Unraveling the Secrets of Rice Wild Species

The world is facing a new challenge with global population predicted to plateau at nine billion people by the middle of this century (Godfray et al. 2010). Increasing food production to feed the world’s population is an even greater challenge considering that agriculture is experiencing greater competition for land, water and energy, as well as, the effects of substantial climate change and the unintended effects of crop production on the environment. Part of the solution to increasing food production on the same or less cultivated land lies in exploiting the subset of genes lost during the domestication process and subsequent targeted breeding. Currently, these valuable genes are found only in the progenitor species genepool for crop cultivars. Cultivated plants having desirable genes were utilized in intensive breeding projects focused on increasing yield for particular environments and management systems but this process has narrowed the genetic diversity (Rausher 2001). For cultivated plants, this unexploited genetic material includes both landraces and the more exotic wild relatives. Improving our under‐ standing of this tertiary gene pool and exploiting it for crop improvement is paramount to meeting the challenges of feeding the world in this century through the integration of classical genetics and genomics-enabled research paradigms.


Introduction
The world is facing a new challenge with global population predicted to plateau at nine billion people by the middle of this century (Godfray et al. 2010). Increasing food production to feed the world's population is an even greater challenge considering that agriculture is experiencing greater competition for land, water and energy, as well as, the effects of substantial climate change and the unintended effects of crop production on the environment. Part of the solution to increasing food production on the same or less cultivated land lies in exploiting the subset of genes lost during the domestication process and subsequent targeted breeding. Currently, these valuable genes are found only in the progenitor species genepool for crop cultivars. Cultivated plants having desirable genes were utilized in intensive breeding projects focused on increasing yield for particular environments and management systems but this process has narrowed the genetic diversity (Rausher 2001). For cultivated plants, this unexploited genetic material includes both landraces and the more exotic wild relatives. Improving our understanding of this tertiary gene pool and exploiting it for crop improvement is paramount to meeting the challenges of feeding the world in this century through the integration of classical genetics and genomics-enabled research paradigms.
The loss of genetic diversity can be more problematic for self-pollinated plant species where the rate of cross pollination is below five percent, thus making it more difficult to reintroduce the lost diversity. In the case of the two major grain crops, rice (Oryza sativa L.) and wheat (Triticum aestivum L.), both self-pollinated, the re-introduction genetic diversity from the wild is central to the continued success of breeding, given that viruses, fungi, and bacteria, three main causal agents of biotic stress, are constantly evolving to cause the breakdown of the host plant's defense mechanisms (Rausher 2001).
Abiotic stress, including salinity, aluminum toxicity and acid sulfate soils, as well as, temperature and drought, complicate the difficulty of improving crop yields, especially in the face of global warming (Brooker 2006;Tilman and Lehman 2001), which makes modern cultivars even more vulnerable. Genetic sources of resistance or tolerance offer the most promising mechanism to protect plants against these unfavorable conditions. Often wild species are not included as parental lines in cultivar development because it is relatively difficult to harness desirable genes by genetic recombination and many undesirable genes are introgressed from the wild parent resulting in inferior yield, undesirable plant architecture, and/or poor grain quality (Tanksley and McCouch 1997). Recent studies, however, in rice (McCouch et al. 2007) and tomato, Lycopersicon esculentum Mill. (Grandillo and Tanksley 2003), have shown that wild species contain genomic components that could result in genetic gains in terms of agronomic performance.
The rapid advancement in molecular technologies allows for genotyping plants much more quickly and inexpensively than ever before. The availability of high resolution genotypic information creates the opportunity to further explore an expanding number of accessions in a greater depth, and harness this information to enhance the efficiency and accuracy of introgression. These developments create opportunities not previously possible, to identify molecular markers associated with desirable traits in wild species and transfer these traits into elite lines and/or varieties, as well as, to unravel multi-genic traits for crop improvement (Tanksley and McCouch 1997;McCouch et al. 2012).
Our main objective is to summarize efforts over the past 15 years to identify useful novel alleles in the Oryza species that were lost during evolution and domestication, genetically dissect the traits encoded by these alleles through chromosome mapping, and incorporate these traits or alleles into an agronomically useful genetic background. To do this we will (a) briefly describe the relationships among the species in the genus Oryza, (b) describe the types of populations that have been developed for mapping desirable traits identified in the wild Oryza species to a chromosome location, and (c) summarize the quantitative trait locus (QTL) studies focused on mapping the useful traits and novel alleles to specific locations in the genomes of Oryza species.

Phylogeny of the Oryza genus
The Oryza genus includes two cultivated species, Asian rice, O. sativa, which is grown throughout the tropical and temperate climates of the world, and African rice, O. glaberrima, which is found in sub-Saharan Africa along the Niger River. The 22 wild species composing the Oryza genus are characterized by eleven different genomes identified as the A-, B-, C-, D-, E-, F-, G-, H-, J-, K-and L-genomes and arranged in the following 10 genome types AA, BB, CC, BBCC, CCDD, EE, FF, GG, HHJJ and KKLL. Four of the wild Oryza species are tetraploid and the remaining 18 are diploid, as well as, the two cultivated species (Table 1). Rice is the only major cereal found in the ancient lineage of the Bambusoideae and is currently placed in the subfamily Erhartoideae. Historically, the grass family, Poaceae, is thought to have evolved about 70-55 mya (million years ago) with the tribes Oryzeae and Pooideae (wheat and oats) diverging about 35 mya [reviewed by Kellogg (2009)

Methods for developing Oryza interspecific mapping populations
Traits are classified as either qualitative or quantitative traits. Qualitative traits are controlled by one or a few genes with major effects while quantitative characters are controlled by many genes with minor effects (Poehlman and Sleper 1994). Identification of genes associated with quantitative traits is always more complicated compared to those involving qualitative traits.
Interspecific and intergenomic hybridization, hybridization between species with the same or different genomes, have been used to transfer desirable genes or QTL associated with simple or complex traits from wild species into a cultivated genetic background (Brar and Khush 1997;Dalmacio et al. 1995;Tanksley and McCouch 1997). Nevertheless hybridization success can be hindered by genomic incompatibilities and sterility barriers (Ishii et al. 1994;McCouch et al. 2007; Wang et al. 2005). The utilization of embryo rescue and other methods of producing viable and fertile hybrids combined with robust molecular markers and associated computational and statistical analyses, led to the successful generation of interspecific genetic populations that were used to link desirable traits to molecular markers and subsequent identification of the actual genes controlling the traits of interest (

Recombinant Inbred Line (RIL) population
RIL populations have been the most common type of mapping population used in rice genetics and breeding when both parents are O. sativa but a limited number of interspecific populations have been reported. To develop a RIL population, two contrasting cultivars or accessions for the trait(s) to be mapped are crossed together to create an F 1 hybrid. By successive self-pollination starting from the F 1 generation, subsequent generations of segregants are produced (up to F 3 ), representing multiple rounds of recombination and eventually fixation to homozygosity towards either of the parental alleles (Fig. 1). This derived population is advanced for several generations by the single seed descent (SSD) method, where a single F 3 seed from each F 2 plant is planted to produce the F 4 generation, subsequently a single F 4 seed is selected from each line to produce the F 5 generation with the SSD method usually continuing until F 8 seed are produced. At the F 7 , the RILs exhibit genetic homogeneity, such that the genomic contribution of each parent is fixed, and together these RILs compose a mapping population. If selections are being made for improved lines with a particular trait(s), this selection often begins in the F 5 -F 6 if individual plants can be selected for the trait; otherwise, the selection is postponed to later generations (F 7 -F n ) (Nguyen et al. 2003, Poehlman andSleper 1995). The procedure continues until the superior lines with desirable traits are produced.
The main advantage of the RIL method is that no backcrossing is necessary but when a wild Oryza species is a parent, often undesirable traits associated with the wild parent, especially shattering and sterility are problematic, thus it is often necessary to backcross. RIL populations are suitable for identifying major gene(s) or QTL(s), and to detect genetic interactions such as epistasis (Fukuoka et al. 2010). Other advantages are, the individual Figure 1. A comparison of the methods for creating primary and advanced bi-parental mapping populations, including recombinant inbred lines (RILs), backcross inbred lines (BILs), chromosome segment substitution lines (CSSLs) and near isogenic lines (NILs) as summarized by Fukuoka et al. (2010). Also shown are the number of backcrosses (BC) required and the genotypes of the lines obtained by each method. Karyotypes of the three CSSLs illustrate how chromosome 1 of the donor can be introgressed into the recurrent parent. The three NIL genotypes are based on the JeffersonNILs, each with a different O. rufipogon (IRGC105491) introgression selected for a different yield QTL (Imai et al. 2013).
RIL may contain more than one introgressed segment in their chromosomes, representing different recombination events and a higher recombination frequency. As a result, fewer progeny lines are required to cover the complete donor genome as compared to other types of bi-parental mapping populations that include a backcross generation. Moreover, epistatic effects can be detected in RILs due to the presence of several introgressed segments in each line (Keurentjes et al. 2007). Because several segments of each parent are present in each individual line composing the population, there is less homogeneity in RIL populations as compared to most other types of populations. This heterogeneity is easy to observe and provides an excellent opportunity for phenotypic evaluation. In summary, the RIL method has proven to be useful when both parents are O. sativa but with interspecific and intergenomic crosses, backcrossing is often necessary (Fukuoka et al. 2010).
Commonly used softwares for creating the linkage map from the genotypic (molecular marker) data of the population for QTL analyses include MapMaker-QTL (Lander and Botstein 1989), JoinMap (Van Ooijen 2006) and MapDisto ). The possible chromosome location of the QTL for the trait being evaluated is based on the QTL having a significant LOD [logarithm (base 10) of odds] score with the LOD score detecting linkage between the molecular marker and the trait of interest. Several softwares are freely available for conducting the QTL analysis, including MapMaker-QTL (Lander and Botstein 1989), QTLCartographer ), QGene (Joehanes and Nelson 2008), MapDisto  and QTLNetwork . It is important to confirm that the software being used for QTL analysis can correctly analyze the population type since some cannot be used with BC 2 F 2 populations based on differences in fundamental assumptions. Most recent QTL analyses with rice have been performed using either composite interval mapping (CIM) (Zeng 1994) or multiple interval mapping (MIM) (Kao and Zeng 1999) with single point analysis (SPA) (Tanksley et al. 1982), marker regression (Kearsey and Hyne 1994) and interval mapping (IM) (Haley and Knott 1992; Lander and Botstein 1989) being used in earlier analyses.

Advanced Backcross (AB) population
The advanced backcross (AB)-quantitative trait locus (QTL) analysis is a powerful strategy to map desirable trait(s) discovered in the wild species (Tanksley and Nelson 1996). This method was first applied to QTL mapping in tomato, and subsequently to several other crops, including rice (Grandillo and Tanksley 2003;McCouch et al. 2007). In the process of developing the AB populations used for QTL analysis, plants or lines with unfavorable genes derived from donor parents like sterility and sometimes shattering, are often eliminated from the population after phenotypic and genotypic evaluation. Due to artificial selection in favor of lines with desirable alleles and the genetic background from the recurrent parent, the distribution can be skewed toward the recurrent parent, therefore, after the BC 3 generation, the power of the statistical analysis to detect QTL decreases. Since sequential backcrossing in AB-QTL removes epistatic interactions, the chance of detecting QTLs with epistatic interactions among alleles from the donor parent decreases, while the ability to detect additive QTLs increases (Tanksley and Nelson 1996;Grandillo and Tanksley 2003).
To create an AB mapping population, one parent, usually the wild Oryza species, identified as the donor parent, is crossed with the recurrent parent, usually an elite cultivar, which will be crossed with the hybrid parent in subsequent crosses (illustrated in Ali et al. 2010). Often the donor parent is used as a male and the recurrent parent as the female to avoid the cytoplasmic male sterility and because it is usually easier to emasculate the cultivated parent. The F 1 plant(s) is one parent in the second generation and it is crossed with the recurrent parent, which is defined as backcrossing. The resulting first backcross generation (BC 1 ) may be backcrossed again with the recurrent parent to generate a BC 2 population. If the BC 2 progeny are sterile, it is best to advance the population to the BC 3 generation by crossing the BC 2 plants to the recurrent parent a third time. After the progeny lines are advanced to the BC 2 (or BC 3 ) generation and allowed to self pollinate, these BC 2 F 2 (or BC 3 F 2 ) progeny plants are grown to collect phenotypic and genotypic data for the QTL analysis. After the AB-QTL mapping, the AB population can be advanced by (a) allowing all the progeny lines to self-pollinate and be advanced by SSD for three to four additional generations, thus developing a BIL population or (b) backcrossing the progeny lines additional generations to develop a library of CSSLs or NILs for targeted traits (Fig. 1).

Backcross Inbred Line (BIL) population
BIL populations are used to introgress desirable traits from the wild Oryza species donor into rice with the potential of improving the agronomic performance of elite cultivars and develop mapping populations (Fig. 1). After backcrossing, as described in the aforementioned AB population development, the individual lines, BC 1 , BC 2 or BC 3 generation, are self-pollinated for about four generations to the BC 2 F 5 , as described in the RIL population development. If a specific trait is being selected, the BILs will be screened for that trait and backcrossed as The advantages of utilizing BILs are that the method is relatively straightforward and the lines are more homogeneous, having less linkage drag and fewer untargeted segments from the donor parent as compared to RILs. Furthermore, BIL populations can be used to identify major QTLs and single genes, detect QTLs with epistatic or additive effects, as well as, provide an accurate estimation of genotype x environment interactions. It takes more time to develop a BIL population than a RIL population but less time than developing CSSLs and NILs because there are fewer backcrosses to do and less emphasis on targeted segments (Fukuoka et al. 2010;Fulton et al. 1997;Jaquemin et al. 2013). Some disadvantages of this method are the genetic background of the donor parent is higher in the BILs as compared to the CSSLs and NILs, and the lines require more phenotypic evaluation but less genotypic characterization. As a result, mapping in a BIL population is more labor intensive and costly compared to RILs but less costly than NILs and CSSLs. Unfortunately, only limited success has been reported for improving quantitative traits with low heritability and identifying minor QTLs. Also, it is difficult to transfer a relatively large number of genes or QTLs associated with the desirable traits from the wild donor to an elite cultivar using lines selected from a BIL population.

Chromosome Segment Substitution Line (CSSL) library
A CSSL "library" is a set of near isogenic lines, often ranging from 26 to 80 lines, which cover the entire donor genome when the segments included in each introgression line are in the background of the recurrent parent ( Fig. 1; Ali et al. 2010). The concept of CSSL libraries was initially proposed by Eshed and Zamir (1995) as introgression lines and Ghesquière et al. (1997) as contig lines. To develop CSSLs, the initial crossing follows the same scheme as described for AB and BIL populations where the wild, unadapted Oryza species is the donor parent and the recurrent parent is usually an elite cultivar. To confirm the entire donor genome is included in the CSSL library, a set of polymorphic markers is often used to assist in selecting lines for each generation, beginning with the BC 1 F 1 generation. To develop a CSSL library usually requires backcrossing to the recurrent parent for three to four additional generations (BC 4 F 1 or BC 5 F 1 ). The set of polymorphic markers can be used each generation to confirm the targeted segment is present in each line composing the CSSL library as illustrated in Ali et al. (2010). Alternatively, several hundred lines can be backcrossed for 4 to 5 generations and a CSSL library can be selected after genotyping in the BC 4 or BC 5 generation. Once the desired BC 4:5 F 1 lines are selected, the lines are self-pollinated to achieve homozygosity and the lines homozygous for the individual targeted segment are selected from the BC 4:5 F 2 progeny lines. The BC 4:5 F 3 seed is used to establish the CSSL library composed of a set near isogenic lines covering the entire donor genome Fukuoka et al. 2010).
A CSSL library has several advantages compared to BILs or an AB mapping population in that it can be used for fine mapping, to identify both major and minor QTLs, and validate genetic interactions. Also, due to the recurrent parent background in CSSLs, linkage drag and its negative effects on the QTL studies are significantly reduced or eliminated. This uniform genetic background enables one to make rapid progress in linkage mapping of targeted QTLs. Lastly, individual CSSLs which carry a specific trait can be used for fine

Near Isogenic Lines (NILs)
The procedure for developing a set of NILs is similar to CSSLs except the number of backcrosses is unlimited because the focus is on incorporating a single segment with the trait(s) of interest identified in the Oryza species donor into the background of the recurrent parent ( Fig.  1). With NILs, the focus is on a particular set of lines for the trait(s) of interest, not covering the entire donor genome as with a CSSL library. As with CSSLs, once the targeted segment is introgressed into the recurrent parent background, the pre-NIL lines are allowed to selfpollinate, so that the NILs will be homozygous for the targeted segment. Molecular markers, such as SSRs and SNPs, are used to select for the targeted segment and determine the number of chromosomal segments from the donor parent remaining in the background (Fukuoka et al. 2010).
NILs are often developed to fine map QTLs identified in primary mapping populations, like RIL or BIL, because the QTLs can be mapped precisely as single Mendelian factors (McCouch et al. 2007). Use of NILs, like CSSLs, increases the power to detect small-effect QTL and can overcome or minimize genetic incompatibility, linkage drag, cytoplasmic sterility and epistatic effects, all of which are common obstacles in wide hybridization efforts because the genetic background is more or less uniform. Although developing NILs, like CSSLs, is labor intensive, time consuming, and expensive, NILs are a valuable tool for exploring the genes underlying QTLs because the epistatic effects are removed or minimized making it easier to measure gene expression (Keurentjes et al. 2007). Finally, those NILs with valuable genes introgressed from the wild Oryza species donor, can be used as parental lines in breeding programs.

Multi-parent Advanced Generation Inter-Cross (MAGIC) population
Recently, some efforts have turned to MAGIC populations (Cavanagh et al. 2008;Kover et al. 2009) which can serve the dual purpose of permanent mapping populations for precise QTL mapping, and for direct or indirect use in variety development, especially when the parents used to develop the population are the source of agronomically useful traits (Bandillo et al. 2013). MAGIC populations are developed by systematically crossing several F 1 hybrids involving four to sixteen different parental lines to create a set of double crosses, then systematically crossing the double cross hybrids to create a set of 4-, 8-or 16-way crosses. As the final step, the lines composing the population are advanced four or more generations by single seed descent to obtain a set of advanced intercrossed lines (AILs). Bandillo  . In most cases, for this population, the initial crosses had O. sativa as the female parent, and the goal is to produce 16-way crosses with highly mixed genomes.

Yield enhancing QTL from exotic Oryza genomes
Several plant traits directly or indirectly affect rice grain yield including days to heading and maturity; plant height; panicle length; number of panicles per plant, spikelets per panicle and grains per panicle; seed set; grain weight; grain size and shape; and shattering (  Thomson et al. 2003). Yield improvement can be achieved as a result of the vast allelic diversity for these traits found in interspecific populations, especially number of grains per panicle which has proven to have the greatest relevance for rice breeding programs Liu et al. 2008;Tian et al. 2006).
Modern rice varieties are developed after an extensive selection process to improve a few targeted traits related to cultivation and end-use quality but primarily those associated with yield components, such as resistance to shattering, compact growth habit and improved seed germination (Tanksley and McCouch 1997). This prolonged breeding procedure can lead to a reduction in the genetic variability found in modern cultivated rice (Rangel et al. 2008), thus identifying genetic sources for agronomically important traits from wild Oryza species and introgressing them into cultivated rice is desirable and necessary. Although wild Oryza species are inferior in grain yield, especially when compared to cultivated rice, transgressive segregation resulting from a cross between cultivated rice and a wild To evaluate the effect of O. minuta (IRGC101141) with a BBCC genome on yield components, a single plant, WH79006, was selected from the Hwaseongbyeo x O. minuta BC 5 F 3 families and selfed (Jin et al. 2004). QTL analysis of Hwaseongbyeo x WH79006 F 2:3 progeny identified four QTLs, sw7 (seed width), sl11 (seed length), tsw7 (1000-seed weight) and lw10 (seed length to width ratio). Similarly, WH29001 was selected from the BC 5 F 3 families, selfed and by QTL analysis the co-located QTLs for days to heading, dth6 and dth8, and awn length, awn6 and awn8, were identified on chromosomes 6 and 8, respectively (Linh et al. 2006). Subsequently, a new QTL, spp7, for spikelets per panicle, was detected on the long arm of chromosome 7 with the allele attributed to the O. minuta parent and validated in the F 3  Awns are an important trait in wild rice species because it protects the seeds from birds and other animals. By contrast, the majority of modern rice cultivars have short awns so that it is easier to harvest the seed. This trait is reported to be controlled by several genes located in different chromosomes, including

Genes for sterility
Reproductive barriers, such as crossability, hybrid seed inviability, hybrid sterility and hybrid breakdown, have significantly limited the success of interspecific hybridization between O. sativa and non-A genome Oryza species. Several studies reported the production of F 1 seeds by crossing male sterile lines and Oryza species (Huang et al. 2001;Luo et al. 2000). The crossability rate between O. sativa and other Oryza species vary; however, the rate of crossability between A-genome and non-A genome diploid Oryza species is higher than with tetraploid Oryza species, none of which has an A-genome (Jena andKhush 1989 &1990;Yasui and Iwata 1991).
The phenomenon of transmission ratio distortion (TRD) where one allele is transmitted more frequently than the opposite allele in interspecific and intraspecific hybrids has been discovered in a broad range of organisms and is often a reproductive barrier (

Grain quality traits
Acceptable rice grain quality is a major goal of rice breeding programs worldwide because it determines the acceptability of cooked rice to the consumer. Grain quality is a combination of several components including milling efficiency, physical appearance, cooking and eating characteristics, and nutritional quality (Aluko et al. 2004;. A few interspecific populations were evaluated for grain quality traits (Table 3). These studies showed the Oryza parent affects the apparent amylose content, alkali spreading value, protein content, rice bran percentage, milled rice percentage and seed size. What is desirable for these traits is determined for the most part, by consumer preference and marketing classes. When selecting for these traits, often the grain quality of the recurrent parent is preferred.
Most interesting was the BC 3 F 1 progeny of the Caiapó x O. glaberrima (IRGC103544, MG12) doubled haploid population (Aluko et al. 2004). For this population, the QTL analysis revealed 27 QTLs associated with rice quality of which seven QTLs including percent rice bran, percent milled rice, alkali spreading value (inversely related to gelatinization temperature), percent protein and grain dimensions (length to width ratio), were traced to alleles originating from the O. glaberrima parent.  Abbreviations for types of markers are: AFLP, amplified fragment length polymorphism; CAPS, cleaved amplified polymorphic sequence, InDel, insertion-deletion polymorphism, RAPD, random amplified polymorphic DNA; RFLP, restriction fragment length polymorphism; SNP, single nucleotide polymorphism; SSR, simple sequence repeat; STS, sequence-tagged site. Table 3. Summary of QTLs for grain quality, biotic stress tolerance, abiotic stress tolerance and biomass attributed to the non-O.sativa parent.

Disease resistance
Pathogenic microorganisms, such as fungi, oomycetes, viruses and bacteria, and pests, such as insect herbivores, significantly reduce rice seed yield and quality. The most destructive rice diseases include bacterial blight caused by Xanthomonas oryzae pv. oryzae Ishiyama Dye (Cheema et al. 2008), blast caused by the fungus Magnaporthe oryzae B. Couch (Couch and Kohn 2002), and sheath blight caused by the soil-borne fungus Rhizoctonia solani Kühn (Zhang 2007). The first reported successful introduction of an agronomically important trait from a wild Oryza species was the introgression of grassy stunt virus resistance from the AA-genome species O. nivara into the cultivated O. sativa genetic background ). The resistance mechanism was subsequently introduced into several other rice cultivars (Sanchez et al. 2013). Since this first introduction, wild Oryza accessions have been screened as a potential source of resistance genes to biotic and abiotic stresses, as well as, yield and yield components, as previously discussed. These screening efforts, including successful introduction of stress resistance genes from Oryza species were recently summarized by Ali et al. (2010), Brar and Singh (2011) and Sanchez et al. (2013). Table 3 summarizes the efforts to identify the chromosome location of stress resistance genes introduced from the wild Oryza species by QTL and fine mapping analyses.
Seed yield losses due to bacterial blight were reported to be up to 75% in India, Indonesia, and the Philippines, and 20 to 30% in Japan (  Stem rot, a fungal disease caused by Sclerotium oryzae Catt., causes yield losses by reduced tillering, more unfilled grains per panicle, chalky grain, lower milling yields and increased lodging (Ou 1985

Insect resistance
Genetic resistance is an effective method of protecting rice from insect pests in Asia and the Americas (Kiritani 1979; Way 1990) and avoids the possible environmental contamination associated with chemical control (Zhang 2007 The wild Oryza species have been used successfully as a source of novel alleles conferring resistance to both rice diseases and insect pests because in many instances these alleles could be transferred to O. sativa by backcrossing and screening the progeny. In fact, several of these alleles were successfully transferred even before the advent of molecular markers. With molecular markers, it is now possible to expedite the introduction of these novel alleles because marker-assisted breeding techniques can be used. With the molecular tools currently available, it should be possible to unravel those resistances which are quantitatively inherited like sheath blight.

Genes for abiotic stress
Abiotic stresses, including drought and flood, high and low temperatures, high salinity, high aluminum and acid sulfate soils, have a negative impact on rice productivity worldwide. Rice, like other plant species, has evolved to adapt to different environmental stresses using different mechanisms and strategies with multiple sensors. When the sensors identify a stress, a signal transduction pathway is invoked, which activates genes conferring the initial response for short term or long term tolerance to the stress (Grennan 2006;Lexer and Fay 2005). Recent studies identified many genes involved in plant tolerance to abiotic stress, which are classified into two groups based on their products. The first group includes genes that protect the cells via synthesis of chaperones, a group of proteins that help non-covalent folding and unfolding of other proteins in the cell under stress conditions, and enzymes for protecting metabolites and proteins. The second group includes those genes that regulate stress responses acting as transcriptional factors to control stress genes or by producing hormones (Grennan 2006 . Three Ctss-QTLs were detected with those on chromosomes 3 (qCtss 3) and 11 (qCtss11) attributed to the O. rufipogon parent, and on chromosome 10 (qCtss10) to Guang-luai 4. The major QTL, qCtss11, explained 40% of the phenotypic variation and using backcross progenies, it was fine-mapped to a 60kb candidate region defined by markers AK24 and GP0030 with Os11g0615600 and/or Os11g0615900 hypothesized as the causal gene(s) for cold tolerance.
Seedling cold tolerance was measured in the M-202 (medium grain, U.S. temperate japonica) x O. nivara (IRGC100195) AB-QTL population using a slant board method [Jones and Petersen 1976;Eizenga et al. (accepted)]. In this study, QTLs for increasing coleoptile length and shoot length were identified in the same region on chromosome 5 and attributed to the O. nivara parent. QTLs for increased shoot length and root length were found on chromosome 8 and 6, respectively, and attributed to the M-202 parent.

Tolerance to aluminum and acid soils
Aluminum toxicity is another abiotic stress that causes grain yield reduction especially when rice is grown in an acidic soil (IRRI 1978). If the soil pH falls below 5.5, aluminum will more likely separate from the soil colloids and come into a solution phase. Aluminum at toxic levels slows root development, reduces the plant's ability to take up water and nutrients, and decreases plant growth, consequently reducing grain yield and grain quality (Foy 1992). Application of lime to the soil, reduces soil acidity and improves soil fertility but the results have showed limited success in overcoming the effects of aluminum toxicity. Aluminum tolerance is a quantitative trait and varies among rice species. Both additive and dominance effects contribute to the genetic heritability of aluminum tolerance as documented by the importance of both general combining ability and specific combining ability (Howeler and Cadavid 1976; Wu et al. 1997).
In the past decade, one O. rufipogon (IRGC106424) accession found growing in an acid sulfate soil in Vietnam (Sanchez et al. 2013) has proven to be valuable for improving tolerance to both aluminum and acid sulfate soils in cultivated rice. Initially, Nguyen et al. (2003) evaluated 171 F 6 RILs derived from IR64 (indica, susceptible) x O. rufipogon (IRGC106424, tolerant) for aluminum tolerance. QTL analysis revealed QTLs for root length under stress conditions attributed to the O. rufipogon parent in six different chromosomal regions on chromosomes 1, 2, 3, 7, 8 and 9 that individually explained 9.0-24.9% of the phenotypic variation and were controlled by additive effects. The major QTL on chromosome 3, explaining 24.9% of the variation, was found to be conserved across cereal species. During the same time period, the tolerance to acid sulfate soils identified in this O. rufipogon accession was introgressed into the IR64 background through breeding efforts. The selected introgression line, IR73678-6-9-B, was released by IRRI as variety AS996 (Sanchez et al. 2013). AS996 is currently grown on 100,000 ha in the Mekong Delta and described as moderately tolerant to acid sulfate soils and tolerant to brown planthopper and blast.
Even though traits associated with abiotic stress are more difficult to evaluate because of environmental effects and interactions between genes, the development of the AS996 variety is an exciting success story. The release of Arizona-1 and Arizona-2 could make significant contributions to improving rice yields in areas where high temperatures routinely lower yield. With the improved molecular techniques for dissecting these traits and the gene functions related to abiotic stress, more significant advances should be made in the near future, especially as the scientific community provides the tools for rice producers to deal with global climate change.

Conclusions
The repositories of Oryza species accessions found around the world are a storehouse of novel alleles and traits lost during the evolution and domestication of cultivated rice as we know it today. The fact that introgression lines derived from crosses between Asian rice and it's ancestral species, O. rufipogon and O. nivara, exhibited notable improvement in yield and yield components through the phenomenon known as transgressive variation, was surprising and unexpected. The identification of novel alleles related to biotic stress, especially insect pests like brown planthopper and bacterial leaf blight, and more recently abiotic stresses like acid sulfate soils and drought, underscore the importance of mining these collections. The advent of molecular marker technology and development of mapping populations, especially AB-QTL and CSSL, have made it possible to map many of these alleles to chromosome location and begin to dissect the interactions between various genes. The fact that high quality genome sequences are now available or will soon be available, make it possible to interrogate the wild Oryza species accessions at a level that was not possible before. These resources will allow us to move swiftly beyond the first step of QTL identification to fine mapping traits of interest; introgressing desirable traits into elite breeding lines using markers within the gene, thus decreasing linkage drag; identifying genotype by environment interactions; determining the effect of epistasis (non-allelic genes) on traits of interest; discovering epigenetic effects such as histone modification or DNA methylation; and finally unraveling other genetic phenomenon like gene silencing. In summary, the interspecific and intergenomic mapping populations available or soon to be available, and the increased availability of SNP data, resequencing data and advanced statistical software, create even more opportunities to investigate novel alleles for agronomically important traits discovered in the Oryza species and increase our understanding of the mechanisms underlying these traits to deal with the challenges of climate change and feeding nine billion people.