Sorghum MSD3 Encodes an ω-3 Fatty Acid Desaturase that Increases Grain Number by Reducing Jasmonic Acid Levels

Grain number per panicle is an important component of grain yield in sorghum (Sorghum bicolor (L.)) and other cereal crops. Previously, we reported that mutations in multi-seeded 1 (MSD1) and MSD2 genes result in a two-fold increase in grain number per panicle due to the restoration of the fertility of the pedicellate spikelets, which invariably abort in natural sorghum accessions. Here, we report the identification of another gene, MSD3, which is also involved in the regulation of grain numbers in sorghum. Four bulked F2 populations from crosses between BTx623 and each of the independent msd mutants p6, p14, p21, and p24 were sequenced to 20× coverage of the whole genome on a HiSeq 2000 system. Bioinformatic analyses of the sequence data showed that one gene, Sorbi_3001G407600, harbored homozygous mutations in all four populations. This gene encodes a plastidial ω-3 fatty acid desaturase that catalyzes the conversion of linoleic acid (18:2) to linolenic acid (18:3), a substrate for jasmonic acid (JA) biosynthesis. The msd3 mutants had reduced levels of linolenic acid in both leaves and developing panicles that in turn decreased the levels of JA. Furthermore, the msd3 panicle phenotype was reversed by treatment with methyl-JA (MeJA). Our characterization of MSD1, MSD2, and now MSD3 demonstrates that JA-regulated processes are critical to the msd phenotype. The identification of the MSD3 gene reveals a new target that could be manipulated to increase grain number per panicle in sorghum, and potentially other cereal crops, through the genomic editing of MSD3 functional orthologs.


Introduction
Grain number per panicle is a major determinant of grain yield in sorghum (Sorghum bicolor (L.) Moench) and other cereal crops [1][2][3][4][5][6]. Increased grain number and grain size, which are directly related to improved grain yield, are common goals during domestication of cereal crops, resulting in selection Table 1. Complementation tests between msd3, msd2, and msd1 mutants. The msd3 mutants were crossed to each other and to msd1-1 and msd2-1 mutants by hand emasculation and manual pollination. The panicle phenotypes of the F 1 plants were evaluated during the grain filling stage.

Identification of the MSD3 Gene
The MSD3 gene was identified through bulk segregant analysis of whole-genome sequencing data with an in-house bioinformatics pipeline, as described in rice and sorghum [18,19]. Previously, we crossed p6, p14, p21, and p24 to BTx623 and derived four F2 populations. After bioinformatics analysis of the four bulked F2 pools, we identified only one gene (Sorbi_3001G407600) that carried homozygous mutations in all four bulked F2 pools ( Figure 2). The genomic sequence of MSD3 is 3132 bp in length, with a CDS of 1356 bp that encodes a protein of 451 amino acids. Two putative msd mutants, p21 and p24, harbored a mutation that converted the G residue at Chr01_69163608 to A, creating a splice site mutation at the junction of the third exon and the third intron ( Figure 3). Both were renamed as msd3-1. The mutant p14, renamed msd3-2, harbored a G-to-A transition at Chr01_69163762 that created a nonsynonymous mutation, R240W, in the MSD3 protein. Mutant p6, renamed msd3-3, harbored a G-to-A transition at Chr01_69164229, resulting in a premature stop

Identification of the MSD3 Gene
The MSD3 gene was identified through bulk segregant analysis of whole-genome sequencing data with an in-house bioinformatics pipeline, as described in rice and sorghum [18,19]. Previously, we crossed p6, p14, p21, and p24 to BTx623 and derived four F 2 populations. After bioinformatics analysis of the four bulked F 2 pools, we identified only one gene (Sorbi_3001G407600) that carried homozygous mutations in all four bulked F 2 pools ( Figure 2). The genomic sequence of MSD3 is 3132 bp in length, with a CDS of 1356 bp that encodes a protein of 451 amino acids. Two putative msd mutants, p21 and p24, harbored a mutation that converted the G residue at Chr01_69163608 to A, creating a splice site mutation at the junction of the third exon and the third intron ( Figure 3). Both were renamed as msd3-1. The mutant p14, renamed msd3-2, harbored a G-to-A transition at Chr01_69163762 that created a nonsynonymous mutation, R240W, in the MSD3 protein. Mutant p6, renamed msd3-3, harbored a G-to-A transition at Chr01_69164229, resulting in a premature stop codon 4 of 15 (W321*). Subsequently, we identified another mutation in the MSD3 gene from the sequenced mutant library (ASR106, 25M2-1370) at Chr01_69165175, which created a splice site mutation at the junction of the seventh exon and the seventh intron [20]. The homozygous mutant (p37) exhibited the expected msd phenotype and was named as msd3-4. codon (W321*). Subsequently, we identified another mutation in the MSD3 gene from the sequenced mutant library (ASR106, 25M2-1370) at Chr01_69165175, which created a splice site mutation at the junction of the seventh exon and the seventh intron [20]. The homozygous mutant (p37) exhibited the expected msd phenotype and was named as msd3-4. Figure 2. Identification of the Msd3 gene by MutMap. Four F2 populations, BTx623*p6, BTx623*p14, BTx623*p21, and BTx623*p24, were subjected to sequencing of 20 bulked F2 mutants. The number in each circle represents the number of homozygous deleterious mutations. Only one gene, Sorbi_3001G407600, was commonly mutated in all four populations. In p21 and p24, the MSD3 gene was mutated at the same position. To determine whether these mutations in the MSD3 gene resulted in the msd phenotype, we made pairwise crosses among these four msd3 mutants; all F1 plants from these crosses exhibited the msd phenotype (Table 1). By contrast, F1 plants resulting from crosses between msd3 and msd1, msd3 and msd2, or msd2 and msd1 exhibited the wild type panicle phenotype. Together, these results indicate that Sorbi_3001G407600 is the MSD3 gene. Furthermore, we tested the co-segregation of the Identification of the Msd3 gene by MutMap. Four F2 populations, BTx623*p6, BTx623*p14, BTx623*p21, and BTx623*p24, were subjected to sequencing of 20 bulked F2 mutants. The number in each circle represents the number of homozygous deleterious mutations. Only one gene, Sorbi_3001G407600, was commonly mutated in all four populations. In p21 and p24, the MSD3 gene was mutated at the same position. codon (W321*). Subsequently, we identified another mutation in the MSD3 gene from the sequenced mutant library (ASR106, 25M2-1370) at Chr01_69165175, which created a splice site mutation at the junction of the seventh exon and the seventh intron [20]. The homozygous mutant (p37) exhibited the expected msd phenotype and was named as msd3-4. Figure 2. Identification of the Msd3 gene by MutMap. Four F2 populations, BTx623*p6, BTx623*p14, BTx623*p21, and BTx623*p24, were subjected to sequencing of 20 bulked F2 mutants. The number in each circle represents the number of homozygous deleterious mutations. Only one gene, Sorbi_3001G407600, was commonly mutated in all four populations. In p21 and p24, the MSD3 gene was mutated at the same position. To determine whether these mutations in the MSD3 gene resulted in the msd phenotype, we made pairwise crosses among these four msd3 mutants; all F1 plants from these crosses exhibited the msd phenotype (Table 1). By contrast, F1 plants resulting from crosses between msd3 and msd1, msd3 and msd2, or msd2 and msd1 exhibited the wild type panicle phenotype. Together, these results indicate that Sorbi_3001G407600 is the MSD3 gene. Furthermore, we tested the co-segregation of the To determine whether these mutations in the MSD3 gene resulted in the msd phenotype, we made pairwise crosses among these four msd3 mutants; all F 1 plants from these crosses exhibited the msd phenotype (Table 1). By contrast, F 1 plants resulting from crosses between msd3 and msd1, msd3 and msd2, or msd2 and msd1 exhibited the wild type panicle phenotype. Together, these results indicate that Sorbi_3001G407600 is the MSD3 gene. Furthermore, we tested the co-segregation of the msd3-4 mutation by Kompetitive allele-specific PCR (KASP) (Figure 4). Among 63 individual F 2 plants derived from the cross of BTx623 * msd3-4 (p37), 15 plants were scored as AA at the mutation site, and all 15 lines exhibited the expected msd3 phenotype. Fifteen plants were scored as GG, and 33 were scored as GA; these two genotypic classes, GG (wild-type, WT) and GA (heterozygote), exhibited that the wild-type panicle structure indicating that the msd3 mutation is recessive. These results also showed that the causal mutation in the msd3-4 co-segregated perfectly with the msd phenotype.  (Figure 4). Among 63 individual F2 plants derived from the cross of BTx623 * msd3-4 (p37), 15 plants were scored as AA at the mutation site, and all 15 lines exhibited the expected msd3 phenotype. Fifteen plants were scored as GG, and 33 were scored as GA; these two genotypic classes, GG (wild-type, WT) and GA (heterozygote), exhibited that the wild-type panicle structure indicating that the msd3 mutation is recessive. These results also showed that the causal mutation in the msd3-4 co-segregated perfectly with the msd phenotype.
FAD7 appears to be the major ω-3 fatty acid desaturase in sorghum. To determine the relative transcript abundance of the two plastidial ω-3 fatty acid desaturases in the developing panicle of the wild type BTx623, we compared the transcript levels of FAD7 and FAD8 using qPCR with three biological and three technical replicates. Comparing with the internal gene control EIF4α, the average cycle threshold (Ct) to detect FAD8 transcript was 24.2 cycles, while the Ct number to detect FAD7 transcript was 19.8. This difference in Ct translated to a relative abundance of FAD7 transcript 21fold over FAD8 according to the calculation method as described previously [26]. This relative abundance was largely consistent with the trends of online data ( Figure 3S).
To determine which of two plastidial ω-3 fatty acid desaturases were affected in the msd3 mutants, we compared the abundance of FAD7 and FAD8 transcripts in the msd3-1 and msd3-3 SNP (single nucleotide polymorphism) analysis. The x-axis represents the endpoint fluorescence data at 483-533 nm (allele X or wild type allele) and the y-axis represents the endpoint fluorescence data at 523-568 nm (allele Y or mutant allele). Genotype for individual F 2 plants was determined by cluster analysis of the endpoint KASP assay. AA, homozygous msd3-4; GG, homozygous wild-type (WT); and GA, heterozygous WT.
FAD7 appears to be the major ω-3 fatty acid desaturase in sorghum. To determine the relative transcript abundance of the two plastidial ω-3 fatty acid desaturases in the developing panicle of the wild type BTx623, we compared the transcript levels of FAD7 and FAD8 using qPCR with three biological and three technical replicates. Comparing with the internal gene control EIF4α, the average cycle threshold (Ct) to detect FAD8 transcript was 24.2 cycles, while the Ct number to detect FAD7 transcript was 19.8. This difference in Ct translated to a relative abundance of FAD7 transcript 21-fold over FAD8 according to the calculation method as described previously [26]. This relative abundance was largely consistent with the trends of online data ( Figure S3).
To determine which of two plastidial ω-3 fatty acid desaturases were affected in the msd3 mutants, we compared the abundance of FAD7 and FAD8 transcripts in the msd3-1 and msd3-3 mutants relative to the wild type BTx623 ( Figure 5). The abundance of FAD7 was reduced to 10% in msd3-1 and 30% in msd3-3 in comparison to the wild type FAD7 levels. Because the primers of FAD7 and FAD8 were designed based on the 3 -UTR sequences, which lie outside of the mutation sites, the low levels of the FAD7 and FAD8 transcripts may be due to truncated RNAs that do not encode active proteins. On the other hand, the transcript levels of FAD8 were increased by 3.8-fold and 1.5-fold in the msd3-1 and msd3-3 mutants relative to the levels of BTx623, respectively, indicating that only the transcript abundance of the FAD7 gene was significantly reduced in the msd3 mutants. The slight increase in the abundance of FAD8 gene suggested that the loss of function mutations in FAD7 gene may be partially compensated by the elevated expression of FAD8 ( Figure 5). This result suggested that expression of FAD7, but not FAD8, was greatly reduced in the msd3 mutants. mutants relative to the wild type BTx623 ( Figure 5). The abundance of FAD7 was reduced to 10% in msd3-1 and 30% in msd3-3 in comparison to the wild type FAD7 levels. Because the primers of FAD7 and FAD8 were designed based on the 3′-UTR sequences, which lie outside of the mutation sites, the low levels of the FAD7 and FAD8 transcripts may be due to truncated RNAs that do not encode active proteins. On the other hand, the transcript levels of FAD8 were increased by 3.8-fold and 1.5-fold in the msd3-1 and msd3-3 mutants relative to the levels of BTx623, respectively, indicating that only the transcript abundance of the FAD7 gene was significantly reduced in the msd3 mutants. The slight increase in the abundance of FAD8 gene suggested that the loss of function mutations in FAD7 gene may be partially compensated by the elevated expression of FAD8 ( Figure 5). This result suggested that expression of FAD7, but not FAD8, was greatly reduced in the msd3 mutants.
In addition to the two plastidial ω-3 fatty acid desaturases, the sorghum genome encodes two microsomal ω-3 fatty acid desaturases, Sorbi_3005G002800 and Sorbi_3008G003200, homologous to the Arabidopsis FAD3 gene [25,27]. We searched our mutation database from the 256 sequenced mutant lines [20] and identified nonsynonymous mutations in all three genes (Table 2S). However, none of the sequenced lines that harbored a mutation in one of the three ω-3 fatty acid desaturase genes exhibited a phenotype similar to the msd mutants.

Mutation in MSD3 Gene Dramatically Reduced the Levels of Linolenic Acid
To understand how mutations in the MSD3 (FAD7) gene lead to the dramatic change in panicle architecture and restore the fertility of the pedicellate spikelets, we first assessed the effect of the msd3 mutations on lipid composition of leaves and panicles of the msd3-1 mutant. In WT BTx623 leaves, 18-carbon fatty acids account for over 92% of total leaf lipids ( Figure 6, Table 2S). Levels of seven lipid species with molar percentage >1% are plotted in Figure 6. Galactolipids, including monogalactosyl diacylglycerol (MGDG) and digalactosyl diacylglycerol (DGDG), are the major lipids in leaves, accounting for over 90% of the total lipids. The lipid species with 36 carbon and six double bonds (36:6) consist of two linolenic acid molecules, whereas those with four double bonds (36:4) consist of two linoleic acid molecules, and species with five double bonds (36:5) consist of one linoleic acid and one linolenic acid. Sorghum has very low levels of 16-carbon fatty acids, and hexadecatrienoic acid (16:3) was not detectable in either leaves or panicles (Table 3S).
As can be seen from Figure 6, mutation in MSD3 decreased the levels of 36:6 lipid species and concomitantly increased the levels of 36:4 species. The ratio of linolenic acid to linoleic acid was 13.43 in the WT BTx623 leaves, falling to 2.48 in msd3-3 leaves (Table 3S). In general, panicles contained much less linolenic acid than in leaves. In WT BTx623 panicles, the ratio of linolenic acid to linoleic acid was 0.62, falling to 0.08 in the msd3-3 mutant panicles. Thus, msd3-3 panicles contained very little In addition to the two plastidial ω-3 fatty acid desaturases, the sorghum genome encodes two microsomal ω-3 fatty acid desaturases, Sorbi_3005G002800 and Sorbi_3008G003200, homologous to the Arabidopsis FAD3 gene [25,27]. We searched our mutation database from the 256 sequenced mutant lines [20] and identified nonsynonymous mutations in all three genes (Table S2). However, none of the sequenced lines that harbored a mutation in one of the three ω-3 fatty acid desaturase genes exhibited a phenotype similar to the msd mutants.

Mutation in MSD3 Gene Dramatically Reduced the Levels of Linolenic Acid
To understand how mutations in the MSD3 (FAD7) gene lead to the dramatic change in panicle architecture and restore the fertility of the pedicellate spikelets, we first assessed the effect of the msd3 mutations on lipid composition of leaves and panicles of the msd3-1 mutant. In WT BTx623 leaves, 18-carbon fatty acids account for over 92% of total leaf lipids ( Figure 6, Table S2). Levels of seven lipid species with molar percentage >1% are plotted in Figure 6. Galactolipids, including monogalactosyl diacylglycerol (MGDG) and digalactosyl diacylglycerol (DGDG), are the major lipids in leaves, accounting for over 90% of the total lipids. The lipid species with 36 carbon and six double bonds (36:6) consist of two linolenic acid molecules, whereas those with four double bonds (36:4) consist of two linoleic acid molecules, and species with five double bonds (36:5) consist of one linoleic acid and one linolenic acid. Sorghum has very low levels of 16-carbon fatty acids, and hexadecatrienoic acid (16:3) was not detectable in either leaves or panicles (Table S3). linolenic acid (7% vs. 38% in WT panicles). This result confirms that MSD3 is an ω-3 fatty acid desaturase that catalyzes the desaturation of linoleic acid to linolenic acid in both leaves and panicles. Moreover, the dramatic effect of msd3 mutations on linoleic acid desaturation was consistent with the idea that MSD3 is sorghum FAD7, as FAD7 has major effects on linoleic acid desaturation in other plant species [22][23][24].  Table 3S. The difference between WT and msd3-1 in each lipid species was analyzed by t-test. * p < 0.05; ** p < 0.01.

The msd3 Phenotype Was Reverted by Treatment with MeJA
Because linolenic acid (18:3) is a substrate for JA biosynthesis [28], we next asked whether the msd phenotype of the msd3 mutants was due to a deficiency of JA. For this purpose, we then measured the level of JA in developing panicles of msd3-1 and BTx623. The JA level was reduced from 709 ng/g fresh weight (FW) in the wild type to 409 ng/g FW in msd3-1 ( Figure 4S).
To further demonstrate if the decrease in JA led to the msd phenotype, we applied MeJA to the whorls of BTx623 and the msd3-1 mutant every other day, starting from when the plants had eight fully expanded leaves until the plants had 10 fully expanded leaves. Because the pedicellate spikelets never produce viable grains in BTx623 and other natural accessions, the reversion of the pedicellate spikelets from grain-bearing perfect flowers to sterile spikelets is a reliable indicator to demonstrate the effect of JA treatment. As shown in Figure 7, the pedicellate spikelets of the wild type BTx623 were sterile while that of msd3-3 mutant were fertile and produced viable grain. After treatment with MeJA, the pedicellate spikelets of msd3-3 mutants became sterile spikelets as in the wild type BTx623. Furthermore, the grain size of the msd3 mutant also restored to similar size as the wild type BTx623 (Table 4S). This result indicates that the lack of sufficient JA in the msd3 panicle resulted in full fertility in the pedicellate spikelets.  Table S3. The difference between WT and msd3-1 in each lipid species was analyzed by t-test. * p < 0.05; ** p < 0.01.
As can be seen from Figure 6, mutation in MSD3 decreased the levels of 36:6 lipid species and concomitantly increased the levels of 36:4 species. The ratio of linolenic acid to linoleic acid was 13.43 in the WT BTx623 leaves, falling to 2.48 in msd3-3 leaves (Table S3). In general, panicles contained much less linolenic acid than in leaves. In WT BTx623 panicles, the ratio of linolenic acid to linoleic acid was 0.62, falling to 0.08 in the msd3-3 mutant panicles. Thus, msd3-3 panicles contained very little linolenic acid (7% vs. 38% in WT panicles). This result confirms that MSD3 is an ω-3 fatty acid desaturase that catalyzes the desaturation of linoleic acid to linolenic acid in both leaves and panicles. Moreover, the dramatic effect of msd3 mutations on linoleic acid desaturation was consistent with the idea that MSD3 is sorghum FAD7, as FAD7 has major effects on linoleic acid desaturation in other plant species [22][23][24].

The msd3 Phenotype Was Reverted by Treatment with MeJA
Because linolenic acid (18:3) is a substrate for JA biosynthesis [28], we next asked whether the msd phenotype of the msd3 mutants was due to a deficiency of JA. For this purpose, we then measured the level of JA in developing panicles of msd3-1 and BTx623. The JA level was reduced from 709 ng/g fresh weight (FW) in the wild type to 409 ng/g FW in msd3-1 ( Figure S4).
To further demonstrate if the decrease in JA led to the msd phenotype, we applied MeJA to the whorls of BTx623 and the msd3-1 mutant every other day, starting from when the plants had eight fully expanded leaves until the plants had 10 fully expanded leaves. Because the pedicellate spikelets never produce viable grains in BTx623 and other natural accessions, the reversion of the pedicellate spikelets from grain-bearing perfect flowers to sterile spikelets is a reliable indicator to demonstrate the effect of JA treatment. As shown in Figure 7, the pedicellate spikelets of the wild type BTx623 were sterile while that of msd3-3 mutant were fertile and produced viable grain. After treatment with MeJA, the pedicellate spikelets of msd3-3 mutants became sterile spikelets as in the wild type BTx623. Furthermore, the grain size of the msd3 mutant also restored to similar size as the wild type BTx623 (Table S4). This result indicates that the lack of sufficient JA in the msd3 panicle resulted in full fertility in the pedicellate spikelets.

Discussion
In this study, we characterized a new group of msd mutants, msd3, that exhibits increases in the size and number of the primary branches and fertile pedicellate spikelets, similar to the panicle structure previously reported in the msd1 and msd2 mutants [13,14]. Using bulk segregant analysis of four independent F2 populations, we identified that all causal mutations are located within the FAD7 (Sorbi_3001G407600) gene. Because FAD7 is a major plastidial fatty acid desaturase, loss-of-function mutations in the FAD7 gene resulted in a dramatic reduction in the content of linolenic acid, a substrate of JA, in both leaves and panicles. The reduction in JA levels in msd3 panicles and the reversion of the msd3 phenotype to the wild type by application of MeJA indicate that the msd phenotype in msd3 mutants is probably due to insufficient levels of JA to activate the programmed cell death pathway leading to the arrest of the pedicellate spikelets. This conclusion is consistent with the previously reported mechanisms in msd1 and msd2 mutants [13,14].
MSD3 (FAD7) appears to the predominant form of ω-3 fatty acid desaturase in sorghum. Most plants, including sorghum, have multiple forms of microsomal and plastidial ω-3 fatty acid desaturases that catalyze the formation of linolenic acids from linoleic acid. Arabidopsis has one microsomal fatty acid desaturase, originally known as FAD3, and two closely related plastidial ω-3 desaturases, FAD7 and FAD8 (plastidial) [21][22][23][24]29]. Mutations in all three genes, FAD3, FAD7, and FAD8, are required to reduce the linolenic acid levels sufficiently to affect JA-regulated processes such as male sterility and the defense response to insect attack [29][30][31]. However, loss-of-function mutation in FAD7 alone blocks the JA-induced responses in tomato [22]. The sorghum genome encodes two microsomal ω-3 fatty acid desaturases, Sorbi_3005G002800 and Sorbi_3008G003200, which are homologous to Arabidopsis FAD3 [25,27], and two plastidial fatty acid ω-3 desaturases, Sorbi_3001G407600 and Sorbi_3002g430100, which are homologous to Arabidopsis FAD7 and FAD8 [27,32]. Loss-of-function mutations in MSD3 (FAD7) alone decreased the ratio of linolenic acid to linoleic acid from 13.43 to 2.48 in leaves, and from 0.62 to 0.08 in panicles, resulting in dramatic changes in panicle architecture, and restoration of the fertility of the pedicellate spikelets. As noted

Discussion
In this study, we characterized a new group of msd mutants, msd3, that exhibits increases in the size and number of the primary branches and fertile pedicellate spikelets, similar to the panicle structure previously reported in the msd1 and msd2 mutants [13,14]. Using bulk segregant analysis of four independent F 2 populations, we identified that all causal mutations are located within the FAD7 (Sorbi_3001G407600) gene. Because FAD7 is a major plastidial fatty acid desaturase, loss-of-function mutations in the FAD7 gene resulted in a dramatic reduction in the content of linolenic acid, a substrate of JA, in both leaves and panicles. The reduction in JA levels in msd3 panicles and the reversion of the msd3 phenotype to the wild type by application of MeJA indicate that the msd phenotype in msd3 mutants is probably due to insufficient levels of JA to activate the programmed cell death pathway leading to the arrest of the pedicellate spikelets. This conclusion is consistent with the previously reported mechanisms in msd1 and msd2 mutants [13,14].
MSD3 (FAD7) appears to the predominant form of ω-3 fatty acid desaturase in sorghum. Most plants, including sorghum, have multiple forms of microsomal and plastidial ω-3 fatty acid desaturases that catalyze the formation of linolenic acids from linoleic acid. Arabidopsis has one microsomal fatty acid desaturase, originally known as FAD3, and two closely related plastidial ω-3 desaturases, FAD7 and FAD8 (plastidial) [21][22][23][24]29]. Mutations in all three genes, FAD3, FAD7, and FAD8, are required to reduce the linolenic acid levels sufficiently to affect JA-regulated processes such as male sterility and the defense response to insect attack [29][30][31]. However, loss-of-function mutation in FAD7 alone blocks the JA-induced responses in tomato [22]. The sorghum genome encodes two microsomal ω-3 fatty acid desaturases, Sorbi_3005G002800 and Sorbi_3008G003200, which are homologous to Arabidopsis FAD3 [25,27], and two plastidial fatty acid ω-3 desaturases, Sorbi_3001G407600 and Sorbi_3002g430100, which are homologous to Arabidopsis FAD7 and FAD8 [27,32]. Loss-of-function mutations in MSD3 (FAD7) alone decreased the ratio of linolenic acid to linoleic acid from 13.43 to 2.48 in leaves, and from 0.62 to 0.08 in panicles, resulting in dramatic changes in panicle architecture, and restoration of the fertility of the pedicellate spikelets. As noted above, MSD3 encodes one of the four ω-3 fatty acid desaturases, but the roles of the other three enzymes in fatty acid desaturation remains unclear. Based on their relative expression levels ( Figure S3), these genes may not contribute much to the desaturation of linoleic acid except at specific developmental stages. Thus, MSD3 (FAD7) plays a major role in the desaturation of linoleic acid to linolenic acid in sorghum.
Linolenic acid (18:3) and hexadecatrienoic acid (16:3) are the substrates for the lipid-derived JAs [16,17]. Because sorghum has non-detectable levels of hexadecatrienoic acid, linolenic acid serves as the main substrate for JA biosynthesis. We propose that the dramatic decrease in linolenic acid in the msd3 mutants may result in deficiency of JA, which failed to arrest the development of the pedicellate spikelets in sorghum. Thus, we determined the JA levels in the developing panicles of BTx623 and msd3-1 mutant. Although JA level in the msd3-1 mutant panicle was significantly reduced to 409 ng/g FW from 709 ng/g fresh weight (FW) of the wild type, the msd3 panicles still contained substantial levels of JA. Next, we tested if the msd3 phenotype, such as fertile pedicellate spikelets, can be reversed by JA treatment. As shown in Figure 7 and Figure S5, the fertile pedicellate spikelets on the msd3 panicle was reverted to sterile spikelets as in the BTx623. Thus, it is more likely that the dramatic reduction in linolenic acid in msd3 mutants may lead to lower levels of JA, which failed to trigger the programmed cell death pathway and allowed the pedicellate spikelets to develop viable grains. This conclusion is also supported by our previous reports on msd1 and msd2 mutants [13,14].
The role of JA in the mediation of the msd phenotype in sorghum may be closely related to its role in maize tassel development by functioning at the level of sexual determination. Maize is monoecious with separate staminate (male) and pistillate (female) inflorescences called the tassel and the ear, respectively. During early floral development, the florets in both the tassel and the ear spikelets are bisexual. The monoecious nature is conferred by the selective abortion of pistillate organs in tassel florets [15]. A large number of sex-determination mutants (tasselseed) have been identified in maize [33]. The TS1 encodes a lipoxygenase that catalyze the conversion of free linolenic acid to 13(S)-hydroperoxylinolenic acid, the first committed step in JA biosynthesis. The sorghum MSD2 gene is an ortholog of the maize TS2 gene [14]. The maize TS2 gene encodes a short-chain alcohol dehydrogenase, probably involved in JA biosynthesis [34]. Another tasselseed mutant was created by knocking out the duplicated orthologs of OPR3, a major OPR (12-oxo-phytodienoic acid reductase) gene in Arabidopsis that acts in JA biosynthesis [35]. The resulting maize opr7opr8 double mutant is phenotypically similar to ts1 and ts2 mutants and can be reverted to normal tassel phenotype by treatment with MeJA [36]. The maize TS5 gene encodes a wound-inducible gene, CY94B, which serves as an enzyme that inactivates the biologically active JA-isoleucine [33]. Although the causal mutation in the TS5 gene has not been identified, the mutation of ts5 is dominant and leads to the overexpression of CY94B, decreased JA-isoleucine, and a concomitant increase in the deactivated 12OH-JA-isoleucine. As with the maize ts1 and ts2 mutants, the tasselseed phenotype was reverted by JA treatment. It is well-established that JA plays a critical role in the maintenance of the staminate state of the tassel by suppressing the development of female flower organs through JA-induced programmed cell death [15].
Sorghum is hermaphroditic and produces perfect flowers with both male and female floral organs. Like many other grasses, sorghum produces two types of spikelets depending on the mode of attachment to the inflorescence axis. The sessile spikelets, directly attached to the inflorescence axis, contain perfect flowers and produce viable grains. The pedicellate spikelets, attached to the inflorescence axis through a short petiole called pedicel, occasionally produce anthers but never produce mature ovaries, and thus, become sterile spikelets in mature panicles. In sorghum, msd mutants produce both sessile and pedicellate spikelets that contain perfect flowers and can develop viable grains. Sorghum may employ similar mechanisms to control the fertility of the pedicellate spikelets as reported in maize in controlling the ovary development in tassels. In the wild type BTx623, both sessile and pedicellate spikelets have both male and female floral organs initially but the floral organs, especially the ovary, in the pedicellate spikelets become arrested in a late development stage [13]. The abortion of the ovary has been shown to be regulated by JA-induced programmed cell death as in maize tassels [13,15]. As previously reported, the MSD1 gene encodes a TCP-transcription factor that activates many genes involved in JA biosynthesis and MSD2 encodes a lipoxygenase that catalyzes the first committed step of JA biosynthesis from linolenic acid [13,14]. The discovery of the MSD3 gene as a predominant form of ω-3 fatty acid desaturase responsible for the conversion of linoleic acid to linolenic acid highlights the importance of the JA regulatory module(s) in the control of the fertility of the pedicellate spikelets in sorghum.
The mechanisms determining the staminate state of tassels in maize and that suppressing the development of the pedicellate spikelets in sorghum may also differ. All maize tasselseed mutants are male sterile due to the ectopic growth of the silks. However, the sorghum msd mutants develop both anthers and ovaries. Male sterility was never observed in the sorghum msd mutants. In addition, JA may also contribute to panicle development in sorghum, because all msd mutants exhibit increases in the number and size of the primary inflorescence branches [11,13,14].

Plant Materials
Sorghum (Sorghum bicolor (L.) Moench) msd mutants were identified from a pedigreed sorghum mutant library that was developed by mutagenizing the sorghum inbred line BTx623 seeds with the chemical mutagen ethyl methane sulfonate (EMS) [37]. The wild type BTx623, the msd mutants, and their backcrossed F 2 populations were planted in the Agricultural Experiment Station at Lubbock, TX (33 39" N, 101 49" W) of the Agricultural Research Service of the United States Department of Agriculture (USDA-ARS) in May 2017. During the late grain-filling stage, when the msd phenotype could be easily observed, leaf samples were collected from each of the confirmed homozygous msd mutants to prepare genomic DNA as described previously [38].
The msd mutants isolated from the library harbored a high density of EMS-induced mutations [20]. To decrease the number of background mutations, these mutants were backcrossed (BC) to WT BTx623 for three generations before use in this study. Moreover, the new msd mutants were also crossed to the known msd1 and msd2 mutants, as well as to each other, to determine whether they represented alleles of known msd loci or new locus.

Gene Identification by Next-Generation Sequencing
Homozygous msd mutants were identified from the backcrossed F 2 populations of p6, p14, p21, and p24 during the grain-filling stage, during which the msd phenotype could be easily scored. Genomic DNA pooled from the homozygous msd mutants from each F 2 population was sequenced to >20× coverage on an Illumina HiSeq 2000 through a service from BGI (https://www.bgi.com/us/). Causal mutations in these populations were identified by an inhouse bioinformatic pipeline described previously [19]. Briefly, low-quality reads, adapter sequences, and contamination were excluded from the raw reads, and then the clean reads were aligned to the sorghum reference genome v3.0 with Bowtie2 [39]. SNP (single nucleotide polymorphism) calling was performed by Samtools and Bcftools using only reads with mapping and sequencing quality greater than Q20 [40]. The read depth for true SNPs was set from 3 to 50. Because EMS induces only G/C-to-A/T transition mutations [41], only homozygous G/C to A/T SNPs were used for prediction of effects on gene function by the Ensembl Variant Effect Predictor [42]. Homology analysis and functional annotation of candidate genes were obtained from Gramene database release 39 [43] (http://www.gramene.org).

Confirmation of the Causal Mutation with KASP
The BC 3 F 2 population consisting of 63 F 2 offspring was utilized for this study. Based on putative SNPs discovered through the sequencing analysis described above, three types of SNP-KASP primers were designed [44,45] and synthesized by Integrated DNA Technologies (Coralville, IA, USA). Genotyping analyses were performed using extracted genomic DNA from leaf tissues of sorghum plants collected from the field. SNP genotype in each offspring was determined by Kompetitive Allele Specific PCR (KASP) chemistry (www.lgcgenomics.com) with some modifications [44]. Analysis of genotypes was conducted at the Plant Stress and Germplasm Development Unit at Lubbock, TX, USA. Phenotype data for the population were gathered at the full maturity stage of the panicle. Co-segregation analyses were conducted based on the correspondence of genotype to phenotype and chi-square test.

Lipidomic Assay
Samples were obtained from~5 cm 2 from the middle lamina of the first mature leaf (the youngest leaf with a visible leaf collar) and from developing panicles (approximately 3 cm long). The samples were quickly immersed in liquid nitrogen and stored at −80 • C until use. Lipids were extracted and analyzed following the procedure established at the Lipidomic Center of Kansas State University (https://www.k-state.edu/lipid/). The original protocol was modified as follows: instead of cutting leaves into small pieces, the leaf tissue was completely ground into a fine powder with liquid nitrogen chilled mortars and pestles. The plant material was then treated with hot isopropanol and 0.01% butylated hydroxytoluene (preheated to 75 • C). The resultant lipid extract was dissolved in chloroform and stored at −80 • C. Lipid extracts were dried with a stream of nitrogen gas, packed in dry ice, and shipped to the Lipidomic Center of Kansas State University for lipid analysis. The tissue used for extraction was dried in an oven at 65 • C for 1 week to estimate the total dry weight of tissue used for extraction of each lipid sample.

Jasmonic Acid Determination
Panicles at stage 4, when the fate of the pedicellate spikelets is determined [13,14], were collected and immediately frozen in liquid nitrogen. Five biological replicates were solvent-extracted, methylated, collected on a polymeric adsorbent using vapor-phase extraction, and analyzed using GC/isobutene chemical ion mass spectrometry (GC/CI-MS) [46]. For metabolite quantification, d 5 JA (Sigma-Aldrich, St. Louis, MO, USA) was used as an internal standard. The JA level in each sample was normalized to the mass of the panicle, and expressed as ng/g fresh weight (FW).

Quantitative Gene Expression Analysis of FAD7 and FAD8
Sorghum msd3 and BTx623 plants were grown in a greenhouse with temperatures maintained at 28 • C (12 h day)/25 • C (12 h night) cycle. Panicle samples at stage 4, about 3 cm long, were collected, flash frozen in liquid nitrogen, and stored at −80 • C until use. Total RNA was extracted from a single panicle with Trizol reagent (Thermo-Fisher Scientific, Waltham, MA, USA), followed by column purification using the RNA Extraction kit from Sigma-Aldrich (Sigma-Aldrich, St. Louis, MO, USA). After purification, genomic DNA contaminants were removed by On-Column DNase I Digestion supplied in the Sigma kit (Sigma, St Louis, MO, USA). RNA concentration was measured using a BioSpectrophotometer ® Kinetic (Eppendorf, Hamburg, Germany). RNA quality was checked by electrophoresis on a 1% agarose gel. First-strand cDNA was synthesized from 1 µg RNA per sample using the Superscript 2 Reverse Transcriptase kit (Invitrogen, Carlsbad, CA, USA). The reaction mixture was incubated at 42 • C for 50 min followed by heat inactivation at 70 • C for 15 min. The resultant cDNA is used as a template for quantitative analysis of FAD7 and FAD8 genes with control eukaryotic translation initiation factor.
Real-time polymerase chain reaction (PCR) analysis was used to determine which plastidial fatty acid desaturase was affected in the msd3 mutants. The Primer3 software was used to design primers for the FAD7 and FAD8 genes from sorghum. Combinations of forward and reverse primers with Tm above 60 • C were selected for real-time PCR analysis, and the annealing temperature was set to 53 • C. Because of the high sequence similarity among fatty acid desaturase genes, primers were designed within the 3 -UTR region to prevent non-specific binding to genes in the same family. 3 -UTR regions unique to FAD7 (FAD7 forward primer: 5 TCC CTC AAA TCC CAC ATT 3', FAD7 reverse primer: 5 GAA GAG CAC CCG ACT TCT TT 3') and FAD8 (FAD8 forward primer: 5 TGC ATG GAG GTT CAT ATA CTG C 3', FAD8 reverse primer: 5 AAT TCT GTT CTG TTT GGT TGG TG 3') were used to design primer pairs that amplified a product of~100 base pairs. Internal control primers were designed against the gene encoding eukaryotic translation initiation factor 4 α (EIF4α) (SB_EIF4α Forward: 5 CAA CTT TGT CAC CCG CGA TGA 3 SB_EIF4αReverse: 5 TCC AGA AAC CTT AGC AGC CCA 3 ).
The cDNA was diluted five times with nuclease-free water, mixed with Roche Fast Start SYBR Green ® (Roche, San Francisco, CA, USA) along with specific primers and run on a Roche LightCycler 96. Three biological replicates (control and mutant samples) were used for gene expression analysis. The melt curve from the Light Cycler was analyzed by exporting the ∆Ct and ∆∆Ct values to an Excel file. EIF4α was used as an internal control for analyzing the relative expression level differences of FAD7 and FAD8 genes.

Reversion of the msd3 Panicle Structure by MeJA
Treatment with MeJA was performed as described [13,14]. Briefly, 1 mL of 1 mM MeJA in 0.05% Tween-20 in water was applied to the central whorl every other day, starting when the plants had eight fully expanded leaves until the plants had 10 fully expanded leaves. Controls were treated with 1 mL of 0.05% Tween-20.

Statistical Analyses
Statistical analyses, such as one-way ANOVA and t-test, were run in Excel with the XRealStats add-in (http://www.real-statistics.com). Option 'Tukey's HSD' was used for multiple comparisons.

Conclusions
We identified MSD3 as FAD7 (Sorbi_3001G407600), a major plastid-targeted ω-3 fatty acid desaturase. Loss-of-function mutations in MSD3 (as in msd3-1, msd3-3, and msd3-4) result in low levels of linolenic acid, the substrate for JA biosynthesis. The msd panicle architecture was restored to the wild-type panicle phenotype by treatment with MeJA. The discovery that MSD3 is FAD7 provides further support for the idea that JA-regulated modules are responsible for suppressing the development of pedicellate spikelets in sorghum, as described previously in our characterization of the msd1 and msd2 mutants. Compared with msd1 and msd2 mutants, msd3 mutants have higher grain weight indicating that msd3 mutants may be better adapted to increasing sink capacity, and therefore may have better potential to increase grain yield. These findings of this study provide a new approach for genetic manipulation of grain numbers per panicle with the goal of increasing grain yield. Funding: This research was funded in part by the United Sorghum Checkoff. ZX was also partly supported by USDA-ARS 3096-21000-020-00-D, and DW was partly supported by USDA-ARS 8062-21000-041-00D.