Introduction

Sorghum is the fifth most widely grown cereal crop worldwide (faostat.fao.org). Its drought and heat tolerance make this crop especially important in semi-arid regions. Sorghum is a C4 grass with a diverse germplasm that has been selected for many uses including production of grain, forage, sugar, and biomass for bioenergy. In its native Africa, sorghum grows 4–5 meters tall and many genotypes are photoperiod sensitive, resulting in delayed flowering in long day environments. Upon introduction to temperate locations, photoperiod insensitive varieties that flower early were selected for production of grain1. Additionally, shorter grain varieties were selected to reduce lodging and to aid mechanical harvesting. In contrast, sorghum genotypes with longer stems and delayed flowering enhance biomass and sugar production2, 3. In sweet sorghum, stem length is associated with higher sugar yield because stems accumulate high levels of sucrose post floral initiation3,4,5. In energy sorghum, 83% of the shoot biomass accumulates in the stem6. Therefore, increasing our knowledge of stem growth will aid the improvement of sorghum hybrids for bioenergy production.

Plant height is determined primarily by the length and number of stem internodes. The number of internodes produced by a plant is a consequence of growth duration and the rate of internode production. Quinby and Karper7 identified four loci (Dw1-Dw4) that control internode length by measuring the height of the stem from the ground to the flag leaf. At each Dw locus the dominant allele increased internode length. Recessive alleles of Dw1 and Dw2 were identified in Milo lines, while recessive alleles of Dw3 were identified in Kafir backgrounds, and dominant alleles at Dw4 were only found in broomcorns7. Dw2 was shown to have pleiotropic effects on panicle length, seed weight, and leaf area8, 9. In addition to internode length, Dw3 influences grain yield, tiller number10, and leaf angle11.

Dw3 was the first dwarfing gene to be cloned in sorghum12. Dw3 encodes a homolog of the maize Br2 gene and is an ATP-binding cassette type B1 (ABCB1) auxin efflux transporter. This is in contrast to dwarfing or semi-dwarfing genes in other important crops, such as rice and wheat, which have mutations in genes involved in the gibberellin pathway13, 14. Dw1 was mapped to a region on chromosome 9 between 56.8–57.1 Mb15. The gene corresponding to Dw1 was recently identified as Sobic.009G229800 by map-based cloning16, 17. This gene regulates internode cell proliferation17 and encodes a putative membrane protein not previously assigned a function16. The recessive dw1 allele in Dwarf Yellow Milo (DYM), first identified by Quinby and Karper7, contains a stop codon in exon 2 that results in protein truncation16. The dw1 allele originating from Dwarf Yellow Milo has been used extensively in grain sorghum breeding programs.

Dw2 has also been used extensively in grain sorghum breeding programs to reduce plant height. Dw2 is linked to Ma1, an important flowering time gene that confers photoperiod sensitivity1. Ma1 is located on chromosome 6 at ~40.3 Mb and encodes PRR3718. Dw2 was previously mapped to a location near Ma1 at ~42 Mb in several QTL mapping studies19,20,21,22 and was associated with a SNP marker at 42.7 Mb in a GWAS study23. In another study, Dw2 was suggested to be a histone deacetylase (Sobic.006G067600) based on GWAS analysis21. Recessive alleles of Ma1 and the dwarfing genes were used in the Sorghum Conversion Program to convert tall late flowering landraces from Africa into short, early flowering genotypes that are useful for grain sorghum breeding. The landraces were crossed to BTx406 (dw1dw2dw3dw4) to introduce one or more of the recessive alleles at the Dw loci into landrace backgrounds19. Recent analysis of the sorghum conversion lines has shown that large portions of chromosome 6 have been introgressed from BTx406 into landrace accessions during conversion and that the peak of introgression frequency aligned with Dw2 24.

In the current study, Dw2 was map-based cloned using two RIL populations: BTx623 (dw1Dw2dw3dw4) x IS3620c (dw1dw2Dw3dw4) and BTx642 (dw1dw2dw3dw4) x Tx7000 (dw1Dw2dw3dw4). Dw2 was identified as a protein kinase whose closest homolog in Arabidopsis is the kinesin-like calmodulin-binding protein (KCBP)-interacting protein kinase (KIPK), a member of the AGCVIII subfamily that also includes PINOID (PID) and PHOTOTROPIN1 and 2 (PHOT1 and 2).

Results

Comparison of DYM and DDYM internode lengths

The recessive dw2 allele present in Double Dwarf Yellow Milo (DDYM), the original source of dw2, arose as a mutation in Dwarf Yellow Milo (DYM)19, 25. Comparison of DYM and DDYM stem internode lengths at anthesis showed that the recessive allele of dw2 in DDYM caused a reduction in the length of every elongated internode compared to the corresponding internodes in DYM (Supplementary Fig. S1). The dw2 allele found in DDYM was used extensively in U.S. grain sorghum breeding programs and the Sorghum Conversion Program19 to reduce the length of stems of sorghum genotypes such as IS3620c and BTx642 that were used in this study to clone Dw2.

QTL Mapping using a RIL population derived from a cross of BTx623 x IS3620c

QTL for total stem length, average internode length, the length of each internode numbered from the peduncle, and the length of the peduncle were mapped using the BTx623 x IS3620c RIL population (Fig. 1, Supplementary Fig. S2, Table 1). As expected the population segregated for Dw2 on chromosome 6 (~42.7 Mb) and for Dw3 on chromosome 7 (~59.8 Mb) and these loci affected both total stem length and internode length. An additional QTL (Dw03_67.5) at ~67.5 Mb on chromosome 3 affected total stem length (Fig. 1). The influence of Dw2 and Dw3 on the length of the eight internodes was analysed to determine if the action of these genes varies with development (Table 1). Dw3 affected the length of all eight internodes measured. Dw2 influenced the length of the first five internodes but had minimal impact on the length of internodes 7–8. There is an additional QTL on chromosome 6 (48.6 Mb, Dw06_48.6) near Dw2 segregating for the length of the sixth internode below the peduncle. However, the peaks for the fifth and sixth internode are broad and the 2-LOD interval for the peak on chromosome 6 for both internodes includes both Dw2 and Dw06_48.6 (Table 1, Supplementary Fig. S2). The additive effect of Dw2 and Dw3 on internode length varied with internode number (Fig. 2). The additive effect for Dw2 was highest for the internode immediately below the peduncle. The additive effect of Dw3 on the length of the same internode was similar to that of Dw2. However, Dw3 influenced the length of internodes formed earlier in development more than Dw2. The additive effect of Dw3 decreased from the sixth to eighth internodes (Fig. 2). QTL for peduncle length did not align with Dw2 or Dw3 (Supplementary Fig. S2).

Figure 1
figure 1

QTL identified using the BTx623 x IS3620c RIL population. The RIL population was grown in the greenhouse and genotyped using DG. Stem length (a) was measured from the base of the plant to the base of the panicle. Genetic map generation and QTL mapping were performed in R/qtl using interval mapping (IM). The x-axis is the markers along the chromosomes and the y-axis is the LOD score. The significant QTL peaks are labeled with the Dw locus and location (Mb). Stem length (a), average internode length (b), and the length of the first internode below the peduncle (c) are shown.

Full size image
Table 1 QTL Segregating for Stem Traits in the BTx623 x IS3620c RIL Population.
Full size table
Figure 2
figure 2

Additive effects of Dw2 and Dw3 on the length of each internode (BTx623 x IS3620c RIL population). The RIL population was grown in the greenhouse and the length of each internode (numbered from the peduncle) was measured. Additive effects were determined as part of QTL mapping performed in R/qtl using IM. The BTx623 allele of Dw2 increases internode length, whereas the IS3620c allele of Dw3 increases internode length.

Full size image

There was no strong statistical evidence of a genetic interaction between Dw2 and any of the other loci from the multiple-QTL mapping (MQM) analysis (Supplementary Table S1). For the best model for each phenotype, the only phenotype that included interactions in the model with the highest LOD is the length of internode 7. There are two interactions in this model, one between a QTL on chromosome 5 and Dw3 (chromosome 7 at 59.8 Mb) and another between a QTL on chromosome 1 and a QTL close to Dw3 (10.7 cM from Dw3 at 61.2 Mb) (Supplementary Table S1). The composite multiple-QTL model included both interactions and revealed interesting trends between the internode length traits with internodes further from the peduncle having better support for the two interactions (Supplementary Table S2, Supplementary Fig. S3). Additionally, composite model analysis clarified the effects of the two QTL on chromosome 6. Dw2 affects the length of internodes 1–6, but starting at internode 4 and continuing through internode 7 the QTL at ~49 Mb on chromosome 6 also affected internode length (Supplementary Table S2, Supplementary Fig. S3).

Dw2 fine mapping and gene identification

Dw2 was fine mapped in a second RIL population derived from BTx642 x Tx7000 that was expected to segregate for alleles of Dw2 in a background fixed for recessive Dw1, Dw3, and Dw4. QTL analysis of the BTx642 x Tx7000 RIL population for total plant height revealed a major QTL aligned with Dw2 as expected (Fig. 3a). The QTL corresponding to Dw2 showed a peak located on chromosome 6 at ~43.2 Mb. The 2-LOD interval containing Dw2 in the BTx642 x Tx7000 RIL population spanned a region of ~756 kb on chromosome 6. Eight RILs with recombination breakpoints in this region were identified and targeted for higher resolution analysis of breakpoint locations. Sequence polymorphisms within the target interval identified using high resolution DG analysis and by targeted gene sequencing were used to fine map the breakpoints in the eight fine mapping lines (Fig. 3b). Four RILs with breakpoints closest to Dw2 were phenotyped in a greenhouse during the winter. Phenotyping in the winter under low light conditions revealed that Dw2 had a large impact on the length of the internode below the peduncle. As a consequence, RILs containing Dw2 could be readily distinguished from RILs encoding dw2 by phenotyping eight plants from each genotype for the length of the internode below the peduncle (Fig. 3c and d, Supplementary Fig. S4). The information from lines with breakpoints delimited the Dw2 locus to a region spanning ~98.1 kb containing ten genes (Fig. 3, Table 2, Supplementary Table S3). The genes within this region were annotated in Phytozome as encoding a PPR repeat protein, an rRNA N-glycosylase, an F-box protein, a glycogen branching enzyme, a phosphatase, a histone deacetylase, a kinase, and three genes of unknown function.

Figure 3
figure 3

Fine mapping of Dw2 in the BTx642 x Tx7000 RIL population. (a) QTL map of total plant height (2009) with Dw2 labeled. Plant height was measured as the length of the plant from the base of the stem at ground level to the top of the panicle. Genetic map construction and QTL analysis were performed in R/qtl using IM. The x-axis is the markers along the chromosomes and the y-axis is the LOD value. (b) Diagram of fine mapping in BTx642 x Tx7000. The diagram shows the location of the recombination breakpoints in the 2-LOD region in the eight fine mapping lines (numbers at bottom), two of these lines had more than one recombination breakpoint in the region. The markers found through DG using NgoMIV are labeled as “Ngo_”. The markers found with Sanger sequencing are labeled with “SNP_ _” with the last five digits of the gene name. The red, dashed-line box shows the refined region of Dw2. For both (b) and (c), asterisk indicates the approximate location of Dw2. (c) Diagram of the haplotypes of the four fine mapping lines with breakpoints closest to the refined region. The region between Ngo1 and Ngo3 is shown. Blue indicates that the RIL has the BTx642 allele, red is the Tx7000 allele, and grey is the region where the breakpoint is located. Dashed lines flank the refined region of Dw2. (d) The length of the first internode below the peduncle in the same lines shown in (c). Blue indicates that the line is dw2 while red is Dw2. Average (n = 4) and standard deviation is shown.

Full size image
Table 2 Genes in the Delimited Region of Dw2.
Full size table

Sequence analysis of genes in the Dw2 locus

The gene corresponding to dw2 is expected to contain a mutation(s) that decreases function; therefore, all of the genes in the delimited Dw2 locus (Table 2) were sequenced from DYM and DDYM. Only one polymorphism was found in the delimited Dw2 locus that distinguished DYM from DDYM, an indel in Sobic.006G067700 located in the first exon at 549 bp that causes a frameshift resulting in a stop codon at 573 bp. This mutation changed the amino acid sequence after E183 resulting in a truncated polypeptide containing 190 amino acids instead of the 809 amino acids present in the full-length protein. The indel mutation in Sobic.006G067700 that causes protein truncation was also present in BTx642 and IS3620c, genotypes that acquired dw2 by introgression from DDYM, and not present in BTx623 (Dw2) and Tx7000 (Dw2) (Tables 3 and 4). None of the parental lines contain polymorphisms in the coding region of the histone deacetylase (Sobic.006G067600), a gene previously proposed as a candidate for Dw2 21. A number of sequence variants in the Dw2 delimited region were identified that distinguished the parental mapping lines (Supplementary Table S4); however, none of these variants differentiated DYM (Dw2) from DDYM (dw2), the source of the recessive allele of dw2.

Dw2 alleles in sorghum germplasm

The distribution of Dw2 alleles in historically important sorghum genotypes was investigated by sequencing Sobic.006G067700 from the genotypes listed in Tables 3 and 4. Many of the genotypes identified by Quinby and Karper7 as dw2 contain the indel in Sobic.006G067700 derived from DDYM. For example, 80 M is a maturity standard with the reported genotype dw1dw2Dw3dw4 1. 80 M and the other maturity standards were selected from a cross of Early White Milo (Dw2) and DDYM (dw2) and contain the DDYM dw2 allele (Table 4). BTx406 also contains the DDYM dw2 allele, consistent with a previous study showing that the dw2 allele in BTx406 was derived from DDYM based on pedigree analysis and haplotype-graphical genotyping19. BTx406 was used in the conversion of IS3620 to the early flowering short genotype IS3620c, explaining why IS3620c contains the DDYM dw2 allele. SC170 and BTx642 (formerly B35) were also crossed to genotypes containing the DDYM dw2 allele during their construction26. Several genotypes contained polymorphisms in exons that changed the protein sequence encoded by Sobic.006G067700. However, SIFT27 analysis predicted that those polymorphisms would be tolerated and not disrupt function (Table 3). As expected, the haplotypes of the two dw2 recessive RIL population parents, BTx642 and IS3620c, were the same as the progenitor lines DDYM and BTx406. The haplotypes of the two Dw2 dominant RIL population parents, BTx623 and Tx7000 were the same as the progenitor line Texas Blackhull Kafir (Table 4). Six other genotypes identified by Quinby and Karper7 as Dw2 also encoded full length proteins (Table 4). Hegari was reported to contain a recessive dw2 allele1, 7. However, subsequent analysis showed that Hegari has the dwarfing genotype Dw1, Dw2, dw07_55.1, Dw3 16. Consistent with the revised genotype, Hegari encoded a full length Dw2 protein and lacked the DDYM indel in Sobic.006G067700 that disrupts protein function (Table 4).

Table 3 Polymorphisms in Sobic.006G067700.
Full size table
Table 4 Selected Genotypes Scored at the Polymorphisms Listed in Table 3.
Full size table

Dw2 is homologous to the AGCVIII protein kinase KIPK

Fine mapping, sequence analysis, and gene annotation indicates that Dw2 is a protein kinase encoded by Sobic.006G067700 (Phytozome). Genes in other plants with the greatest sequence similarity to Sobic.006G067700 include LOC_Os12g29580 (rice), GRMZM2G412524 (maize), GRMZM2G128319 (maize), and At3G52890 (Arabidopsis) (Phytozome). At3G52890 encodes an ACGVIII kinase called KIPK, a KCBP-interacting protein kinase28. In Arabidopsis there are 23 members of the AGCVIII kinase subfamily that has been further subdivided into four groups, AGC1–AGC4. KIPK and D6 PROTEIN KINASE/D6 PROTEIN KINASE LIKEs (D6PK/D6PKLs) are members of the AGC1 group. A BLAST search of the Arabidopsis AGCVIII kinase gene family to the sorghum genome identified 21 sorghum homologs (Supplementary Table S5). Among these genes, Dw2 was the best BLAST hit for the Arabidopsis KIPK1, KIPK2 (AGC1–9) and AGC1-8. KIPK1 and KIPK2 also aligned well with a related gene in sorghum, Sobic.008G096200. Since the correspondence between AtKIPK1, AtKIPK2 and the two sorghum homologs could not be assigned, we designated Dw2 as SbKIPK and Sobic.008G096200 as SbKIPK-like. The relationship among the 21 members of the sorghum AGCVIII subfamily was analysed by constructing a phylogenetic tree (Fig. 4). The sorghum genes clustered into four groups, as in Arabidopsis, though the closest sorghum homolog to AGC1-12 (Sobic.005G036500) groups with the AGC3s. If this gene is excluded from the sorghum AGC1 subfamily, then sorghum has three fewer members of the AGC1 group than Arabidopsis. Interestingly, while similar phylogenetic trees of the Arabidopsis AGC1 subfamily showed KIPK1 and KIPK2 grouping with AGC1-829,30,31, the sorghum AGC1 family has only two genes on that branch, Sobic.006G067700 (Dw2, SbKIPK) and Sobic.008G096200 (SbKIPK-like) (Fig. 4). The sorghum AGC1 group also includes a cluster of four sorghum genes that correspond to the four Arabidopsis genes that encode D6PK/D6PKLs. The sorghum AGC3 group has five members, including the AGC1-12 homolog, with two genes matching with PID and one gene corresponding with the WAGs. The remaining two groups of sorghum genes corresponding to AGC2 and AGC4 are similar to Arabidopsis (Fig. 4 & Supplementary Table S5).

Figure 4
figure 4

Phylogenetic tree of the AGCVIII subfamily in sorghum. The tree of the 21 sorghum AGCVIII genes was generated in MEGA6 using Maximum Likelihood. Dw2 is bolded. The four different groups, AGC1-4, are labeled and colored. The names in parenthesis are the best hit from a BLAST search of the Arabidopsis genome using that sorghum gene as a query. *The best hit for Sobic.008G170500 is PHOT2 but the score is much lower than Sobic.007G105500 to PHOT2 (203.4 vs. 1122.1 for the Dual Affine Smith Waterman alignment score). Further, Sobic.008G170500 is the best BLAST match of the maize PID homolog, BARREN INFLORESCENCE2, in sorghum.

Full size image

Plant AGC kinases contain a catalytic core consisting of 12 conserved subdomains32. A comparison of Dw2 (Sobic.006G067700) with KIPK and other members of the AGC1-kinase group showed that Dw2 contains a conserved GxGxxG sequence in the P-loop of sub-domain I of the N-lobe, an activation segment in the C-lobe that includes the Mg++ binding sequence DFDLS, an insertion domain typical of plant AGC-kinases, and a T-loop and activation domain [SxxSFVGTxYxAPE] that is a site of phosphorylation32 (labelled in Supplementary Fig. S5). The protein has a C-terminal FxxF sequence found in many AGC-kinases that binds 3-phosphoinositide-dependent kinase 1 (PDK1), a highly conserved member of the AGC kinase family that phosphorylates several AGC kinases30. AGC kinases vary significantly in the length, sequence, and function of their N-terminal domains that often mediate interaction with other proteins. KIPK1 and 2 and AGC1-8 have N-terminal domains of 546–549 amino acids, significantly larger than other members of the AGC1 kinase subfamily32. When the N-terminal 423 amino acid domain of Dw2 was used to search for matches in the Arabidopsis genome (Phytozome), it aligned best with the N-terminal domain of KIPK and next best to KIPK2 (AGC1-9). Multiple sequence alignment of Dw2, rice and maize homologs of Dw2, and Arabidopsis KIPK showed regions of sequence similarity throughout the N-terminal domain and several deletions relative to Arabidopsis KIPK that explain the difference in overall length of the N-terminal protein sequences (423 versus 545 amino acids) (Supplementary Fig. S5).

Expression of Dw2

Dw2 RNA abundance was examined in tissues of BTx623 (Dw2) by analysis of RNAseq profiles that are part of the sorghum RNA Atlas (Phytozome). Dw2 is annotated as having two transcripts that differ in the 5′UTR. The primary transcript (Sobic.006G067700.2) has a UTR with no introns that extends 537 bp before the start codon, while the secondary transcript (Sobic.006G067700.1) has one intron and extends 923 bp. The analysis of Dw2 expression shown in Fig. 5 utilized tissues collected from plants at ~10 days post-floral initiation, when upper leaves, leaf sheaths, internodes, nascent panicles and peduncles are growing. The expression of Dw2 was relatively high in developing panicles, peduncles, growing internodes and leaf sheaths, with lower expression in fully expanded internodes, leaf blades and the lower portion of the root system that includes root tips and fully elongated roots (Fig. 5 & Supplementary Fig. S6). The expression of sorghum KIPK-like (Sobic.008G096200) was higher than Dw2 in roots and lower in leaf tissues, the peduncle, and panicle (Fig. 5).

Figure 5
figure 5

Expression of Dw2 and Sobic.008G096200 in various tissues. Gene expression data is from the publicly available RNA-seq GeneAtlas on Phytozome v11. Tissues are from BTx623 (dominant Dw2) at 44 Days after Emergence (DAE). The leaf tissue was taken from the last ligulated leaf, so the base is still growing whereas the tip is maturing.

Full size image

Discussion

In this study, Dw2, an important dwarfing locus used in grain sorghum breeding, was mapped as a QTL in two populations. Using map-based cloning, the gene corresponding to Dw2 was identified as a protein kinase whose closest homolog in Arabidopsis is KIPK, a member of the AGCVIII protein kinase family.

Dw2 QTL analysis and fine mapping were performed using two different RIL populations. In the first population derived from BTx623 x IS3620c, alleles of the dwarfing loci Dw2 and Dw3 were segregating. Analysis of average internode length identified a QTL aligned with Dw2 at ~42.7 Mb on chromosome 6 and a QTL corresponding to Dw3 located on chromosome 7 at ~59.8 Mb. Dw2 was the only dwarfing (Dw) locus segregating in the second population derived from BTx642 x Tx7000, genotypes recessive for dw1dw3dw4. Indeed, the only QTL segregating for total height in this population was a QTL corresponding to Dw2 (~43.2 Mb). The location of the Dw2 QTL mapped in this study corresponds to most previous reports of the location of Dw2 19, 21. Higgins et al.22 also identified QTL for plant height in this region of chromosome 6 with peaks at 44.3–44.5 Mb or 42.1 Mb depending on the population and QTL model. The authors suggested that variation in QTL location was due to the linkage between Dw2 and Ma1 since both influence plant height22. In the current study, the influence of Ma1 alleles is minimal because the BTx642 x Tx7000 RIL population is segregating for a weak allele and null allele of Ma1, respectively33, and BTx623 and IS3620c each contain null alleles of Ma1 18, 19. During the analysis of Dw2 a nearby QTL located at 48.6 Mb on chromosome 6 was identified that modified the length of internode 6 according to single QTL mapping. MQM revealed that this QTL also affected the length of internodes 4–7; however, Dw2 had a greater impact on the length of the fourth and fifth internode. This additional QTL could also have confounded the location of Dw2 in the study of Higgins et al.22.

QTL analysis in the BTx623 x IS3620c population showed that Dw2 and Dw3 influence internode length differentially during development. Dw2 had the greatest additive effect on the length of the internode immediately below the peduncle. The additive effects of Dw2 and Dw3 on the length of this internode were similar. The influence of Dw2 gradually decreased in the internodes below the top internode and there was no detectable impact of Dw2 on the length of internodes 7–8 below the peduncle in this population. Dw3 had a much greater effect than Dw2 on internodes 2–5 below the peduncle with reduced but significant impact on the length of internodes 6–8 (Table 1, Fig. 2). Similarly, in maize, Br2, the homolog of Dw3 that encodes an ABCB1 auxin transporter, had a greater influence on elongation of the lower stem internodes compared to upper internodes that elongate post-floral initiation12. RILs from the BTx642 x Tx7000 population that are null for Dw3 and differ in Dw2 alleles showed a large difference in length of the internode below the peduncle when grown in low light in the greenhouse during the winter (Supplementary Fig. S4). A comparison of the yellow milos (DYM: dw1 Dw2 Dw3dw4 and DDYM: dw1 dw2 Dw3dw4) showed that Dw2 has an effect on the length of all of the ~12 internodes produced by plants grown in the greenhouse during the fall under short day conditions (Supplementary Fig. S1). Taken together these results indicate that Dw2 affects the length of internodes produced by plants during the vegetative phase and the last 6–7 internodes produced after floral initiation. The relative impact of Dw2 on the length of sorghum stem internodes depends on genetic background, stage of development, and environmental conditions.

Fine mapping narrowed the region encoding Dw2 to a ~98.1 kb region of chromosome 6 containing ten genes. Fine mapping was more effective in the BTx642 x Tx7000 RIL population compared to the BTx623 x IS3620c RIL population even though the population size was smaller. Map-based cloning of Ma1 (PRR37), a gene located near Dw2 on chromosome 6 (40.3 Mb), was also more effective in one of the three populations used in that study18. The reason for differences in recombination efficiency among populations used to map Ma1 and Dw2 is unknown. However, it is possible that sequence divergence or differences in chromatin/DNA methylation in this region of chromosome 6 between parental genotypes affects local recombination frequency. In both studies, the most useful population for fine mapping involved parental lines of the durra race that were crossed to parental lines of the kafir race (BTx642 (durra) xTx7000 (kafir) (Dw2); 100 M (durra) x Blackhull kafir (Ma1)). Larger populations and additional crosses will be required to determine if the association between genetic background and local recombination frequency is significant.

One of the ten genes in the delimited Dw2 locus encoded a histone deacetylase that was previously suggested to be a candidate for Dw2 21. However, the deacetylase did not contain polymorphisms in the coding regions that distinguish the parental genotypes used for fine mapping, or DYM (Dw2) and DDYM (dw2). DDYM was reported to have originated as a shorter mutant in a field of DYM25. Thus, these two yellow milos should be isogenic except at Dw2. All of the other genes in the delimited Dw2 region were sequenced from DYM and DDYM. Only the kinase encoded by Sobic.006G067700 had a polymorphism that distinguished DDYM from DYM in the delimited Dw2 locus. This polymorphism resulted in a frameshift mutation and a premature stop codon in the first exon. This results in a protein of only 190 amino acids instead of 809 amino acids found in DYM. The kinase domain is located between 424–763 amino acids; therefore, the mutant protein found in DDYM would lack kinase activity.

The closest homolog of sorghum Dw2 in Arabidopsis is KIPK, a member of the AGC family of kinases. The AGC family is named after the cAMP dependent protein kinases, cGMP dependent protein kinases, and protein kinase C and also includes PDK1 and the ribosomal protein S6 kinases. The plant-specific AGCVIII subfamily includes PID, PHOT1 and 2, and the D6PK/D6PKLs34. Each of these kinases has been shown to regulate auxin efflux transporters, including ABCB1 and PIN1, with PHOT1 and 2 doing so in a blue-light dependent manner32, 35. In Arabidopsis, KIPK has a close homolog, KIPK2 (also known as AGC1-9 and At2g36350) and the closely related kinase, AGC1-830, 36. In sorghum, Dw2 has one closely related homolog, Sobic.008G096200, and these two genes form their own branch on the phylogenetic tree (Fig. 4). As some of the members of the AGCVIII subfamily have been shown to regulate auxin transport, Dw3, the sorghum homolog of Arabidopsis ABCB1, was initially considered a potential target of Dw2 action. However, while Dw2 was expressed in growing internodes, MQM analysis provides no genetic evidence for interaction between Dw2 and Dw3. Furthermore, the Dw2 allele positively affects the length of the upper most internode in a dw3 background, indicating that Dw2 can act at least partially through pathways independent of Dw3.

In Arabidopsis KIPK was so named due to its interaction with KCBP, a plant-specific kinesin-like calmodulin binding protein that functions in cell division and trichome formation28. KCBP has a C-terminal motor and calmodulin-binding domain, and is unusual among kinesins in its ability to interact with microtubules and with actin, the latter interaction mediated by a MyTH4-FERM tandem that occurs in myosin37. Type-VI kinesin-14 dimers in Physcomitrella patens, homologs of KCBP, are highly processive, and transport vesicles/cargo long distances when clustered38. KCBP contains a calmodulin binding domain and is down-regulated by calcium via calmodulin as well as the KCBP interacting Ca2+-binding protein (KIC)39, 40. While KIPK did not phosphorylate the N-terminal end of KCBP under experimental conditions, it is possible that it phosphorylates KCBP under other conditions, and it is possible that KCBP transports KIPK within the cell28.

Subsequent work has also shown that Arabidopsis KIPK1 and 2 directly interact with members of the proline-rich extensin-like receptor-like kinase (PERK) family, specifically PERK8, 9, 10, and 1336. Other PERK-genes, such as PERK1, mediate growth inhibition, possibly in response to cell wall signals41. In Arabidopsis, KIPK1 and 2 double mutants did not produce shoot phenotypes although there were differences in root elongation when plants were grown on elevated sucrose36. Different parts of the N-terminal domain of KIPK1 and 2 mediate the direct interactions with KCBP and the various PERKs36. The 423 amino acid N-terminal sequence of Dw2 aligned well with the ~545 amino acid N-terminus of KIPK despite several deletions that account for the difference in overall length of this domain. The sequence similarity of the N-terminal domains of KIPK and Dw2 indicates that Dw2 has likely retained the ability to interact with one or more members of the PERK family. The best BLAST hits to Arabidopsis PERK8 and 10 (At5g38560 and At1g26150, respectively) in sorghum (Sobic.003G100700, Sobic.003G289800, and Sobic.009G000300) were expressed in stem internodes (Phytozome). Therefore, it will be of interest to determine if Dw2 interacts with sorghum PERK8 or 10 homologs.

If Dw2, like Arabidopsis KIPK, interacts with PERKs and KCBP, the interactions with these proteins may modulate growth regulation and serve other regulatory functions. For example, because KCBP transports vesicles/cargo long distances38 potential Dw2 interactions with PERKs and KCBP in sorghum could regulate growth and the flow of materials to the cell wall during and after organ elongation. Alternatively, in trichomes KCBP has been found to organize cytoskeleton components37, thus KIPK may be involved cytoskeletal regulation that is associated with cell elongation. This more general coordinating function may explain why Dw2 is expressed in growing zones of leaf blades, leaf sheaths, stems, and panicles. Lack of growth phenotypes in all organs where Dw2 is expressed (i.e., peduncle) could be due to the presence of a second KIPK-like gene in sorghum (Sobic.008G096200). In fact, Sobic.008G096200 is more highly expressed than Dw2 in the roots, and both genes are highly expressed in the panicle, peduncle, and internodes (Fig. 5). One other possibility could be that KIPK is involved in a PERK signalling pathway. Another member of the PERK family, PERK4, has been shown to regulate cell elongation in roots as part of an abscisic acid (ABA) signaling pathway42.

While Dw2 is a homolog of Arabidopsis KIPK, Dw2 has an important role in regulating stem length in sorghum, a function not observed in Arabidopsis KIPK mutants36. This may be because grass stem growth occurs by sequentially elongating internodes adjacent to intercalary meristems located just above nodes, a mode of stem growth that is unique to grasses. The first sorghum dwarfing locus cloned, Dw3, also had a more severe stem phenotype than mutants affecting the Arabidopsis homolog, ABCB1. Multani et al.12 showed that mutation of Dw3, an auxin efflux carrier, results in short internodes in sorghum whereas the corresponding ABCB1 single mutant in Arabidopsis had little effect on stem length43. Knoller et al.44 showed that brachytic2, the maize homolog of sorghum Dw3, is expressed in stem nodes but not in stem internodes, whereas Arabidopsis lacks intercalary meristems. This difference in physiology between Arabidopsis and the grasses helps explain the differences in ABCB1 mutant phenotypes. It may also explain the differences in phenotypes between the Dw2 and KIPK mutants in sorghum and Arabidopsis, respectively. Alternatively, the difference in phenotype could be due to differences in functional redundancy and/or expression within the AGCVIII subfamily.

Dw2 has been used extensively in grain sorghum breeding in the U.S. to create lines and hybrids with reduced stem length. A recessive allele of dw2 derived from DDYM was used in the Sorghum Conversion Program to reduce the height of lines that were being converted for use in temperate grain sorghum breeding programs19. Dw2 is linked to Ma1 (PRR37), another important gene in grain sorghum and energy sorghum development18. In addition to its historical significance, a better understanding of Dw2 function may enable the design of improved sorghum crops.

Methods

Phenotypic Analysis of DYM and DDYM Stems

The progenitor genotypes Dwarf Yellow Milo (DYM; Dw2) and Double Dwarf Yellow Milo (DDYM; dw2) were grown to examine the internode length phenotypes caused by the two Dw2 alleles. For each genotype, three plants were individually grown in 3.8-gallon pots (Custom2000) containing Metro Mix 900 (Sun Gro Horticulture) with supplemental fertilizer (Peters 20-20-20) in the greenhouse in short days during the fall. At anthesis, the plants were harvested and the total stem length and length of each internode were measured.

QTL Mapping of Dw2 in a RIL population derived from BTx623 and IS3620c

The BTx623 x IS3620c RIL population was used for mapping Dw2 45. Seed for the population was obtained from the USDA-ARS Plant Genetic Resources Conservation Unit (Griffin, GA). BTx623 is dw1Dw2dw3dw4 and IS3620c is dw1dw2Dw3dw4; therefore, the population segregated for both Dw2 and Dw3. The population (n = 380) was grown in the greenhouse in the summer of 2013 with natural day lengths. Three plants of each RIL were grown per pot, one pot per line in the same manner as DYM and DDYM. Plants were harvested at grain maturity. For each plant, the total length of the plant (base of the plant to the base of the panicle) and the length of each internode and peduncle were measured. Internodes were numbered from the peduncle. Plants differed for flowering time, with earlier flowering lines producing fewer elongated internodes. As a consequence, the 6th, 7th, and 8th internodes below the peduncle had smaller sample sizes than the other traits measured (n = 375, n = 356 and n = 296, respectively). Genotyping and genetic map construction (n = 398) were performed as described in Truong et al.46 except the DG marker sequences were mapped to version 3 of the sorghum reference genome assembly (Sorghum bicolor v3.1 DOE-JGI, http://phytozome.jgi.doe.gov/), using BWA47, and INDEL realignment and joint variant calling were performed with the GATK using the naive pipeline of the RIG workflow48,49,50,51. QTL mapping was performed in R/qtl using interval mapping (IM) with 1000 permutations and an α = 0.0552. Both the genetic map and QTL mapping were performed as an F7 instead of a RIL population due to excess heterozygosity.

MQM was performed using the same phenotypes, except peduncle length, and genotypes that were used for IM, except the genetic map was thinned to obtain a marker set with at least 1 cM spacing between markers. Also, measurements of the length of each internode, average internode length, and total internode length were normalized using Empirical Quantile Normal Transformation prior to QTL mapping with R/qtl52,53,54. Penalties (main effect, heavy interaction, and light interaction) for all normalized phenotypes were calculated from 25,000 permutations of two-dimensional genome scans using the TIGGS-HPC cluster at Texas A&M; penalties calculated were negligibly different between phenotypes (i.e. same to the tenths place). Significant QTL identified from an initial IM analysis (alpha = 0.05, main effect LOD = 3.2) were used to seed multiple-QTL model selection analysis (maximum number of QTL in a model was restricted to 7; main effect LOD = 3.2, heavy interaction LOD = 4.3, light interaction LOD = 1.9)52, 54. The best scoring multiple-QTL model from model selection of each phenotype was then merged into a composite multiple-QTL model. The composite multiple-QTL model was generated by merging all overlapping 2-LOD intervals into one QTL and designating the position of the MLOD (maximum LOD) marker as the QTL position55, where loci with an epistatic interaction were merged independently of strictly additive loci.

QTL Mapping of Dw2 in a RIL population derived from BTx642 and Tx7000

BTx642 is dw1dw2dw3dw4 and Tx7000 is dw1Dw2dw3dw4; therefore, the population derived from a cross of these genotypes will segregate for alleles of Dw2. The BTx642 x Tx7000 RIL population (n = 89) was grown in the field in the spring and summer of 2009. It was planted in a Norwood silty clay loam (fine-silty, mixed (calcareous), thermic Typic Udifluvent) in duplicate in a randomized block design at the Texas A&M Research Farm located near Snook, TX on 03/04/2009. The blocks were arrayed in 20 rows 4.6 m long and spaced 76 cm apart with two buffer rows on each end of the block. Each block was offset from the next by approximately 1.5 m. The plants emerged on 08/04/2009 and were thinned to a within-row spacing of 10 cm at 16 days after emergence (DAE). The average daily maximum temperature was 33.3 °C and the average daily minimum temperature was 21.1 °C. The population received 24.9 cm of natural rainfall during the growing season with supplemental flood irrigation as needed. The population was harvested on 23/06/2009 (76 DAE), approximately at anthesis for the population. Three plants of each RIL and parental lines from each of two replicates were harvested. For QTL mapping, the average of the two replications was used. Plants were phenotyped for total height, which was measured from the base of the plant to the top of the panicle.

DNA was extracted from leaf tissue harvested from each RIL and processed using ZR Plant/Seed DNA MiniPrep (Zymo Research). Digital Genotyping (DG) was performed as previously described20 using the enzyme NgoMIV to digest genomic DNA. Reads were mapped to the reference genome and variants were processed as described for the BTx623 x IS3620c RIL population. The genetic map was constructed using R/qtl (n = 93) after removing any markers that did not define a recombination breakpoint. QTL mapping was also performed in R/qtl using IM with 1000 permutations and an α = 0.0552.

Fine Mapping of Dw2

The BTx642 x Tx7000 RIL population was used for fine mapping Dw2. Lines that had recombination breakpoints in or near Dw2 were used to delimit the locus to the extent possible using additional DG genotypes and SNPs identified by Sanger sequencing genes in the region. Primers used for Sanger sequencing are listed in Supplementary Table S6. All PCR amplification was done with Phusion© High-Fidelity DNA Polymerase (New England BioLabs, Inc.) using the standard conditions. The PCR product was gel purified using QIAquick Gel Extraction Kit (Qiagen) and prepared for capillary sequencing with BigDye© Terminator v3.1 Cycle Sequencing Kit (Applied Biosystems) using standard reaction conditions. Sequencing was performed with the ABI 3130xl Genetic Analyzer (Applied Biosystems) and the results were analyzed with Sequencher v4.8 (Gene Codes Corp.).

RILs with recombination breakpoints in the delimited Dw2 region were grown to confirm stem and internode length phenotypes. Two pots containing two plants from each RIL were grown in two different greenhouses for a total of eight plants per RIL; otherwise the RILs were grown in the same manner as DYM and DDYM. At anthesis, the plants were harvested and the total length of the stem (measured from the base of the plant to the base of the panicle) and the length of each internode were recorded.

Sequencing of Genes in the Genomic Region Spanning Dw2

Once the region encoding Dw2 was delimited to the extent possible with available genetic resources, the genes in this region were sequenced to search for functional mutations that distinguish DYM (Dw2) from DDYM (dw2). The genes in the Dw2 locus were identified using the sorghum reference genome version 3.1 gene set (Sorghum bicolor v3.1 DOE-JGI, http://phytozome.jgi.doe.gov/). The primers for sequencing the genes are listed in Supplementary Table S7 and capillary sequencing was performed as with fine mapping SNPs. DDYM was identified as a short plant in a field of DYM and alleles of Dw2 differentiate the two genotypes25. For Sobic.006G067600 only the exons were sequenced, for all other genes, the entire gene was sequenced. Sobic.006G067700 was further sequenced in the other important breeding lines to examine the distribution and extent of allelic variation in Dw2.

Whole Genome Sequencing

Whole genome sequencing was used to identify polymorphisms that distinguish the parents of the two populations used to map Dw2. Tx7000 and BTx642 seeds were obtained from Dr. W.L. Rooney (Dept of Soil and Crop Sciences, TAMU). IS3620c seed (PI 659986 MAP) was obtained from the USDA-ARS Plant Genetic Resources Conservation Unit (Griffin, GA). Seeds were soaked in 20% bleach for 20 minutes and washed extensively in distilled water for one hour. Seeds were germinated on water-saturated germination paper in a growth chamber (14 hr light; 30 °C/10 hr dark; 24 °C). Genomic DNA was isolated from 8-day old root tissue using a FastPrep DNA Extraction kit and FastPrep24 Instrument (MP Biomedicals LLC, Solon, OH, USA), according to the manufacturer’s specifications. DNA template (350 bp average insert size) was prepared using a TruSeq® DNA PCR-Free LT Kit, according to the manufacturer’s directions. Paired-end sequencing (125 × 125 bases) was performed on an Illumina HiSeq2500. Sequence reads were mapped to version 3 of the sorghum reference genome assembly (Sorghum bicolor v3.1 DOE-JGI, http://phytozome.jgi.doe.gov/), using BWA v0.7.1247. Base quality score recalibration, INDEL realignment, duplicate removal, joint variant calling, and variant quality score recalibration were performed using GATK v3.3 with the RIG workflow48,49,50,51. Whole genome sequence of Tx7000, BTx6424, and IS3620c are available at the Sequence Read Archive (www.ncbi.nlm.nih.gov/sra).

Protein Sequence Analysis

Each of the AGCVIII proteins in Arabidopsis was aligned with the sorghum genome using BLAST and the best hits were recorded. The resulting sorghum AGCVIII protein family was used to make a phylogenetic tree in MEGA656. The sequences were aligned using the MUSCLE algorithm57, 58. The tree was estimated using maximum likelihood with the substitution model developed by Le & Gascuel59 and the Gamma distribution. To estimate the reliability of the branches, 1000 boostraps were performed. Protein alignments were performed in Jalview v2.060 using the TCoffee algorithm61 with defaults.