Introduction

Sorghum (2n = 20; ~ 730 Mb) is the fifth most significant cereal grain crop in production behind maize, rice, wheat and barley (USDA-NASS; www.nass.usda.gov). Originating from tropical and subtropical regions of Africa and later Southeast Asia, this climate resilient, C4 crop can now also be found in temperate growing regions of Australia, Europe and the Americas1,2,3. The grain is a popular gluten free substitute, the stalks are juiced for syrup in ethanol production and biomass is used as forage4,5,6. Lignocellulosic biomass from bioenergy sorghums are promising feedstocks for production in areas with marginal fertility7. These regions are not suitable for other commodity crops like corn and soybeans because they are prone to periods of extreme heat, drought, or sporadic rainfall8,9,10.

In sorghum, inflorescence meristem development is controlled by circadian clock, light quality, phytohormones, developmental stage and temperature11. Grain sorghum hybrids flower early (42–90 days after planting) to reduce the risk of exposure to abiotic stress during the reproductive and maturity phases. Sweet sorghums have longer vegetative growth periods (flowering 70–100 days after planting) that allows greater potential for sugar accumulation. Photoperiod sensitive forage and bioenergy sorghums flower extremely late (> 120 days after planting) in temperate environments12,13. African sorghum landraces (bicolor, guinea, caudatum, kafir, and durra) were converted from short-day flowering plants (photoperiod of < 11 h) to temperate photoperiod (days with > 12 h daylight), through spontaneous loss of function mutations at six Maturity (Ma1, Ma2, Ma3, and Ma414,15,16 and Ma5 and Ma612) loci (Table 1 and Fig. 1). Sorghum accessions with 5 out of 6 functional alleles at the Ma loci flower extremely late, approximately 120 days after planting under a temperate photoperiod,12,13. Different combinations of alleles at these loci have been used to tune the flowering set period from May 1st to October 1st in temperate zones (latitudes between 31°N and 45°N).

Table 1 Sorghum bicolor Maturity and Dwarf loci information.
Full size table
Figure 1
figure 1

A predicted model of flowering time pathway in sorghum based on GIGANTEA-CONSTANS-FLOWERING LOCUS T (GI-CO-FT) regulatory module, found in Arabidopsis and rice. Light regulated and gated by circadian clock proteins, LATE ELONGATED HYPOCOTYL (LHY), CIRCADIAN CLOCK ASSOCIATED (CCA1) and TIMING OF CAB1 (TOC1). SbPhyB may stabilize and interact with SbPhyC to inhibit flowering in long days (> 12 h) by activating expression of SbPRR37 and SbGhd7, which results in repression of floral activators Early Heading Date 1 (SbEHD1), and FT-like genes CENTRORADIALIS 8 and 12 (SbCN8 and SbCN12) required for floral initiation. SbPhyB facilitates repression of SbCN15. Under long days Ma2 may have an epistatic interaction with Ma4 to delay flowering by inducing expression of SbPRR37 and SbCO to co-repress the expression of SbCN12. Figure adapted and modified from Yang et al.17 and Casto et al.18. Created using BioRender.com.

Full size image

Most of the Ma genes have been shown to encode components that are involved in sensing photoperiod or transmitting a floral repressive signal under nonpermissive conditions, except for Ma4 that has not been identified, but maps to chromosome 10 (Fig. 1)18. Ma1 is located on chromosome 6 (Sobic.006G057866) and encodes a PSEUDORESPONSE REGULATOR 37 protein, that includes an N-terminal pseudoreceiver (residues 99–207) and C-terminal CCT [CONSTANS (CO), CO-like, and TIMING OF CAB1 (TOC1)] motif (residues 682–727) (Fig. 2). This locus strongly affects flowering time photoperiod sensitivity19. Ma2 is located on chromosome 2 (Sobic.002G302700) and encodes a SET (Suppressor of variegation, Enhancer of Zeste, Trithorax) and MYND (Myeloid-Nervy-DEAF1) motif containing protein, which is a member of SMYD (SET and MYND together) protein family (Fig. 2)18. The SET motif can methylate histone lysine residues with roles regulating chromatin state, transcription, signal transduction, and cell cycling20. The MYND motif comprises a DNA binding zinc-finger21. Ma3 and Ma5 encode phytochromes B and C apoproteins and are found on chromosome 1 (Sobic.001G394400 and Sobic.001G087100 respectively) (Fig. 2)17,22. These photoreceptors allow plants to detect red and far-red light23. The fully assembled holoprotein includes a chromophore covalently attached to the apoprotein. An N-terminal photosensory motif comprised of PAS (PER, ARNT and SIM) and GAF (cGMP phosphodiesterase, adenylate cyclase, Fh1A) to transduce a light signal, along with PHY (phytochrome-specific GAF-related), form a “light-sensing knot” (Fig. 2)24. The C-terminal dimerization moiety includes two PAS motifs and HKRD (histidine-kinase-related domain) for dimerization and nuclear localization (Fig. 2)17. Ma6 located on chromosome 6 (Sobic.006G004400) encodes a Grain number, plant height and heading date 7 (Ghd7) ortholog25. Like PRR37, GHD7 contains a CCT motif whose protein family members are involved in the transcriptional complex that regulates flowering (Fig. 2)26.

Figure 2
figure 2

Sorghum bicolor lollipop plot schematics for Maturity genes exons (Ma1, Ma2, Ma3, Ma5, Ma6) with predicted protein motifs27 labeled below. Select deleterious genomic variants are flagged at known locations along the length of the peptide (scale below each plot)28. Triangles represent insertions (point down) and deletions (point up), insertion or deletion a diamond, circles indicate a point mutation and color corresponds to the predicted peptide changes and effect on protein (blue missense, red nonsense, orange located in splice region).

Full size image

Genes from the flowering time pathway are found in most vascular plants, even non-flowering plants; however, their regulation has diverged. SbPhyB inhibits flowering in long days by activating expression of SbPRR37 and SbGhd7, resulting in repression of floral activators Early Heading Date 1 (SbEHD1), and FT-like genes CENTRORADIALIS 8 and CENTRORADIALIS 12 (SbCN8 and SbCN12) required for floral initiation17,19,29,30,31. PhyB increases PhyC stability and chromophore-containing PhyB:PhyC heterodimers are required for PhyC activity in Arabidopsis and rice. SbPhyC is also epistatic to SbPRR37 and SbGhd717. SbPRR37 together with SbGHD7, modulates photoperiod sensitivity and floral repression in an additive mechanism. Under long days, Ma2 delays flowering by inducing the expression of SbPRR37 and CONSTANS (SbCO), which Ma4 may encode32,33, and their gene products co-repress expression of SbCN1217. Under long days, SbGhd7 acts as a strong repressor of flowering, increasing photoperiod sensitivity by inhibiting expression of floral activators SbEHD1, SbCN12 and SbCN8, hence is light dependent and gated by circadian clock34,35. SbGhd7 is a known component of the photoperiod GIGANTEA-CONSTANS-FLOWERING LOCUS T (GI-CO-FT) regulatory module for short day rice and long day Arabidopsis25,36,37. There is a synergistic effect between constituents of SbGhd7 and SbPRR37 loci to enhance photoperiod sensitivity and delay flowering when functional alleles are present at both loci. In contrast, sorghum flowers early when loss of function alleles are present at both loci (Fig. 1)17.

Flowering time is positively correlated with height in sorghum, thus early flowering plants have short statures, reducing lodging risk and enabling machine harvest. Reduced height is achieved with shorter internode lengths while leaf area and maturity are unchanged38. The genes encoded at three of the four Dwarf (Dw1, Dw2, and Dw3) loci38 have been identified (Table 1 and Fig. 3). The Dw4 locus occurs at approximately 6.6 Mb on chromosome 6 but has not been further characterized39,40. Dw1 (Sobic.009G229800), located on chromosome 9, encodes a positive modulator of brassinolide (BR) signaling41 by inhibiting nuclear localization of signaling repressor, BRASSINOSTERIOID INSENSITVE 2 (BIN2) that prevents cell proliferation in internodes42. The dw1 mutants have reduced internode cell proliferation activity, a synergistic phenotype with Dw3, which can also result in reduced internode length43. The Dw2 locus is located on chromosome 6 (Sobic.006G067700) and encodes a protein kinase with similarity to kinesin-like calmodulin-binding protein (KCBP)-interacting protein kinase (KIPK) and is a member of the AGC protein kinase family in Arabidopsis44. Dw2 phosphorylates proteins involved in lipid signaling, endomembrane trafficking, hormone, light, and receptor signaling, and photosynthesis45. Dw3 is located on chromosome 7 (Sobic.007G163800) and encodes an ATP-binding cassette type B1 auxin efflux transporter (ABCB1)46. The height reduction phenotype found in dw3-ref mutant is from a loss-of-function P-glycoprotein; that decreases polar auxin transport in seedlings, reduces stalk height (from shortened lower internodes), increased stem thickness, and alters stalk vasculature47.

Figure 3
figure 3

Sorghum bicolor lollipop plot schematics for Dwarf gene exons (Dw1, Dw2, and Dw3) with predicted protein motifs27 labeled below. Select deleterious genomic variants are flagged at known locations along the length of the peptide (scale below each plot)28. Triangles represent insertions (point down) and deletions (point up), circles indicate a point mutation and color corresponds to the predicted peptide changes and effect on protein (blue missense, red nonsense, orange located in splice region).

Full size image

Current and emerging biotechnological advancements offer promising opportunities for improving sorghum bioenergy and green chemical applications. Next-generation sequencing has become an essential tool for obtaining large amounts of genetic information; although, available studies generally discuss the whole genome and not individual loci and phenotypes. Two recent sorghum whole genome sequencing studies were published with a diverse, worldwide collection (n = 499) and sorghum association panel (SAP; n = 400)48,49. Both studies used Illumina short read (2 × 100–150 bp) technology: sequencing depth ranged from 0.03 to 116.06\(\times\) (average 17.73\(\times\)) for the Lozano study and 25 to 72\(\times\) (average 38\(\times\)) for the Boatwright study. In this study, genomic variants in the known Maturity and Dwarf loci were analyzed to identify novel loss of function alleles and characterize the sorghum germplasm using the predicted variants. The sorghum germplasm presented is a valuable breeding resource for manipulating flowering time and plant height to temperate environments and functional ideotypes: grain, forage/biomass, and sweet.

Results

To analyze the sorghum loci controlling flowering time and plant height, SNPs and InDels identified in Maturity (Ma1, Ma2, Ma3, Ma5, and Ma6) and Dwarf (Dw1, Dw2, and Dw3) loci from two whole genome sequencing studies48,49 were used (see Figs. 4, 5 and Supplementary Tables S1–S9 online). Across all loci 1445 gene variants were identified. In the Lozano dataset48, 960 variants were found across these eight genes and 37 were predicted to have a high protein impact or were considered deleterious via SIFT prediction (Figs. 4 and 5)50. The low sequencing depth from several lines in this study did not impact the identification of novel alleles. For the Boatwright study49, 864 variants were reported; 58 had protein impact or were considered deleterious (Figs. 4 and 5)50. The Lozano study had 482 unique sorghum lines, whereas the Boatwright study included the entire 400 sorghum association panel. When compared together an overlap of 22 sorghum lines were found (see Supplementary Table S9 online). This resulted in 860 total sorghum lines for this study.

Figure 4
figure 4

Set diagrams indicating total number of whole-genome sequencing predicted variants from Lozano et al.48 (left set) and Boatwright et al.49 (right set) for Ma1, Ma2, Ma3, Ma5, Ma6, Dw1, Dw2, and Dw3 loci. The bottom set is a count of deleterious variants and intersecting the unique deleterious variants in each study50.

Full size image
Figure 5
figure 5

Number of SNPs and InDels for each predicted variant effect. The whole-genome sequencing data from Lozano et al.48 (left bar) and Boatwright et al.49 (right bar) for Ma1, Ma2, Ma3, Ma5, Ma6, Dw1, Dw2, and Dw3 genes.

Full size image

Deleterious genomic variants of maturity genes

Ma1 has seven previously characterized19,51 mutant alleles (Sbprr37-1 or prr27Milo, Sbprr37-2 or prr37Kafir-1, Sbprr37-3 or prr37Kafir-2, prr37Sudangrass, prr37Feterita, prr37Durra, and prr37Broomcorn) that can be identified using the resequencing datasets. Between the two studies, 530 unique variants were predicted from a total 652 variants (Fig. 4). Nomenclature from previous studies was used to label the corresponding alleles19,51. Sbprr37-1 or prr37Milo contains a 1-bp deletion upstream of the pseudo-receiver motif, likely resulting in a null or amorphic allele (Fig. 2). Sbprr37-2 or prr37Kafir-1 has a missense mutation p.(Lys184Asn) whose substitution of an uncharged asparagine for a positively charged lysine likely alters functionality of the pseudoreceiver motif. Sbprr37-3 or prr37Kafir-2 contains both a nonsense mutation p.(Gln270Ter) before the CCT motif, and a missense mutation p.(Lys184Asn). The Sbprr37-3 or prr37Kafir-2 allele is the result of a nonsense mutation at p.(Gln292Ter), and other characterized nonsense variants include prr37Durra p.(Ser443Ter) and prr37Broomcorn p.(Gln494Ter). Several other race specific alleles are also characterized20: prr37Sudangrass p.(Ile126Lys), prr37Feterita p.(Gly177RfsTer4), prr37Durra p.(Ser443Ter), and prr37Broomcorn p.(Gln494Ter). The prr37Sudangrass and prr37Feterita alleles are located in a conserved pseudoreceiver motif. An additional 11 alleles with predicted deleterious mutations (Figs. 4 and 5) include four nonsense, two 1-bp insertions, one 1-bp deletion, two splice site, and eight missense variants (see Supplementary Table S1 online). The nonsense mutation p.(Cys90Ter) in PI 562717 results in a predicted truncated protein, beginning at the pseudoreceiver motif (Fig. 2), which also contains nonsense mutation p.(Ser443Ter). Two 1-bp insertions p.(Val95GlnfsTer16) and p.(Gly177ArgfsTer4) and a 1-bp deletion p.(Gly209GlufsTer74) were predicted to result in frameshifts, found in 22 sorghum accessions. Splice acceptor site variants are predicted to affect splicing outside of the consensus sites in the + 3 to + 5 range at the beginning of introns and from -3 to -10 at the end of introns52. Two missense splice region variants in Ma1 were predicted after intron 1 p.(Val106Asp) in lines PI 656035 and PI 656050 and intron 2 p.(Met160Leu) in line PI 35038. Six additional missense amino acid changes were predicted: p.(Asn117Thr), p.(Ile126Lys), p.(Arg187Cys), p.(Ser210Cys), p.(Asp236Tyr), and p.(Trp264Arg). These missense mutations were estimated to have a deleterious effect on protein function due to their homology to conserved residues in orthologous genes and nine variants are in the conserved pseudoreceiver motif. Less than half (355) of the 860 lines sequenced in these studies predicted a fully functional Ma1 allele, while the Sbprr37-1 or prr27Milo, Sbprr37-2 or prr37Kafir-1, and Sbprr37-3 or prr37Kafir-2 alleles19,51 were found in 182, 156 and 45 lines, respectively (see Supplementary Table S1 online). The allelic composition at Ma1 locus was unable to be determined for 80 lines due to poor or missing data.

There are two characterized alleles of ma2. A likely amorphic allele with a nonsense mutation p.(Leu141Ter) in the third exon outside the characterized MYND motif (38 M, 44 M, SM60, 60 M, SM 80, and 80 M)18 was found in seven lines. A second characterized missense allele is predicted p.(Met83Thr) to be deleterious by PROVEAN (Protein Variation Effect Analyzer) and is only found in IS3614-218. Of the 119 gene variants identified, 5 were predicted to deleteriously affect Ma2 protein (Figs. 4 and 5). We identified four new variants p.(Gly23Ala), p.(Leu112Ile), p.(Tyr196Phe), and p.(Arg338Trp) in 21 accessions (see Supplementary Table S2 online). None of these missense mutations were predicted to alter amino acid residues within MYND-type zinc-finger motifs (Fig. 2). Most lines (832) in the datasets appeared to have a functional allele at Ma2 (see Supplementary Table S9 online).

Ma3 has two characterized loss of function alleles17 and only 40 lines were predicted to have ma3 loss of function alleles (see Supplementary Table S9 online). SbphyB-1 (previously Ma3R or Ma3Ryer; 58 M) allele has a 1-bp deletion p.(Asn1023MetfsTer11) that causes a frameshift resulting in a termination codon 30-bp downstream; one line was predicted to have this Sbphyb-1 allele. The SbphyB-2 (IS3620C) mutation contains a 3-bp in-frame insertion p.(His31dup) or deletion p.(His31del) and two missense mutations p.(Asp308Gly) and p.(Leu1113Val) (Fig. 2). Eight lines were predicted to have a Sbphyb-2 allele (see Supplementary Table S9 online). The first substitution p.(Asp308Gly) changes an amino acid in the conserved GAF motif; the residue change has a ‘Sorting Intolerant From Tolerant’ (SIFT) prediction score of 0.1 indicating moderate intolerance17,50. There were 236 total gene variants between the two studies for Ma3 gene, but only 6 mutations were predicted to have deleterious effects (Figs. 4 and 5). Five deleterious variants p.(Asp636Tyr), p.(Pro346Leu), p.(Arg337Gln), p.(Lys214Asn), and p.(Leu132Phe) were found in 13 lines. A large insertion of varying length (89–121 bp) within the GAF motif results in several frameshift variants, p.(Glu295AlafsTer16), p.(Glu295AlafsTer19), or p.(Glu295GlyfsTer25), detected in five heterozygous lines (see Supplementary Table S3 online). Two missense variants p.(Lys214Asn) and p.(Leu132Phe), were predicted to alter amino acids in the PAS fold-2 motif, and two other missense variants, p.(Pro346Leu) and p.(Arg337Gln), were predicted to alter amino acids in the GAF motif (Fig. 2). There were no predicted deleterious Ma3 alleles in 820 of the 860 accessions analyzed.

The SbphyC-1 (R.07007) allele is the only Ma5 mutant characterized17. There are four missense mutations: two located in the PAS motif p.(Gly124Val) and p.(Gly162Arg), one in the PAS-GAF loop motif p.(Val190Ala) of exon 1, and one in the HKRD motif p.(Glu922Asp) of exon 2 (Fig. 2). However, the phenotypes of only two of the missense mutations p.(Gly124Val) and p.(Glu922Asp) differ from the wildtype phenotypes of 90 M and 100 M, where both p.(Gly162Arg) and p.(Val190Ala) are present; thus, would likely not confer a change to the flowering time phenotype17. There were 44 genomic variants identified in Ma5, but only nine of them were predicted to have a deleterious effect on the protein (Figs. 4 and 5). The characterized Ma5 allele, SbphyC-117 was found in six lines (see Supplementary Table S9 online). Six newly identified, missense variants p.(Gly981Asp) [8 lines], p.(Thr653Ser) [PI 656078], p.(Met437Thr) [25 lines], p.(Thr371Ile) [PI 586430], p.(Gln274Glu) [4 lines], and p.(Ser49Tyr) [8 lines], were predicted to be deleterious (see Supplementary Table S4 online). The p.(Met437Thr) SNP is in the GAF motif, and p.(Thr653Ser) is in the PAS fold motif of the protein (Fig. 2).

Of the 159 total gene variants identified in Ma6, only six are predicted to have deleterious effects on the protein (Figs. 4 and 5). Amorphic allele Sbghd7-1 includes a 5-bp insertion (GTCGA) in exon 1 resulting in a frameshift before the CCT motif towards the end of exon 1 p.(Glu94AspfsTer6) (Tx623, 100 M, SM100, BTx406) (Fig. 2)25. We identified 385 sorghum lines with the Sbghd7-1 mutation whereas 281 lines had wildtype alleles at Ma6, but 165 lines were unable to be characterized due to incomplete or missing data (see Supplementary Table S9 online). An alternate Sbghd7-1 allele p.(Glu94SerfsTer77) located at the same position is the result of a 4-bp deletion and was identified in three lines. Sbghd7-2 is a hypomorphic allele resulting from a large insertion within the second intron (Red Kafir, Hegari, Double Dwarf Feterita, and Rio), whose size and impact have not been precisely determined25. Our analyses identified 21 lines with three variable length insertions (18–56 bp, 10 bp and 30–49 bp, respectively) in the second intron region at a nearly identical position (Chr06: 698,157, 698,160, and 698,179) to the ‘Rio’ Sbghd7-2 allele, where the large insertion was previously described25. These small insertions are likely hallmarks of the Sbghd7-2 allele, but the large insertion could not be fully characterized using short read resequencing mapped to a reference genome (BTx623). Newly identified deleterious variants found include two missense mutations p.(Cys6Tyr) [2 heterozygous lines PI 576347 and PI 576348] and p.(Arg220Gln) [PI 533965], two frameshift deletions p.(Cys14ThrfsTer154) [PI 534046 and PI 576437] and p.(Arg220GlyfsTer2) [3 lines] and a nonsense variant p.(Gln76Ter) [PI 655981] (see Supplementary Table S5 online). The CCT motif of the protein has two mutations predicted to affect the protein (Fig. 2): the 1-bp deletion p.(Arg220GlyfsTer2) and missense p.(Arg220Gln).

Deleterious genomic variants of dwarf genes

The Dw1 locus had the fewest number of total gene variants (31) of the accessions analyzed (Figs. 4 and 5). The dw1 allele contains a nonsense mutation at p.(Lys199Ter)41, which is likely an amorphic allele (Fig. 3). However, only one additional predicted deleterious variant p.(Gln398His) was detected. The p.(Gln398His) missense variant was heterozygous in two lines (PI 576375 and PI 609456), but no homozygous alleles were identified (see Supplementary Table S6 online). There are 287 lines with the dw1 nonsense mutation p.(Lys199Ter) among the two resequencing datasets, 554 lines wildtype for Dw1, and 17 lines unable to be characterized due to missing or incomplete data (see Supplementary Table S9 online).

There were 105 total gene variants found in Dw2, but only 2 were predicted to have deleterious effects on the gene product (Figs. 4 and 5). The single characterized dw2 allele p.(Leu184IlefsTer8) contains 2-bp deletion between amino acid position 183 and 184 in exon 1 resulting in a frameshift, hence likely encodes an amorphic allele (Fig. 3)44. Almost one third (261) of the lines were identified as having the dw2 allele while 571 were wildtype and 21 were not characterized due to missing or incomplete data (see Supplementary Table S9 online). A second allele p.(Glu183AspfsTer27) was identified in two lines PI 576396 and PI 656033 at the same location as the dw2 allele. One newly identified missense variant p.(Ser160Cys) was found in 14 lines (see Supplementary Table S7 online).

There are three previously characterized dw3 alleles found in exon 5. One (dw3-ref) contains an 882 bp duplication event, which reverts at a frequency 0.1–0.5% likely due to unequal crossing-over during DNA replication46. The others dw3-sd1 p.(Leu1435GlyfsTer120) a 2-bp deletion and dw3-sd2 p.(Gln1174_Arg1175del) a 6-bp deletion located in the C-terminal ABC transporter signature motif (‘LSGGQ’), a highly conserved region with no amino acid sequence variation among eukaryotes47,53. There were 221 gene variants identified in the Dw3 gene, with 21 predicted to have a deleterious effect on the gene product (Figs. 4 and 5). Given the large size of the duplicated repeat (882 bp) in the previously characterized dw3-ref allele46, we are unable to identify the reported duplication event in any lines from these two datasets (see Supplementary Table S9 online). Two lines (PI 533810 and PI 533927) were heterozygous for the dw3-sd1 allele p.(Leu1435GlyfsTer120)46, but no lines were identified as homozygous for this allele. There were five frameshift variants, a 137 bp deletion p.(Arg311GlnfsTer37) [5 lines], one 2- to 4-bp frameshift deletion p.(Cys17TrpfsTer123) [7 lines] or p.(Cys17AspfsTer54) [9 lines] respectively), a 2-bp deletion p.(Met599AsnfsTer956) [heterozygous PI 576426], and a 1-bp deletion p.(Gly307AlafsTer18) [5 lines]. Three nonsynonymous SNP variants p.(Gly887Arg) [3 lines], p.(Arg403Pro) [6 lines], and p.(Glu64Val [PI 291246 and PI 656047], two nonsense SNPs p.(Gln475Ter) [4 lines] and p.(Glu322Ter) [4 lines] where a stop codon is gained, and one splice donor variant in intron 4 (PI 656078) were predicted to be deleterious across 38 accessions (see Supplementary Table S8 online). Two missense variants p.(Glu64Val) and p.(Gly887Arg) and one 2 p.(Cys17TrpfsTer123) to 4-bp p.(Cys17AspfsTer54) deletion causing a frameshift are in the ABC transporter type 1, transmembrane motif (Fig. 3). The p.(Glu322Ter) and p.(Gln475Ter) nonsense mutations and p.(Arg403Pro) missense mutation are variants predicted to have an impact on the ABC transporter-like, ATP-binding motif. The analyses also confirmed the presence of dw3-sd2 allele in PI 533957, PI 655978 and PI 65599853.

Only 4 accessions among the 860 lines had deleterious mutations at four of the five maturity loci (see Supplementary Table S9 online). PI 533839 has mutant alleles Sbprr37-1, Sbphyb-2, p.(Gly981Asp) (ma5), Sbghd7-1, dw1, dw2, and p.(Cys17TrpfsTer123) (dw3) and a wildtype allele at Ma2. PI 656021 has mutant alleles Sbprr37-1, ma2, Sbphyb-1, Sbghd7-1, dw1, dw2 and wildtype alleles SbPHYC-2. PI 656081 has mutant alleles Sbprr37-2, Sbphyb-2, p.(Gly981Asp) (ma5), Sbghd7-1, dw2, p.(Cys17TrpfsTer123) (dw3) and wildtype alleles Ma2 and Dw1. PI 656116 has mutant alleles Sbprr37-2, p.(Arg388Typ) (ma2), Sbphyb-2, Sbghd7-1, dw2, p.(Cys17TrpfsTer123) (dw3) and wildtype alleles SbPHYC-1 and Dw1. These lines would likely make ideal candidates for development of improved grain sorghum due to the 2 or 3 dwarf alleles found in each line and a relatively short time to flowering (60–70 days) under a temperate environment.

Deleterious alleles and phenotype

To determine the allelic effect on sorghum phenotype, plant height and days to anthesis data was gathered from USDA-ARS Germplasm Resources Information Network (www.ars-grin.gov) and Mural et al.54. An Analysis of Variance (ANOVA) test was used to group various allelic means and a t-test to determine the statistical significance (α = 0.05).

Plant height had a significant (p < 0.001) association with the dwarf genes and the coefficients used to determine the differences among the alleles (see Supplementary Table S10 online). For Dw1 the p.(Gln398His) was not found to be significantly different (p = 0.397) compared to reference (Dw1) and mutant (dw1) phenotype. All the Dw2 alleles were significant (p < 0.001). The Shapiro–Wilk’s test for normality (W = 0.98786, p = 0.008686)55 provides evidence that the data is not normally distributed. A Kruskal–Wallis rank sum test was performed (p < 0.001) (supplementary Fig. S1)56.

The number of days to anthesis did not have a considerable association to all maturity alleles (see Supplementary Table S11 online). Each individual maturity loci had significant alleles (p < 0.001) for days to anthesis. However, when combining all 5 loci, only the Ma1 alleles were significant (p < 0.001) and some alleles were not viable for statistical inference due to singularities from the dataset. Based on the Shaprio-Wilk’s test (W = 0.99231, p = 0.667)55 it can be assumed the population is normally distributed. This result is likely due to the multigenic interaction of six characterized loci controlling the maturity phenotype in sorghum.

Discussion

The sequence variants characterized in the Maturity and Dwarf loci are a useful resource for sorghum germplasm development. In these accessions, previously identified alleles along with the genomic variants identified here can be used to manipulate flowering time and plant height. Flowering time is critical for developing plants that flower early to avoid the extreme heat of summer or mature before a killing frost in the fall. Grain sorghum hybrids in the U.S. have a flowering time between 42 and 90 days (photoperiod sensitive) after planting and are relatively short (typically 3-dwarf) for mechanical harvesting and lodging avoidance38. Grain hybrids are the result of combining recessive, loss of function alleles at the same Dw and Ma loci, so the resulting hybrid is homozygous recessive. Some production areas can achieve two or three harvests in a year using short season sorghum15. Forage sorghums generally have functional alleles at the Dw loci, which maximizes internode length and translates to greater biomass57. Forage hybrids may also be photoperiod sensitive, which prevents flowering at temperate latitudes resulting in longer vegetative growth periods and increased biomass production7. Lines considered dual purpose produce harvestable grain and stover that may be used in ensiling, and breeders target a longer vegetative growth period with a flowering time towards the end of the growing season to produce both grain and biomass57. Each of the sorghum ideotypes (grain, forage/biomass or sweet) were developed for their specific conditions and purposes.

By using available whole genome sequencing and examining these genes, 425 ma1, 22 ma2, 40 ma3, 74 ma5, 414 ma6, 289 dw1, 268 dw2 and 45 dw3 alleles were identified, which include ones previously reported. Prior to this work, there have been relatively few alleles characterized at Ma and Dw loci in sorghum. This study utilized whole genome sequencing datasets of over 800 lines to describe insertion, deletion, missense (nonsynonymous variants), nonsense (stop gain), and splice site mutations that are predicted to impair the function of the encoded protein. Day length has the greatest impact on flowering time in sorghum18. Ma1 had the most deleterious, highly impactful genomic variants compared to the other maturity genes. However, the previously characterized mutant alleles (Sbprr37-1, Sbprr37-2, and Sbprr37-3) are predicted most frequently in the accessions analyzed. SbPRR37 is a central component of the flowering regulatory pathway and is influenced by Ma2 and possibly Ma4, downstream of SbPHYB (and SbPHYC) and co-represses SbEhd1 with SbGhd717,18,19. Therefore, any substantial change to this protein would strongly affect flowering time. Ma2 is the least studied of the Maturity genes and its role in the flowering time pathway and its interaction with Ma4 has not been identified18. For the Ma loci, the fewest genomic variants and predicted deleterious alleles were detected at the Ma2 locus. This observation may result from the use of the parental line BTx406 (Sbprr37-1, Sbghd7-1, and recessive for all four dwarf loci) in the sorghum conversion program where this line was used as the recurrent parent to convert photoperiod sensitive exotic germplasm into photoperiod insensitive, dwarf lines58,59,60. The Ma3 locus has several predicted variants; however, many of the lines with ma3 mutant alleles are the previously characterized Sbphyb-1 allele. The Sbphyb-2 allele has three genomic variants corresponding to the reduced day-length sensitivity phenotype in which p.(His31dup) or p.(His31del) and p.(Leu1113Val) are predicated to not impact the protein as severely as p.(Asp308Gly), which alters a conserved residue of the GAF motif, thus likely affecting the function of the protein. Although several genomic variants were identified for the Ma5 locus that encodes for phytochrome C, very few of these variants were predicted to be loss of function alleles. This result may be because amorphic alleles of Ma5 are not tolerated in sorghum, and some residual level of function must be maintained in the protein due to the important biological role phytochromes play. The lack of nonsense mutations identified in Ma5 supports this hypothesis. The Ma6 locus had the largest InDels (5 and 10 bp) of the Ma loci analyzed; a 5-bp deletion was found in the previously characterized Sbghd7-1 allele25. Ma6 encodes the smallest protein (246 amino acids), and 159 deleterious genomic variants were identified in our analysis, unexpectedly, given the short coding region. Only one new deleterious genomic variant was identified in Dw1 and Dw2, but the previously characterized alleles (dw1 and dw2) were identified in many accessions by our analyses. Many accessions that were sequenced included their common progenitor in their pedigree; for example, Dwarf Yellow Milo (ma2, Sbghd7-1 and dw1) is a common progenitor for several lines, which may contribute to the lack of diversity of ma2, Sbghd7-1 and dw1 mutations among the lines analyzed.

Next-generation resequencing allows researchers to identify single nucleotide variants, insertions/deletions, and copy number changes across many genomes at minimal cost. However, there are limitations to using short sequencing read technologies and assembling these reads to a reference genome to identify variants, for example, the inability to detect repeat regions60. Identifying large structural variants substantially greater than the average read length is difficult or not possible without de novo genome assembly. For example, the dw3-ref allele contains an 882 bp tandem repeat that was not detectable due to limitations in short read sequencing (100–150 bp) with assembly to a reference genome (Tx623). Several new predicted alleles at Dw3 were identified in addition to dw3-ref, dw3-sd1 and dw3-sd246,53. The dw3-ref allele is unstable due to the tandem repeat and reverts at a frequency of 0.1–0.5%46; these predicted alleles identified could be used to stabilize this dwarfing phenotype.

Availability of sorghum genomic data is vital for elucidating the genetic architecture of traits and propelling genomics-assisted breeding. Accessions containing previously identified alleles as well as genomic variants not previously-characterized with corresponding germplasm can be used for line development or to study the impact at a locus. This public resource is valuable for fully utilizing the huge variety of sorghum germplasm to develop improved parental lines and hybrids. Developing hybrids with modified flowering and height genes gives breeders flexibility to adapt grain, sweet, and forage/biomass sorghums to specific uses and environments61. Having multiple loss of function alleles at critical loci controlling flowering time and plant height will allow for biallelic combinations at each of these loci to maintain heterosis in hybrids.

Methods

Whole genome sequencing data analyses

The exon/intron junctions, translation start and stop sites, were obtained from SorghumBase (www.sorghumbase.org) and literature. Additional selected SNP/InDel mutations of all genes were collected from previously published studies48,49. In the Lozano study48, 499 sorghum lines from a diversity panel were sequenced and about 41 M variants were identified (~ 35 M SNPs and ~ 3.5 M InDels). After quality filtering, ~ 13 M SNPs and ~ 1.8 M InDels were recorded using the reference genome (BTx623)62. The Boatwright study49 utilized the sorghum association panel (SAP) of 400 accessions and found almost 44 million variants (~ 38 M SNPs and ~ 5 M InDels) but after filtering ~ 19.7 M SNPs and ~ 2.6 M InDels remained. In total, 860 unique accessions were included between both sequencing studies with 22 lines overlapping from both studies. The VCF (Variant Call Format) files were uploaded to the Holland Computing Center at University of Nebraska, Lincoln. UNIX command line operation of intersect was used from the BEDtools utilities suite63 to parse the reported variants based on chromosome location.

Genomic variant analysis

The ‘Sorting Tolerant From Intolerant’ (SIFT) algorithm50 predicts whether an amino acid substitution is deleterious to the corresponding protein product for every non-synonymous single nucleotide polymorphism in a coding gene. SIFT uses protein sequence homology to identify conserved amino acids throughout evolution and provides a score of the putative deleterious effect of all possible substitutions at each position in the protein sequence. These scores range from 0 to 1, and positions with a SIFT score < 0.05 are predicted to be deleterious50. Variants were separated into four categories: same sense (synonymous) mutations (mutations that do not change the encoded amino acid), tolerated mutations (nonsynonymous missense mutations, SIFT > 0.05), nonsense mutations, and deleterious missense mutations (SIFT < 0.05). Sorghum SIFT annotations were calculated with the EnsemblPlants Variant Effect Predictor Web Tool, a database using the Sorghum_bicolor_NCBIv3 assembly (plants.ensembl.org). Only primary transcripts were considered. Descriptions of sequence variants are based on nomenclature recommendations from the HGVS nomenclature v20.05 (varnomen.hgvs.org). This includes p. for prediction of reference amino acid, codon number, alternative amino acid, del for deletion, ins for insertion, fs for frameshift, and Ter for the introduced stop codon and number of codons from the frameshift. Some figures were created using R Statistical Software (v4.2.3)64. Lollipop gene schematics were generated using trackViewer R package (v1.34.0)65. Bar charts were graphed using the ggplot2 R package (v3.4.2)66.

Statistical analysis of genotype and phenotype

R Statistical Software (v4.2.3)64 was used to calculate the linear models for each allele and the corresponding phenotype. The Shaprio-Wilk’s test55 for normality was performed and Kruskal–Wallis test56 as a non-parametric alternative. Phenotypic data was collected from USDA-ARS Germplasm Resources Information Network (www.ars-grin.gov) and Mural et al.54.