Identification of quantitative trait loci for yield and yield related traits in groundnut (Arachis hypogaea L.) under different water regimes in Niger and Senegal

Yield under drought stress is a highly complex trait with large influence to even a minor fluctuation in the environmental conditions. Genomics-assisted breeding holds great promise for improving such complex traits more efficiently in less time, but requires markers associated with the trait of interest. In this context, a recombinant inbred line mapping population (TAG 24 × ICGV 86031) was used to identify markers associated with quantitative trait loci (QTLs) for yield and yield related traits at two important locations of West Africa under well watered and water stress conditions. Among the traits analyzed under WS condition, the harvest index (HI) and the haulm yield (HYLD) were positively correlated with the pod yield (PYLD) and showed intermediate broad sense heritability. QTL analysis using phenotyping and genotyping data resulted in identification of 52 QTLs. These QTLs had low phenotypic variance (<12 %) for all the nine traits namely plant height, primary branching, SPAD chlorophyll meter reading, percentage of sound mature kernels, 100 kernel weight, shelling percentage, HI, HYLD and PYLD. Interestingly, few QTLs identified in this study were also overlapped with previously reported QTLs detected for drought tolerance related traits identified earlier in Indian environmental conditions using the same mapping population. Accumulating these many small-effect QTLs into a single genetic background is nearly impossible through marker-assisted backcrossing and even marker-assisted recurrent selection. Under such circumstances, the deployment of genomic selection is the most appropriate approach for improving such complex traits with more precision and in less time.


Introduction
Groundnut or peanut (Arachis hypogaea L.), an important oilseed and confectionary crop, is cultivated in [100 countries of the world. It covers 24.6 million Electronic supplementary material The online version of this article (doi:10.1007/s10681-015-1472-6) contains supplementary material, which is available to authorized users. ha global area with the total harvest of 41.3 million tons and productivity of 1676 kg ha -1 (FAOSTAT 2014). The Asia (11.6 mha, 47.15 %) and the Africa (11.7 mha, 47.56 %) together (23.3 mha, 94.71 %) cover maximum global area. Despite development of several new improved cultivars together with available management practices, the Asia and the Africa could achieve mere productivity of 2217 and 929 kg ha -1 , respectively as compared to the Americas (3632 kg ha -1 ) (FAOSTAT 2014). Engagement of mostly smallholder farmers in groundnut cultivation under rainfed condition with limited resources and input is the main reason behind poor performance. In addition, majority of the cultivated area falls under semi-arid tropic (SAT) regions which faces frequent erratic rainfall mostly during flowering and pod filling stages (Mahmoud et al. 1992;Fletcher et al. 1992). Groundnut, as a legume species, is no less important in the nutrition for grazing animals as it is for human. Thus, apart from pod yield, genotypes with biomass production are desirable in SAT zone, which can provide a source of additional revenue for farmers and important source of feed for animals. Niger and Senegal are the major contributors in West Africa and drought stress has been the main challenge in getting pod yield up to full genetic potential of popular and widely cultivated cultivars. Unpredictable environment due to climate change has further made the task of breeders very difficult to improve these complex traits.
Drought tolerance and pod yield are the two highly complex traits and genotype (G) 9 environment (E) plays critical role in deciding the ultimate pod yield achieved under drought stress. The complexity of drought is further aggravated in the SAT due to occurrence of high temperature, high levels of solar radiation and poor soil health. The above mentioned variables/factors have always posed difficulty in identifying the specific physiological traits needed to be improved for enhancing crop performance under drought stress condition. The breeding community has always relied on selections based on phenotyping i.e., pod yield achieved under drought stress in a specific environment. These genotypes not only failed to perform in other environments but also sometimes even comparable performance could not be achieved in farmers' field in the same environment.
Understanding trait genetics and identification of genetic factors with stable performance across environments based on precise phenotyping is essential to harness the benefits of modern genomics technology and the process is referred as genomicsassisted breeding (GAB) (Varshney et al. 2005(Varshney et al. , 2006. Last decade witnessed sharp progress in development of large genomic resources in groundnut along with their successful utilization in molecular breeding for trait improvement. Molecular markers have already been deployed at large scale in several genetics and breeding applications in groundnut Varshney et al. 2013aVarshney et al. , 2014. In order to integrate genomics in breeding, identification of genomic regions/genetic markers for component traits of pod yield and drought tolerance is essential. Identification of main-effect (M-QTLs) as well as epistatic QTLs (E-QTLs) is the key for GAB to improve above mentioned traits. Of the several successful examples of integration of genomics into breeding for improving drought tolerance, the most notable efforts include introgression of desirable QTL alleles for drought tolerance in rice (Oryza sativa) (Shen et al. 2001;Courtois 2007), QTLs for higher yield under drought stress in maize (Zea mays L.) (Ribaut et al. 2008) and ''QTL-hotspot'' harbouring QTLs for several drought tolerance traits in chickpea (Cicer arietinum) (Varshney et al. 2013b).
In the case of groundnut (A. hypogaea L.), despite presence of large morphological variation in germplasm accessions, low genetic variation was detected at the molecular level by DNA markers (Hopkins et al. 1999;Stalker and Mozingo 2001;He et al. 2003;Ferguson et al. 2004). During the last 5 years, several genetic maps were constructed and QTLs were identified for yield components and traits potentially related to drought adaptation (Varshney et al. 2009;Ravi et al. 2011;Fonceka et al. 2012;Gautami et al. 2012) including genome-wide association studies . Even though many QTLs were identified in above mentioned studies but these QTLs had low phenotypic variance. The major bottleneck remains in introgression of large number of such QTLs. In addition, complex traits are highly influenced by environments and QTLs identified at one specific location may not be valid for another location with varied environmental conditions. Thus, the QTLs identified under Indian environmental conditions (Varshney et al. 2009;Ravi et al. 2011;Gautami et al. 2012;Pandey et al. 2014) may not hold true for other SAT regions such as Eastern and Southern Africa (ESA) and West and Central Africa (WCA). Therefore, identification of QTLs for drought component traits specific to West Africa such as Niger and Senegal, where drought stress is a challenge, will have greater implications on groundnut improvement. Therefore, this study was conducted to phenotype an existing recombinant inbred line (RIL) population (TAG 24 9 ICGV 86031) for yield related traits and examine their relationship with yield under well watered (WW) and water stress (WS) conditions in Niger and Senegal, and then compare to QTLs previously identified in an Indian context. We examined reproductive characteristics (shelling percentage, maturity percentage and one hundred kernel weight) and biomass features (plant height and number of primary branches) related to haulm yield under normal and stress conditions. The genotyping and phenotyping data were used for conducting QTL analysis to identify location specific and common QTLs across locations. The identified environment specific QTLs and linked markers will improve breeding efficiency and reduce time duration for developing cultivars with enhanced stress resilience for West Africa.

Plant material
A RIL mapping population derived from a cross TAG 24 9 ICGV 86031 with 318 individuals had been phenotyped earlier at ICRISAT, Patancheru (India). The phenotyping data generated for several drought tolerance related traits at ICRISAT, Patancheru (India) revealed large variation for transpiration efficiency and its related traits (Krishnamurthy et al. 2007). Most interestingly, this population also holds uniqueness of being the basis for development of first SSR-based genetic map for cultivated groundnut and improved version of this map now possesses 191 mapped loci (Varshney et al. 2009;Ravi et al. 2011). This population was also used for QTL analysis based on the phenotyping data generated at ICRISAT, Patancheru (India). In the present study, multi-seasonal phenotyping data was generated to identify QTLs at two important locations situated in two different WCA countries i.e., Niger and Senegal. Realizing the huge variations in environmental conditions of these two locations, the soil features and environmental statistics were also recorded in detail to draw reliable inference from this study.

Growth conditions
The population was evaluated at the ICRISAT Sahelian Centre in Sadore (Niger) during 2009 and 2010, and at Centre National de Recherches Agronomiques (CNRA) at Bambey (Senegal) during 2009. The experiment was conducted under WW and WS conditions in the field. Local varieties (drought tolerant and drought sensible) were included as checks to assess the severity/extent of drought stress prevailed in a particular location/season in all the field experiments.
In Niger, the experiment was conducted during the rainy season (2009 and 2010) at ICRISAT Sahelian Centre, Sadore (13°6 0 N; 2°2 0 E; altitude 549 m) at a late planting date (August-December) to avoid the bulk of rains. These experiments have been referred to as Sad09 and Sad10 further in this manuscript. It is important to note that the soils at Sadore were arenosols (World reference base) with low pH, low water holding capacity, low inherent soil fertility and organic matter content. Therefore, fertilizer NPK (15-15-15) at a rate of 200 kg ha -1 and farmyard manure (2000 kg ha -1 ) was applied after completion of field plowing in order to grow a healthy crop. In both the experiments, a total of 284 RILs along with five check varieties were evaluated (a total of 289 entries) in a 17 9 17 alpha lattice design (LSD) with four replications. The experimental plot was consisted of two rows of 2 m length (row to row 50 cm and plant to plant 10 cm) achieving plant density of 20 plants m -1 .
In Senegal, the experiment was conducted towards the end of the rainy season 2009 (October-January 2010) at the ISRA-CNRA center, Bambey (14°42 0 N; 16°28 0 W; altitude 20 m), further referred to as Bam09 in this manuscript. The soil of the experiment had a sandy texture (910-940 g kg -1 ) with\55 g kg -1 clay content, 2.7-3.4 g kg -1 organic matter, 1.7-2.2 cmol kg -1 carbon exchange capacity and 6.5 water pH. The field was previously cultivated with pearl millet and plowed before sowing. In this experiment, no fertilizer was added at any stage of the crop. At Bambey, the trial was carried out with 294 RILs and six varieties as checks (a total of 300 entries). The test was conducted under two conditions i.e., well watered (WW) and water stress (WS) conditions. All 300 entries were planted in an alpha lattice design comprising 10 blocks and 30 plots per block with three replications. Each plot consisted of two rows of 2.10 m length (row to row 50 cm and plant to plant 15 cm) achieving the plant density of 13 plants m -1 .

Irrigation management for imposing stress
In Senegal (Sadore), planting of the mapping population was done at the end of August in 2009 and 2010. In these years, in the first 4 weeks after planting, the experimental station received 240 and 224 mm rain in 2009 and 2010, respectively. The experiment station did not receive any rain after 4 weeks of plant stand. In Senegal (Bambey), the experiment station did not receive any rain during the entire experiment. In brief, as none of the three experiments received any rain during the crucial growth stages of the plants, we had ideal condition for conducting experiments under well watered (WW) and water stress (WS) conditions.
The drought stress was imposed after flowering (i.e. at 35 days after sowing) by irrigating water stress (WS) plots alternatively (only once every two times) while the well watered (WW) plots were irrigated at regular interval. The first irrigation up to 40 mm was provided for all the plots (WW and WS) at the time of flowering. The following irrigation was supplied about 7 days later to only WW plots based on the estimated evapotranspiration. The third irrigation was supplied to all the plots (both WW and WS). The time interval for third irrigation was based on a leaf wilting assessment of the WS plots i.e., irrigation being supplied when the wilting score of a majority of WS plots reached a value of 3.0. The score of 3.0 corresponded to a stage when majority of the plants of most plots showed wilting symptom in the early afternoon hours of the day (Hamidou et al. 2012). The fourth irrigation was supplied to only WW experiment plots while the fifth irrigation supplied again to both WW and WS experimental plots. This scheme was followed until maturity. Following this irrigation scheme, the irrigation of WS plots was half of that in the WW plots after the initiation of the water stress.

Phenotypic measurements
Number of primary branches (PBr) and plant height (PH) were recorded at 60 days after sowing (DAS) at Bambey. At this stage, it was assumed that maximum plant development had reached and hence corresponds to highest transpiration rate. Plant height was estimated by measuring the length of the main stem from the plant collar to the plant apex and the same plants were used to count the number of primary branches. Phenotype of each genotype per replication was estimated as the value mean of the two sampled plants.
The SPAD chlorophyll meter reading (SCMR) measures absorbance by plant tissues of wavelengths in the visible spectrum and serves as a measure of the relative internal concentration of chlorophyll a and b. The SCMR was recorded using SPAD-502 (Minolta Corp., Ramsey, NJ, USA) in all the experiments with two plants per plot and on two fully developed leaves per plant in all the replications. Each reading was recorded on each of the four leaflets, care being taken to avoid the midrib, and averaged to correct for possible non-homogeneous distribution of chlorophyll throughout the leaf (Sheshshayee et al. 2006). The SCMR was evaluated at 15 days after stress initiation, at this time, scoring of wilting symptoms recorded was three in plots under water stress conditions.
Harvesting was carried out at about 100 DAS. Haulms can be defined as that part of the plant remaining after removing the pods. Pod weight as well as the haulm weight was estimated after air-drying the total biomass collected from each plot. Pod yield (PYLD) and haulm yield (HYLD) were expressed in grams per square meter (g -2 ). Pod weight as well as the haulm weight was used to estimate the yield performance of each genotype in all the three experiments. The harvest index (HI) indicates the efficiency of translocation of assimilates from the vegetative tissues to reproductive tissues. The HI was computed as the ratio between pod yields over the total biomass of a given genotype.
In the experiment conducted at Bambey, yield components such as shelling percentage (SP), 100 kernels weight (100 KW) and the percentage of sound mature kernels (SMK %) were also recorded. All these traits were recorded after complete drying of pods in an oven at 65°C. The percentage of maturity was based on the ratio between the numbers of mature pods out of 20 pods randomly sampled from each plot. The maturity of a pod was assessed by the blackening of the internal inner parenchyma (Miller and Burns 1971). The shelling percentage was estimated as the percentage by weight of kernels from 200 g of pods sampled from each replication. The 100 kernels weight was obtained by handpicked selections.

Genotyping of mapping population with SSR markers
A total of 3215 simple sequence repeat (SSR) markers available in the public domain and/or accessed through collaborators were earlier used to screen the parental genotypes of this population. Genotyping was done with polymorphic markers for this population which resulted in development of an improved genetic map with 191 markers loci. The complete details on parental polymorphism, genotyping of mapping population and construction of genetic map is provided in Varshney et al. (2009) and Ravi et al. (2011).

Statistical analysis of phenotypic assessments
All the phenotypic variables were tested for normal distribution. The iMAS ver 2.0 statistical package was used for analysis of variance (ANOVA) using a ReML (restricted maximum likelihood) mixed model analysis with RILs and replications 9 block as random effects. Broad sense heritability (H 2 ) was computed under the same program as follow: where r 2 g = genetic variance and r 2 e = residual variance.
The significance of genetic variance r 2 g is assuming the ratio Z ¼ s 2 g =S:E:ðs 2 gÞ to follow normal distribution asymptotically. Relationships between traits among environments were estimated by the Pearson correlation coefficients.
Quantitative trait locus (QTL) analysis QTL analysis was performed using existing genetic map information for 191 marker loci. Composite interval mapping (CIM) was conducted to identify main-effect QTLs (M-QTLs) and epistatic interaction analysis (EIA) to identify epistatic interactions between different genomic regions (epistatic-QTLs, E-QTLs) using WinQTL Cartographer, v. 2.5 (Wang et al. 2007) and QTLNetwork 2.0 (Yang et al. 2005), respectively. The threshold of LOD C2.5 was chosen for claiming putative QTL detected by WinQTL Cartographer. The mapping strategy, one-dimensional scan followed by two-dimensional search on the whole genome, was used for QTL detection with single-locus effect and epistasis effect, respectively. In each of the mapping procedures, permutation test was exploited to control genomewide false positive rate. Relative contribution was calculated as the proportion of variance caused by a specific genetic source in the total phenotypic variance.
In addition, EIA was also carried out using genotype matrix mapping (GMM 2.1) (Isobe et al. 2007, http://www.kajusa.or.jp/GMM) for identification of interactions between different loci. Two and three loci interactions were identified using GMM. All the QTLs (M-and E-QTLs) detected were anchored on the genetic map. All the M-QTLs identified and mapped were added a suffix which refers to the water regime (WW and WS), seasons (''09'' refers to 2009 and ''10'' refers to 2010) and location (letter ''B'' for Bambey and ''S'' for Sadore).

Results
The study identified several genomic regions (M-QTLs and E-QTLs) controlling yield component and drought tolerance related traits for Niger (Sadore) and Senegal (Bambey). The detailed information on the results is presented below.

Phenotypic variation and broad-sense heritability
Large variation within the RIL population was observed for all the traits across the experiments (Table 1, Figs. 2, 3). For all the experiments under water stress conditions, a significant decrease in PYLD, HYLD and HI was observed while small or no decrease was observed for remaining traits. The ANOVA revealed significant genotypic components of variance (r 2 g) and varied range of broad-sense heritability (H 2 ) (0.11-0.73) for the RIL population depending on the trait, water regime and the experiment location (Table 1). At Bambey during 2009, the heritability of PYLD, HYLD, SCMR and HI was lower under WS than under WW while it was reverse at Sadore location during 2009 and 2010. The heritability of SP, 100 KW and SMK % was higher under WS condition than achieved under WW condition. Similarly at Sadore during 2009 and 2010, the heritability of the HYLD and SCMR was higher under WS condition than WW condition. The results for heritability of PYLD and HI were not consistent at Sadore during 2009 and 2010 in either of the water regimes.

two-loci interaction for SCMR (between a locus on LG_AhI and a locus on
LG_AhXIII) detected by the QTL Network program was also detected with the GMM program. With this later program, two three-loci interactions were detected for the HYLD at Sadore during 2010. Among these interactions, single three-loci interaction involved one locus each mapped on LG_AhIII (gi-4925), LG_AhV (GM633) and LG_AhVIII (IPAHM229). The second three-loci interaction involved the loci (gi-4925, IPAHM229 and RN6F05) mapped on LG_AhV. For SCMR at Bambey during 2009, a single E-QTL identified by QTL Network was confirmed with the GMM. In addition to that E-QTL, a three-loci interaction was observed between loci mapped on LG_AhV, LG_AhXI and LG_AhXX. For the same trait but measured at Sadore during 2009, the GMM identified three different three-loci interactions and all these E-QTLs were different to those identified for SCMR at Bambey during 2009. On the contrary to the QTLNetwork program which identified no QTL, GMM program revealed two three-loci epistasic interactions for the PYLD at Sadore during 2009 (Fig. 5). In addition, GMM also revealed single twoloci interaction and two three-loci interactions for PBr while a single three-loci interaction for SP.

Discussion
Drought tolerance is one of the major constraints towards achieving full yield potential from improved groundnut cultivars. In depth understanding of this complex trait and its component traits will be beneficial in order to develop stress resilient cultivars through different crop improvement approaches. Getting precise and consistent phenotyping data on such complex traits in each varied season is very difficult due to highly fluctuating environmental and soil conditions. Selections and release of new varieties based on unreliable and inconsistent phenotyping data often leads to failure in adoption by the farmers. Under such circumstances, integration of genomics tools with ongoing genetic improvement programme will be key to develop improved cultivars with high yield under drought stress conditions. In this context, effort was made through present study to identify genomic regions associated with drought tolerance related traits which can be deployed in improving pod yield under drought stress through GAB. The results indicated involvement of several M-QTLs and E-QTLs for drought tolerance related traits for Niger and Senegal.

Phenotypic variation and broad-sense heritability
The  humidity leading to comparable lower water stress than Bambey. The above results clearly showed that variable microenvironments lead to different level of water stress which resulted in wide range of yield losses in these regions. Varied heritability was observed for different traits across different water regimes and the experiment locations. It was interesting to note that the trait heritability for PYLD, HYLD, SCMR and HI was lower under WS than WW at Bambey location during 2009. More interestingly, the above trend was in reverse direction at Sadore during 2009 and 2010. Similar results for heritability (H 2 ) in different locations were also reported for physiological, yield and yield related traits in groundnut (Painawadee et al. 2009). Not only groundnut, such results were also been reported for grain yield under WS condition in two important cereal crops i.e., rice (Lanceras et al. 2004) and wheat (Eid 2009). The heritability either remained unchanged or increased in case of SP, 100 KW and SMK % under WS condition at Bambey during 2009. This result indicated that traits such as PYLD, HYLD and HI were more prone and easily affected by water stress than SP, SMK % and 100 KW. Similarly, the heritability of the SCMR was found quite similar across different experimental sites and water regimes. Similar observations were reported in earlier experiment in groundnut (Krishnamurthy et al. 2007) which indicates that SCMR is also less prone to water stress. The above results confirm the current understanding that the heritability of complex traits show more fluctuations with the varied environmental conditions as compared to simple inherited traits. Because of this reason, the breeding lines selected in a  Phenotypic correlations and factors contributing to pod yield Results showed strong significant positive correlation between two important traits i.e., PYLD and HYLD across the experiments and water regimes. In addition, PYLD and HYLD showed significant and positive correlations with the morphological traits (PH and PBr). These correlations indicate increased pod and haulm yield due to increased plant height and number of primary branches. In all the three experiments, HYLD showed strong negative correlation with HI and SMK % in both the water regimes. In this study, no correlation between SCMR and PYLD was found in any of the experiment and water regimes. It indicates no direct role of SCMR towards PYLD. As expected, the correlation between HI and PYLD was found to be higher under WS condition than under WW condition. This may be due to better allocation of carbohydrates for pod filling in groundnut under WS condition. This correlation was reinforced by the positive and significant correlation between the HI and the SP (i.e. the ratio of the kernel weight to pod weight). Lanceras et al. (2004) observed in rice that the correlation between grain yield and HI increased with the increase in water stress levels. No significant change was observed for remaining traits.
Consistently, there was a significant and positive correlation between PYLD under WW and PYLD under WS. Therefore, good agronomic performance under water stress condition could be partially explained by good potential yield. This was in agreement with previous studies in groundnut (Songsri et al. 2008a, b;Painawadee et al. 2009;Ntare and Williams 1998;Hamidou et al. 2012). Furthermore, this study showed highly significant and positive relationship for all the six traits (PYLD, HYLD, HI, SMK %, SP and SCMR) across water regimes and locations. Consistently over all the experiments, correlation was always lower for PYLD as compared to other traits indicating the highly influential nature of this trait. The pod yield under water-limited condition was mainly determined by the HI and in some extent by the above biomass (HYLD). We observed a significant and positive relationship of HI with SMK % and SP. This indicates that the genotypes with better efficiency of translocation would have a better pod maturity percentage and/or a higher shelling percentage. In this study, we also noted a very highly significant (P \ 0.001) and negative correlation between the HI under WS condition and the PBr under WW condition. From a breeder's point of view, HI can serve as a means to predict yield and seed quality in groundnut. This result is in agreement with previous report on groundnut (Ratnakumar and Vadez 2011;Hamidou et al. 2012). These results are consistent with those reported in other crop species such as maize (Edmeades et al. 1999;Lafitte and Edmaedes 1997) and rice (Lanceras et al. 2004). Contrary to the earlier study (Painawadee et al. 2009) that reported correlation between SCMR and PYLD under WS condition, we did not found any significant correlation. However, we found a significant and positive correlation between HI and SCMR which indicates HI may have an indirect correlation with PYLD under WS via its effect on the HI.
The performance of RILs varied significantly between different locations (Fig. 2) as well as between seasons of the same location (Fig. 3). This is due to large genotype by environment (G 9 E) interactions for complex traits such as PYLD and HYLD. Therefore, since a combined environment QTL analysis was not logical under such circumstances, QTL analysis was carried out for each single environment. These results also caution selecting drought tolerant genotypes across locations with such large variations (Wright et al. 1996;Hamidou et al. 2012).

Main-effect QTLs (M-QTLs)
QTL analysis resulted in identification of a total of 51 main-effect QTLs (M-QTLs) for nine different traits under WS condition (27 M-QTLs) and WW condition (24 M-QTLs). Many QTLs detected for different traits were overlapping in some specific genomic regions. The PVE of these M-QTLs were low indicating minor effect QTLs. Small-effect QTLs for complex traits such as yield were also been identified in rice (Oryza sativa L.) (Zhou et al. 2011) and maize (Zea mays L.) (Ribaut et al. 1997). These results confirm the current understanding that the complex traits are highly polygenic in nature and are the outcome of several small-effect genetic factors. Indeed the contribution of each locus may be negligible but the total contribution is usually significant.
Of the total QTLs identified, a QTL for SCMR detected at Bambey on the LG_AhVIII coincided with a QTL detected at Patancheru (India) by Ravi et al. (2011) using the same mapping population. Another QTL for SCMR detected at both Sadore 2009 and 2010 was also detected at Patancheru (India) by Ravi et al. (2011). No consistency for QTLs across locations and years was observed for any trait considered in this study. Such inconsistency of the QTLs between water regimes, locations and years was also previously reported (Ribaut et al. 1997;Martin et al. 1994). The results indicated inconsistency of QTLs of a given trait between experiments even under quite similar conditions. This inconsistency across seasons and locations was due to presence of high environmental influence in addition to varied soil moisture conditions. It was observed that the G 9 E variance component was larger than genotypic variance under water stress condition and been observed previously in groundnut (Hamidou et al. 2012), rice (Pantuwan et al. 2002) and tropical maize (Cooper et al. 2006). In addition to climatic conditions, which were quite different from one experiment to another in this study, the physical and chemical soil properties may have a great effect on the G 9 E of the yield and its components. These results suggest there is a need to understand the cause for these interactions.

Pleiotropism and linked genes
Most importantly, this study identified genomic regions containing several QTLs for different yield and yield component traits i.e., QTL clusters at several specified genomic regions. Despite significant correlations between different traits, common QTL genomic regions were not identified for highly correlated traits indicating distribution of these small-effect QTLs across the genome. If pleiotropism is the major reason, the coincidence of both the locations of QTLs for related traits and the directions of their effects can be expected. In the study, therefore, pleiotropism would explain the relation between PH and the HYLD under WW condition as indicated in LG_AhVI,LG10 and LG_AhXVII. If the simple close linkage is the major reason, the directions of the genetic effects for different QTLs may be different, although the coincidence of the locations can still be expected (Zhuang et al. 1997). It seems to be the case in our study between the QTL for the SMK % and PBr under WW condition on the LG_AhIX, between the QTL for SP under WS and the QTL for PH under WW. The detection of putative QTLs for different traits, in common regions, under the same or different environmental conditions, highlights the relation between these parameters and suggests that these regions are involved in several aspects of the determinisms to drought tolerance. Classical quantitative genetics assumes that trait correlation can be attributed to the effect of pleiotropy or to the tight linkage of genes (Hittalmani et al. 2003).

Epistatic effect QTLs (E-QTLs)
The epitasis analysis with QTLNetwork revealed additive 9 additive interactions for different traits while the GMM software revealed two-loci and threeloci interactions for some traits. It was important to note that the same two-loci interaction for SCMR (between a locus on LG_AhI and a locus on LG_AhXIII) was detected by both the softwares (QTLNetwork and GMM). Similar to M-QTLs, the E-QTLs also showed low PVE range indicating minor E-QTLs and also been earlier detected at Indian conditions (Ravi et al. 2011). The low PVE is adding weight to evidence that all these traits are quantitatively inherited including the morphological traits. In addition, inconsistency of the E-QTLs between locations and years was also observed similar to M-QTLs. Isleib et al. (1978) previously reported inconsistency of allelic interactions detected between locations and years. This result is also in agreement with the findings of Upadhyaya and Nigam (1998) who reported epistasis in 11 vegetative and reproductive traits in groundnut and interactions between epistasis and environment using a triple test cross.

Conclusions
Heritability and correlation information provide guidelines for selection of traits in crop improvement.
Among the traits analyzed under water stress condition, the harvest index and the haulm yield were more correlated with the pod yield and have intermediate broad sense heritability. Based on QTL analysis, all the nine traits analyzed were controlled by additive and multiplicative effects. The QTLs detected explained a low proportion of the phenotypic variance. Some of these QTLs over-lapped with previously reported drought tolerance QTLs detected in this population. Therefore, ignoring E-QTLs could result in significant bias which may limit progress in any breeding program for these traits. Fortunately, new approach like marker-assisted recurrent selection (MARS) and genomic selection (GS) allowed mining and advancing best lines to avoid as much as possible the effects of undesirable alleles.