Introduction
Billions of people worldwide, especially in developing countries, suffer from the sinister form of hunger called micronutrient malnutrition or hidden hunger, a particular form of unnoticed undernutrition. Micronutrient deficiencies can exist in populations even where the food supply is adequate in terms of meeting energy requirements. Humans need more than 22 mineral elements; some of them are required in large amounts, but others, such as Fe, Zn, Cu, I, and Se, are required in trace quantities (Grusak and Cakmak, 2005).
Micronutrient malnutrition resulting from the dietary deficiency of critically important minerals such as iron (Fe) and zinc (Zn) has been reported to be a major food-related primary health problem among populations of the developing world—including those in India—that are heavily dependent on cereal and legume-based diets and have limited access to meat, fruits and vegetables (Sandstead, 1991; Gibson, 1994; Welch and Graham, 2002). Worldwide, one out of seven people suffers from hunger; and most of them are poor people, particularly women, infants, and children. Micronutrient malnutrition persists as a major problem that not only affects vital growth in children but also actively damages the cognitive development of children, resulting in blindness, lowering disease resistance and reducing the likelihood that mothers survive childbirth (Sharma et al., 2003).
To address the occurrence of mineral deficiencies in human populations, plant scientists are devising methods of applying fertilizers and/or using plant breeding strategies to increase the concentrations and/or bioavailability of mineral elements in agricultural produce (Cakmak et al., 2004; Graham et al., 2007; Pfeiffer and McClafferty, 2007; Cakmak, 2008; White and Broadley, 2009). These approaches are termed ‘agronomic’ and ‘genetic’ biofortification, respectively. The biofortification approach involves a set of one-time, fixed costs in developing breeding methodologies, breeding nutritional quality traits into current crop varieties, and adapting these varieties to diverse environments. Breeding for higher trace mineral content in the consumed plant parts may not incur a yield penalty (Graham and Welch, 1996; Graham et al., 2007). It is known that biofortified crops can produce nutrient dense grains even in relatively poor soils with target nutrients within the critical range (Kanatti et al., 2014; Valentinuzzi et al., 2015).
Pearl millet [Pennisetum glaucum (L.) R. Br.] is a highly nutritious cereal with high levels of metabolizable energy, protein and micronutrients, a more balanced amino acid profile than wheat and rice and may other cereals, and low glycemic index (Singh et al., 1999; Sehgal et al., 2004). It is an important and low-cost source of food for the poor in West Africa and India. Its contribution of micronutrients, especially iron [Fe] and zinc [Zn], is higher varying from 30 to 50% of the intake of these micronutrients from cereals (Rao et al., 2006). An accelerated incorporation of the gene(s)/QTLs that enhance the nutrient content in staple crops will largely determine the success of various nutritional breeding schemes (Bohra et al., 2016). This study was undertaken with a hypothesis that measurable genetic variation exists in pearl millet, and such variation could be utilized for mining and mapping of QTLs for grain Fe and Zn content. In this study, for the first time, we report linkage relationship and phenotypic effects for grain Fe and Zn content QTLs in pearl millet.
Materials and Methods
Mapping Population
Recombinant inbred line (RIL) population advanced by single seed descent method from F2 to F6 was used in the present study. The population was derived from an intraspecific cross: ICMB 841-P3 × 863B-P2; comprised 144 lines, and was segregating for grain [Fe] and [Zn] contents, respectively. The parent ICMB 841 was bred by pure-line selection for downy mildew resistance within an outcrossed seed lot of elite seed parent maintainer line MS 5141B. ICMB 841 is the seed parent maintainer line of several popular dual-purpose hybrids released in India (Singh, 1990; Govila et al., 1997). Inbred maintainer line 863B was bred by pure-line selection at ICRISAT-Patancheru within a sample of Iniadi landrace material from Togo in West Africa (Rai et al., 2008). Compared to ICMB 841, 863B has larger grain size (and associated broader leaf blades, thicker stems, and thicker panicles), and better drought tolerance, downy mildew resistance, and stover quality. It was originally selected as a parent for mapping population development based on its combination of agronomic eliteness and superior combining ability for grain filling under terminal drought stress (Yadav et al., 2004; Bidinger et al., 2007). Standard agronomic practices and need-based plant protection measures were adopted uniformly to raise the crop.
Experimental Conditions
The experiment was conducted in an alpha-lattice design on Alfisol fields at ICRISAT-Patancheru (17.51°N; 78.27°E) with two replications in two environments during 2009 late Kharif (E1) and 2010 Summer (E2). Randomization was performed using IRRISTAT 5.0. The trials were conducted with a total of 120 entries (106 RILs + 2 parents + 4 checks). Each entry was sown in two-row plots (length 4 m) with 60 cm inter- and 15 cm intra-row spacing to produce bulks of selfed seeds as well as open-pollinated seeds for mineral element analysis. On the basis of seed setting, 15–20 panicles were harvested manually to produce the self-bulk seed samples while 5–10 panicles were harvested to produce the open-pollinated bulk seed samples, from each plot.
Mineral Analysis
Well dried F6 panicles (Self-and open-pollinated) were threshed using a thresher (Winter Steiger, Germany) to avoid any metal contamination. Briefly, cleaned grains were dried in hot air oven at 80°C for 24 h followed by grinding of 10 g of whole grain using a cyclone sample mill (Udy Corporation, USA). The grain powder (1 g) was digested in fume-hood with 10 ml triacid mixture of 5:2:1 by volume of concentrated nitric acid, sulphuric acid, and perchloric acid until clear white residue was obtained. After overnight cold digestion, the samples were digested at 120°C for 1 h, followed 2-h digestion at 230°C temperature. Required volume (75 ml) was made after completion of the digestion process, and digests were analyzed using an atomic absorption spectrophotometer (Model: SpectraAA 20, Varian, USA) according to the (Velu et al., 2006). Each sample was measured thrice and the means of these triplicate observations were used to represent the sample’s mineral elements contents. The mean observation for each sample was converted into the amount of mineral elements (mg kg-1) in each sample using Microsoft Excel. Grain Fe and Zn contents were computed as parts per million (ppm). Trait correlations between iron and zinc contents were determined using SAS software.
Genotyping of Mapping Population
DNA was extracted following Mace et al. (2003) and was normalized to 5–10 ng μL-1 concentrations for SSR, whereas, for DArT marker-based genotyping, DNA was diluted to 50 ng μL-1. All the RILs and their two parents were genotyped by both simple sequence repeat (SSR) and DArT markers. For SSR analysis, microsatellite markers from different sources [EST-SSRs (Xicmp and Xipes series), gSSRs (Xpsmp and Xctm series), and STS (Xpsmp(sts)] were used to identify polymorphism between ICMB 841-P3 and 863B-P2 (Allouis et al., 2001; Qi et al., 2001, 2004; Budak et al., 2003; Senthilvel et al., 2008; Rajaram et al., 2013). The forward primers of Xipes series SSRs were synthesized with an m13-sequence (5′CACGACGTTGTAAAACGAC3′) tail on the 5′ end. SSRs were amplified in a 10 μl PCR reaction mixture containing 10–15 ng of genomic DNA, 2 pmol of each primer, 2 mM MgCl2, 0.4 mM of each dNTP, 1× reaction buffer, and 0.2 U Taq polymerase (Bioline). PCR conditions were as follows: denaturation at 94°C for 5 min, followed by 10 cycles of denaturation at 94°C for 15 s, annealing at 61°C to 51°C (touch-down cycles) for 30 s, and extension at 72°C for 30 s, followed by 40 cycles of denaturation at 94°C for 10 s, annealing at 54°C for 30 s, and extension at 72°C for 30 s, followed by final extension at 72°C for 20 min. PCR amplification was checked on 1.2% agarose gels and PCR products were separated by capillary electrophoresis on an ABI3730xl sequencer and their sizes were determined using GeneMapper v4.0 software (Applied Biosystems, USA).
For DArT genotyping, DNA samples (50 ng μL-1) was submitted to the Genotyping Service Laboratory (GSL) at ICRISAT-Patancheru, where RIL population was genotyped using an array of 6912 DArT clones developed by PstI/BamII complexity reduction. DArTsoft, a software package developed at DArT P/L, was used to automatically analyze the output data. The DArTsoft-generated 0–1 scores of the polymorphic DArT markers detected among the inbred lines and parents of the particular RIL population were converted to A–B scores and used as input for mapping.
To construct the map, 38 RILs (26% of the population) were discarded from the original set of 144 lines because they were contaminated or had loci with non-parental alleles. Further, we used only the remaining 106 RILs for mapping of the DArT and EST-SSR markers (Xipes series).
Linkage Map Construction
A standard χ2-test was employed to test the segregation at each marker locus for deviation from the expected Mendelian segregation. Deviations of marker loci from the expected proportion of heterozygous alleles were tested by a t-test. Residual heterozygosity was considered while mapping. Linkage analysis was performed by using the MAPMAKER/EXP 3.0 program (Lander et al., 1987). The critical logarithm of odds (LOD) score for the test of independence of marker pairs was set at 3.0 and maximum recombination fraction (𝜃) of 0.49. In a preliminary analysis of the data, markers were deleted that had high segregation distortion (P ≤ 0.01). Subsequently, the linkage analysis of genotyping data for all the markers including those with distorted segregation was used for linkage analysis. Further, the RECORD software (Van Os et al., 2005) was used to get the most likely marker order in each linkage group. The order of markers in each linkage group was finalized by RECORD software (Van Os et al., 2005), and distances between markers loci calculated using the Haldane mapping function. A graphical representation of the map was drawn using MapChart software (Voorrips, 2002). The consensus maps developed by Rajaram et al. (2010), and DArT marker-saturated linkage maps developed by Supriya et al. (2011) were considered as a reference map for grouping of the markers and to construct the map in the present study.
Trait Analysis
The phenotypic variance was partitioned using the residual maximum likelihood (ReML) algorithm with a mixed model, where replication and block were considered to be fixed effects, while genotypes were random effects, to obtain the best linear unbiased predictions (BLUPs) of the performance of genotypes for each observed trait (Patterson and Thompson, 1971). Pearson correlation coefficients were carried out for studied traits; namely, Fe and Zn content (ppm) using PROC CORR in SAS (SAS Institute Inc., 1999). Because a normal distribution could not be assumed for all observed variables, Spearman’s rank correlation (rs) was also used as a robust estimation of the correlation coefficient. Similarly, genotypic correlations were also estimated. The significance of the correlation coefficients at P ≤ 0.05 and 0.01 was indicated as ∗ and ∗∗, respectively. Broad-sense heritability on plot means basis was computed from the estimates of genetic (σg2 ) and residual (σe2 ) variances using progeny means across RILs in each environment for both traits using PROC MIXED in SAS (SAS Institute Inc., 1999). Heritability (H2) was estimated as explained by Falconer (1989).
QTL Detection
Composite interval mapping (CIM) (Zeng, 1994) was used to search for QTL using the BLUP data of each trait. CIM was performed in PLABQTL (Utz and Melchinger, 1996) on an RIL population of 106 lines with 2 cM increments. To declare a putative QTL as statistically significant, a minimum LOD score of 3 was fixed according to the Bonferroni correction. The critical LOD threshold was analyzed empirically for each trait using 1,000 permutation runs. The proportion of the phenotypic variance explained by the QTL was determined by the estimator R2adj. The putative QTLs detected for each of the respective traits were assigned to linkage groups based on the map positions of their flanking markers. QQE interactions were identified using QTLNetwork 2.0.
Discussion
Bio-fortification — breeding to develop cultivars with dense levels of micronutrients, especially Fe and Zn content in edible parts of staple foods — provides a low-cost, sustainable strategy for reducing levels of micronutrient malnutrition (Bouis, 2002). Bio-fortification is a valuable and better alternative to alleviate micronutrient deficiencies. The major advantage of growing biofortified crops is that these may produce nutrient dense grains even in relatively poor soils with target nutrients just within the critical range (Kanatti et al., 2014; Valentinuzzi et al., 2015). Presently rice, wheat, maize, common bean, sorghum and pearl millet are the crops in which research toward bio-fortification to reduce micronutrient malnutrition is going on under the umbrella of HarvestPlus Challenge Program of CGIAR. These nutritionally rich cereals will improve the health, both directly by enhancing micronutrient availability and indirectly through improved agronomic performance and crop yields (Welch, 1999; Gómez-Galera et al., 2010). In addition, seedlings grown from micronutrient-dense seeds will be vigorous and healthy hence crops will be more disease resistant and stress tolerant, ultimately resulting in improved agricultural production (Welch, 1986, 1999).
The male parent of the population was developed from the Iniadi landrace. This study compared estimates of mineral micronutrient content of selfed vs. OP seed samples in each of the screening environments and across seasons. The average selfed grain [Fe] was higher than the selfed grain [Zn]. Similar results were also detected by Voorrips (2002). For RILs of the (ICMB 841-P3 × 863B-P2)-derived mapping population, the average [Zn] in OP seeds was similar in both environments while the average [Fe] in OP seeds was lower in summer. In contrast, mineral contents in selfed seeds were higher than those in OP seeds and the effect of seasonal variation was also greater on selfed seeds, perhaps as a result of differences in selfed seed set in the two seasons (but this was not measured). In the summer season, the mineral contents in selfed seeds were higher indicating possible effects of environment on selfed seed set, mineral distribution, and uptake. In contrast to the present results, a previous study conducted by Velu et al. (2007) in pearl millet demonstrated that the rainy season was more favorable for mineral content in seeds than summer indicating the effect of season on seed mineral content. Vreugdenhil et al. (2004) reported that nutrient contents of plant seeds depend on environment factors. Alterations in the environment or physiology of a plant can affect the accumulation of multiple elements simultaneously. Variation in mineral uptake in different environments has been described in Arabidopsis thaliana (Loudet et al., 2007; Ghandilyan et al., 2009a,b) and Silene vulgaris (Ernst et al., 2000). Moreover, the nutrient availability in the environment not only affects the nutrient concentration of the vegetative plant parts, but also of the economic parts of a plant, and different soil types also result in micronutrient content variability (Ernst, 1974; Tyler and Zohlen, 1998; Ernst et al., 2000). Sankaran et al. (2009) also detected variation in mineral contents of seed between environments and emphasized the importance of environmental factors on quantitative traits. However, for the present study, there is no information on differences in soil conditions as both screens were conducted on similar Alfisol fields and no soil tests of micronutrient content or availability were performed, so it is not possible to determine whether soil or climatic factors were responsible for the observed differences.
There have been concerns about the possibility that high grain mineral contents may result from “concentration effects” Gomez-Becerra et al. (2010) as a consequence of small seeds, reduced seed set (which could be associated with larger selfed seed size if there is significant yield component compensation), and/or low yield capacity. Being a protogynous cross-pollinated crop, in which stigmas are normally fertilized by wind-borne pollen from other sources prior to pollen shed from flowers of the same panicle, the number of selfed seed per panicle were less in RILs and their parents compared to OP seeds per panicle (not shown). The differences in mineral contents of selfed and OP seed samples could be due to differences in numbers of seed set on per panicle dilution effect (Garvin et al., 2006). However, many studies conducted to examine the correlation between numbers of fruit and mineral/nutrient partitioning that shown the fraction of assimilates allocated to the fruits actually increased with fruit load (Hurd et al., 1979; Marcelis, 1992; Heuvelink, 1997). OP seeds were developed by a mixture of selfing and outcrossing due to the action of the wind. The effect of pollen source, xenia effect, may be another possible reason for observed differences in mineral contents between selfed and OP seed samples as the genotype of embryos and the triploid endosperm is decided after pollination. Thus OP seeds could not only be more numerous (per panicle) than self-pollinated seeds but also show xenia effects on TGM and grain mineral contents.
The analysis of variance revealed that genotypic variance was highly significant in individual environments, and broad-sense operational heritabilities were generally high enough to permit effective QTL mapping. In addition, joint analysis across the two screening environments for the (ICMB 841 × 863B)-based RIL population indicated moderately high to high proportions of genetic variance for the traits studied. In this joint analysis, G × E variance values were significant but substantially lower than genetic variance values, suggesting that the observed traits were under genetic control with limited influence of genotype × environment interaction effects, which implied there was no need for G × E partitioning. None-the-less, operational heritability estimates from these joint analyses were substantially less than those of the corresponding single-environment analyses of this RIL. These findings were largely in agreement with previous pearl millet studies conducted at ICRISAT by Ali et al. (2001) and Kannan et al. (2014) which found that G × E interactions were significant for grain yield, flowering time, plant height and panicle length. Assessment of the environmental stability of micronutrients is important in a crop improvement program aimed at enhancing the nutritional quality of food crop plants (Bidinger et al., 2007). High G × E interactions for grain Fe and Zn concentrations, which affect the rank of genotypes across the environments, have been reported in various cereal crops (Oikeh et al., 2004; Morgounov et al., 2006; Oury et al., 2006; Bidinger et al., 2007; Gomez-Becerra et al., 2010).
The broad-sense operational heritability values of 62% for selfed seed Zn content and 66% for selfed seed Fe content observed in the (ICMB 841 × 863B)-based RIL population are very promising for future application in breeding programs. Hence, based on observations of RIL progeny means, G × E variances, trait heritabilities, and variation between parents suggests that the male parents of RIL population would be highly suitable to use as donors for improving the mineral micronutrient contents in elite lines.
Correlations
The correlation between grain Fe and Zn content has been studied in several crops, with the results, by a large, showing similar trends. In both populations, Fe and Zn contents were strongly and positively associated. This may point to common molecular mechanisms controlling the uptake and metabolism of these minerals in grains or common transporters controlling the movement of these minerals within plants (Vreugdenhil et al., 2004; Ghandilyan et al., 2006). Co-segregation of QTLs for traits might be the reason of strong association between the grain Fe and Zn content in the RIL population. The direction and intensity of association in the RIL population suggest there are good opportunities for simultaneous genetic improvement of both micronutrients (Velu et al., 2008) by co-transferring superior alleles controlling these traits into the genetic backgrounds of elite lines.
Segregation Distortion
Distorted segregation in RIL populations derived via single-seed descent (SSD) represents the cumulative effect of both genetic and environment factors on multiple generations, and G × E interactions become more pronounced with the progress of selfing (Wang et al., 2003; Xu, 2010). In the present study, a big portion of the distorted markers was skewed toward female parent alleles. Groups of these closely linked skewed loci together make a segregation distortion region (SDR). SDRs have been reported in many crops in previous studies (Xu, 2010). Similarly, many researchers observed segregation distortion in favor of the female parent alleles (Singh and Ghai, 2007), however, segregation distortion favoring alleles from a male parent has been reported previously in pearl millet (Liu et al., 1994; Azhaguvel, 2001; Kolesnikova, 2001; Gulia, 2004; Supriya, 2010). Markers accommodated by LG1 were skewed toward male parent alleles (863B) in the (ICMB 841 × 863B)-based RIL population, while the greatest portion of distorted markers on LG2 was skewed toward female parent (ICMB 841B) alleles. In addition to genetic backgrounds, population sizes can also play and important role in the distortion phenomenon (Xian-Liang et al., 2006). Favoring of marker alleles toward one parent seems to be common and not unexpected in nature. In male gametes, pollen killers or pollen abortion, more frequently results in marker distortion as compared to disturbances in female gametes (Taylor and Ingvarsson, 2003). It has been suggested that the protogynous nature of pearl millet also contributes to segregation distortion (Liu et al., 1994; Supriya, 2010) also observed similar results in two pearl millet RIL populations genotyped with SSR and DArT markers. Finally, residual heterozygosity may also have the reason as previously demonstrated (Livingstone et al., 1999).
Linkage Map
The published genetic maps in pearl millet have in general been created primarily with co-dominant RFLP and SSR markers. However, the linkage maps constructed in the present investigation were generated with codominant SSRs and dominant DArTs and span 1748.7 (ICMB 841 × 863B map consisting of 305 markers). Most of the SSRs used in the present study were previously mapped and published, which made easy because of the ease of use and their reproducibility. In general, different map lengths reported previously in pearl millet were the result of variation in recombination frequencies of different population structures, numbers of mapped markers, population generation, and population size. Recently Vengadessan (2008) reported the longest SSR-based skeleton linkage map for F2-derived F3 progenies with a length of 1018.7 cM accommodating only 44 well-distributed markers. Gulia (2004) also reported a large map using SSRs with map length of 748 cM for an F2 population. The shortest F2-based map (287.7 cM) reported so far Liu et al. (1994) had 181 RFLP Markers. Using a (ICMB 841 × 863B) population (ancestral to one of the populations in the current study) Yadav et al. (2004) incorporated 91 marker loci in a population of F2 individuals, with a map length of 617.4 cM. The same F2 population was used by Senthilvel et al. (2008) to obtain a map length of 677 cM with 112 loci. Using the same population but in the F6 generation, the present map spans 1748.7 cM. This can be explained, by the presence of a very high number of markers and small size of population (106 RILs), and a more advanced generation (expected to have twice the recombination and therefore map length as the corresponding F2 population). Of the 144 RILs available, only 106 could be used in present study RIL, as the remainder were excluded from the investigation because of detection of non-parental alleles (a result of outcrossing during the inbreeding generations), with a consequent reduction in population size. Several reasons can be attributed to the presence of non-parental alleles, with outcross contamination of RILs during generation advance being the most likely. The map length appears to indicate that good genome coverage has been achieved in the present study using SSRs and DArTs and that addition of more molecular markers would not serve to substantially increase the map length. As compared to previous maps in pearl millet where several regions were had marker intervals more than 50 cM, the present (ICMB 841 × 863B)-based map contains very few big gaps due to the filling of previously reported gaps leading to increment in total map length. Even with good numbers of markers and good genome coverage, the distribution of markers was not uniform. Instead, markers tended to cluster, especially at the distal ends of all chromosome arms. This suggests still there is limited chance to increase map length by increasing the number of markers.
Map length of LG1 for the (ICMB 841 × 863B) RILs in the present study was drastically increased by the addition of flanking markers. Compared with the SSR map of Senthilvel et al. (2008) two additional markers could be mapped after Xctm27 while 46 markers were mapped before Xpsmp2273, which explains the observed increase in map length. One large gap of more than 50 cM remained in LG1 similar to the maps of Senthilvel et al. (2008) and Supriya (2010), using the (81B-P6 × ICMP 451-P8)-based RIL population).
LG2 also has shown an increment in map length over previous estimates ranging from 36.2 cM (Liu et al., 1994) to 179 cM (Yadav et al., 2004). The length of LG2 was comparable to that of reported by Yadav et al. (2004) as a later generation of the same cross (ICMB 841 × 863B) was used. The occurrence of additional meioses during generation advance explains the increase in the length of LG2. Senthilvel et al. (2008) reported 81.4 cM length of LG2 with a large gap of 48 cM for an F2 population of (ICMB 841 × 863B). Gulia (2004) also mapped two SSRs showed larger inter-marker distances in this linkage group.
LG3 was reported as the smallest among all linkage groups of pearl millet in all previous studies. But in the present study, LG3 was the second smallest after LG6. This indicates that marker numbers were not sufficient in previous investigations to cover the whole genome of LG3. In contrast to previous studies, segregation of almost all LG3 loci in both populations was distorted, which may explain part of the increase in LG3 length. Similar to previous studies reported by Liu et al. (1994), Gulia (2004), and Vengadessan (2008), a big inter-marker gap was also detected on LG3 of the (ICMB 841 × 863B)-based population in the present study.
In present study, the length of LG4 of (ICMB 841 × 863B)-based RILs was comparable with previous maps generated by Devos et al. (2000), Kolesnikova (2001), and Gulia (2004) without any large inter-marker gap like those previously reported by Liu et al. (1994) and Supriya (2010) for RILs of cross 81B-P6 × ICMP 451-P8) including above researchers. This indicates that length of LG4 of the (ICMB 841 × 863B)-based RILs will not likely increase by much with additional markers as it has good coverage with a high number of markers.
The lengths of LG5 for both populations in the present study were comparable and higher than previously reported lengths for this linkage group. The span is in agreement with that of the previously mapped F2 population of the cross (ICMB 841 × 863B) that showed a length of 103 cM, largest among all pearl millet populations mapped to that time Yadav et al. (2004).
LG6 of the RILs from the cross (ICMB 841 × 863B) was split into three segments in which the major segment (LG6B) accommodated 32 markers across a length of 94.2 cM. Using the F2 from this same population, Senthilvel et al. (2008) also detected these two big inter-marker spaces in LG6. Interestingly, the cumulative length of LG6A and LG6B is comparable with the length reported by Yadav et al. (2004) using the same F2 population. The consensus map generated by Qi et al. (2001) also showed that most of the loci are mapped in the center of this LG with big inter-marker spaces at distal ends and these gaps explain the segmentation of LG6 into three pieces in the present study. This indicates the there is a need to map more markers in this LG to better fill these gaps and link the three segments.
LG7 of the (ICMB 841 × 863B)-based RILs was found to be the second largest linkage group (294 cM), which is in agreement with Yadav et al. (2004). In contrast to this Gulia (2004) reported it as the pearl millet largest linkage group (195 cM). Using am (ICMB 841 × 863B)-based mapping population, Senthilvel et al. (2008) positioned marker Xicmp3092 at the top end in LG7, while in the present study it is in position second whereas Xctm8 was the last marker on the other end in both maps, which indicates that these markers may be in telomeric regions. There is clear cut indication of an increase in length because additional markers have been added to this linkage group compared to the earlier published maps in pearl millet. There were many large inter-marker distances observed in previous maps of this linkage group (Gulia, 2004; Yadav et al., 2004; Senthilvel et al., 2008). However, in the present map, only two big inter-marker gaps (<29 cM) were detected, which were also found in previous maps.
All linkage groups revealed a common feature of scanty marker loci and large gaps in distal ends of all linkage groups. This is likely due to hot spots of recombination in the gene-rich distal/telomeric regions of chromosomes (Devos et al., 2000; Qi et al., 2001; Senthilvel et al., 2008). The number of large gaps has decreased in the present study compared to previous linkage maps, although still there is an opportunity to map more markers to reduce or eradicate these gaps. However, (Qi et al., 2001) hypothesized that the large gaps in distal regions of pearl millet linkage groups represent the high recombination regions, rather than insufficiency of markers in these regions.
In the current study, clustering of DArTs in distal as well as central regions was more frequent than that of SSRs (Supriya et al., 2011). It seems that DArT markers may have a stronger tendency than genomic SSR and AFLP markers, in particular, to map to such gene-rich regions (Vuylsteke et al., 1999). In the present study, this occurrence may be due to usage of the methylation-sensitive restriction enzyme PstI during library construction for the DArT array, and subsequent preparation of DNA samples for hybridization to the array. DArT marker clustering in distal regions of chromosome arms was observed in previous DArT mapping studies on wheat (Akbari et al., 2006; Semagn et al., 2006), barley (Wenzl and Carling, 2004), and pearl millet (Supriya et al., 2011).
Mapping Quantitative Trait Loci (QTLs)
QTLs for Self-Pollinated Grain Fe Content
Although six putative QTLs were detected for this trait across individual screening environments for the (ICMB 841 × 863B)-based RILs, only one major on LG3 was consistently detected across the individual screening environment analyses and the two across-environment analyses. This suggests that the remaining variation may be the cumulative effect of many different genomic regions contributing small percentages of the phenotypic variation, with the small size of this population preventing their detection in the present study. This suggests that the transport and accumulation of minerals in seeds is a complex trait and highly influenced by environment as significant QEI was detected for self-pollinated grain Fe content in the present study. The occurrence of several QTLs for grain Fe content was also detected in various previous studies in different crops like rice (Peleg and Cakmak, 2009), Medicago (Sankaran et al., 2009) and wheat (Lu et al., 2008). No common QTL for Fe content in OP seeds was detected in analyses of the three data sets making all putative QTLs detected with this population for this trait environment-specific. Similarly, no common loci were observed for Fe content in self-pollinated and OP seeds of this population. This indicates that environment and pollen source play important roles on Fe content in seeds consequently different positions of detected QTLs.
QTLs for Self-Pollinated Grain Zn Content
In the present study, the positions of a putative major QTL on LG3 influencing the Zn content in self-pollinated as well as OP seeds were same as that detected for Fe in self-pollinated seeds. This co-localization of QTLs explains the positive correlations between contents of these two minerals. Co-localization of QTLs affecting different traits suggests either a single pleiotropic locus is involved in controlling of multiple traits or that several separate loci affecting independent traits are in close proximity (Ding and Yang, 2010). The same QTL position for Zn in self-pollinated and OP seeds also suggests that probably this QTL is less affected by variation in seed set or pollen source. Similar to Fe, a number of environments specific QTL was detected for Zn content. For example, against the five QTLs detected for Zn in OP seeds for this RIL population in the E2 screen, no QTL was detected for this trait in E1. Failure to detect QTL associated was previously reported for RIL populations of tomato (Saliba-Colombani et al., 2001), tobacco (Julio et al., 2005), and Chinese cabbage (Wu et al., 2008).
Co-mapped QTLs for Grain Fe and Zn Content
Quantitative traits affected by pleiotropism and linkage tend to exhibit correlations among them. This, in turn, often leads to detection of co-mapped QTLs. However, it is not easy to differentiate between linkage and pleiotropy until the QTN (Quantitative Trait Nucleotide) responsible for the phenotypic variation of each trait has been identified (Mackay, 2001). In the present study, QTLs for Fe and Zn content in the (ICMB 841 × 863B)-based RIL population were co-mapped on LG3 and favorable alleles for these QTLs were contributed by 863B (Tables 5 and 6). The co-localization of QTLs for contents of multiple elements was previously observed in wheat (Wu et al., 2008; Chatzav and Peleg, 2010) Brassica oleracea (Broadley and Hammond, 2008), Arabidopsis (Vreugdenhil et al., 2004), rice (Shimizu and Guerta, 2005; Stangoulis and Huynh, 2006; Ishikawa and Abe, 2009) and Medicago (Sankaran et al., 2009). These results suggest that some QTL regions appear to affect multiple traits. The co-mapped QTLs demonstrate the existence of genes or gene clusters with major effects on related traits. These loci might contain a gene for a common transporter such as a ZIP gene family member, which is capable of transporting Zn in addition to Fe, or a gene controlling synthesis for nicotinamide, a metal chelator involved in Fe, Zn, Cu, and Mn homeostasis (Curie and Briat, 2003; Delhaize, 2003; Sankaran et al., 2009). The stability of grain nutrient constituents across environments as well as co-localization of QTLs is of interest for both crop breeding and commercial production. In the present study, several QTLs have been identified for both minerals using the (ICMB 841 × 863B)-based RILs, with one being identified in both years and co-mapped, thereby indicated that this QTL is robust and can be targeted for marker-assisted breeding. This study also suggested that both RIL populations can be exploited for agronomic as well as mineral traits for molecular breeding.
Epistasis
No epistasis was detected for QTLs influencing either mineral micronutrient, which suggests that marker-assisted breeding for grain Fe and Zn content would not be adversely affected by epistatic interactions associated with the putative QTLs detected from ICMB 841 and/or 863B. Assessment of QTL × QTL and QTL × QTL × Environment interactions by two-locus analysis using QTLNetwork-2.0 for the RIL population derived from cross ICMB 841 × 863B indicated that major QTLs are not involved in interactions. The main effects of the epistatic loci did not account for any phenotypic variance, but their interactions with each other produced very small and negligible effects on phenotypic variation for the observed traits. Hence interactions can be neglected during marker-assisted breeding for the observed traits using either 863B or ICMB 841 as donor parents. Epistasis analysis of this RIL population revealed that Fe content in self-pollinated seeds is largely controlled by the single main effect QTL. Only one pair of epistatic loci was identified for this trait, and their interaction did not have a big effect on phenotypic variance. Interactions of screening environments and epistasis effects also were found to be non-significant and had no effect on phenotypic variation. One reason for the absence of epistasis in the present study could be that we investigated a population developed by intra-specific cross of two well-adapted inbreds. In this case, there should be less opportunity to disrupt co-adapted epistatic gene complexes in the parents as one might expect for the inter-specific crosses that are often employed in QTL mapping studies (Melchinger et al., 1998).