Translational Genomics in Agriculture: Some Examples in Grain Legumes

Abstract

Recent advances in genomics and associated disciplines like bioinformatics have made it possible to develop genomic resources, such as large-scale sequence data for any crop species. While these datasets have been proven very useful for the understanding of genome architecture and dynamics as well as facilitating the discovery of genes, an obligation for, and challenge to the scientific community is to translate genome information to develop products, i.e. superior lines for trait(s) of interest. We call this approach, “translational genomics in agriculture” (TGA). TGA is currently in practice for cereal crops, such as maize (Zea mays) and rice (Oryza sativa), mainly in developed countries and by the private sector; progress has been slow for legume crops. Grown globally on 62.8 million ha with a production of 53.2 million tons and a value of nearly 24.2 billion dollars, the majority of these legumes have low crop productivity (<1 ton/ hectare) and are in the developing countries of sub Saharan Africa, Asia and South America. Interestingly, the last five years have seen enormous progress in genomics for these legume crops. Therefore, it is time to implement TGA in legume crops in order to enhance crop productivity and to ensure food security in developing countries. Prospects, as well as some success stories of TGA, in addition to advances in genomics, trait mapping and gene expression analysis are discussed for five leading legume crops, chickpea (Cicer arietinum), common bean (Phaseolus vulgaris), groundnut (Arachis hypogaea), pigeonpea (Cajanus cajan) and soybean (Glycine max). Some efforts have also been outlined to initiate/ accelerate TGA in three additional legume crops namely faba bean (Vicia faba), lentil (Lens culinaris) and pea (Pisum sativum).

Keywords:

• biotic and abiotic stresses
• genomic resources
• genomics assisted breeding
• legumes
• next generation genomics
• predictive markers

I. INTRODUCTION

The United Nations Department of Economic and Social Welfare in its demographic estimates and projections, predicts that the global human population will reach 9 billion by 2050. To feed this growing population, agricultural production must be 60% higher than current production. This will be a challenge in the face of climate changes such as higher temperatures, shifting seasons, more frequent and extreme weather events, flooding, and drought. The dramatic increases in food production and land productivity during 1960–1990s, due to the “green revolution”, has reached a plateau (www.fao.org). Scientists have been striving to use conventional breeding approaches to enhance production and productivity of crop plants, thus emphasizing the need for new approaches to meet food and nutritional security for the future (Tester and Langridge, 2010).

With the advent of molecular markers and recent advances in genomics research, it has been possible to utilize genomics for enhancing the precision and efficiency of crop breeding. This process has been referred as “genomics-assisted breeding” (GAB; Varshney et al., 2005). GAB includes deployment of structural, functional as well as comparative genomics to identify molecular markers including functional markers, candidate genes, and predictive markers for breeding. Several kinds of molecular markers such as simple sequence repeats (SSR) and single nucleotide polymorphism (SNP) markers have been used for trait dissection and for enhancing precision in selection in plant breeding programs (Xu et al., 2012). Further, advanced research institutes (ARIs) and private sector organisations, mainly in developed countries, have made significant progress towards deploying GAB for enhancing yield under abiotic and biotic stresses (Varshney et al., 2013a). For instance, downy mildew resistant pearl millet (HHB67-2 improved; http://www.icrisat.org/impacts/impact-stories/icrisat-is-hhb67.pdf), submergence tolerant rice (Swarna; Neeraja et al., 2007), bacterial blight resistant rice (Triguna; Sundaram et al., 2009) and quality protein maize (VivekQPM9; Gupta et al., 2009; 2013) are some products of marker-assisted selection (MAS) or GAB in public sector. Improved lines have been developed, in fact, for many different crop plants (see Gupta et al., 2009; Kulwal et al., 2011).

Understanding the structure, organization and dynamics of genomes in plant species can provide insights into how genes have been adapted by natural and artificial selection to respond to environmental constraints and the potential for their manipulation for crop improvement. In pursuit of these goals, several plant genomes have been sequenced. Because of the high cost of Sanger sequencing and limited expertise in the plant science community, only a few plant species were initially selected as ‘models’ and targeted for both sequencing and the development of functional genomics resources. One purpose of generating genome sequences for model plant species was to better understand genome architecture and to create a ‘parts list’ for a plant. There were high hopes that much of this information could be transferred and utilized in crop species, generally referred as ‘comparative genomics’ (Ahn and Tanksley, 1993; Gale and Devos, 1998). The idea was prevalent that ‘model to crop’ translation would be a powerful tool for the application of genetics and genomics in crop plants; occasionally this approach was also referred to as ‘translational genomics’ (Elma et al., 2007). In part the idea that ‘model to crop translation’ could solve the problems of agriculture was driven by an idea, prevalent in the 1990s, that transgenic solutions to the constrains that face crop species would be simple to find without the need for extensive genetic and genomic analyses of these target crop species. This was a seductive simplification with obvious attractions for lowered research costs.

As a result of emphasis on model plants, large-scale genome sequence information was first generated for Arabidopsis thaliana (The Arabidopsis Genome Initiative, 2000), followed by rice (Oryza sativa) both as model as well as crop (International Rice Genome Sequencing Project, 2005), poplar (Populus trichophora; Tuskan et al., 2006), grape (Vitis vinifera; Jallion et al., 2007), maize (Zea mays; Schnable et al., 2009) and later the first legume genome sequence, of soybean (Schmutz et al., 2010). Recent advances in next generation sequencing (NGS) technologies, with increased throughput and reduced sequencing costs, dramatically changed the sequencing scenarios for plant genomes (Deschamps and Campbell, 2010; Thudi et al., 2012; Varshney et al., 2009a). To date, the genome sequences for more than 55 plant species have been produced (Michael and Jackson, 2013) and many more are being sequenced (http://www.onekp.com/). The genome sequence information available has enabled the identification and development of genome-wide markers and a range of marker genotyping platforms including KASPar assays, Sequenom, BeadXpress System, GoldenGate assays, Infinium assays, genotyping-by-sequencing, etc. In addition, several organizations have started to re-sequence large-scale germplasm collections (Branca et al., 2011; Lam et al., 2010; Varshney et al., 2013a; Xu et al., 2011). It is important to mention that advances have not been limited to genome sequences; data at the transcriptome and epigenome level have also contributed greatly. NGS technology has become routine now for transcriptome sequence (Hiremath et al., 2011), gene expression analysis (Hoen et al., 2008; Wang et al., 2012), and epigenetics (Jackson et al., 2011), not only for model plants but in crop species too.

It is evident from the above that we are living in sequence-rich or genome information era. For the majority of crop species, this transition occured during the past 10 years. Before this transition, the breeding community could not initiate genomics-assisted breeding due to lack of genome information. Moreover, it is still difficult for the breeding community today to navigate and interpret sequence data in order to devise strategies for crop improvement. In our opinion, the plant science community has a challenge as well as an obligation to translate this genomic information for improved agriculture to feed the anticipated 9 billion people in the next 35 years. We propose to call the systematic application of genome wide sequence information in support of crop improvement as ‘translational genomics for agriculture’ (TGA). Because of its importance and the attention of global plant research community, including the private sector, TGA has been successfully applied in several cereals; maize and rice are the crop species where TGA is most advanced.

In addition to cereals, legumes play important roles both in food and nutritional security as well as in the sustainability of agricultural systems. Among legume species, major food legumes are chickpea (Cicer arietinum), common bean (Phaseolus vulgaris), cowpea (Vigna unguiculata), faba bean (Vicia faba), groundnut (Arachis hypogaea), lentil (Lens culinaris), pea (Pisum sativum), pigeonpea (Cajanus cajan) and soybean (Glycine max). Globally these legumes are grown on 62.8 million ha with a production of 53.2 million tons and fetch about 24.2 billion dollars (http://www.icrisat.org/crp/CRP3.5_Grain_Legumes_15Aug12.pdf). However, the crop productivity of these legumes crops in farmers’ fields in developing countries is generally less than 1 ton per hectare. These statistics indicate that in parallel to improving agronomic practices, advancing genomics research as well as enhanced deployment of TGA in these legume crops is urgently needed.

This review discusses recent advances in genomics research as well as the prospects and potential of TGA in eight leading legume crops namely chickpea, common bean, faba bean, groundnut, lentil, pea, pigeonpea and soybean.

II. NEXT GENERATION GENOMICS

With an objective to understand basic plant biology and use that information to understand and improve important crop species, a few plant species were considered as “model systems.” Their ease of manipulation, simple genome organization, patterns of inheritance, rapid life cycle, desirable experimental properties etc., were important features for a model system (Koebner and Varshney, 2006). Model species were supposed to share reasonably good synteny and orthology with genomes of related crop species and help to facilitate discovery of genes and association of genes with phenotypes.

Arabidopsis was the first model plant species for which a genome sequence was produced (Arabidopsis Genome Initiative, 2000). However, with the argument that Arabidopsis may not be good model (Devos et al., 1999) and, in particular, does not represent monocot plant species, rice was selected as a second model system for monocot plant species, especially cereals. As rice is also a crop species, the crop community got excited by the prospect of sequencing the rice genome so that rice genes, especially those involved contributing to yield, could be used for crop improvement in this and other plant species. As a result, two draft genomes, one each for japonica and indica rice, were released simultaneously (Goff et al., 2002; Yu et al., 2002).

Legumes are quite different from other plant species in certain features such as symbiotic nitrogen fixation, protein rich physiology and secondary metabolism. Therefore it was argued that the model systems, i.e. Arabidopsis and rice, did not serve as good models for legumes (Cook, 1999; Li et al., 2012). Most notably because of the connection between mycorrhization and nodulation in legumes (Duc et al., 1989), the former being of wide significance, but not accessible using A. thaliana as a model, and also for the understanding of plant architecture regulation by strigolactones (Gomez-Roldan et al., 2008). For these reasons, two legume plant species, Medicago truncatula (Barker et al., 1990) and Lotus japonicus (Handberg and Stougaard, 1992) were selected as model systems for legumes. Subsequently after spending several years time, and much money, genome sequences were assembled for Lotus (Sato et al., 2008) and Medicago (Young et al., 2011).

Although genome sequencing used to be restricted to model systems, the advent of NGS technologies and increased expertise in sequence assembly in the scientific community has led to genome projects for other crops. The main rationale behind targeting crop species for sequencing is the often complex genomic composition of crop species and past experience has shown that comparative genomics, epitomized by the ‘model to crop translation’ approach, was not very useful for crop improvement (Koebner and Varshney, 2006). An illustration of the sequencing of some key model and crop plant species is shown in Figure 1.

FIG. 1. A snapshot on sequencing status of key model plant/ crop genomes. This figure shows sequencing status of some key model plant as well as crop species including legumes. Initiation of the sequencing project has been shown by dotted lines with the year and % genome assembled has been showed as filled circles.

After work began on the model legumes Medicago and Lotus discussed above, soybean was the first crop legume for which reference genome was assembled (Schmutz et al., 2010). Legumes like chickpea, common bean and pigeonpea, constitute an important dietary component for the vast majority of populations in many developing countries but did not immediately benefit from the genomic revolution (Varshney et al., 2010a). However, as a result of community efforts, draft genome sequences have become available recently for pigeonpea (Varshney et al., 2012a) and chickpea (Varshney et al., 2013a). These legumes, once considered orphan crops, are no-longer so and now have joined the group of genome resource-rich species (Varshney et al., 2010b). Recently the common bean community has developed genome assemblies for two accessions (Scott Jackson, personal communication). Significant efforts are underway in sequencing other important legume genomes such as peanut (http://www.peanutbioscience.com/), pea (http://www.coolseasonfoodlegume.org/pea_genome) and lentil (Judith Burstin, Bert Vandenberg, personal communication). The genome sequences of model legumes, together with those of the above mentioned crops, serve as references to help assemble and organize sequence scaffolds from the related legume crop species (Cannon et al., 2009).

In addition to the generation of reference genome sequences, NGS has facilitated re-sequencing of genomes of several varieties/ cultivars/ accessions in order to identify genomic variation, such as SNPs or structural variants, and to develop haplotype maps (Imelfort et al., 2009; Lorenc et al., 2012). For example in the model legume Medicago, 26 genomes from diverse accessions were sequenced and identified >3 million SNPs and characterized sequence polymorphisms and patterens of linkage disequilibria (LD) (Branca et al., 2011). In the case of soybean, 17 wild and 14 cultivated soybean genomes were re-sequenced identifying 205,614 SNPs that revealed patterns of genetic variation between wild and cultivated soybeans (Lam et al., 2010). Re-sequencing of 90 chickpea genomes revealed 4.4 million vari­ants (SNPs and INDELs) (Varshney et al., 2013a). Re-sequencing enables rapid identification of SNP markers, development of high density genetic maps, and creation of haplotype maps (HapMap) for genome wide association studies (GWAS).

III. TRANSCRIPTOME ASSEMBLIES, FUNCTIONAL MARKERS AND GENE EXPRESSION ATLAS

Global gene expression studies provide insights into the transcriptome at the whole genome level in terms of the response to a particular biological conditions or constraints. These studies, even in the absence of a complete genome sequence, can help in understanding gene function, transcriptional programs, and the molecular basis of various cellular processes. Much work was undertaken in the area of transcriptome and gene expression analysis even before genome sequences became available. Initially, and for the majority of cases, either Sanger sequencing based ESTs or microarray based gene expression analysis was conducted. Using gene expression studies and gene cloning, a number of candidate genes for resistance/ tolerance to several biotic/ abiotic stresses have been isolated or cloned from several legume crops (Table 1). In recent years, due to the advent of the NGS technologies, it has become possible to develop de novo transcriptome assemblies with low cost and indeed NGS technologies are being used for gene expression analysis. Transcriptome assemblies developed using either Sanger sequencing or NGS technologies, or a combination of both, have been used for gene annotation for genome sequencing projects. These transcriptome assemblies provide a large number of functional markers (derived from genes/ transcripts) such as SSRs, SNPs, conserved orthologous sequence (COS) and intron spanning region (ISR) markers (Dubey et al., 2011; Hiremath et al., 2011; Jhanwar et al., 2012; Kudapa et al., 2012).

TABLE 1 Some candidate genes identified in five legume crops

Crop	Candidate gene(s)	Stress/trait(s)	Approach	Reference
Chickpea	CarNAC3	Drought and developmental processes	cDNA library	(Peng et al., 2009)
	CarF -box1	Drought, salinity and developmental processes	Molecular cloning	(Jia et al., 2012)
Common bean	αAI-Pv	Bruchid beetles	Overexpression	(Shade et al., 1994)
Groundnut	AhLPAT	Triacylglycerol biosynthesis	Homology based	(Chen et al., 2009)
Pigeonpea	CcHyPRP	Drought	Overexpression	(Priyanka et al., 2010)
	CcCYP	Drought, salinity and extreme temperatures	Overexpression	(Sekhar et al., 2010)
	CcMT1	Heavy metal stress	Overexpression	(Sekhar et al., 2010)
Soybean	GmRPS1, GmRPS6, GmRPL37	Cold stress	SuppressionSubtractiveHybridization	(Kim et al., 2004)
	P5CR	Drought and heat	Overexpression	(de Ronde et al., 2004)
	GmDREBa, GmDREBb, GmDREBc	Salinity, drought and cold	Expression analysis	(Li et al., 2005a)
	GmMIPS1	Seed development	RNAi-mediated silencing	(Nunes et al., 2006)
	GmNAC	Cold and developmental processes	Molecular cloning	(Meng et al., 2007)
	GmEREBP1	Defense response	Overexpression	(Mazarei et al., 2007)
	GmbZIP44, GmbZIP62 or GmbZIP78	Salt and freezing tolerance	Overexpression	(Liao et al., 2008)
	GS52	Nodulation	RNAi-mediated silencing	(Govindarajulu et al., 2009)
	Rpp4C4	Asian soybean rust	virus-induced gene silencing	(Meyer et al., 2009)
	GmDREB3	Cold, drought and salt	Overexpression	(Chen et al., 2009)
	GmOLPb, P21e	Salinity, methyl jasmonate and salicylic acid	Expression analysis	(Tachi et al., 2009)
	Glyma14g38510, 38560 and 38580	Soybean mosaic virus	Homology based	(Wang et al., 2011)
	Glyma03g28610/ Glyma03g 28630	Iron deficiency chlorosis	Transcriptome sequencing	(Pieffer et al., 2012)
	Glyma07g05410/ Glyma16g01980	Flowering time/maturation	Homology based	(Kim et al., 2012)
	GmDGAT2C/ GmDGAR1B	Seed yield	Homology based	(Eskandari et al., 2013)

Using the Roche 454-FLX platform, a transcriptome assembly in M. truncatula was generated with a total of 184,599 contigs (Cheung et al., 2006) and this assembly was found to contain 610 SSRs. In the case of Lotus, a tissue-specific transcriptome analysis containing 21,495 transcripts was generated (Takanashi et al., 2012). Among legume crop species, several versions of transcriptome assemblies were generated by different research groups (Table 2). For instance, in the case of chickpea, several transcriptome assemblies based on different sequencing platforms have been developed (Garg et al., 2011a; 2011b; Hiremath et al., 2011). These assemblies together with unpublished data from the University of Saskatchewan have been used for defining a comprehensive trasncriptome assembly comprising of 46,369 contigs with an average N50 of 1,726 bp (Kudapa et al., 2014). Details of some key transcriptome assemblies based on NGS technology or combination of NGS technologies with Sanger ESTs in select legume species have been given in Table 2. The Legume Information System (LIS, http://www.comparative-legumes.org/) hosts transcriptome assemblies for several legume species.

TABLE 2 NGS-based transcriptome assemblies and genomic resources

Crops	Platform	Unigenes/ transcripts	SSRs	SNPs	Reference
Chickpea	Illumina	53,409	4,816	-	(Garg et al., 2011a)
	454FLX	34,760	4,111	495	(Garg et al., 2011b)
	454FLX and Illumina	1,03,215	26,252	26,082	(Hiremath et al., 2011)
	454FLX	1,60,883	-	1,022	(Gaur et al., 2012a)
	454FLX and Illumina	-	-	14,454	(Hiremath et al., 2012)
	454FLX	37,265	4,072	36,446	(Jhanwar et al., 2012)
	454FLX and Illumina	43,389	5,409	39,940	(Agarwal et al., 2012)
Common bean	454FLX and Illumina	1,60,036	-	3,487	(Hyten et al., 2010)
	454FLX	59,295	1,516	-	(Kalavacharla et al., 2011)
Groundnut	454FLX	7,43,232	2,884	-	(Guimarães et al., 2012)
	Illumina	59,077	5,883	-	(Zhang et al., 2012)
Pigeonpea	454FLX	43,324	3,771	-	(Dutta et al., 2011)
	454FLX and Illumina	127,754	50,566	12,141	(Dubey et al., 2011)
	Illumina	-	-	17,113	(Saxena et al., 2012)
Soybean	Illumina	5,02,020	-	1,682	(Deschamps et al., 2010)
	Illumina	-	-	205,614	(Lam et al., 2010)

Mining of transcriptome assemblies with SSR search tools has provided large numbers of SSR markers and these may be validated as genetic markers in wet lab experiments. For example, 250,393 SSRs were identified from 247 faba bean accessions (Yang et al., 2012) using 454/FLX sequencing technology. Beyond providing SSR markers, these transcriptome assemblies have been used for identification of SNPs from NGS datasets generated from two or more genotypes (Table 2). For instance, in the case of soybean, a total of 1,682 SNPs were identified by deep re-sequencing of the reduced representation libraries using Illumina Genome Analyzer (GA) (Deschamps and Campbell, 2010). Similarly, 26,082 SNPs were discovered from two contrasting drought responsive genotypes in chickpea (Hiremath et al., 2011), 36,188 SNPs for two contrasting salinity genotypes in pea (Leonforte et al., 2013) and 12,141 SNPs from 10 pigeonpea genotypes representing five mapping populations segregating for Fusarium wilt and sterility mosaic disease (Dubey et al., 2011) were identified. In addition, COS markers have found wide application in cross-genome comparative studies in legume species. For example, 387 COS markers were developed in chickpea from the transcriptome dataset (Hiremath et al., 2011). In some cases, transcriptome assemblies were analyzed for the identification of flanking intron junctions based on alignment to cognate or related genome sequences. As a result many ISR markers have been identified in legumes; for example, 2,088 and 10,009 ISRs were identified in chickpea (Hiremath et al., 2011) and pigeonpea (Kudapa et al., 2012), respectively and approximately 1,000 have been identified for pea using a comparative genomics approach (http://bioweb.abc.hu/mt/pisprim) and about 5,500 unigenes have been used to generate a consencus genetic map for pea (Bordat et al., 2011).

Although vast genomic resources have been developed rapidly in recent years for legumes, gene expression profiling data is insufficient to correlate the developmental processes to transcriptomes, leaving a large gap to associate the phenotype and genome/gene sequence. However, systematic mutant populations have been developed in several species in order to address this issue. TILLING populations are available for L. japonicus, pea, M. truncatula and chickpea (Dalmais et al., 2008; Le-Signor et al., 2009; Perry et al., 2003; Varshney et al., unpublished), Fast Neutron generated deletion mutant populations for M. truncatula, pea and soybean (Bolon et al., 2011; Hofer et al., 2009; Rogers et al., 2009; Smykal et al., 2012) and insertion mutant populations generated using exogenous (Medicago, d’Erfurth et al., 2003) and endogeneous retrotransposons (Lotus, Urbański et al., 2012) have been generated and have contributed to gene discovery in these species (see Domoney et al., 2013; Domonkos et al., 2013; Urbański et al., 2013).

Plant growth and development are controlled by the programmed (by time, tissue and abundance) expression of suites of genes in response to exogeneous or endogeneous queues. Hence, there is a need to generate genome-wide expression data from a range of tissues/ developmental stages in order to understand gene expression profiles in order to relate phenotypes and genotypes in legumes. A gene expression atlas was developed from a comprehensive set of developmental stages and tissues for some important crops such as rice (Wang et al., 2010) and maize (Sekhon et al., 2011).

In the case of legumes, the M. truncatula Gene Expression Atlas (MtGEA) was developed covering a range of tissues such as roots, nodules, stems, petioles, leaves, vegetative buds, flowers, seeds and seed pods with detailed developmental time-series for nodules and seeds, using the Affymetrix GeneChip Medicago Genome Arrays (Benedito et al., 2008). It has also incorporated the transcriptome data from plants subjected to various kinds of abiotic and biotic stresses and data from specific cell and tissue types. These data are useful for gene function determination, biological discovery, and molecular breeding efforts (http://mtgea.noble.org/v3/). Similarly, gene expression atlases have been developed in Lotus (http://ljgea.noble.org/) and soybean (Libault et al., 2010; Severin et al., 2010) from diverse tissues and developmental time series. Similar kinds of efforts have been initiated in chickpea, pigeonpea and groundnut at ICRISAT. These gene expression atlases are a powerful resource for legume genomics research, which can be used for understanding the regulatory network of developmental processes, tracing the expression profiles of stress responsive candidate genes etc.

In addition to gene expression atlases, efforts have been started to compile protein sequence information for different plant species. For example, all proteins identified for legumes (Medicago, Lotus and soybean) are available at http://bioinfoserver.rsbs.anu.edu.au/utils/PathExpress4legumes/. This database is a valuable resource for comparative seed proteomics and pathway analysis within and beyond the legume family, and protein sequences are exceptionally useful for genome annotation and determining the nature of genes for which only distant relatives are known. Similarly, another legume protein database (LegProt) has been created containing sequences from seven legume species, i.e., M. sativa, M. truncatula, L. japonicas, G. max, P. vulgaris, P. sativum and L. albus (Lei et al., 2011; http://bioinfo.noble.org/manuscript-support/legumedb). In addition the Soybean Proteome Database has been generated exclusively for soybean (http://proteome.dc.affrc.go.jp/Soybean/).

IV. COMPREHENSIVE GENETIC MAPS AND PREDICTIVE MARKERS

As previously mentioned, the availability of large-scale genomic resources has led to development of large numbers of molecular markers, particularly SNPs that can be assayed on a range of genotyping platforms. For instance, a variety of SNP genotyping platforms including GoldenGate (Kassa et al., 2012), VeraCode (Roorkiwal et al., 2013) and Competitive Allele Specific PCR (KASPar) assays (Saxena et al., 2012) have been developed for pigeonpea. Such genotyping platforms in combination with classical SSR genotyping platforms have been used to generate large-scale marker segregation data on mapping populations and have led to comprehensive genetic maps (Saxena et al., 2012). Several legume crop species did not have good genetic maps until recently. For example, the first SSR-based genetic linkage map for cultivated groundnut was recently developed for the mapping population, TAG 24 × ICGV 86031 (Varshney et al., 2009b). Subsequently five more genetic maps were developed based on mapping populations segregating for drought and foliar diseases (Gautami et al., 2012a; Ravi et al., 2011; Sujay et al., 2012). A consensus map for drought tolerance related traits with 293 SSR loci (Gautami et al., 2012a) and foliar disease resistance with 225 SSR loci (Sujay et al., 2012), respectively have been developed. Furthermore, international collaborative efforts have resulted in the construction of consensus map with 897 SSR marker loci using genotyping data from 11 mapping populations. This consensus map was upgraded with maps for two more populations and the updated version map covers 2,651 cM with 3,693 marker loci anchored to 20 consensus linkage groups (LG) corresponding to the A and B genomes (Shirasawa et al., 2013). Another example of development of comprehensive genetic maps is in cowpea, where a 1,536-SNP GoldenGate genotyping array was developed and applied to 741 recombinant inbred lines from six mapping populations. 1,375 out of 1,536 SNPs provided dependable markers and a subset of 928 were incorporated to develop a consensus genetic map spanning 680 cM with 11 linkage groups and an average marker distance of 0.73 cM (Muchero et al., 2009). This map has been further updated and the consensus map has 1,107 SNP loci (Lucas et al., 2011).

Details on genetic maps in selected legume crops are listed in Table 3. These dense consensus genetic maps are useful resources for the identification of highly informative and well distributed markers to be used for background selection in breeding schemes, the construction of new genetic maps or analysis of genetic diversity. Analysis of genetic maps, together with trait phenotyping on the respective segregating populations provides predictive markers for traits of interest. Both biotic and abiotic stresses are complex in nature, so trait dissection is important for improving component traits through breeding. Often agronomical traits are composites of many underlying phenothypes, but their dissection into component, genetically determined, contributions represent the underlying biology and can be used for marker-based breeding. Both bulked segregant analysis (BSA) as well as quantitative trait loci (QTL) mapping based marker analysis, have been used to identify markers linked to these genetic determinants of traits. In fact, quantitative traits have been studied in legumes since Mendel. In his 1866 paper he commented on the near continuous variation of flower colour in Phaseolus crosses: “Even these enigmatical results, however, might probably be explained by the law governing Pisum if we might assume that the color of the flowers and seeds of Ph. multiflorus is a combination of two or more entirely independent colors, which individually act like any other constant character in the plant” [http://www.mendelweb.org/]

TABLE 3 List of some comprehensive genetic maps for selected legumes

Name of population	Type	Number of loci	References
Chickpea
ICC 4958 × PI 489777	RIL	406	(Choudhary et al., 2012)
ICC 4958 × PI 489777	RIL	1,063	(Gaur et al., 2012b)
ICC 4958 × PI 489777	RIL	1,328	(Hiremath et al., 2012)
ICC 4958 × PI 489777	RIL	1,291	(Thudi et al., 2011)
ICC 4958 × PI 489777	RIL	300	(Gujaria et al., 2011)
ICC 4958 × PI 489777	RIL	521	(Nayak et al., 2010)
Common Bean
DOR3646 × BAT477	RIL	291	(Galeano et al., 2011)
BAT93 × Jalo EEP558	RIL	275	(McConnell et al., 2010)
BAT93 × Jalo EEP558	RIL	413	(Hanai et al., 2010)
BAT93 × Jalo EEP558	RIL	275	(McConnell et al., 2010)
DOR364 × G19833	RIL	215	(Córdoba et al., 2010)
Xana × Cornell 49242	RIL	294	(Pérez-Vega et al., 2010)
Faba bean
Vf 6 × Vf 136	RIL	277	(Díaz-Ruiz et al., 2009a)
Vf 6 × Vf 27	RIL	258	(Cruz-Izquierdo et al., 2012)
Groundnut
A-genome genetic maps
A. duranensis × A. duranensis	F2	2,319	(Nagy et al., 2010)
A. duranensis × A. stenosperma	F2	369	(Leal-Bertioli et al., 2009)
B-genome genetic maps
A. batizocoi PI 298639 × A. batizocoi PI 468327	F2	449	(Guo et al., 2010)
AB genome genetic maps
Tifrunner × GT-C20	F2	381	(Wang et al., 2013; 2012)
Satonoka × Kintoki	F2	516	(Shirasawa et al., 2013; 2012)
Nakateyutaka × YI-0311	F2	293	(Shirasawa et al., 2013; 2012)
Tifrunner × GT-C20	RIL	239	(Qin et al., 2012)
A. duranensis × (A. ipäensis × A. duranensis)	BC1F1	298	(Foncéka et al., 2009)
ICP 28 × ICPW 94	F2	239	(Bohra et al., 2011)
ICP 28 × ICPW 94	F2	910	(Saxena et al., 2012)
Pusa Dwarf × HDM04-1	F2	296	(Kumawat et al., 2012)
Lentil
ILL 5588 × L 692-16-1(s)	RIL	283	(Hamwieh et al., 2005)
Eston × PI 320937	RIL	207	(Tullu et al. 2006; 2008)
Digger (ILL 5722) × NorthWeld (ILL 5588)	RIL	211	(Gupta et al., 2012a)
CDC Robin × 964a-46	RIL	543	(Sharpe et al., 2013)
CDC Robin × 964a-46	RIL	577	(Fedoruk et al., 2013)
Pea
Puget × 90-2079	RIL	324	(Pilet-Nayel et al., 2002)
Wt 10245 × Wt 11238	F2	204	(Irzykowska and Wolko, 2004)
Carneval × MP 1401	RIL	207	(Tar’an et al., 2003a; 2004)
DP × JI 296	RIL	206	(Prioul et al., 2004)
Shawnee × Bohatyr	RIL	302	(Loridon et al., 2005)
Orb × CDC Striker	RIL	224	(Ubayasena et al., 2010)
P665 × Messire	RIL	246	(Fondevilla et al., 2008; 2010)
Orb × CDC Striker	RIL	255	(Sindhu et al., 2013)
Soybean
Essex × Forrest	RIL	959	(Xia et al., 2007)
Essex × Forrest	RIL	237	(Kassem et al., 2006)

For simple inherited traits, BSA has been used in several plant species including the legumes discussed in this article. For instance, BSA was used for mapping of Fusarium wilt (FW) resistance with RAPD markers in the case of pigeonpea (Kotresh et al., 2006). On the other hand, in order to understand the complex nature of drought tolerance in chickpea, precise phenotypic data for 20 different root traits was analyzed using a QTL mapping approach. As a result, nine QTL clusters containing QTLs for several drought tolerance traits have been identified that are being targeted for chickpea molecular breeding program. Among these QTL clusters, one QTL cluster corresponding to QTLs for 12 traits and explaining about 60% phenotypic variation has been identified on CaLG04, referred to as a “QTL-hotspot” (Varshney et al., 2013b). Similar mapping efforts have been undertaken for important the biotic stresses like FW (Benko-Iseppon et al., 2003; Cobos et al., 2005; Sabbavarapu et al., 2013), and Aschochyta blight (Anbessa et al., 2009; Aryamanesh et al., 2010; Kottapalli et al., 2009).

One agronomic trait under complex genetic control (Buckler et al., 2009; Mouradoy et al., 2002) that has received much attention is flowering time. This was studied as a quantitative trait in rice and pea at the beginning of the 20th century (Hoshino, 1915), and the physiological genetics of flowering time in pea was a model for this that was eclipsed by the rise of Arabidopsis as an experimental system. The various lineages of angiosperm families diverged about 100 to 150 million years ago (Bell et al., 2010), so it is not surprising that many structural genes that regulate this pathway are common to different families, but that their regulatory circuitry differs. In the legumes (that diversified 55 – 60 million years ago) the study of the regulation of flowering time is a resurgent and the novelty of the genetic control of flowering time in legumes is being revealed (Hecht et al., 2011; Mach, 2011; Weller et al., 2012) with strong indications that floral induction in pea may involve the action of several members of the FT family with distinct modes of action (Hecht et al., 2011). Thus some major determinants of flowering time are known, so the underlying genes can be used (if necessary) to follow this trait. However the Buckler et al. (2009) analysis of the control of flowering time in maize suggests that there may also be many loci of small effect that together contribute greatly to the genetic variation of this trait. Some key studies on identification of QTLs and predictive markers associated with agronomically important traits in addition to tolerance/ resistance to abiotic and biotic stresses in select legumes are summarized in Table 4.

TABLE 4 Summary of some trait mapping studies in select legumes

Trait	Gene/ QTL	*PVE (%)	Reference
Chickpea
Resistance to biotic stresss
Fusarium wilt	foc-0, foc-1, foc-2, foc-3, foc-4, foc-5	15	(Cobos et al., 2005; Iruela et al., 2007)
	QTL		(Gowda et al., 2009)
Ascochyta blight	QTL	14–56	(Anbessa et al., 2009; Cobos et al., 2006; Irula et al., 2006; Kottapalli et al., 2009; Lichtenzveig et al., 2006; Tar’an et al., 2007)
	BSA	-	(Rajesh et al., 2002)
	AR2, ar1, ar1a, ar1b, ar2a, ar2b, Ar19	28	(Cho et al., 200)
Botrytis gray mold	QTL	44	(Anuradha et al., 2011)
Rust	QTL	31	(Madrid et al., 2008)
Agronomic and yield related traits
Agronomic and yield related traits	QTL	19	(Gowda et al., 201)
Days to flower	QTL	26–56	(Cobos et al., 2007; 2009; Lichtenzveig et al., 2006)
Double podding	QTL	-	(Cobos et al., 2005; Cho et al., 2002; Rajesh et al., 2002)
100-seed weight	QTL	52	(Cho et al., 2002)
Flower colour	B/b		(Gowda et al., 2009)
Seed coat colour, thickness, size weight	QTL	49–73	(Cobos et al., 2005; 2009; 2007; Gowda et al., 2011; Hossain et al., 2011; Lichtenzveig et al., 2006)
Common bean
Resistance to biotic stresses
Angular leaf spot	QTL	22–64	(López et al., 2003; Oblessuc et al., 2012; Teixeira et al., 2005)
Anthracnose	Co-genes	-	(Rodriguez-Suarez et al., 2007)
Ashy stem blight	QTL	16–29	(Miklas et al., 2006b)
Bean common mosaic necrosis virus	BSA	-	Strausbaugh et al., 1999)
	bc-3; I	4	(Johnson et al., 1997)
Bean golden yellow mosaic virus	QTL	15	(Miklas et al., 2006)
	bgm	1	(Blair et al., 2007)
Bean pod weevil	QTL	22	(Blair et al., 2006)
Bruchids	Arc gene		(Blair et al., 2010)
Bean rust resistance	BSA	-	(Correa et al., 2000; Johnson et al., 1995; Mienie et al., 2005; Miklas et al., 1993; Park et al., 2008)
Bacterial brown spot	QTL		(Jung et al., 2003)
Beet curly top virus	BSA	-	(Larsen and Miklas, 2004)
Climbing ability and component traits	QTL	25	(Checa and Blair, 2008)
Common bacterial blight (CBB) resistance	QTL	30–73	(Liu et al., 2005; Miklas et al., 2006; Vandemark et al., 2008)
Charcoal rot	QTL	1	(Hernandez-Delgado et al., 2009)
Fusarium wilt	QTL	63	(Fall et al., 2001)
	PvPR2, PvPR1	3	(Schneider et al., 2001)
Halo blight	QTL	-	(Miklas et al., 2011)
	Pse	1	(Miklas et al., 2009)
Thrips	Tpr6.1		(Frei et al., 2005)
White mold	QTL	39–52	(Park et al., 2001; Soule et al., 2011)
Phosphorus uptake	Pup4.1, 10.1 and 2.1		(Beebe et al., 2006)
Rust	UR-6, UR-13	19	(Mienie et al., 2005)
Seed chemical content	QTL	51	(Casanas et al., 2013)
Faba Bean
Ascochyta blight	QTL	24–45	(Avila et al., 2004; Díaz-Ruiz et al., 2009b; Roman et al., 2003)
Broomrape resistance	QTL	33–43	(Díaz-Ruiz et al., 2009a; Gutiérrez et al., 2013; Román et al., 2002)
Rust	Uvf1		(Avila et al., 2003)
Floral characters	QTL	20	(Avila et al., 2005)
Days to flowering	QTL	28	(Cruz-Izquierdo et al., 2012)
Flowering length	QTL	31	(Cruz-Izquierdo et al., 2012)
Pod length	QTL	25	(Cruz-Izquierdo et al., 2012)
Number of ovules per pod	QTL	27	(Cruz-Izquierdo et al., 2012)
Number of seeds per pod	QTL	26	(Cruz-Izquierdo et al., 2012)
Yield characters	QTL	58	(Avila et al., 2005)
Frost tolerance	QTL	40	(Arbaoui et al., 2008)
Fatty acid content	QTL	63	(Arbaoui et al., 2008)
Groundnut
Resistance to biotic stresses
Late leaf spot, Leaf rust, Aspergillus flavus invasion, groundnut rosette disease, Tomato Spotted Wilt Virus	QTL	1.70–82.96	(Khedikar et al., 2010; Liang et al., 2009a; Qin et al., 2012; Sujay et al., 2012; Wang et al., 2013)
Drought tolerance related traits
Transpiration,Transpiration efficiency, Specific leaf area, Leaf area, SPAD chlorophyll meter reading (SCMR), Biomass, Canopy conductance, Total dry matter	QTL	3.48–22.39	(Gautami et al., 2012b; Ravi et al., 2011; Varshney et al., 2009b)
Morphological and yield related traits
Shoot dry weight, Pod weight, Seed weight, Haulm weight, Harvest index, Pod mass/plant, Mature pods/plant, Number of branches, Height of main axis, Stem diameter, Leaf length, width and length /width ratio, Length of main stem, Length of the longest branch, Number of branches, Weight of plant, Weight of mature pod per plant, Thickness of pod, Width of pod, Shape of tip of pods, Number of seeds per a plant, Yield parameters	QTL	3.78–40.10	(Gautami et al., 2012a; Liang et al., 2009; Ravi et al., 2011; Selvaraj et al., 2009; Shirasawa et al., 2012; Varshney et al., 2009b)
Seed and oil quality
Oil content, Oil quality, Protein content, High oleate trait	QTL	1.4–74.03	(Liang et al., 2009; Pandey et al., 2012c; Sarvamangala et al., 2011; Selvaraj et al., 2009; Shirasawa et al., 2012)
Pea
Plant height	QTL	19–65	(Tullu et al., 2008; Tar’an et al., 2003b; Dirlewanger et al., 1994)
Days to 50% flowering	QTL	35	(Fedoruk et al., 2013)
Days to maturity (DM)	QTL	34	(Tar’an et al., 2004)
Plant Maturity	QTL	27	(Timmerman-Vaughan et al., 2004)
Earliness	QTL	46	(Tullu et al., 2008)
Powdery mildew	genes		(Fondevilla et al., 2011)
Fusarium wilt (FW) resistance	QTL	-	(Hamwieh et al., 2005)
Stemphylium blight (SB) resistance	QTL	46	(Saha et al., 2010)
Rust	QTL	63	(Barilli et al., 2010)
Aphanomyces root rot	QTL	21–60	(Hamon et al., 2011, 2013; Pilet-Nayel et al., 2002)
Ascochyta blight (AB) resistance	QTL	35–75	(Fondevilla et al., 2008; Gupta et al., 2012b; Prioul et al., 2004; Rubeena et al., 2006; 2003; Tar’an et al., 2003b; Timmerman-Vaughan et al., 2002; Tullu et al., 2006)
Orobanche	QTL	8–36	(Fondevilla et al., 2010)
Frost resistance/damage	QTL	61	(Dumont et al., 2009; Klein et al., 2014)
Winter hardiness	QTL	43	(Kahraman et al., 2010)
Seed thickness	QTL	38	(Fedoruk et al., 2013)
Seed plumpness	QTL	40	(Fedoruk et al., 2013)
Grain yield	QTL	38	(Tar’an et al., 2004)
Lodging resistance	QTL	58	(Tar’an et al., 2003b)
Seed protein concentration/content	QTL	59	(Tar’an et al., 2004; Burstin et al., 2007)
Yield component and developmental traits	QTL	62	(Timmerman-Vaughan et al., 2005)
Yield-related traits and seed protein content	QTL	56	(Irzykowska and Wolko, 2004)
Pigeonpea
Fertility restoration	QTL	15–24	(Bohra et al., 2012)
Fusarium wilt	BSA		(Kotresh et al., 2006)
Number of pods per plant	QTL	19	(Kumawat et al., 2012)
Plant height	QTL	28	(Kumawat et al., 2012)
Plant type	BSA		(Dhanasekar et al., 2010)
Primary branches per plant	QTL	20	(Kumawat et al., 2012)
SMD resistance	QTL	25	(Gnanesh et al., 2011)
	BSA		(Ganapathy et al., 2009)
Soybean
Resistance to biotic stresses
Cyst nematode, Corn earworm, Brown stem rot, Sclerotinia stem rot, Phytophora root rot, Sclerotinia stem rot, Soybean mosaic virus, Asian rust, Brown stem rot, Sudden death syndrome	QTL	2–28	(Atibalentja et al., 2005; Guo et al., 2006; Guo et al., 2008; Han et al., 2008; Huynh et al., 2010; Kazi et al., 2008; Li et al., 2010; Ruben et al., 2006; Shi et al., 2008; Silva and Danielle, 2008; Vuong et al., 2008; Wang et al., 2010; Winter et al., 2007)
Soybean mosaic virus	Rsv, Rsv	1 to 3	(Shi et al., 2008)
Cyst nematode	rhg1, Rhg4	5	(Guo et al., 2005; Ruben et al., 2006; Vuong et al., 2010; Winter et al., 2007; Wu et al., 2009)
Agronomic and yield related traits
Seed protein oil content, Water use efficiency, Specific leaf weight and leaf size, Seed weight, Flowering time, Yield, maturity, Sprout yield, Seed isoflavone, Seed size, viability of seed, germination rate of seed, water absorbability of seed, ability, frequency, and efficiency of somatic embryogenesis, Vitamin E content, Chlorophyll a fluorescence parameter, Developmental behavior, Browning in soybean seed coats, Domestication, Seed composition, Seed shape, Photoperiod insensitivity, Net-like cracking in seed coats, Cleistogamy, Nodule number and nodule dry weight, Water logging, Oligosaccharide and sucrose	QTL	3–63	(Choi et al., 2010; Funatsuki et al., 2005; Githiri et al., 2006; Githiri et al., 2007; Khan et al., 2008; Kim et al., 2006; Lee et al., 2009; Li-H. et al., 2010; Liu et al., 2007; Liu and Abe, 2010; Nicolas et al., 2006; Oyoo et al., 2010; Panthee et al., 2006; Qi et al., 2008; Reinprecht et al., 2009; Salas et al., 2006; Song et al., 2010; Sun et al., 2006; Tuyen et al., 2010; Yin et al., 2010)
Early maturity	E8	8	(Cober et al., 2010)
Phosphorus deficiency	fsw1, fsw2, rp1, fsw3, rp2, lp1, lp2	12	(Li et al., 2005)

V. SUCCESS STORIES OF TGA IN LEGUMES

Molecular breeding is a direct strategy to translate genome information for developing products i.e. superior lines for traits of interest. In general, in addition to MAS, three main approaches namely marker-assisted backcrossing (MABC), marker-assisted recurrent selection (MARS) and genomic selection (GS) can be used in molecular breeding. Several success stories utilizing molecular breeding have been reported in crop species such as rice, maize, wheat (see Kulwal et al., 2011). With the notable exception of soybean in the commercial sector, for legumes, in general, molecular breeding has been delayed due to limited availability of genomic resources. However, recent advances in genomics especially in the legumes discussed here have opened the avenues for accelerating practicing of TGA. An overview of integrative genomics for accelerated crop improvement is illustrated in Figure 2.

FIG. 2. Integrative genomics for accelerated crop improvement. The figure depicts the use of genomic resources for identification of marker /candidate gene. Subsequently the markers/genes identified may be used to detect gene-trait associations thus by enhancing molecular breeding program by deploying different approaches like MABC, MARS and GS. Ultimately integrating genomic resources through translational genomics would result in improved lines or developed cultivars.

MAS has been extensively deployed for development and release of improved lines/varieties in legumes, some key examples are summarized in Table 5. In the case of soybean, in addition to developing improved lines or varieties for resistance to different biotic stresses like soybean cyst nematode (SCN) races (Cahill and Schmidt, 2004; Concibido et al., 1996), phytophthora root rot and brown stem rot (Cahill and Schmidt, 2004), insect resistance (Narvel et al., 2001; Walker et al., 2002; Warrington et al., 2008), mosaic virus (Saghai-Maroof et al., 2008; Shi et al., 2009), MAS products were also developed for low linolenic acid content (Sauer et al., 2008) and yield (Concibido et al., 2003). Furthermore, SNP markers tightly linked to soybean root-knot nematode resistance QTL have enhanced the efficiency of MAS (http://www.plantmanagementnetwork.org/pub/php/news/2007/Shortcut/). Similarly, in case of common bean, efforts at CIAT focused mainly in deploying single genes for enhancing resistance to viral diseases. As a result, the cultivar Perola with enhanced resistance bean Anthrocnose in Brazil (Raganin et al., 2003), pinto beans in the United States (Miklas et al., 2003) and Andean climbing bean in Mexico / Colombia (Garzon et al., 2008) have been developed. In cowpea, SCAR markers have been used in discriminating Striga resistant and susceptible lines (Omoigui et al., 2012).

TABLE 5 Key success stories of TGA in legume crops

Crop	Trait	Progenies developed/ cultivars released	Reference/source
Chickpea	Fusarium wilt resistance	Back cross progenies from (C 214 × WR 315)	(Varshney et al., 2013c)
	Aschochyta blight resistance	Back cross progenies from (C 214 × ILC 3279)	(Varshney et al., 2013c)
	Drought tolerance	Back cross progenies from (JG 11 × ICC 4958)	(Varshney et al., 2013b)
		Back cross progenies from (Chefe × ICC 8261), (KAK2 × ICC 8261), (ICCV 10 × ICC 4958)	(ICRISAT, India)
		Back cross progenies from (Ejere × ICC 4958), (Arerti × ICC 4958)	(EIAR, Ethiopia)
		Back cross progenies from (ICCV 97105 × ICC 4958), (ICCV 95423 × ICC 4958)	(EU, Kenya)
Common bean	Disease resistance	USPT-ANT-1, ABCP-8, ABC-Weihing	(Miklas et al., 2003; Mutlu et al., 2005, 2008)
	Resistance to white mold	Back cross lines	(Carneiro et al., 2010)
Groundnut	Nematode resistance	COAN cultivar, NemaTam	(Simpson et al., 2003)
	Nematode resistance and High O/L ratio	Tifguard High O/L	(Chu et al., 2011)
	Rust resistance	Back cross progenies from (JL 24 × GPBD 4), (TAG24 × GPBD 4), (ICGV 91114 × GPBD 4)	(Varshney et al., 2014)
	High O/L ratio	Back cross lines	(Koilkonda et al., 2012)
Soybean	Disease resistance	JTN-5503, JTN-5303	(Arelli et al., 2006; 2007)
	Soybean mosic virus	F4:5 lines, Near Isogenic Line (NILs)	(Saghai-Maroof et al., 2008; Shi et al., 2008)
	Low phytate	Back cross lines	(Landau-Ellis and Pantalone, 2009)
	Soybean cyst nematode resistance	JTN-5109, DS-880	(Smith et al., 2010)
	Rust resistance	Back cross lines	(Khanh et al., 2013)

MABC is the simplest way to develop superior lines by transfering QTLs with large phenotypic effect using tightly linked molecular markers to a given trait for foreground selection and genome-wide unlinked markers for background selection or recovery of the recurrent parent genome. In particular MABC is useful to capture major QTL(s)/gene(s) with a strong association to the given trait of interest. For example in groundnut, MABC has been used to develop and release an improved variety; markers linked with root-knot nematode (Meloidogyne arenaria) resistance were used for introgression through the amphidiploid pathway into cultivated groundnut (Simpson et al., 2001). Identification of linked markers due to sequence divergence between diploid and tetraploid genomes (Chu et al., 2007; Nagy et al., 2010) was relatively easy in groundnut. The simultaneous introgression of nematode resistance and oil quality was achieved using a recessive AhFAD2B allele (which controls high ratio of oleic: linoleic acid (O/L)) and linked markers for foreground selection. These efforts led to the release of the improved Tiftguard variety “Tifguard High O/L” (Chu et al., 2011). The MABC approach has been initiated to introgress a major QTL contributing ca. 80% of the phenotypic variation for leaf rust into the genetic background of three elite cultivars namely ICGV 91114, JL 24 and TAG 24. By using 2–3 rounds of backcrossing and selfing, BC₂F₃ and BC₃F₂ homozygous lines have been developed at ICRISAT.

Two major MABC projects are underway in chickpea at ICRISAT and its partner organizations. Under the Tropical Legume-I initiative of CGIAR Generation Challenge Programme in collaboration with Bill & Melinda Gates Foundation, significant efforts have been made to develop drought tolerant progenies (BC₃F_3:4) in the genetic background of JG11, a leading variety in India, by transferring a genomic region, “QTL-hotspot,” that contains several QTLs for drought tolerance traits (Varshney et al., 2013b). Detailed phenotypic evaluation of these lines is underway in India, Kenya and Ethiopia and preliminary results are promising (unpublished). The second initiative is part of the Accelerated Crop Improvement Programme (ACIP) of Department of Biotechnology, Government of India. In this initiative, efforts are being made to introgress resistance to different races independently as well as pyramiding of resistance to two races for FW in some elite varieties in India. ICRISAT (India) is pyramiding resistances for Foc1 and Foc3 from WR 315 and 2 QTLs for Ascochyta blight (AB) resistance from ILC 3279 line into C 214 (Varshney et al., 2013c).

However, the majority of traits targeted by breeders e.g. drought tolerance or durable resistance to multiple races of pathogens are controlled by multiple QTLs or genes. In such cases, retaining desirable gene combinations or pyramiding of several QTLs through the MABC approach is a challenging task (Peleman and Voort, 2003). Hence, marker assisted recurrent selection (MARS), has been introduced as an alternative approach and demonstrated experimentally to be more effective in accumulating a large number of smaller effect QTLs (Ribaut et al., 2010; Ribaut and Ragot, 2007). MARS, on the other hand, aims to take advantage of QTL information generated on breeding populations to develop superior lines with an optimum combination of favourable alleles originating from both the parents. QTL alleles are accumulated through successive intercrossing using only genotypic selection. Recombined lines are then subjected to a final phenotypic screening to select the best varieties to release. MARS can be initiated without any prior knowledge of QTLs with the objective of discovering and harnessing the superior QTLs/alleles during the scheme itself, which is not possible in MABC. However, it is observed that existing marker-trait associations enhance the quantum of genetic gains in MARS scheme (Bernardo and Charcosset, 2006). The recurrent-selection method is routinely used mainly in cross-pollinated crops like maize and, this process can be improved with the help of molecular markers, which is why the process is called marker-assisted recurrent selection (MARS). While several multinational companies have been using MARS in crops like maize and soybean, only a few public sector institutes have used MARS in crops likes wheat (Charmet et al., 2001), sorghum (Abdallah et al., 2009) and rice (Grenier et al., 2012). MARS based programs have been initiated in cowpea (see Ribaut et al., 2012) targeting improvement in drought tolerance. Some efforts have been initiated to use MARS in the case of chickpea, for assembling favorable alleles for drought tolerance using ICCV 04112 × ICCV 93954 and ICCV 05107 × ICCV 94954.

Genomic selection (GS) is a novel approach in which genetic markers covering the whole genome are used so that all (QTL) are in linkage disequilibrium with at least one marker. GS focuses on the selection of desirable genotypes on the basis of an index derived from genome-wide DNA marker information. The index is designated as genomic estimated breeding values (GEBVs) and the phenomenon of selecting individuals on GEBV basis is referred as genomic selection (GS) (Heffner et al., 2009). Genotyping and phenotyping of a ‘training population’ is required to train the GS models or extensively calculate GEBV estimates (Nakaya and Isobe, 2012). The GEBV estimates thus obtained are subsequently used for selection of genotypes from ‘breeding population.’ In GS, no phenotyping is required for the ‘breeding population’, however all the individuals are genotyped thoroughly. GS efficiently exploits the high density marker data available at reasonable cost and can also reduce the selection cycle length of a breeding program where it could take several seasons to develop reliable phenotypic data. In brief, GS as an approach relies on cycles of prediction and testing by phenotyping the training population as the process proceeds and this constant checking of the selection index vs the training population overcomes type I errors associated with multiple loci contributing small effects to many quantitative traits.

For estimating the GEBVs with high precision, an important component of GS, use of an appropriate statistical model is very critical. Among different models of GS, GEBVs predicted using either best linear unbiased prediction (BLUP) or Bayesian methods are effective according to simulation studies (Bernardo and Yu, 2007). In addition to a Bayesian method, Bayes B, another method called wBSR (weighed Bayesian Shrinkage Regression) which reduces the computational burden on MCMC-based Bayesian methods is considered to be a method of choice for genomic selection (Takeshi and Hiroyoshi, 2010). Cornell University (Jannink et al., 2010), the University of Minnesotta (Bernardo and Yu, 2007) and Hohenheim University (Piepho, 2009) also developed statistical models and/or pursued applications of GS in breeding of some major crops like maize, wheat. Though GS has not been used in any legume species at present, mostly due to lack of: (i) historical phenotyping data on several breeding lines (that can be used for the training population), (ii) big linkage disequilibrium (LD) blocks in breeding populations, and (iii) genome wide marker genotyping system like DArT or SNP markers, ICRISAT has initiated the efforts to deploy GS in chickpea and groundnut.

VI. CHALLENGES AND OPPORTUNITIES

Whole genome sequences of both model and crop legumes are expected to offer new perspectives for TGA in legume species. From the examples presented above it is evident that ongoing genomics research could accelerate crop improvement programs in legume crops. In the pre-genomics era, comparative genomics approach was successful for identifying homologues/ orthologues or cloning genes in several legumes by using information from model plant systems. For example, conservation of Arabidopsis flowering time genes was studied in legumes and preliminary survey revealed that legume flowering-related genes (LFY and TFL1) identified significant differences in function to Arabidopsis counterparts (Hecht et al., 2005). However, this approach was not very successful for developing superior varieties in legume crops. In the post-genomics era, availability of genome sequence information as well as species specific genomic resources offer greater opportunities for TGA even in the crops like lentil, pea and fababean where limited information is available. However, so far, success stories of TGA are limited in the targeted legume crops. In this context, there is a series of challenges that include: (i) availability of limited predictive molecular markers for target traits, (ii) limited access to marker technologies, (iii) limited availability of high-throughput, cost-effective and precise phenotyping platforms, (iv) lack of capacity both human as well as infrastructure and (v) absence of suitable data management systems (Varshney et al., 2012b). High-throughput sequencing and genotyping facilities available in different countries could benefit from economies of scale. Although sample shipment across national borders sometimes poses logistic and quarantine challenges, especially in the developing world, access to those centers could facilitate deployment of molecular breeding in the legume crops. For modernization of the breeding programs, it is important to deploy decision support tools (computational tools) in every step of the process of breeding. The Integrated Breeding Platform initiative (www.integratedbreeding.net), houses a range of decision support tools that can be used in a distributed way for example by National Agricultural Research Systems (NARS) in their breeding programs.

Although conventional breeding based exclusively on phenotypic selection remains the mainstay for most breeding programs in these legumes, adoption of TGA is increasing and in some cases is superseding conventional approaches. With advances in genomics, the ultimate aim is to increase the rate of genetic gain across target environments, in less time and at lower cost compared to conventional selection. Nonetheless, it is also important to mention the role of quality phenotypic data. Efforts need to be initiated to establish field based high throughput and cost effective phenotyping platforms. As plants are exposed to range of stresses, breeders in general perform selection based on the plant's overall performance and not just for a single trait. Therefore GS seems to be an ideal approach because it involves selection of the plants based not on just one trait but for overall performance of the plant based on whole genome marker profiling data. We believe that centralized genotyping service facilities, together with access to genomics and breeding analytical tools (decision support tools), should enhance implementation of TGA in these legume crops. Furthermore web accessible mutant populations are available for Medicago, Lotus and soybean. Developing such collections in other crop legumes may fasten the breeding programs. It is also important to continue the training of breeders and geneticists in integrated genomics and molecular breeding. Effective implementation of TGA in the legume crops will lead to crop improvement programs for ensuring food security in developing countries.

Additional information

Funding

Authors are thankful to CGIAR Generation Challenge Programme (GCP), the Indian Department of Biotechnology (Centre of Excellence and Accelerated Crop Improvement Programme), the Indian Ministry of Science & Technology (Indo-German Science & Technology Cooperation and Australia–India Science & Research Foundation) and The United States Agency for International Development (USAID) for funding projects related to molecular breeding at ICRISAT. This work has been undertaken as part of the CGIAR Research Program on Grain Legumes. ICRISAT is a member of CGIAR Consortium.