Due to the multiplicity of genes and their partial effects on phenotypic variation, the candidate gene approach is suggested to be more suitable for QTL characterization than genome wide scanning or positional cloning [37, 38]. Molecular-linkage maps based on functional gene markers (molecular-function maps) are a prerequisite for a candidate-gene approach to identify genes responsible for quantitative traits at the molecular level. Therefore, plant function maps have recently been generated in many species [28–30] and candidate genes have been identified for various biotic stress resistance and abiotic stress tolerance QTLs using this approach [39–43]. Considering this, and the fact that hitherto little information on gene-based SNPs is available in pearl millet , the present study was undertaken to develop a resource of mapped gene-based SNPs for pearl millet and to identify putative candidate genes underlying a major validated drought tolerance QTL.
We placed 69 gene-based SNPs and CISPs onto existing SSR-based skeleton map of pearl millet based on the cross H 77/833-2 × PRLT 2/89-33. Although the number of markers mapped earlier on this cross is relatively large, a high percentage of the markers are anonymous sequences and/or exhibit dominant patterns of inheritance [1, 6, 8]. Recently, attempts have been made to enrich the existing pearl millet maps with co-dominant genomic and EST-derived SSR markers [6, 7]. The limitation of genomic SSRs is their low cross-species transferability due to either disappearance of the repeat region or degeneration of the primer binding sites. Although cross-species PCR amplification of EST-SSRs is more successful compared with genomic SSRs, their polymorphism rates are, however, very low. The SNP markers, as developed in this study, provide many benefits over SSRs, including their abundance in the genome, frequent occurrence in coding regions of the genes, and their ease of analysis and unambiguous results across various platforms [35, 45]. Every SNP in single copy DNA is potentially a useful marker.
A total of 228 SNPs were obtained in 30.5 kb sequenced region resulting in an SNP frequency of one SNP per 134 bp (Table 1). In maize, an out-crossing species like pearl millet, SNP frequency of one SNP per 61 bp was observed with 18 gene fragments analysed in 38 inbred lines . In another study on maize, SNP frequency of 1 SNP per 73 bp was observed with analysis of 592 unigenes in 12 inbred maize lines . For out-breeding forage grass Lolium perenne, a frequency of 1 SNP per 54 bp was observed in the analysis of 100 candidate genes . In rye and sugar beet, the estimated SNP frequency was 1 SNP per 58 bp  and 1 SNP per 72 bp , respectively. Thus, the polymorphism frequency determined for pearl millet is lower than those of other out-crossing species. Selection of germplasm is one of the major factors that affect SNP frequency. A contrasting variation in SNP frequency was reported in two different sets of germplasm of maize having different number of accessions [24, 49]. Higher SNP frequencies have generally been reported in studies involving a large number of diverse accessions [24, 26, 48, 50]. Compared to other studies, the number of genotypes used for SNP discovery was very small in the present study. Moreover, two (ICMR 01029 and ICMR 01004) of the four pearl millet genotypes used in this study are QTL near-isogenic lines generated in the background of H 77/833-2. The other important factor that affects SNP frequency between different plant species is the differences in genomic region(s) assayed e.g., coding regions, promoters, introns or untranslated regions (UTRs) . The sequence set in the present study is not complete with respect to any of these categories; in general, this study targeted both exonic regions and 3' -UTRs. Considering this and the fact that the limited number of EST sequences were analysed, the frequency estimates in this study might not reflect the exact picture for pearl millet.
The level of attrition from marker discovery to genetic map assignment in the present study was observed for only 8 SNPs (13.3%). In Lolium perenne, the same approach led to assignment of only 40% of the genes on the genetic map . The addition of gene-based markers extended the genetic map of pearl millet by 125 cM; the total length of the combined SSR, SNP and CISP marker-based map being 815.3 cM. Most importantly, 63 gene-based markers mapped to positions different to framework markers, thus enriching the map with new candidate gene loci (Figure 1). Further, some of these candidate gene loci filled the large gaps present in some linkage groups of the framework map. For instance, the framework map had the largest (54.8 cM) gap on LG4 between markers Xpsmp2086 and Xipes0186. Six gene-based markers (Xibmsp57, Xibmsp5, Xibmsp6, Xibmcp4, Xibmsp17, and Xibmsp22) mapped between Xpsmp2086 and Xipes0186 on LG4 (Figure 1).
Interestingly, the recombinant inbred line (RIL) mapping panel used in the present study revealed high rates of SDRs with both framework SSRs and gene-based SNP and CISP markers. Distortions toward either of the parental allele were observed. Such segregation distortion due to an excess of one of the parental allele has also been reported in essentially all previous studies in pearl millet [5–7, 9]. Qi et al. , for instance, observed SDRs due to an excess of one of the parental allele on LG 4 in the cross 81B × ICMP 451, on LG 4, LG5 and LG7 in LGD 1-B-10 × ICMP 85410, on LG2 and LG4 in PT 732B × P1449-2, and on LG3 and LG6 in ICMB 841 × 863B. However, segregation distortion in the RILs used in the present study was significantly higher compared to all previous studies in pearl millet where F2 populations were used [5–7, 9]. Higher segregation distortions in RILs compared to F2 or other early generations has been reported in other crops as well [52, 53]. For example, in a comparative study carried out to explore segregation distortion of molecular markers in different mapping populations (F2, backcross, doubled-haploid and RILs) in rice, consistently more segregation distortion was found in RILs than in doubled-haploid, backcross, or F2 populations . It has been suggested that more generations result in a stronger effect of selection and therefore segregation distortion accumulates with additional generations of meiosis . Preferential transmission of parental alleles could be caused by an allele-specific advantage in viability or fertility, and gene-based markers may represent or be linked to alleles selected during the six generations used for development of the RILs.
The LG2, the prime target of this study, was saturated with 24 new gene-based markers (Figure 1). Of these, 20 were SNPs derived from genes coding for actin depolymerising factor, PSI reaction centre subunit III, PHY C, actin, alanine glyoxylate aminotransferase, uridylate kinase, acyl CoA oxidase, dipeptidyl peptidase IV, Zn finger C × 8-C × 5-C × 3-H (or CCCH type), serine/threonine protein kinase, a homolog of rice flowering time gene HD3, MADS-box, acetyl CoA carboxylase, ubiquitin conjugated enzyme, photolyase, catalase, serine carboxypeptidase III precursor and three hypothetical proteins. The four CISP markers on LG2 represented genes coding for a chlorophyll a/b binding protein, protein phosphatase 1 regulatory subunit SDS22, transmembrane amino acid transporter and a phosphoglycerate kinase. The pearl millet SNP map for LG2 was generally consistent with chromosome-level pearl millet-rice synteny (pearl millet LG2 syntenic to rice chromosomes 2S, 3L, 6S and 10S) previously determined with RFLP markers . For example, three genes retrieved from rice chromosome 2S (serine/threonine-protein kinase, LOC_Os02g57080; serine carboxypeptidase III precursor, LOC _Os02g02320; phosphoglycerate kinase, LOC_Os02g07260) and three others from 6S (acyl CoA oxidase, LOC_Os06g01390; zinc finger C- × 8-C- × 5-C- × 3-H type, LOC_Os06g21390; transmembrane amino acid transporter, LOC_Os06g12320) mapped on LG2 of pearl millet. In addition to LG2, pearl millet-rice synteny observed in the present study was also consistent with previous study  for other linkage groups. However, a few loci retrieved from rice mapped to nonsyntenic positions on pearl millet linkage groups. For example, uridylate kinase and acetyl CoA carboxylase from rice chromosomes 1S and 5S, respectively, mapped on LG2 of pearl millet. Similarly, vacuolar ATPase subunit C from rice chromosome 5 L mapped on LG4 of pearl millet. Such observations have been reported in other crops such as barley  and sorghum  and are not surprising given that the rice genome has undergone large segmental, as well as individual gene duplications, mostly after the divergence of rice and Triticeae ancestors.
Exploiting markers common between the present consensus map and other linkage maps of pearl millet, the position of major DT-QTLs have been added to the present function map to identify potential candidate genes for drought tolerance (Figure 1). Eighteen gene-based markers were localised in the support interval of major DT-QTL region on LG2 (Figure 1). Of these, ten genes have been reported to play important roles in regulation [transcription factors like zinc finger C × 8-C × 5-C × 3-His type (or CCCH type), MADS-box], signal transduction (serine/threonine protein kinase, protein phosphatase 1 regulatory subunit SDS22), energy and carbon metabolism (genes for photosynthesis, photorespiration and β-oxidation such as PSI reaction center subunit III, chlorophyll a/b binding protein, alanine glyoxylate aminotransferase, acyl CoA oxidase), purine and pyrimidine nucleotide biosynthesis (uridylate kinase), and lipid biosynthesis (acetyl-CoA carboxylase) under drought and osmotic stresses [55–63]. The presence of transcription factors belonging to Zn finger CCCH type and MADS-box gene families in support interval of major DT-QTL region is noteworthy. These transcription factors gene families have been reported to activate cascade of downstream genes that act together in enhancing tolerance to multiple stresses [57, 61–65]. Among the different types of Zn finger families, role of C2H2 type Zn finger gene families in drought stress tolerance has been functionally validated in rice and Arabidopsis [64, 65]. However, CCCH types Zn finger proteins are poorly characterized in plants under drought stress. The best characterized CCCH-type zinc finger proteins in plants are OsDOS in rice , AtSZF1 and AtSZF2 in Arabidopsis  and GhZFP1 in cotton  under salt and fungal stresses. The CCCH type Zn finger in pearl millet shows significant homology with RING finger types OsC3H41 and AtC3H69 of rice (BlastX; 2e-100) and Arabidopsis (BlastX; 1e-65), respectively, the members of which have been reported to be regulated by various biotic and abiotic stresses including water stress induced by mannitol .
The MADS-box family, identified initially as floral homeotic genes, is one of the most extensively studied transcription factor gene families in plants [57, 69]. The most striking feature of the MADS-box gene family is the diverse functions taken up by its members in different aspects of plant growth and development including flowering time control . Different members of MADS family have been reported to be induced under drought stress in rice [57, 70], maize  and wheat . The MADS box gene in pearl millet shows significant homologies with MIKC type MADS-box genes of Triticum aestivum (BlastX; 2e-15), Zea mays (MADS22, BlastX; 2e-15) and Brachypodium (MADS22, BlastX; 4e-15). The homologue of MADS22 in rice, OsMADS22, has been reported to be up-regulated by more than two-fold in response to dehydration stress . In pearl millet, polymorphism in the MADS-box gene MADS11 has been reported to be associated with flowering time variation . Studies have shown that a large number of genes involved in flower development are associated with abiotic stress responses [74, 75]. The significant association of MADS box gene with flowering time and grain yield QTLs in pearl millet under drought stress (Table 3) suggests this to be another strong candidate gene for DT-QTL in pearl millet.
Similarly, the candidacy of serine/threonine protein kinase and acyl CoA oxidase in the DT-QTL interval is supported by expression evidences of these genes in pearl millet . A ~10 fold increase in expression was obtained for both serine/threonine protein kinase and acyl CoA oxidase in pearl millet seedlings subjected to drought stress . Another important gene mapped in the support interval of major DT-QTL was that coding for acetyl-CoA carboxylase (ACC). In plants, ACC isozymes provide malonyl CoA pools used for de novo fatty acid synthesis in plastids and mitochondria, and for fatty acid elongation, flavonoid and stilbene biosynthesis in the cytosol . ACC reaction is the most important regulatory step, controlling metabolite flow in response to stress. From the water-deficit stress tolerance perspective, fatty acids are essential in membrane biogenesis, lipoic acid and cuticular wax synthesis and stress signalling .
The candidate genes, identified in the present study, were significantly associated with QTLs of grain yield, flowering time and leaf rolling under drought stress (Table 3) thus confirming their associations with drought tolerance phenotype(s) in pearl millet (Table 3). Such mapping of candidate genes also offer a range of possible links to the other physiological  and agronomical [4, 5, 15] traits including salt tolerance  that co-map with the major LG2 DT-QTL region. Similar approach has been used in other crops to find positional candidate genes underlying QTLs [28, 40, 42, 81]. For example, a molecular function map with 85 loci was constructed in potato based on 69 genes involved in carbohydrate metabolism and transport to identify the candidate genes for tuber starch content . Similarly, 16 transcription factor genes were integrated on barley framework map and drought and cold tolerance QTLs were positioned on the consensus map to find positional candidate transcription factors for drought and cold tolerance . In the latter study, emphasis was given to transcription factors and upstream regulators, rather than to structural genes. In the present study, however, we have assembled sequences from both structural and transcription factor genes to gain a more complete picture of the distribution of abiotic stress genes around pearl millet DT-QTL interval and across the genome.