The sorghum (Sorghum bicolor L. Moench) accessions ‘SUGARY FETERITA’ (SUF, JP No. 44623) and ‘FETERITA FAYOUMI D.S.21’ (FET, JP No. 246696) were obtained from Genebank Project, NARO (National Agriculture and Food Research Organization) and used for the phenotypic and genetic characterization of the sugary endosperm and gene expression. ‘Nakei MS-3B’ (MS3B) and SUF were used for segregation analysis and rough mapping. For complementation analysis using rice, the sugary-1 mutant SG0807 was obtained from Oryzabase (https://shigen.nig.ac.jp/rice/oryzabase/). Sorghum ethyl methanesulfonate (EMS) mutants (PI 678144, PI 678119, and PI 678151) were obtained from the Functional Gene Discovery Platform for Sorghum (https://www.purdue.edu/sorghumgenomics). For haplotype analysis, the Nagoya University Panel (NUP; 187 sorghum accessions) was obtained from the sorghum diversity research set of NARO (National Agriculture and Food Research Organization) (Shehzad et al., 2009), the US historic sweet sorghum collection (Wang et al., 2009), and the available lines in our lab, were used for this study ( Supplementary Table S2 ).

For rough mapping, seeds exhibiting a wrinkled trait in the F₂ progeny, resulting from a cross between MS3B and SUF, were used. Genomic DNA was isolated from ca. 2-week-old leaves from 96 individuals using the cetyltrimethylammonium bromide (CTAB) extraction method (Murray and Thompson, 1980) with some modifications. Briefly, leaf samples were ground using a MultiBead Shocker (Yasui Kikai, Osaka, Japan) in 2 × CTAB extraction buffer (100 mM Tris-HCl [pH 8.0], 50 mM EDTA, 1.4 M NaCl, 2% [w/v] CTAB, 1% [w/v] PVP). After incubating at 60°C for 30 min, an equal volume of chloroform was added. After centrifugation at 15,000 rpm for 5 min, the supernatant was recovered, and an equal volume of isopropyl alcohol was added. The sample was recovered by centrifugation at 15,000 rpm for 5 min and the pellet was washed with 70% (v/v) ethanol. The DNA pellet was dried and dissolved in 1 × Tris/EDTA (TE) solution (10 mM Tris-HCl [pH 8.0], 1 mM EDTA [pH 8.0]). The purified DNA samples were then used for genotyping using a polymerase chain reaction (PCR). Simple sequence repeat (SSR) markers, as reported by Yonemaru et al. (Yonemaru et al., 2009), were screened and 178 markers showing polymorphisms between MS3B and SUF were selected ( Supplementary Table S3 ). DNA segments were amplified by PCR using the following program: 95°C for 1 min; 30 cycles of 95°C for 30 sec, 55°C for 30 sec, and 72°C for 30 sec; 72°C for 7 min. The PCR products were analyzed by agarose gel electrophoresis.

‘SUGARY FETERITA’ and ‘FETERITA FAYOUMI D.S.21’ were grown in the greenhouse and immature seeds were sampled 0, 7, 14, and 21 days after flowering (DAF). These samples were immediately frozen in liquid nitrogen and crushed with a mortar and pestle until a powder was obtained. Next, RNA extraction buffer (50 mM Tris-HCl [pH 9.0], 20 mM EDTA [pH 8.0], 200 mM NaCl, 1% (w/v) Sodium N-lauroyl sarcosinate, and 5 mM DTT) was added to each sample and centrifuged at 14,000 rpm for 5 min at 4 °C. The supernatant was used in the subsequent steps. Total RNA was isolated from each sample using the TRIzol^® reagent (Invitrogen, Carlsbad, CA, USA) according to the manufacturer’s instructions. Next, 500 ng of total RNA was used to synthesize first-strand cDNA using an Omniscript reverse transcription (RT) kit (Qiagen, Hilden, Japan), according to the manufacturer’s instructions. Quantitative RT-PCR (qRT-PCR) was conducted using the KOD SYBR^® qPCR Mix (TOYOBO, Osaka, Japan) and a real-time thermal cycler (Bio-Rad Laboratories, Hercules, CA, USA). The ubiquitin (UBQ) gene served as the internal reference for the qRT-PCR analyses. Primers used in the analysis were listed in Supplementary Table S3 .

We examined the morphological traits of sorghum kernels using a stereo microscope. Seeds were cut with a razor blade and cross-sections were observed using the same microscope. For histological analysis, seed samples were fixed in FAA fixative (formaldehyde and acetic acid in 50% [v/v] ethanol) under a vacuum and dehydrated in a graded ethanol series. Samples were embedded in Technovit 7100 (Kulzer, Hanau, Germany) and sectioned at a thickness of 1 μm with a manually operated rotary microtome (Leica RM2125RT). Sections were stained with iodine solution (1% [w/v] KI, 0.1% [w/v] I₂) and observed under a light microscope (Olympus BX5). Photographs were taken using a digital camera (Olympus DP74) connected to the microscope. Brix values of the seeds were obtained as follows; first, 0.2 g of seeds were roughly crushed with a hammer and placed in a tube with beads. They were then crushed in a grinding apparatus (Bio-Medical Science Co., Ltd, BMS-A20TP) for at least 20 min at 1,100 rpm. To the powdered sample, 200 µL of water was added and mixed in the grinding device for a further 5 min at 1,100 rpm. Finally, the sample was centrifuged at 15,000 rpm for 5 min and the supernatant was measured with a refractometer (AS ONE, APAL-1).

Genomic DNA fragments (8.6 kb) containing Sobic.007G204600 from MS3B and SUF were introduced into pCAMBIA1300 (Cambia) using the KpnI/EcoRI sites. Each construct was transformed into Agrobacterium tumefaciens EHA105 using an electroporator (Eppendorf) at 2,500 V and used to infect calli of sugary-1 mutant (SG0807), whose genetic background is Oryza sativa (L.) cv. ‘Kinmaze’ as described by Ozawa (Ozawa, 2012). Transformed calli and plants were selected by hygromycin resistance, with regenerants grown in pots in a greenhouse.

All accessions were sequenced using an Illumina Hiseq X Ten sequencer (Illumina, San Diego, CA, USA) by pair-end sequencing to obtain SNPs as previously reported (Hashimoto et al., 2021). For SNP calling, clean reads were aligned with the BTx623 (v3.0.1) reference genome using Burrows-Wheeler Alignment (BWA, v0.7.17) (Li and Durbin, 2009) with default parameters. Variants were called independently and merged using the Genome Analysis Toolkit (GATK, v4.1.8, HaplotypeCaller) (McKenna et al., 2010). After the genomic region of SbSu (Sobic.007G204600) was extracted using VCFtools v0.1.16 (Danecek et al., 2011) from the resulting merged VCF file, haplotypes were categorized based on nonsynonymous SNPs associated with single amino acid substitutions in the coding region.

During the vegetative and heading stages, SUF showed no obvious differences from the wild type FETERITA (FETERITA FAYOUMI D.S.21; FET) ( Supplementary Figure S1 ). However, at the hard dough stage, grains of SUF exhibited wrinkled and translucent phenotypes whereas FET had normal, smooth grains ( Figures 1A–D ). Grains stained with iodine solution revealed that the endosperm of FET stained well in dark blue ( Figures 1E, G ), while that of SUF stained less overall, with no staining in the central region ( Figure 1F, H ). The detailed observation of the cross-sections revealed that SUF had loosely packed and abnormal starch granules compared to the densely packed granules of FET ( Figures 1I-L ). The Brix value of SUF was 2 fold higher than FET ( Figure 1M ). A decrease of grain weight was also observed in SUF ( Figure 1N ). These results indicate that the starch components of the endosperm of SUF differ from FET, resulting in the wrinkled grain phenotype and the accumulation of soluble sugars.

It is known that sugary mutants of maize and rice present wrinkled seeds, therefore, we used this phenotype as a morphological indicator of sugary endosperm in sorghum. For positional mapping, a Japanese sorghum variety, MS3B (a maintainer line with a normal phenotype), and SUF were crossed. The resulting MS3B x SUF (F₁) seeds showed a normal phenotype ( Supplementary Figure S2 ), suggesting that the sugary endosperm phenotype is recessive. We then analyzed the segregation pattern of the 1,627 F₂ seeds and found 1,212 were wild-type seeds and 415 seeds with the wrinkled phenotype. The segregation ratio of normal to wrinkled seeds in the population fit to 3:1 via the chi-square test ( Supplementary Table S1 ), suggesting that the seed wrinkling trait is a recessive, single-gene trait.

Positional mapping using 96 F₂ wrinkled phenotype seeds revealed genomic regions on chromosome 7 (60.6–65.0 Mb) and 9 (55.1–59.2 Mb) with biased segregation ratios ( Supplementary Figure S3 ). Although the bias on the latter region did not fit the segregation pattern described above (the recessive mutation model), the former biased for the SUF homozygotes fit it ( Supplementary Table S1 ). In the region on chromosome 7, we searched candidate genes (i.e., starch synthesis-related genes). Here, we found Sobic.007G204600, which encodes a DBE (SbISA1), localized within 3.4 Mb on the candidate region ( Figure 2A ). It should be noted that Sobic.007G204600 is the sorghum homolog of maize Su1 ( Figure 2A ). Sequencing analysis of the genomic DNA of the Sobic.007G204600 of MS3B and FET revealed that these genes produce proteins with the same amino acid sequences as BTx623, the reference genome ( Figure 2B ). Conversely, the allele in SUF had five SNPs which resulted in nonsynonymous amino acid substitutions in the protein-coding region ( Figure 2B ). One of those in SUF (L421) was next to the highly conserved catalytic residue (D420) of ISA1 ( Figure 2C ) (Sim et al., 2014). A protein structure analysis using Chlamydomonas reinhardtii ISA1 (CrISA1) suggested that L453H (corresponding to L421H in SUF) could sterically hinder the binding of the catalytic residue ( Figure 2D ). Gene expression analysis by qRT-PCR indicated that Sobic.007G204600 was highly expressed in developing seeds ( Figure 2E ). From these results, we supposed that this gene might play a role in starch metabolism in endosperm cells and that mutations therein result in their loss of function. Therefore, it is a candidate for the corresponding gene of SUF.

To confirm that the SUF allele of Sobic.007G204600 is a loss-of-function allele, we performed a complementation test using the rice sugary-1 (osisa1) mutant line (SG0807), which exhibits thin seeds with distorted starch, resulting in a less stained endosperm following iodine staining ( Figure 3 ). A genomic DNA fragment containing the promoter (2 kb upstream from ATG) and the coding region from MS3B or SUF was introduced into SG0807. The results showed that the endosperm of transgenic lines transformed with coding region of MS3B stained positive with iodine ( Figure 3 ) and almost recovered the seed thickness of SG0807. In contrast, the endosperm of transgenic lines complemented with the genomic fragment derived from SUF did not stain the same level as SG0807 ( Figure 3 ). These results suggested that the corresponding gene is Sobic.007G204600 (SbSu), and that the haplotype of SUF (pronounces E68V, L139R, V234I, A251T and L421H mutations) causes sugary endosperm, and the MS3B and FET alleles of this gene are gain-of-function alleles, whereas SUF allele is a loss-of-function allele. This sugary gene Su/su assigned firstly by Karper and Quinby (1963) was redefined as SbSu in the later sections.

To understand how the mutation (L421H) in SUF affects starch synthesis, we examined the expression of 16 starch synthesis-related genes ( Supplementary Figure S4 ) using qRT-PCR. These genes include four DBEs (SbSu, SbIsa2, SbIsa3, and SbPul1) and four SBEs (SbSbeI, SbSbeIIa, SbSbeIIb, and SbSbeIII). The DBEs catalyze the hydrolysis of α-1,6-glucosidic linkages in α-polyglucans, while SBEs catalyze the formation of α-1,6-glucosidic linkages in amylopectin and glycogen. The other starch synthesis-related genes include an SS (SbSsI), four transporters (SbSweet2-1, SbSweet4-3, SbBt1, SbPho1), an AGPase (SbSh2), and an ortholog of se1 in maize (SbSe1).

At the early seedling stage (ca. 14 DAF), the expression levels of the two DBEs (SbSu1 and SbPul) were upregulated in SUF, indicating positive feedback due to the loss-of-function of SbSu1 ( Figure 4 ), although th other DBEs (Sbisa2 and Sbisa3) were not upregulated. During the late stage (21 DAF), the expression of SbSbeI, SbSbeIIa, SbSsI, SbSe1, SbBt1, SbSh1, and three DBEs (SbSu1, SbIsa2 and SbPul1) was significantly downregulated in SUF ( Figure 4 ). The expression of SbSweet4-3 was markedly lower in SUF at 7 and 14 DAF. There were no significant changes for SbSbeIIb and SbSh2 ( Figure 4 ). These observations suggest that a fine-tuning activity works for the appropriate expression of DBEs during the early stage, while during the later stage, the activity disappears in the starch synthesis pathway.

If SbSu is used for modern breeding in grain sorghum, it is estimated that weak alleles were widely used during the establishment of those landraces and varieties. To find weak allels, we did a haplotype analysis using a NUP containing 187 accessions (see Materials Methods). Here, 13 nonsynonymous SNPs were found in the coding region of SbSu, which resulted in 8 haplotypes ( Figure 5 ). Interestingly, the accession containing Hap 8 is the only SUF among the NUP accessions. Subsequently, we measured the Brix values for each haplotype and found that the values were mostly similar except for Hap 8 (SUF), which had a higher score than the rest ( Figure 5 ). These results suggest that although Hap 8 (SUF allele) shows a high Brix value, it is not used or spread in NUP. It is supposed that the sugary endosperm of Hap 8 is too strong of a phenotype, decreasing grain weight and wrinkled seed, thereby decreasing the germination rate. It should be noted that the five amino acid substitutions (E68V, L139R, V234I, A251T and L421H) observed in Hap 8 ( Figure 2B ) compared with Hap 1; however, four non-synonymous substitutions (E68V, L139R, V234I, and A251T) were the same as in Hap 2. Considerd that the Brix score of Hap 8 was high, although that of Hap 2 was almost as low as that of Hap 1, the L421H substitution (caused by the SNP of T2875A) should be the corresponding mutation for the sugary phenotype of SUF.

The previous results reveal the importance of a weak allele for breeding; i.e., showing both a relatively high Brix value and a relatively normal seed shape ( Supplementary Figure S5 ). However, we could not find a weak allele in the NUP accessions. Therefore, we searched the sorghum mutant panel of the Functional Gene Discovery Platform for sorghum (https://www.purdue.edu/sorghumgenomics). The accessions in the platform were mutagenized by EMS, generating 12,000 mutant families in the BTx623. Genome sequences were generated for 586 EMS lines and bioinformatic processing was used to identify all mutations that impact predicted protein sequences. We searched the mutants of Sobic.007G204600 and found three mutants: PI 678144, PI 678119, and PI 678151. The mutations comprised single nonsynonymous amino acid substitutions; D182N (PI678144), P308L (PI678119), and R701Q (PI678151) ( Figure 6A ). It should be noted that these mutations were located in the highly conserved region ( Supplementary Figure S6 ), and especially, D182 and P308 were located in the catalytic domain and carbohydrate-binding module, respectively. To study how these mutations influence endosperm phenotype, the Brix value and grain weight were analyzed and compared to BTx623 (wild-type). Of the three mutants, PI 678144 is an intermediate phenotype of sugary endosperm, with a Brix value of 9.6 compared with 5.9 in wild-type and 14.4 in SUF ( Figure 6B ). For grain weight, the relative value of PI 678144 against wild-type was 74.6, which was higher than that of SUF (69.8) ( Figure 6C ). It was supported by the section stained with an iodine solution of PI 678144, which showed small and distorted starch granules, but seemed a weaker phenotype than that of SUF ( Figure 6D ). The Brix value of PI678119 was slightly higher than that of the wild-type, and the sections also indicated that it had a weaker phenotype than that of PI 678144. The phenotype of PI678151 was similar to the wild-type ( Figure 6B-D ). These results suggested that PI678144 (D182N) is an appropriate weak allele, which could make it valuable for the breeding of sugary endosperm.