −0.09 to 0.092 number per plant. Only three of the detected QTLs had positive additive effects indicating that the increase in tiller numbers was contributed by SS79.

Fresh panicle weight

At a LOD score of 5.62, a major significant QTL associated with fresh panicle weight was located on SBI-06. This QTL explained 12.9% of phenotypic variation and showed additive effect of 4.87 g/plant. Besides, four other significant QTLs were detected at a LOD score ≥2.5 on SBI-01, SBI-02, and SBI-04a. Together, they explained 20.3% of the phenotypic variation.

Stem juice weight

For stem juice weight, one QTL was detected on SBI-02 between 12 and 26 cM with LOD score of 3.54. This QTL explained 7.9% of the phenotypic variation and had additive effect of −17.603 g/plants. The negative additive effect shows that the allele was derived from SS79. Stem juiciness was not segregating well in the population because both parents are juicy, and the trait is also significantly controlled by the environment.

Brix

Four out of 14 detected QTLs significantly influenced the total soluble solids in sorghum stem juice. These significant QTLs were distributed on three chromosomes where chromosome 4 had two of these significant QTLs. Two major QTLs were identified on SBI-02 and SBI-06 at position 82 and 16, respectively. These QTLs explained 21.9 and 22.3% of the total phenotypic variation, respectively. Besides, they had the LOD score ≥10.00 and the additive effects of −0.167 and −0.797, respectively. The negative signs of both alleles show that these alleles were derived from SS79, the parent with high sugar.

Glucose content

A total of ten QTLs and one suggestive QTL were detected for glucose content on seven chromosomes. Out of these 11 QTLs, four QTLs were found significant in various environments. The QTL detected at position 70 of SBI-07 seems to be the major QTL. This QTL had LOD score of 5.99 explaining 13.7% phenotypic variation with additive effect of −0.351 g/l. This QTL was significant in the combined analysis and in two of the four environments. Its negative additive effect shows the contribution of the female parent in glucose content. In general, the detected QTLs showed that both parents contributed alleles for increasing and decreasing glucose content.

Sucrose content

For sucrose content, seven QTLs and two putative QTLs were pin-pointed on seven chromosomes, where four QTLs were found to be significant. These significant QTLs explained 42% of the phenotypic variation where a QTL on SBI-06 explained the highest phenotypic variation (10.6%) compared to the others. Both negative and positive additive effects were observed showing the positive alleles for sucrose content of both parents.

Sugar content

A total of 15 QTLs and two putative QTLs was detected that controls sugar content in stem juice on seven different chromosomes. Out of these 17 QTLs, nine QTLs were found significant in different environments. All detected QTLs explained from 6.6 to 12.5% phenotypic variation. The additive effects of the QTLs were either negative or positive showing that the QTLs derived from both parents.

Epistatic effects and QTL × environment interaction

The digenic epistatic effects are presented in Table 5. Twenty-four significant digenic epistatic effects (aa) were detected for seven traits using combined analysis. This suggests that two locus interactions were widespread in the entire genome: a total of five digenic QTLs were found regarding fresh panicle weight, four for plant height and glucose content, three for number of tillers, sucrose and sugar content, and two for flowering date. There were no significant additive × additive interactions observed for estimated juice weight, brix, and stem diameter.

Table 5 Digenic epistatic loci associated with sugar-related traits
Full size table

Epistasis between the loci affecting both sucrose and sugar content was found between SBI-01 and SBI-4b, and SBI-01 and SBI-06. The total phenotypic variation explained by digenic interactions was as follows: 8.7% for flowering dates (2 pairs), 21.4% for plant height (4 pairs), 25.4% for fresh panicle weight (5 pairs), 18.7% for number of tillers (3 pairs), 21.7% for glucose content (4 pairs), 19.1% for sucrose content (3 pairs), and 17.8% for sugar content (3 pairs). In general, the total phenotypic variation explained by such interactions was smaller than the corresponding main effects. Nevertheless, the importance of additive × additive interaction effects in total genetic effects may well be trait-dependent.

In this study, interactions between QTL and environments were analyzed for all traits (Supplementary Table S1). Significant QTL × environmental interactions were observed for sugar content on SBI-01, fresh panicle weight on SBI-01 and SBI-06, tiller number per plant on SBI-07 and SBI-10, and flowering date on SBI-06. No significant QTL × environment interactions were observed for the other traits studied.

Pleiotropic effects and co-localization

To identify a linkage group as a hotspot, we considered QTL clusters that have at least one QTL explaining more than 5% of the phenotypic variance with both paternal and maternal effects as reported by Rae et al. (2009) in poplar. QTLs for all traits co-localized in all linkage groups except for linkage groups 3 and 5 (Supplementary Table S1; Fig. 1). Hotspots can be declared on SBI-01, SBI-02, SBI-06, SBI-07, and SB1-09 because all traits co-localized on those linkage groups and more than one QTL has been found per trait.

Several QTLs that influence multiple traits were observed in this study (Supplementary Table S1; Fig. 1). A major QTL with multiple effects was detected in linkage group 6 with a support interval 12–18 cM. This QTL has been detected for all traits except for glucose and number of tillers per plant, and was significant for five out of six traits. The additive effects of this QTL differed per trait where the female parent contributed positive allele for flowering date, plant height, total soluble solids, and sucrose, but the male parent contributed positive alleles regarding fresh panicle weight and estimate of sugar content.

Other QTLs with pleiotropic effect have been located on various linkage groups such as E44M48-140 on SBI-04a, E35M49-212 on SBI-07, Xtxp51 on SBI-04a and Stgnhsbm36 on SBI-02. Most of the QTLs for sucrose were also detected either for Brix, or total sugar content or both.

Discussion

The growing importance of bioenergy-derived sugar crops such as sorghum and sugarcane is likely to create local as well as global environmental and economical challenges of fossil fuel usage. Similar to sugarcane as reported by Lingle (1997), our sorghum lines also differed in sucrose content. However, the accumulation mechanisms during growth in our materials are widely unknown, whereas genotypic differences of sucrose enrichment have been described in sugarcane (Lingle 1997). As sugar accumulation and content are quantitative traits controlled by polygenes and significantly affected by the environment, suitable strategies and success of breeding can be based on partitioning the agronomic traits and sugar components into various sub-traits as, e.g., has been done in the case of drought tolerance (Kebede et al. 2001; Hausmann et al. 2002). In the current study, agronomic traits related to sugar content such as flowering date, plant height, number of tillers, stem diameter, fresh panicle weight and stem juice, and sugar components such as glucose, sucrose, brix, and total sugar content were investigated.

To date, there are only a few studies on the identification of QTL for sugar-related traits in Sorghum bicolor. Our own results support some of the results reported this far. For example, several QTLs associated with flowering date were detected on linkage groups (LGs) SBI-01, SBI-04, SBI-06, and SBI-10 (Ritter et al. 2008), SBI-06, SBI-09 (Murray et al. 2008), and SBI-01, SBI-03, SBI-06, and SBI-08 (Feltus et al. 2006). Besides, Chantereau et al. (2004) found few QTLs controlling photoperiod response on LGs C, F, and H that correspond to SBI-01, SBI-02, and SBI-10 in our study. In addition, Pereira and Lee (1995) detected eight QTLs associated with plant height by interval mapping and single factor analysis on four different linkage groups, which correspond to our LGs SBI-06, SBI-07, SBI-09, and SBI-10. Supporting our study, their QTLs accounted for 9.2–28.7% phenotypic variation. In 168 F7 RILs of sorghum, Srinivas et al. (2009b) located only one major QTL on SBI-06 associated with fresh panicle weight, explaining 14.7% of the phenotypic variation having a LOD score of 4.9. Just like in our study, Murray et al. (2008) and Ritter et al. (2008) in two independent studies used different populations derived from crosses between a grain and a sweet sorghum type, respectively. Using 184 recombinant inbred lines (RILs) Ritter et al. (2008) reported 5 QTLs associated with Brix, 2 QTLs with glucose content, 11 QTLs with sucrose content and 10 QTLs associated with estimated sugar content. Their QTLs clustered on LGs SBI-01, SBI-05, SBI-06, and SBI-07. We have located two significant QTLs associated with estimated sugar content on chromosome 6, while Ritter et al. (2008) detected four QTLs on the same chromosome. Murray et al. (2008) used 176 RILs and reported QTLs associated with Brix on SBI-03, SBI-06 and SBI-07, and QTLs associated with glucose content on SBI-02, SBI-05, and SBI-09.

Six maturity genes are known to alter flowering time and period in sorghum: Ma 1 , Ma 2 , Ma 3 , Ma 4 , Ma 5 , and Ma 6 (Aydin et al. 1997; Quinby 1967). The first four genes cause long days to inhibit flowering but allow early flowering under short days, while the last two can inhibit floral initiation regardless of day length. Collinearity between rice chromosomes 7 and sorghum chromosome 2 (in our study) has been observed by Paterson et al. (1995, 2005). In rice, chromosome 7 has been reported to carry photoperiod sensitive genes Se-2 and El (Laurie 1997; Yano et al. 1997). Unfortunately, we have not found a QTL on chromosome 2; chromosome 3 is shown to harbor significant QTL associated with flowering date in three environments and also to the overall mean.

Plant height of sorghum was reported by Quinby and Karper (1954) to be controlled by four independent genes: Dw1, Dw2, Dw3, and Dw4. The dwarf gene Dw2 has a pleiotropic effect on grain yield and panicle length (Graham and Lessmann 1966), and Dw3 exerts pleiotropic effect on the number of tillers and panicle size (Casady 1965). The QTLs for plant height detected in the present study also co-localized with QTLs and had a pleiotropic effect on other morphological and sugar traits (SBI-01, SBI-02, SBI-04, SBI-06, and SBI-07). Congruently, Pereira and Lee (1995), and Rami et al. (1998) observed co-localization of QTL for plant height and stem diameter with other morphological attributes and concluded that these QTLs can be Dw loci. Although the markers reported in the latter two studies are different from ours to verify, this hypothesis is a possible explanation for the results of our study.

Several alignments between sugarcane and sorghum chromosomes has been done in recent years (Dufour et al. 1997; Guimarães et al. 1997; Ming et al. 1998, 2001), which is important to determine collinearity or orthology of sugar-related QTLs. Al-Janabi et al. (1994) suggested that sorghum and sugarcane diverted approximately five million years ago. The map created by Ming et al. (2002) showed that six of the 13 homologous groups identified in the Saccharum–Sorghum consensus map are closely corresponding while the remaining showed a varying level of synteny. Jordan et al. (2004) reported that stalk number and suckering (underground branching or tillering) influence sugar yield in sugarcane. They aligned the maps of sugarcane generated by Grivet et al. (1996) with those of sorghum generated by Paterson et al. (1995) and Boivin et al. (1999). Using the sugarcane RFLP probes, they identified QTLs for rhizome number, rhizome distance, re-growth, and seedling tillers on linkage groups SBI-01, SBI-02, SBI-03, SBI-06, and SBI-07 based on the chromosome nomenclature of sorghum by Kim et al. (2005). The biggest cluster of QTLs was found on chromosome 1. In the current study, QTLs associated with number of tillers were detected on all chromosomes except chromosome 8.

Ming et al. (2001) detected 36 sugar content QTLs in sugarcane using two interspecific segregating populations derived from S. officinarum and S. spontaneum. Sugar content was defined as sucrose at 96% purity, calculated based on Brix and Pol values. Through alignment, the detected QTLs corresponded to eight regions on linkage groups A, B, C, D, F, G, and I of the sorghum genome. By searching for linked markers in http://www.gramene.org and also using the maps of Ming et al. (1998), these linkage groups seem to correspond to LGs SBI-01, SBI-02, SBI-03, SBI-04, SBI-06, SBI-09, and SBI-10. Although in our study the QTLs for brix, sucrose, and sugar content were located on these linkage groups, the exact positions of the QTLs cannot be aligned for comparative analysis. However, as Ming et al. (2001) concluded, a much smaller number of ancestral genes may account for the observed QTLs between the two species. These original genes are assumed to have been multiplied by the duplication of chromosomes that characterized sugarcane genome evolution since its divergence from a common ancestor shared with sorghum.

In our study additive × additive interaction was found in seven out of ten characters studied. There were no significant additive × additive interactions observed for estimated juice weight, brix, and stem diameter. Epistasis is difficult to detect due to most of the experimental variation being assigned to main effects (Purcell and Sham 2004) and epistasis in both negative and positive direction is important for the breeding populations even if the net effects are not observed (Routman and Cheverud 1997). According to Whitlock et al. (1995) detecting epistasis has limitations due to genotype × environmental interactions and linkage disequilibrium. Regarding flowering date, one significant additive × additive interaction between SBI-04 and SBI-08 could be detected with the effect of −1.44 together explaining 3.0% of the phenotypic variance. Ritter et al. (2008) depicted an additive × additive effect of 0.98 between SBI-04 and SBI-08. These interactions delineated in both studies favor the late flowering parents.

Uncertainty in terminology regarding epistasis often arises due to the differences in definitions and types of epistasis being reported. The epistatic interaction effect was not consistent in our study. For example, digenic pairs for plant height on linkage groups SBI-01 and SBI-09 had a negative effect separately but together produced a positive effect in sucrose content; the interaction between SBI-01 and SBI-4b showed a strong negative effect while they showed small effects with different signs separately. The simplest epistatic interactions are duplicate and complementary interactions. Although it is well known that biosynthetic pathways of sugar traits involve series of genes, no conclusions can be made as yet on the type of epistasis reported.

Analysis of epistasis is often carried out in conjunction with QTL mapping, but epistatic effects may be confounded with background genome segregation and QTL × environmental interactions (Holland 2001). In our study, we analyzed QTL × environmental interaction among the detected QTLs. One QTL has been observed for sugar content and flowering date, and two QTLs were found for fresh panicle weight and number of tillers per plant. It is obvious that QTL that interact with the environment or even stable QTL across environments are more desired for marker-assisted selection.

Sugar-related traits in sweet sorghum are mainly influenced by additive genes as in the present study more than five additive QTLs were identified per trait using composite interval mapping. The co-localization and pleiotropic effect of the QTL is also validated by the reported correlation estimates among these traits. QTLs showing multiple effects were located, while the major one was found at 14 cM on SBI-06. This QTL favored the parental allele(s) responsible for increasing the characteristic value in almost all traits except for estimated sugar content. Such a situation would facilitate the combination of favorable alleles in single genotypes. Our study proves that besides major QTLs, breeders should also consider minor QTLs, additive × additive QTLs, and QTL × environmental interactions to improve sugar-related characteristics in S. bicolor. These results are considered to have significant implications for biomass and bioethanol improvement programmes in sorghum and other candidate crops such as maize and sugarcane.