SNP Quality control, LD decay and frequency distribution of phenotypic data
Out of the 21,113 SNPs imputed for the 173 genotypes, 8,507 quality SNPs were retained across 167 genotypes for downstream analysis after filtering and removal of SNPs from unmapped contigs. The distribution of the 8,507 SNPs across the sorghum genome is shown in Fig.1a. The overall SNP density was ~12 SNPs/Mbp. Results also showed that LD decayed at 48.1 Kb (Fig. 1b). Regarding trait data, both traits germination frequency (GF) and radicle length (RL) revealed quantitative distribution patterns that were either kurtosis (GF) or slightly skewed to the right (RL) (Fig. 1c).
Sorghum accessions have distinct race structures
To determine appropriate models for GWAS, we carried out population structure analysis of the sorghum accessions using three criterions: (i) Bayesian information criterion (BIC), (ii) dissimilarity matrix-based neighbor joining (NJ) tree and (iii) hierarchical Bayesian clustering (HBC)
Bayesian Information Criterion (BIC) identified 8 clusters (Fig. 2a), indicated by an elbow curve of BIC values as a function of K (Suppl. Fig. S2). Clusters 1 and 8 were clearly differentiated from the rest of the clusters. Cluster 8 consisted of Guinea margaritiferum (Gma) and Guinea. Other clusters were cluster 2, Kafir, cluster 4 Guinea gambicum (Gga), cluster 5 Durra and Durra-caudatum DC), 6 and cluster 8, wild.
Unlike BIC that grouped the populations to 8, the NJ tree assigned 7 groups according to the major sorghum races (Fig. 2b). There were distinct groups of Kafir, Guinea, Gga, Gma and wild races. Other clusters comprised races combined with sub-races. These were: Caudatum and DC, Durra, and DC. Bicolor could not be placed on any cluster, instead, it appeared to cluster within most groups but more so with the Durra, DC cluster. Noteworthy, Gma formed the most divergent sub-group appearing to form a long branch of its own. Lastly, Gga, wild and Guinea accessions – which also included the sub-races – formed respective clades. This clustering supported observations made using DAPC.
Hierarchical Bayesian clustering (HBC) using ADMIXTURE grouped the sorghum accessions into 7 clusters using the K value with the lowest cross validation (CV) error (K = 7; CV=0.35293) as shown in Fig. 2c. Further ADMIXTURE clustering using K8 (CV=0.35673) revealed 6 clusters of Durra, Caudatum/Guinea Caudatum, Kafir, Gma and Gga while clustering with K9 (CV=0.36229) showed 5 clusters of Kafir, Durra Caudatum, Guinea-caudtum (GC), Gga and Gma. Combined ADMIXTURE plots are shown in Suppl. Fig. S3.
Overall, groups of NJ, DAPC and ADMIXTURE were, in agreement and provided clear evidence of population structuring based on sorghum races. Therefore, subsequent GWAS analysis were carried out with the confounding genetic structuring considerations.
Marker-trait associations and candidate genes
While there were no significant SNPs associated with RL from both analysis models, FarmCPU and MLMM identified 9 and 1 SNPs respectively that were significantly associated with GF (Table 1 and Fig. 3). Among the significant SNP markers identified, 5 were localized within candidate genes including: (i) ADP-glucose phosphatase (S2_4487093; p = 0.00011), (ii) Membrane-associated kinase regulator 3 (S9_1847809; p = 0.00082), (iii) Marspadin like α/β hydrolase (S1_47702467; p = 0.002290, (iv) Abscisic acid-insensitive 5(ABI5) (S3_68089192; p value = 0.00229) and, (v) a disulphide oxidoreductase (S4_4060034; p = 0.00229). A closer look at all the significantly associated SNPs revealed a link to hormone signaling and ABA biosynthesis in the following genes: ABI5, NINJA, RPK1, Maspardin and ADP-glucose phosphatase. These results showed a prevalence of hormone and dormancy regulating factors.
Haplotype blocks suggest further hormone signaling activity during Striga seed germination
Following this analysis, we found two markers in significant LD with ABI5 (Fig. 4). One of them was at position S3_68025304 and located in the coding region of a gene that encode ethylene response factor 2 (ERF2) on chromosome 3. The other SNP that was in significant LD with ABI5 was located at position (S3_68035093 and S3_68025207) in the exon of the Arabidopsis’ ortholog TRANSPARENT TESTA GLABRA1 (TTG1). Two copies of the gene ADP-Glucose-phosphatase were found within the same haplotype block.
The finding that AP2/ERF2 and TT1 are in LD with ABI5 suggest co-inheritance, a further indication of a hormonal crosstalk in the process of Striga germination. And an additional copy of ADP-Glucose phosphatase underscores the importance of the gene in the process of germination.
Germination of parasitic plants seeds (Striga and Alectra) is influenced by ABA, GA in addition to SL
Intrigued by the prevalence of signaling associated SNPs in our GWAS and LD analysis, we sought to test the hypothesis of a possible hormonal crosstalk using Striga germination assays. In the experiments, we determined the effect of conditioning of Striga seeds with the phytohormones ABA and GA. We further investigated the effect on Striga seed germination when ABA was inhibited using fluridone.
To our surprise we discovered that Striga seeds treated with FLU at conditioning, germinated before further treatment with any Striga germination stimulants. Striga seed germination occurred remarkably early – at 4 days after incubation appearing to bypass the requirement for condition. The germination frequency increased in subsequent days and peaked on the 10th day (Fig. 5a and b). Similarly, the mean radicle length of germinated Striga seedlings increased and was longest on the 8th day (Fig. 5a and b). No germination was observed in seeds treated with ABA, ABA + FLU or water. To further determine if any of the hormone treatments had interactions with SL, Striga seeds conditioned for 14 days with ABA, GA, or ABA+FLU were treated with root exudates of a high germination inducer accession (IS2730) as well as that of a low germination inducer accession (IS27146) and compared to induction of Striga germination using the synthetic SL GR24. Results (Fig. 5c) showed that ABA significantly inhibited germination of Striga seeds in all cases. Combination of ABA and FLU marginally increased the Striga germination percentages in root exudate treatments but more notably in the GR24 experiment. In treatments conditioned with FLU, Striga germination was significantly increased in all cases (root exudates of sorghum accessions and GR24). Intriguingly, treatments conditioned with GA indicated notably lower Striga seed germination frequency significantly in root exudates from IS27146 and synthetic SL, GR24.
To corroborate and affirm the findings of hormonal crosstalk in other parasitic plants, we performed germination assays using seeds of Alectra vogelii. For these experiments, we conditioned seeds of Alectra with water, ABA, FLU and GA and determined their germination frequency (GF) as well as radicle length (RL). Remarkably, Alectra seeds conditioned with FLU germinated after 5 days independent of SL induction (Fig. 6 and Table 2). No germination occurred in ABA or GA treated seeds (Table 2). Furthermore, upon treating conditioned seeds (ABA, FLU, and GA), we observed inhibition of germination and radicle growth by ABA (GF = 39.42 ± 1.10 %; RL = 0.57±0.12 mm) as well as GA (GF = 69.13 ± 1.15 %; RL = 0.30 ± 0.12 mm), in the same degree as in the Striga germination experiments compared to water conditioned seeds (GF = 74.40 ± 3.28 %; RL = 0.91 ± 0.12 mm). In the case of GA, notably small radicles were observed (Fig. 6b). This appeared to indicate that GA inhibited radicle growth of Alectra seedlings.
Taken together, these findings provide unequivocal evidence that: (i) ABA causes inhibition of Striga and Alectra seed germination and that obliteration of ABA activity is enough to trigger germination of the parasitic plants seeds in a SL independent manner, (ii) GA interacts with SL to negatively regulate Striga and Alectra seed germination.