Forward genetics is a powerful approach to identifying gene mutations that cause phenotypes of interest, and ultimately elucidating the biochemical and molecular mechanisms and signaling processes underlying plant development and adaptation. The forward genetic approach begins with physical or chemical mutagenesis of a pure (inbred) line, followed by screening for mutants with the desired phenotype. The resultant mutants are outcrossed to another pure line with extensive DNA polymorphism relative to the mutagenized parental line. Linkage of the mutant phenotype with DNA markers is analyzed, using as many markers as possible, until the mutation is mapped to a very small region flanked by two markers. The success of this process, also referred to as map-based cloning, depends on two critical factors: a large segregating mapping population and dense DNA markers with the resolution to delimit the mutation into a small region. Several hundreds to thousands of accurately phenotyped F₂ plants must be analyzed with large number of DNA markers in order to narrow the mutation to a region harboring only a few genes (Jander et al., 2002). The entire region is then sequenced in order to identify the gene that carries the causal mutation. To confirm the identity of the gene, the wild-type gene is introduced into the mutant through transformation to determine whether it can complement the mutant phenotype. This last step can be bypassed if two or more independent mutant alleles are identified that carry unique mutations in the same gene because the odd of finding two independent mutations by chance on the same gene that lead to the same phenotype is near zero (Jander et al., 2002).

Next-generation sequencing (NGS) techniques provide massive high-resolution genotypic data very quickly (Metzker, 2010), and can thus replace the costly and time-consuming linkage analysis required by conventional map-based cloning (Schneeberger and Weigel, 2011). Two general approaches have been developed for using NGS to identify causal mutations. The first strategy is represented by two similar approaches, mapping-by-sequencing (ShoreMap) (Schneeberger et al., 2009; Hartwig et al., 2012) and next-generation mapping (NGM) (Austin et al., 2011). These approaches are very similar to conventional map-based cloning. First, a mapping population is generated by outcrossing the mutant to a divergent line with sufficient DNA polymorphism, and homozygous mutants are selected from the segregating F₂ population. Then, equal quantities of genomic DNA from each of the selected lines are pooled for NGS analysis, according to the principle of bulk segregant analysis (BSA) (Michelmore et al., 1991). Because only homozygous mutants selected from the mapping F₂ population are subjected to whole-genome sequencing, the single-nucleotide polymorphic (SNP) markers surrounding the causal mutation will be derived predominantly from the mutant parent. Consequently, the region containing the causal mutation will form a “valley” devoid of polymorphic markers from the divergent parent. The region containing the causal mutation can be narrowed further with sophisticated bioinformatics tools, sometimes, to a single gene (Austin et al., 2011; Schneeberger and Weigel, 2011).

The second approach is represented by isogenic mapping-by-sequencing (MutMap), a variation of ShoreMap, designed to map mutant phenotypes subject to modification by genetic background (Abe et al., 2012; Hartwig et al., 2012; Zhu et al., 2012). In MutMap, the mutant is crossed to a wild-type plant of the same genetic background as the mutant, rather than to a divergent line as in ShoreMap, to avoid interference from a divergent genetic background. Homozygous mutants are selected from the backcrossed F₂ population and bulked for whole-genome sequencing. After identification of SNPs, the SNP ratio (i.e., the proportion of all short reads that contain the mutated SNPs) is plotted along each chromosome. The causal mutation for the phenotype is expected to have a SNP ratio of 1 (i.e., 100% mutant), whereas unlinked background mutations are expected to have ratios around 0.5. For mutations close to the causal mutation, the SNP ratio should vary from <0.5 to 1 depending on genetic distance. This approach is particularly useful for mapping induced mutations (Abe et al., 2012; Zhu et al., 2012). Because major mutations capable of causing phenotypes are rare in induced mutant populations (Greene et al., 2003; Henry et al., 2014; Jiao et al., 2016; Krasileva et al., 2017), it is often possible to narrow down the candidates to one or a few mutations using 20–50 homozygous mutants selected from a segregating F₂ population (Abe et al., 2012; Takagi et al., 2015).

Sorghum is the fifth most important crop in the world, providing food, feed, and forage to humans and agricultural animals in many areas around the world (http://www.fao.org/home/en/). In addition, due to its excellent tolerance of drought and high-temperature stress, high water use efficiency, and high productivity, it has become an increasingly important crop for bioenergy production (Rooney et al., 2007; Vermerris, 2011; de Siqueira Ferreira et al., 2013; Mullet et al., 2014). Sorghum accumulates higher levels of epicuticular wax (EW) on its aerial surface than most other crops, and this trait is considered to make an important contribution to sorghum's high water use efficiency and drought tolerance (Jordan et al., 1984; Jenks et al., 1992; Premachandra et al., 1994; Samdur et al., 2003; Burow et al., 2008; Uttam et al., 2017). Many sorghum mutants that are devoid of EW, named “bloomless” (bm), have been isolated over the years but few genes responsible for the deposit of massive EW on its aerial surface have been identified (Weibel, 1986a,b; Peters et al., 2009; Jiao et al., 2016).

Here, we report an efficient method for identifying causal gene mutations by whole genome re-sequencing of bulked F₂ populations, and demonstrate the approach using two independent alleles of sorghum bm mutants devoid of EW. We chose the bm mutants for allelic mutant mapping because we have isolated a large collection of bm mutants from the sorghum mutant library generated with ethyl methane sulfonate (EMS) in an elite inbred line BTx623 and the characteristic shining green sheath that can be easily distinguished from the sheath of wild type sorghum (Burow et al., 2008, 2009; Peters et al., 2009). Our method offers a tool to identify most Bm genes in sorghum. These Bm genes may facilitate improvement of water use efficiency and drought tolerance, not only in sorghum, but also in other crops (Weibel, 1986a,b; Peters et al., 2009; Jiao et al., 2016; Uttam et al., 2017).

Here, we demonstrated that sequencing of two independent allelic mutants represents a powerful approach to identifying causal gene mutations in sorghum. Our approach is very similar to MutMap through bulked segregant analysis, described in rice (Abe et al., 2012). Conceptually, MutMap is a straightforward way to identify the causal homozygous mutation in bulked F₂ plants selected for a phenotype of interest. Because the mutant phenotype is preselected, the causal mutation is expected to present in all short sequence reads covering the mutation site, whereas the ratio of mutant to WT SNPs will be 0.5 for all unlinked mutations. For linked mutations, this ratio varies from 1 to <0.5 depending the distance to the causal mutation. In practice, however, the situation is more complicated. Sequencing errors in the reference genome and pre-existing SNPs in BTx623 prior to mutagenesis will all be identified as homozygous mutations (Figure 3A), making it challenging to determine which homozygous mutations are causal. In a previous study, we sequenced BTx623 and 256 selected lines from a mutant population generated in this line (Jiao et al., 2016). From sequenced BTx623 and mutant lines, we annotated over 80,000 SNPs that differ from the sorghum reference genome sequence (Paterson et al., 2009). These SNPs serve as a powerful filter for removing background SNPs from BSA. For example, during the mapping of both bm40-1 and bm40-2, huge number of mutations (13,419 in bm40-1 and 11,677 in bm40-2) mutations were identified in the bulked bm40-1 and bm401-2 populations. After removing background mutations by subtracting the 80,000 SNPs annotated from the sequenced mutant lines, the numbers of raw mutations were greatly reduced (Figure 3A).

Another challenge of BSA is identification of the causal mutation from closely linked mutations, but again, this hurdle can be overcome by sequencing two or more allelic mutants. In our study, we identified 5 linked homozygous mutations in bm40-1 and 13 in bm40-2. Comparison of these two populations revealed that only one gene, Sobic.001G228100, had homozygous mutations in both bulked F₂ sequences. Furthermore, the mutation site in bm40-1 bulked F₂ was a missense mutation in the conserved GDSL domain, whereas the mutation in bm40-2 bulked F2 altered the splice junction from the donor side. It is less likely for two allelic mutants with same phenotypes to have mutations in different genes. Thus, this gene is most likely the candidate for Bm40. Based on the average number of deleterious mutations of 147 per line in the sequenced mutant library (Jiao et al., 2016), the gene mutation rate per line per generation would about 0.004 (147/34,000) by chance. For two mutations occurred randomly in one gene would be 1.6 × 10⁻⁵, a nearly impossible event. Similar approach has been reported in identifying the causal mutations that affect germination of lettuce seeds under high temperatures (Huo et al., 2016). Because of the advantage of sequencing two independent alleles by BSA in overcoming the limitations of MutMap, This method should be named allelic MutMap. This method is an efficient and cost-effective approach to identifying causal gene mutations when an allelic mutant series is available. Based on current sequencing prices, the cost to sequence two bulked samples to 10 Gb data or ~15x coverage is <$500 in sorghum.

For a large number of mutants with similar phenotypes, such as the sorghum bm mutants, it is impractical to perform pairwise allelic test. It may be possible to sequence many bulked F₂ populations first without prior knowledge of allelic relationship. If two or more bulked F2 population harbor unique mutations on a common gene, complementation tests can then be performed only among mutants with the same mutated gene to confirm that they are allelic. Reciprocal crosses and cosegregation analysis maybe required to further confirm the identification of the causal gene. In this manner, the time and expense required for complementation testing among large numbers of mutants could be reduced substantially. In light of the continuing decrease in sequencing costs and increase in data output and quality, it will soon be feasible to use this approach to sequence large numbers of bulked F₂ populations. It is likely that some bulked F₂ populations possess several linked gene mutations but no common mutated genes are found from other bulked F₂. In these cases, complementation of the mutant phenotype with the wild type gene through transformation will be needed to confirm the identification of the causal mutation.

The importance of epicuticular wax to the drought tolerance in sorghum has long been recognized. Many bm mutants have been registered and genetically tested (Weibel, 1986a,b; Peters et al., 2009). A total of 38 bm loci were identified through genetic complementation (Peters et al., 2009), and a bm locus from a natural mutation was recently mapped (Uttam et al., 2017), bringing the total number of bm loci to 39. From the pedigreed mutant library, we identified 107 independent bm mutants (Jiao et al., 2016). Despite the large number of sorghum bm mutants collected thus far, few causal genes have been identified. Based on the Arabidopsis gene networks involved in EW accumulation (Kunst and Samuels, 2003; Bernard and Joubès, 2013; Fich et al., 2016), we searched the 256 sequenced core mutant lines and identified seven sorghum homologs in which mutations cause bm phenotypes (Jiao et al., 2016). Due to the massive accumulation of EW in sorghum, it is reasonable to speculate that sorghum employs multiple genes to regulate wax accumulation in its aerial surface. Identification of these genes and elucidation of their regulation may facilitate manipulation of surface wax accumulation in other crops with the goal of increasing their drought tolerance. To identify the genes involved in EW accumulation in sorghum, we established an efficient forward genetic method to identify the causal mutations defined by the vast collection of sorghum bm mutants, which can potentially identify all causal mutations of sorghum bm mutants in the near future.

Another GDSL-like lipase gene, Sobic.001G269200, was implicated in a bm QTL mapped from a natural accession RS647 (Uttam et al., 2017). However, because Sobic.001G269200 is more than 13 Mb from the bm40 on chromosome 1 and belongs to a different clade (Figure 4A), these two GDSL lipases may represent two separate genes. To determine whether this gene is also involved in EW accumulation, we searched the sequenced mutant database (Jiao et al., 2016). We identified 10 mutations in the gene, including one stop-gain mutation in the sequenced mutant line ARS192. However, none of these 10 lines segregated for the bm phenotype (Supplementary Table 1). We also identified a heterozygous mutation in Bm40, Sobic.001G228100, in ARS29. This line was shown to segregate for the bm phenotype as a single recessive gene mutation (Supplementary Table 1). It is unclear if the bm QTL mapped from RS647 is related to Bm40. Nevertheless, Bm40 is likely to be a functional orthology of the tomato CD1 based on phenotypic similarity and protein sequence identity, but further evidence, such as enzyme assays or lipidomic profiling, is required to firmly establish whether Bm40 is indeed a cutin synthase. The allelic mutants and identification of the causal gene provide critical materials for such future studies.

Using allelic MutMap, we showed that the causal gene in bm40 mutants encodes a GDSL-like lipase, similar to tomato cutin synthase 1. The identity of the Bm40 gene is supported by two independent allelic mutations and a nonsynonymous mutation from a sequenced mutant library. Our results demonstrate that the allelic MutMap is an efficient and cost-effective approach for identifying causal gene mutations. This approach could be used to identify causal mutations for other phenotypes for which allelic mutants are available. Identification of the causal gene for the bm phenotype provides critical insight into the accumulation of massive amounts of EW on the aerial surface of sorghum plant and the superior drought tolerance of this crop.

Mention of trade names or commercial products in this article is solely for the purpose of providing specific information and does not imply recommendation or endorsement by the U.S. Department of Agriculture. USDA is an equal opportunity provider and employer.

ZX and DW: Conceived the idea of this work; ZX: Isolated the mutants; ZX, GB, NG, VA-M, JC, and JB: Conducted the phenotype analysis; YJ: Analyzed the genomic data for the identification of the gene; YJ and ZX: Drafted the manuscript. All authors participated in the revision and agree with the final manuscript.