Genetic diversity for starch quality and alkali spreading value in sorghum

Sorghum is an important food crop in many parts of Africa and Asia. Landraces of sorghum are known to exhibit variation in food quality traits including starch and protein content and composition. In this study, a panel of diverse sorghum breeding lines and 788 sorghum conversion (SC) lines representing the global germplasm diversity of the crop were evaluated for variation in starch quality based on alkali spreading value (ASV). A small number of genotypes with stable expression of the ASV+ phenotype across seasons were identified; mostly representing Nandyal types from India. Genetic studies showed the ASV+ phenotype was inherited as a recessive trait. Whole genome resequencing of ASV+ donor lines revealed SNPs in genes involved in starch biosynthesis. A genome wide association study (GWAS) identified a significant SNP associated with ASV near Sobic.010G273800, a starch branching enzyme I precursor, and Sobic.010G274800 and Sobic.010G275001, both annotated as glucosyltransferases. Physiochemical analyses of accessions with contrasting ASV phenotypes demonstrated an environment dependent lower starch gelatinization temperature (GT), amylose content of approximately 22%, and good gel consistency. The starch quality attributes of these lines could be valuable in food products that require good gel consistency and viscosity.


INTRODUCTION
Sorghum [Sorghum bicolor (L.) Moench] serves as a staple food crop for millions of people in Africa and Asia (FAO, 2016;House, Gomez, Sun, Murty, & Verma, 2000;Kimber, 2000). It is commercially important in many regions of the world but especially important in low-income and fooddeficient countries (FAO, 2019). The utilization of sorghum is regionally and culturally dependent and of great economic value for smallholder farmers because it is less expensive than other imported cereals (House et al., 2000;Taylor & Dewar, 2000).

of 17
GRIEBEL ET AL.
The sorghum germplasm collections represent tremendous genetic diversity with more than 36,000 accessions available and stored at ICRISAT in India (Billot et al., 2013;Kimber et al., 2013) and more than 40,000 accessions at the USDA National Center for Genetic Resources Preservation (USDA-ARS-PGRCU) (Kimber et al., 2013). Most of the exotic sorghum accessions stored at USDA-ARS-PGRCU are photoperiod sensitive so the Sorghum Conversion Program was initiated by Stephens, Miller, and Rosenow (1967) to make the collection more readily available for research and breeding in the temperate zone. The program converted alien sorghums from the world collection into photoperiod insensitive, shortstatured, early-maturing sorghums (Rosenow et al., 1997a(Rosenow et al., , 1997bStephens et al., 1967). Over 800 sorghum conversion lines (SC lines) are currently available (Hayes et al., 2015;Kimber et al., 2013). The collection of SC lines was developed to represent the genetic diversity of the crop and provides the basis for sorghum diversity panels such as the Sorghum Association Panel (SAP) (Adeyanju, Little, Yu, & Tesso, 2015;Boyles et al., 2017;Casa et al., 2008;Cuevas, Rosa-Valentin, Hayes, Rooney, & Hoffmann, 2017;Mace et al., 2013;Morris et al., 2013;Shenstone et al., 2018;Sukumaran et al., 2012) and the Sorghum Diversity Panel (SDP) (Hayes et al., 2015).
The sorghum diversity panels are important resources used in genome wide association studies (GWAS). Several different panels have been genotyped using genotype-by-sequencing (GBS) strategies and these markers are associated with phenotypes to identify quantitative trait loci (QTLs) of interest (Adeyanju et al., 2015;Boyles et al., 2017;Morris et al., 2013). The sorghum linkage disequilibrium (LD) decay varies among collections and is reported to be approximately 15 kb (Kimber et al., 2013) with slower LD decay in inbred lines (19.7 kb) than in landraces (10.3 kb) (Mace et al., 2013) and can be as fast as 1 kb (Morris et al., 2013). Mixed linear models are commonly used for GWAS (Adeyanju et al., 2015;Huang et al., 2010;Morris et al., 2013;Zhao et al., 2011). Recently, the Logistic Mixed Model (LMM) was proposed for analysis of binary traits in maize and sorghum diversity panels (Shenstone et al., 2018).

Core Ideas
• Diverse sorghum accessions exhibit variation in alkali spreading value (ASV) • A cluster of Nandyal type durra sorghums from India exhibit the ASV+ phenotype • Testcrosses demonstrated recessive inheritance for the ASV+ phenotype • GWAS identified significant SNPs and candidate genes for ASV The alkali spreading phenotype (ASV) has been reported to identify sorghum genotypes with contrasting starch thermal properties (Griebel et al., 2019a). The key gene for ASV in rice is the ALK (SSII-3) gene that also controls starch gelatinization temperature (GT) (Gao et al., 2011;Tian et al., 2009). GWAS for ASV has been conducted in rice with several genomic regions associated with the trait (Huang et al., 2010;Huang et al., 2012;Mogga, Sibiya, Shimelis, Lamo, & Yao, 2018;Song, Arif, Zhang, Sze, & Zhang, 2019;Zhao et al., 2011). Only a few of the associations were close to candidate genes such as one on chromosome 7 that was close to amylase inhibitors (Huang et al., 2012), chromosome 6 near the ALK (SSII-3) gene (Huang et al., 2010;Zhao et al., 2011), chromosome 8 within 200 kb of the SSII-2 (Zhao et al., 2011), and chromosome 5 (Song et al., 2019). The ASV hits were not confirmed across years (Zhao et al., 2011). Griebel et al. (2019b) showed that SSIIa and SBEIIb are key genes controlling the alkali spreading phenotype in EMS mutant populations of sorghum. These genes have contrasting effects on starch GT, amylose content, and paste viscosity profiles (Griebel et al., 2019b).
At present, nothing is known about the expression, inheritance, or genetic architecture for ASV in the standing variation of sorghum. Thus, we aim to (1) evaluate a diverse collection of more than 800 breeding and SC-lines of sorghum for ASV, (2) conduct a GWAS to determine the genetic architecture and candidate genes for ASV, (3) characterize the starch phenotypes of accessions with ASV+ phenotypes, and (4) determine the mode of inheritance for the ASV+ trait.

Plant material
A sorghum diversity panel of 840 lines (SC lines and breeding lines) was grown in West Lafayette in 2013 and 2017. Plants were self-pollinated each year and seeds for each line were used for subsequent analyses. SC489, SC491, SC587 F I G U R E 1 Scree plot of principal components (x-axis) and their contribution to the variance determined from GBS data. Threshold (red line) at the characteristic elbow was used to estimate the optimal number of PCs and SC589 were found to exhibit a stable ASV+ phenotype and seeds of these accessions were produced in West Lafayette 2017 and 2018. Three biological replicates per line were evaluated in 2017 and four biological replicates were evaluated in 2018. Tx623, Sepon82 and Macia were used as controls in both seasons with two biological replicates each year.

Alkali spreading test
The alkali test was conducted as described in Griebel et al. (2019a). Preliminary testing of ASV at 1.5%, 1.8%, and 2.0% KOH was used to identify the optimal conditions for testing the genotypes in the diversity panel. Sorghum seeds were cut longitudinally and each half was placed into a well of a 96 well plate. The wells were filled with 1.8% KOH and treated for 24 h. Each genotype was evaluated with two seeds in West Lafayette 2013 (preliminary study) and eight seeds in West Lafayette 2017. A positive ASV phenotype was scored as 1 (complete gelatinization) and negative ASV, similar to Tx623, Sepon82, and Macia, as 0 (no swelling, no disintegration). SC489, SC491, SC587 and SC589 and their crosses from genetic studies were screened for ASV using 32 seeds each.

Mode of inheritance for ASV
SC489, SC491, SC587 and SC589 were crossed to ATx623 in West Lafayette 2018 to produce F 1 hybrid seeds. The F 1 seeds were screened for ASV using 32 seeds each.

CTAB DNA extraction
DNA was extracted from individual plants using a CTAB extraction protocol (Griebel et al., 2019b;Murray & Thompson, 1980

Next generation sequencing (NGS) and analyses
Next generation sequencing analyses including reads filtering, contaminants cleaning, mapping to the reference genome, and SNP calling using GATK were conducted as described in Griebel et al. (2019b).

Physiochemical analyses
Flour preparation, starch extraction and starch thermal properties using differential scanning calorimetry (DSC) were analysed as described in Griebel et al. (2019a). The samples were milled into a fine flour using a ball mill. Seeds were milled in 30s intervals to avoid starch damage. Starch extraction protocol is explained in detail in Griebel et al. (2019a) and Benmoussa and Hamaker (2011). In short, flour samples were treated with NaCl and sodium metabisulfite followed by a sonication step. The samples were than filtered through a standard 230 mesh-screen and sonicated again. After centrifugation, a sucrose density solution (65%) was used to obtain starch pellets. Starch thermal properties were conducted in duplicate for each starch sample using differential scanning calorimetry (DSC, Q2000-1223, TA instruments, New Castle, DE, USA) (Griebel et al., 2019a). Tzero hermetic aluminium pans and lids were used containing a starch:deionized water ratio of 1:3 and samples were stored at 4˚C overnight before being analysed. The heating profile was chosen from 20˚C to 120˚C at a rate of 5˚C/min. TA instruments universal analysis software was used for data analysis. Moisture determination, viscosity analyses, and amylose and amylopectin analyses were determined following the F I G U R E 2 Principal components showing clustering of the sorghum diversity panel by (a) race, (b) geographic origin, and (c) ASV phenotype in 2D plots with principal components 1, 2 and 3. Genotypes were evaluated in West Lafayette in 2013 (WL13) and 2017 (WL17) F I G U R E 3 ASV phenotypes at 1.5%, 1.8% and 2.0% KOH for known SbEMS mutants and four SC lines having an ASV+ phenotype and Tx623 having an ASV-phenotype F I G U R E 4 Origin and genetic relationships among sorghum lines exhibiting ASV phenotypes -(a) Panicles from a subset of lines exhibiting ASV+ phenotypes -SC489, SC491, SC587 and SC589. (b) SC lines exhibiting an ASV+ phenotype in West Lafayette 2013 and 2017. (c) A neighbor joining tree depicting genetic relationships among accessions with ASV+ phenotypes procedures described in Griebel et al. (2019b). Viscosity analysis required adjustment for moisture content using a standard oven method (Bradley Jr., 2014;Griebel et al., 2019b). The Rapid Visco Analyzer (RTE-100, model 4, Newport Scientific, Australia) was used to run paste viscosity analysis in the setting of standard method 1. The samples were prepared as water-flour dispersions of 11.86% (Bouvier & Campanella, 2014). Amylose and amylopectin analysis was conducted with the K-AMYL Kit from Megazyme.

Proximate analysis NIRS
NIRS was used to predict protein (db), moisture and starch (db) of whole sorghum seeds from samples produced in Puerto Rico 2016/2017 and West Lafayette 2017 using a Perten NIRS (Model DA 7250, Serial Number 1211581). Two technical replications were conducted for each sample.

Statistical analysis for starch quality traits
Linear mixed model spreading analysis of variance (ANOVA) with the Tukey-Kramer multiple comparison adjustment was performed in SAS 9.4 using the GLIMMIX procedure to compare starch traits across genotypes and environments as described in Griebel et al. (2019b). Parts of the data analysis for this paper were generated using SAS software, Version 9.4 of the SAS System for Windows (SAS Institute Inc., Cary, NC, USA).

Principal component analysis
The race designations and country of origin for the Sorghum Conversion lines were kindly provided by R.R. Klein (personal communication, 2019) with additional information summarized from the USDA GRIN database (USDA-ARS, 2019). The countries of origin were grouped by continent for the minority of lines from Asia and the Americas and kept a single group for India. The majority of the lines are from the African continent and those countries were grouped into regions based on United Nation classifications (United Nations, 2019). The non-SC lines were treated as breeding lines and were not given a race or country designation. Principal component analysis was conducted using GAPIT (based on prcomp) (R. Lipka, personal communication, 2019) with 5 PCs. The PCs were plotted using R v. 3.5.1 and the package "ggfortify" R core Team, 2013;Tang, Horikoshi, & Li, 2016).
T A B L E 1 Whole genome re-sequencing analyses show SC487, SC491, SC587, and SC589 share common SNPs in starch biosynthesis genes

Genome wide association analysis
The imputed genotype-by-sequencing data were kindly provided by Dr. Patrick Brown and raw data are available at University of Illinois database, https://doi.org/10.13012/B2IDB-7570206_V1 (P. Brown, personal communication, 2019;Brown, 2017). The GBS data were aligned to the sorghum reference genome version 3 from Phytozome11 (Goodstein et al., 2012, version 11) and imputed using the Beagle 4 software (P. Brown, personal communication, 2019). 80,103 SNP markers at a minor allele frequency (MAF) of 0.025 were selected from the total set of markers and used in the genome wide association study.
The ASV data were prepared as binary data with 0 = ASVand 1 = ASV+. ASV− were only lines that had no swelling and disintegration at all in any of the seeds tested. Trait The Plant Genome F I G U R E 5 Linkage Disequilibrium (LD) decay in the sorghum diversity panel correlations were calculated using SAS 9.4 and the proc corr procedure. The ASV data from 2013 and 2017 were used as input phenotypes for the GWAS using a mixed linear model (MLM) (Lipka et al., 2012) with 80,103 SNP markers. The Genome Association and Prediction Integrated Tool (GAPIT) (Lipka et al., 2012;Zhang et al., 2010) was used for the mixed linear model (MLM). The PCs were calculated using GAPIT and set to 5. The MLM in GAPIT was run with the default settings and the Van Raden kinship algorithm. The significant SNPs were determined using a bonferroni (Bonferroni, 1936) threshold of 6.242E-07 and at an FDR of adjusted p-values at 0.05 (Benjamini & Hochberg, 1995). Manhattan plots were created using the GWAS results and the R package "qqman" (R Core Team, 2013; Turner, 2014).

Race and geographic classification of the sorghum diversity panel
The sorghum diversity panel consisted of 788 SC lines from the sorghum conversion program and 51 sorghum breeding lines included as checks. The SC lines originated from diverse countries before being converted in the sorghum conversion program (Supplemental Table S1). The countries of origin are predominantly India (171 SC lines), Ethiopia (139 SC lines), Sudan (116 SC lines) and 305 SC lines from 25 other African countries (Supplemental Table S1). The races represented in the diversity panel are predominantly bicolor (37), caudatum (268), durra (171), durra-bicolor (61), guinea (124), kafir (21) and kafir-caudatum (22) (Supplemental Table S1).
The GBS data (P. Brown, personal communication, 2019;Brown, 2017) were used to conduct principal component analysis to determine the covariates used for GWAS and to understand how the sorghum lines in the diversity panel clustered based on ASV, race, and origin. A scree plot from the principal component (PC) analysis showed that the first five PCs explained much of the variation (approximately 23%) from the SNP markers ( Figure 1). The red line was drawn at the characteristic elbow as suggested by Wang et al. (2013).
The first 3 PCs were plotted against each other and colored by their race classification, region of origin and ASV phenotype (Figure 2; Supplemental Table S1). PCs 4 and 5 explained the least variation and were not plotted in these comparisons. The PC1 captured most of the variation and differentiated the durra types from India and some durra-bicolors from eastern Africa (Figure 2). The PC2 differentiated caudatum types originating from northern and eastern Africa as well as guineas and some caudatums from West Africa. The PC3 differentiated kafirs from eastern and southern Africa. The PC4 was dominated by the durra-bicolor cluster and some bicolors from eastern Africa (not shown). PC5 did not show a clear distinction between groups but differentiated a group of guineas from West Africa listed as a second group in PC2 (not shown).

Variation in the ASV phenotype in the sorghum diversity panel
The ASV assay described by Griebel et al. (2019a) was effective in distinguishing variation in starch quality in the standing variation of sorghum ( Figure 3). Comparing concentrations of 1.5%, 1.8% and 2.0% KOH showed that Tx623 did not exhibit any ASV phenotype. Three SbEMS mutants with known ASV+ phenotype (Griebel et al., 2019a) and four SC lines clearly showed the expected ASV+ phenotype at 1.8% and 2.0%, while 1.5% KOH was not effective in differentiating the ASV phenotype.
Variation in the ASV phenotype was found in the sorghum diversity panel in 2013 and 2017 but the observed trait correlations between years was low (r = 0.23). The study in 2013 was set up as an exploratory study with 2 samples per genotype to determine whether variation for ASV could be detected among lines. There were 24 lines that exhibited the ASV phenotype in 2013 (salmon dots in Figure 2c; Supplemental Table S1). Subsequent studies showed that the expression of ASV was more variable in diverse sorghum accessions than in genetic mutants (Griebel et al., 2019b)  The passport data of accessions exhibiting stable ASV+ phenotypes over years showed that 11 of the 13 lines were durra types representing the working group 46(1) Nandyal (Figure 4a, b, and c). The other two accessions were a guinea sorghum representing working group 1: Roxburghii and another accession that had no clear race classification (R. Klein, personal communication, 2019). A neighbor-joining tree created in TASSEL (Bradbury et al., 2007) was used to depict the relationships among lines and showed that the Nandyal accessions clustered together and were closely related (Figure 4c). We selected four SC-lines, SC489, SC491, SC587 and SC589 for further genetic, genomic and physiochemical analysis.

ASV+ sorghum accessions share SNPs in starch branching enzymes
Whole genome re-sequencing of SC489, SC491, SC587 and SC589 revealed more than 200 common homozygous SNPs in candidate genes for starch biosynthesis. However, only 107 SNPs were shared in all four lines and 59 of these SNPs were in genes for starch branching enzymes, like SBEIIb Sobic.004G163700 (Table 1).

Genome-wide association study of ASV in sorghum
The genome wide linkage disequilibrium (LD) decay in the SDP was estimated using GAPIT and shown to be approximately 12 kb ( Figure 5). GWAS using the MLM model detected 53 SNPs associated with ASV in 2013 and 8 SNPs in 2017. Using the FDR (Benjamini & Hochberg, 1995) adjusted p values at 0.05, 205 SNPs exhibited significant associations for ASV in 2013 and 14 SNPs in 2017 ( Figure 6). We did not run a combined analysis across environments due to the low trait correlation of 0.23 between environments. Therefore, we report GWAS with environment specific loci associated with ASV.
All significant SNPs from 2017 and the top ten significant SNPs from 2013 including S10_60808604 and S04_67849488 were further evaluated using the integrated genome viewer (IGV; Robinson et al., 2011;Thorvaldsdóttir, Robinson, & Mesirov, 2013). The IGV was used to identify flanking genes that were approximately 15 kb up or downstream of a significant SNP, related to the estimated genome-wide LD decay of approximately 12 kb in this study. Those genes and their annotations from phytozome (Phytozome 12, version 12.1, 2019, Goodstein et al., 2012) are listed in Table 2. One significant SNP (S10_60808604) on chromosome 10 was approximately 113 kb away from Sobic.010G273800, a 1,4-alpha-glucan branching enzyme I precursor (Table 2; Campbell et al., 2016). The SNP S10_60808604 is also flanked by two other genes, Sobic.010G274800 (approximately 9 kb away) and Sobic.010G275001 (approximately 15 kb away), that are predicted to be glycosyltransferases (Goodstein et al., 2012, version 12.1) and may be involved in starch biosynthesis. More work is needed to validate the role of these genes, if any, in expression of the ASV phenotype.

ASV exhibits recessive inheritance
SC489, SC491, SC587 and SC589 exhibited stable expression of the ASV+ phenotype. These donors were crossed to ATx623 (cytoplasm sterile) and seeds from the crosses were evaluated to determine if the ASV is inherited as a dominant or recessive trait (Table 3). Very few of the hybrid seeds exhibited the ASV+ phenotype indicating that ASV is inherited as a recessive trait.

Instability of the ASV+ phenotype
Four SC lines were evaluated for ASV in 2017, and 2018 (Table 4). These lines exhibited the ASV+ phenotype in each year; however, the expression of the ASV+ phenotype was variable over years. For lines exhibiting the ASV+ phenotype, fewer than 22 seeds of 32 seeds tested exhibited the ASV+ phenotype depending on the individual environment, panicle source, or season. Genotypes exhibiting the ASV-phenotype exhibited a consistent phenotype over years.

Physiochemical analysis
NIRS predictions indicated that SC489, SC491, SC587, and SC589 and controls exhibited total starch (db) values ranging between 70% and 80% and protein (db) between 9% and 14% in 2013 and 2017 with values being slightly higher in 2017 (Table 5). G × E interactions were detected for amylose, starch thermal properties, and paste viscosity profiles (Figure 7). The amylose values of the four SC-lines varied between approximately 21% and 24% with the ASV+ lines being numerically lower than the controls in 2017 and 2018 (Figure 7). Significant differences for amylose were observed only for SC489, SC587 and SC589 in the season 2017 in comparison to Macia and for SC491 and SC589 in 2018 in comparison to Tx623. The starch thermal properties for SC489, SC491, SC587, and SC589 were similar to the control genotypes in 2017 but exhibited significantly lower starch GT for onset (T o ) and peak (T p ) in 2018 (Figure 7).
Evaluation of paste viscosity behavior showed similar profiles of the four SC-lines in comparison to the controls (Figure 7). However, Macia and Sepon82, behaved differently than Tx623 and the SC lines and exhibited no breakdown and lower final viscosity in 2018.

DISCUSSION
The sorghum diversity panel (SDP) included nearly 800 sorghum conversion lines and represents the genetic diversity of the crop (Stephens et al., 1967;Rosenow et al., 1997aRosenow et al., , 1997b. The SDP is larger than the commonly used Sorghum Association Panel (SAP) (Adeyanju et al., 2015;Boyles et al., T A B L E 2 GWAS results from MLM with significant SNPs and candidate genes for West Lafayette 2013 (WL13) (10 most significant SNPs and S10_60808604, S04_67849488) and 2017 (WL17) at FDR 0.05 2017; Casa et al., 2008;Cuevas et al., 2017;Mace et al., 2013;Morris et al., 2013;Shenstone et al., 2018;Sukumaran et al., 2012). The race and regional classification from our principal component analysis confirmed previous reports on the different sorghum races and geographic origins. The SDP is well-represented by durra types and durra-bicolor types from India and Eastern Africa, guinea types from Western Africa, kafir types from Eastern and Southern Africa, and caudatum types from diverse environments across Africa (Mor-ris et al., 2013;Billot et al., 2013;Sukumaran et al., 2012;Harlan & de Wet, 1972;Dahlberg, 2000). The bicolor types exhibited no clear geographical classification (Morris et al., 2013;Deu et al., 2006;Brenton et al., 2016;Billot et al., 2013). Recent genetic studies of sorghum EMS mutants demonstrated the ASV phenotype was generally stable across years and locations (Griebel et al., 2019b). The pattern of expression for ASV was less consistent in the SDP. The ASV phenotyping experiment in 2013 was set up as an exploratory study and 24 accessions exhibiting the ASV trait were identified ( Figure 3). Follow-up studies of these accessions indicated more variable expression of ASV so the phenotyping study of the sorghum diversity panel was repeated in 2017 with eight seeds per genotype. Thirteen accessions exhibited the ASV+ phenotype across two growing seasons (Figure 4). Analyses of genetic diversity confirmed that 11 of these SC lines cluster together as durras representing the Nandyal working group originating from India ( Figure 2). The other lines exhibiting a stable ASV+ phenotype included a guinea from Western Africa and a line of no clear race nor geographic classification. The race durra is reported as having ovate panicles, produce little hairs, and are very stiff and densely packed in comparison to the bicolors ( Figure 4) (Dahlberg, 2000). The durra race is sub-divided into the working groups 50, 51 and 52 (Dahlberg, 2000). The working group 51 describes durra types named Nandyal (Dahlberg, 2000). The Nandyal types are reported as short season durras (Harlan & de Wet, 1972) producing bold, glutinous grains (Golomb et al., 1976). The Nandyal sorghums included in this study exhibited densely packed panicles, creamy to yellow seed color, a strong glutinous and pearly endosperm, and medium to tall plant size (Dahlberg, 2000).
Glutinous seeds are often called waxy (Martin & Jenkins, 1950). An earlier study by Waniska (1976) was able to distinguish between waxy and non-waxy sorghum genotypes by treating the whole, pearled sorghum kernels with NaOH. Glutinous rice is desired in many Asian cultures and countries (Roder, Keoboulapha, Vannalath, & Phouaravanh, 1996) as it becomes soft and sticky after cooking (Wang et al., 2007;Golomb, 1976;Martin & Jenkins, 1950), slightly sweet (Golomb et al., 1976) and tender and glossy when eating, which is a result of low amylose content, low starch GT (Wang et al., 2007) and high gel consistency (Wang et al., 2007;Golomb et al., 1976). The four selected sorghum genotypes SC489, SC491, SC587 and SC589 exhibited similar physiochemical behaviours with slightly lower amylose content, high gel consistency and in some seasons a slightly lower starch GT (Figure 7). These sorghums may be suitable for use in sweet dishes and alcoholic beverages and less suitable for non-sticky and dishes where less viscosity is desired (Golomb et al., 1976).
The ASV trait has been shown to be controlled by the ALK gene, a starch synthase (SS) II in rice (Gao et al., 2011;Wang et al., 2007;Tian et al., 2009) and by a SSIIa and SBEIIb in sorghum (Griebel et al., 2019b). SC489, SC491, SC587 and SC589 have several unique SNPs in starch biosynthesis genes compared to Tx623 that may be involved in the expression of the ASV+ phenotype (Table 1). The 107 SNPs in common in the four lines are not necessarily due to the ASV+ phenotype but due to the close relatedness of the four lines from the same working group (Figure 4c). However, of the 107 SNPs, 59 are in starch branching genes, one being SBEIIb, reported previously as a key gene causing ASV in sorghum (Griebel et al., 2019a). Several genes control amylose content and gel consistency (GC) in rice (Tian et al., 2009;Gao et al., 2011). Genes associated with amylose content in rice include Wx, ALK, SSIII-2, SSI, PUL and for gel consistency ISA, SBE3 (Tian et al., 2009). In sorghum, ASV and starch GT are influenced by SSIIa and SBEIIb (Griebel et al., 2019b). The physiochemical behaviour of the four SC lines in the current study is similar to our previous study, where low amylose content, high gel consistency, and lower starch GT was observed in sorghum ssIIa mutants (Griebel et al., 2019b). The previously reported information on genetics and physiochemical behaviour could help to narrow down the putative candidate genes in the current study and may point to the gene that causes the ASV variation in the SDP. This is the first GWAS reported on the ASV phenotype in the standing variation of sorghum. The GWAS for ASV in sorghum produced results similar to studies in rice with poor trait correlations and inconsistent performance across years (Zhao et al., 2011). This might be influenced by fewer kernels tested in 2013 than in 2017, which could be evaluated in a follow up study in multiple environments and with a larger number of seeds per genotype being tested. The LD decay in rice is slower (Huang et al., 2010) than in sorghum with reports of LD ranging from 1 kb up to 20 kb (Figure 5; Morris et al., 2013;Mace et al., 2013). However, some candidate genes in rice were close to the ALK (SSII-3) gene (Huang et al., 2010;Zhao et al., 2011) and the SSII-2 (Zhao et al., 2011). In this study, the GWAS results do not show a significant SNP in close proximity to Sobic.004G163700, a SBEIIb, and Sobic.010G093400, a SSIIa, reported as the causal genes for ASV in sorghum (Griebel et al., 2019b). The GWAS identified three candidate genes for ASV. The SNP S10_60687678 was 9 kb away from Sobic.010G274800, 15 kb away from Sobic.010G275001, and 113 kb away from Sobic.010G273800 (Table 3). Sobic.010G273800 is annotated as a 1,4-alpha glucan branching enzyme I precursor (Table 3) and a known gene involved in starch biosynthesis in T A B L E 5 Near-infrared spectroscopy predicted moisture, starch, and protein contents of grain samples from sorghum landraces from 2013 and 2017  The common enzymes involved in starch biosynthesis are starch synthases, which are ADP-glucose-dependent (alpha 1,4-glucan-4-glucosyl transferases; Boyer, 1985) transferases, and starch branching enzymes (SBEs) being 1,4-alpha-glucan 6-glucosyl transferases (Tetlow & Emes, 2014;Boyer, 1985).

CONCLUSIONS
Variation in the ASV phenotype can be found in the standing variation of sorghum. The accessions that exhibit stable ASV across seasons are mainly durra types from the working group Nandyal described in previous studies as glutinous varieties. The four selected SC lines exhibited normal amylose content, high GC, and environmentally dependent lower starch GT. These starch characteristics may be valuable for the food industry for product development with different levels of stickiness and viscosity. GWAS showed significant associations for one SNP in proximity to a starch biosynthesis gene Sobic.010G273800 and two candidate genes Sobic.010G274800 and Sobic.010G275001 described as glucosyltransferases. Whole genome resequencing (WGS) revealed several SNPs in genes in the starch biosynthesis pathway that need to be further evaluated. We do recommend a follow up study in additional environments and with a larger number of seeds being tested per genotype to evaluate the effect of growth environment on the ASV expression in the sorghum diversity panel. However, with more than 800 lines it might be an elaborate and costly work to do even so the ASV test itself is cheap.

D A T A AVA I L A B I L I T Y S T A T E M E N T
The ASV phenotypic data used in GWAS studies are provided in Table S1. The imputed genotype-by-sequencing data were kindly provided by Dr. Patrick Brown and raw data are available at University of Illinois database, https://doi.org/10. 13012/B2IDB-7570206_V1 (P. Brown, personal communication, 2019;Brown, 2017).

AU T H O R C O N T R I B U T I O N S
S.G. designed and performed all areas of research, M.R.T. contributed seeds for the sorghum diversity population, A.A. helped with genotyping, S.G. wrote the manuscript; coauthors reviewed and edited the manuscript, and M.R.T. served as the major advisor and project lead.

C O N F L I C T O F I N T E R E S T None.
O R C I D Mitchell R. Tuinstra https://orcid.org/0000-0002-5322-6519