The EMS (ethyl methane sulfonate) mutant Dek line AK-3537 originated from the wheat variety AK58. The Dek phenotype was stably inherited after eight generations of self-pollination. For genetic analysis and mapping, AK-3537 was crossed with AK58 to produce an F₂ population of 130 individuals. Wheat materials were cultivated in experimental fields and greenhouses at Sichuan Agricultural University, Chengdu Chongzhou (103° 38’ E, 30° 32’ N), Wenjiang (103° 51’ E, 30° 43’ N), and Xishuangbanna (99° 56’ E, 21° 08’ N), China. All field trials were well irrigated and managed following local standard practices, and all AK-3537 × AK58 F₂ plants were grown in a greenhouse at 20°C with a 16-h/8-h light/dark cycle.

In the field, a total of 20 plants from AK58 (10 individuals) and AK-3537 (10 individuals) were randomly selected to investigate agronomic traits, including plant height (PH), tiller number (TN), heading date (HD), flag leaf length (FLL), and flag leaf width (FLW), using the method reported in previous studies. Thousand kernel weight (TKW), grain length (GL), and grain width (GW) were also measured using previously reported methods (Liu et al., 2017). Excel 2019 (Microsoft, Redmond, WA, USA) was used to calculate the phenotypic data. Analysis of variance was conducted, and individuals were ranked through Duncan’s test and plotted using GraphPad Prism V9.0.0. R software (version 4.2.1) was used as a plotting tool to calculate the wheat Dek phenotypic data.

After maturity, wheat grains were harvested and threshed by hand. One hundred grains were randomly sampled from each individual, and the sampling was repeated three times. The wheat grains were visually examined for the Dek phenotype, and the incidence (percentage of grain Dek phenotype) was calculated. To better observe the wheat grain Dek phenotype, the grains of AK58 and AK-3537 were stored at 4°C in FAA (Formalin-Aceto-Alcohol) fixative (ensuring the kernels did not float on the surface of the fixative solution) during the wheat grain filling stage. Subsequently, the cell structures of the normal phenotype and Dek phenotype were observed using frozen and free-hand sections (Leica CM1860) and x-ray computed tomography (Micro-CT).

(AK-3537 × AK58) F₁ grains were observed at the mature grain period in the natural field. F₂ populations were used for genetic analysis. We then performed a chi-square test (χ²) to test phenotypic data (grain Dek phenotype) for a goodness of fit to the ratio of 3:1 expected for a single gene (or semi-dominant) genetic basis in Excel 2019 by CHISQ.TEST function (p > 0.05 means no deviation from expectations of 3:1). Using the combined approaches of BSE-seq and BSR-seq, AK58, AK-3537, and the F₂ population (grains with normal phenotype and Dek phenotype, with a mix of 30 random individuals each) were selected for DNA and RNA segregant pools. Leaves were collected for DNA extraction using the Plant Genomics DNA Extraction Kit (BIOFIT^®, DN32-100, Chengdu, China), and grains (10–20 days post-anthesis, DPA) were sampled to extract total RNA using the Plant RNA Extraction Kit (BIOFIT^®, RN34050, Chengdu, China). RNA sequencing generating 150 bp paired-end reads was performed on the Illumina HiSeq™ × platform. Clean RNA-seq data were mapped onto the Chinese Spring reference genome (RefSeq 2.1) and the AK58 genome (Jia J et al., 2021) using the software Bowtie2, and SNP calling was performed with the SAMtools software. The newly designed exome capture probe panel (Dong et al., 2020) and ΔSNP-index algorithm (Abe et al., 2012) were used to map the grain Dek gene in wheat rapidly. High-quality reads were aligned to the Chinese Spring reference genome (RefSeq 2.1) and the AK58 genome (Jia J et al., 2021) with default parameters. The parental AK58 sequencing data (DNA and RNA sequencing data) were used as a “background” to identify the causal mutation based on the assumption. Calculations were analyzed on PlantGmap (Zhang et al., 2021) (http://183.223.252.63:3333/).

Lastly, considering the characteristics of EMS mutagenesis, certain variations were filtered (Dong et al., 2020). Only the candidate regions identified in BSE-seq and BSR-seq were considered. Then, the Chinese Spring reference genome (RefSeq 2.1) was used to obtain the candidate gene, gene sequence, and gene annotation. Homologous analysis and gene expression patterns were evaluated on WheatOmics 1.0 (http://202.194.139.32/). Arabidopsis (TAIR10) and rice (IRGSP-1.0) genomes were used for comparative genomics analyses. A phylogenetic tree was constructed using MEGA11, and Geneious was employed to assemble high-quality reads.

Grains of AK-3537 and AK58 during the grain filling stage (10 DPA) were collected. RNA extraction, library construction, and RNA sequencing were performed as described previously. Differentially expressed gene (DEG) analysis, Gene Ontology (GO) annotation, and Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis were adopted to confirm the putative biological functions and biochemical pathways of DEGs (AK-3537 and AK58) on the OmicsShare Tools (https://www.omicshare.com/).

DNA extraction was performed as previously described, and KASP markers were used for genotyping in parents and the F₂ population. The experimental method for KASP markers was referenced from previous reports (Li et al., 2020). The quantitative reverse transcription polymerase chain reaction (qRT-PCR) experiment was conducted to detect the expression levels of key genes involved in the wheat starch synthesis pathway, including GBSSI, GBSSII, SBEI, SBEII, SSI, SSII, SSIIIa, SSIV, TaAGPL1, TaAGPS1a, TaBEI, TaBEIIa, TaBEIIb, and TaBEIII. The experimental method for qRT-PCR was referenced from previous reports (Li et al., 2022). Actin was selected as the reference gene (Wang et al., 2014), and the relative quantification formula (2^−ΔΔCt) ± standard error of the mean (SEM) was used to evaluate quantitative variation further. Three biological replicates were tested for each sample. All primers used in this study are listed in Supplementary Table 1 .

In a screening of wheat grain, we identified a Dek mutant, AK-3537, with grain shrunken from an ethyl methyl sulfide (EMS) mutant library in the AK58 background ( Figure 1A ). The grains of the wild-type AK58 were plump and normal (AK58 type), whereas AK-3537 showed a 100% grain Dek phenotype rate across all environments (AK-3537 type) ( Figure 1 ). To further establish the genetic basis of this phenotype, a cross was performed between AK-3537 and AK58. All F₁ plants exhibited a normal grain phenotype, suggesting the presence of a recessive gene controlling the wheat grain Dek phenotype. To assess AK-3537 mutation segregation, (AK-3537 × AK58) F₁ plants were self-pollinated and an F₂ segregating population (130 individuals) was developed. The F₂ population was segregated into two categories: 95 plants exhibited an AK58-type phenotype, and 35 plants showed an AK-3537-type phenotype ( Table 1 ). The results were in agreement with a 3:1 segregation ratio [χ² = 0.26 < χ² (0.05, 1) = 3.84, p = 0.61], suggesting that a single gene regulates grain Dek.

Compared with that in AK58, a large hollow area was observed in the endosperm of AK-3537 by frozen and free-hand section, and the hollow phenotype was also observed in the Dek grain of AK-3537 by X-ray 3D tomography ( Figure 1B ). The results showed that AK-3537 grain filling was significantly affected. Finally, we evaluated the agronomic phenotypes of AK58 and AK-3537 in the field. Compared with AK58, FLW, TN, TKW, GL, and GW were all significantly decreased in AK-3537, whereas HD increased considerably ( Figures 2A, B ).

For Dek gene mapping, two pooled samples, each comprising 30 normal and Dek phenotype F₂ segregants, were constructed. Approximately 20 Gb of sequence data for the DNA pool and approximately 6 Gb for the RNA pool were generated and compared with the Chinese Spring reference genome (RefSeq V2.1) and AK58 genome. Then, we used the ΔSNP-index algorithm (Abe et al., 2012) for Dek gene mapping. ΔSNP-index higher than 0.7 in BSE-seq and BSR-seq Dek gene candidate region (DCR) was defined conservatively as the union of BSA-seq and BSR-seq credible intervals for candidate gene identification. In the AK58 genome, two DCRs, DCR1 and DCR2, were mapped on chromosome 7A by BSE-seq and BSR-seq ( Table 2 ), among which DCR1 was located between 279.98 and 287.93 Mb on Chr7A with two SVs (structural variation) in two genes, and DCR2 was located between the range of 565.34 and 568.58 Mb on Chr7A with two SVs in two genes ( Table 2 ). In the Chinese Spring reference genome (RefSeq V2.1), five DCRs, namely, DCR3, DCR4, DCR5, DCR6, and DCR7, were identified on chromosomes 1B, 7A, and 7B. DCR6 and DCR2 are the same candidate region and contain the same variant genes ( Table 2 ). Because AK58 is a wheat–rye 1B/1R translocation line material, the genetic background of the 1B chromosome is very different between the Chinese Spring reference genome (RefSeq V2.1) and the AK58 genome (Jia J et al., 2021). It has also been reported that wheat chromosome 7B underwent structural rearrangement (Chen et al., 2020). Therefore, we speculated that DCR1 and DCR2 located on chromosome 7A were candidate regions for the wheat grain Dek phenotype, which was predicted to be a moderate functional effect (e.g., missense mutation).

Since a single gene regulates the Dek phenotype, we annotated the genes in the two candidate regions based on the wheat Chinese Spring RefSeq v2.1 genome. In DCR1, a gene encoding 3-hydroxy-3-methylglutaryl-CoA synthase (TraesCS7A03G0625900) and a gene encoding a kinesin-like protein (TraesCS7A03G0631200) were annotated. In DCR2, an FBD-associated F-box protein (TraesCS7A03G0922700) and an S-adenosyl-L-methionine-dependent methyltransferase superfamily protein (TraesCS7A03G0929200) were annotated ( Figure 3 ; Table 2 ). Expression patterns of the candidate genes in the WheatOmics 1.0 (http://202.194.139.32/) database revealed that all four candidate genes were expressed in seeds.

To further exclude SNP variations caused by sequence assembly errors, exome capture and RNA-seq data were used to assemble the sequences of the mutated genes in the DCR1 and DCR2 regions. The results showed that the SNP variations at TraesCS7A03G0631200 and TraesCS7A03G0922700 could be detected in both exome capture and RNA-seq data. However, the SNP of TraesCS7A03G0929200 could only be detected in exome capture data, while the SNP of TraesCS7A03G0625900 could only be detected in RNA-seq data ( Figure 4 ; Supplementary Figure 1 ). To validate the segregation of these SNP variations in the F₂ population, KASP markers were developed based on these four SNP variations in this study and validated in the F₂ population. The results showed that only the SNP variation at position 1,049 of the TraesCS7A03G0625900 coding region co-segregated with the grain normal and Dek phenotype in the F₂ population ( Figure 5B ). A query of the Chinese Spring RefSeq v2.1 genome found that the coding sequence of the TraesCS7A03G0625900 gene was 1,410 bp in length, and the full-length genome contained 12 exons and 11 introns. Sequencing analysis of the TraesCS7A03G0625900 coding region in AK-3537 revealed a G-to-A mutation at position 1,049, the 11th exon of the coding region, which resulted in an amino acid substitution from glycine (GGC, Gly, G) to aspartic acid (GAC, Asp, D) ( Figure 5A ). The target gene was tentatively termed HMGS-7A.

In the transcriptome analysis of wheat grains of AK58 and AK-3537, a total of 12,655 DEGs were identified. Enrichment analysis of these DEGs in the wheat grain Dek regulatory pathway showed significant differences in gene expression levels between the mutant AK-3537 and the wild-type AK58, with 6,618 genes downregulated and 6,037 genes upregulated. GO and KEGG analyses were further performed on the screened DEGs to understand the functions and pathways of these DEGs. GO analysis showed that these DEGs were mainly concentrated in processes involved in carbohydrate metabolism (GO:0005975 and GO:0044723) ( Figure 6A ). KEGG analysis showed significant enrichment (p ≤ 0.05) of energy metabolism and starch synthesis pathway (Ko00500) ( Figure 6B ). This indicates that, compared to AK58 grains, AK-3537 grains have significant differences in wheat carbon metabolism, photosynthesis product synthesis, and starch synthesis pathways. These DEGs may cause the AK-3537 grain Dek phenotype. In previous reports, starch synthase also regulated similar phenotypes (Morell et al., 2003; Li et al., 2011). Therefore, this study also found that the expression levels of GBSSII and SSIIIa in the Ko00500 pathway were significantly increased in AK-3537 ( Figure 7 ).