SUMMARY
Drought is a major constraint on plant productivity globally. Sorghum (Sorghum bicolor) landraces have evolved in drought-prone regions, but the genetics of their adaptation is not yet understood. Loci underlying stay-green post-flowering drought tolerance (Stg), have been identified in a temperate breeding line, but their role in drought adaptation of tropical sorghum is to be elucidated.
We phenotyped 590 diverse sorghum accessions from West Africa under field-based managed drought stress, pre-flowering (WS1) and post-flowering (WS2) over several years and conducted genome-wide association studies (GWAS).
Broad-sense heritability for grain and biomass yield components was high (33-92%) across environments. There was a significant correlation between stress tolerance index (STI) for grain weight across WS1 and WS2. GWAS revealed that SbZfl1 and SbCN12, orthologs of maize flowering genes, likely underlie flowering time variation under these conditions. GWAS further identified associations (n = 134) for STI and drought effects on yield components, including 16 putative pleiotropic associations. Thirty of the associations colocalized with Stg1–4 loci and had large effects. Seven lead associations, including some within Stg1, overlapped with positive selection outliers.
Our findings reveal natural genetic variation for drought tolerance-related traits, and suggest a broad role of Stg loci in drought adaptation of sorghum.
INTRODUCTION
Unpredictable rainfall and drought are major limitations to plant productivity worldwide. Improving crop adaptation to water limitation is critical for establishing food security in developing countries where smallholder farmers are vulnerable to climate change (Mundia et al., 2019). From an agronomic perspective, drought adaptation is the ability to maintain yield under agronomic water limitation (Blum, 2010). An understanding of the genetic architecture of grain yield and its components across various drought scenarios can facilitate crop breeding to increase production. However, collecting good phenotypic data under well-managed water stress environments and integrating phenotypes with genotypes remain major constraints. The genetic dissection of yield components under various drought scenarios would provide favorable natural variants for drought tolerance.
Sorghum (Sorghum bicolor) is a staple cereal crop in drought-prone regions worldwide, including many developing countries of the semi-arid tropics as well as industrialized countries in the temperate latitudes. Sorghum is among the most drought-resilient crops, but the physiological and genetic basis of its drought tolerance is not yet understood (Mullet et al., 2014). Several quantitative trait loci (QTL) associated with drought tolerance variation in sorghum have been identified, but no genes have been cloned. The best studied of these QTL are stay-green loci (Stg1–Stg4) linked to post-flowering drought tolerance in biparental families and near-isogenic lines (Tuinstra et al., 1997; Xu et al., 2000; Harris et al., 2007; Borrell et al., 2014b; Hayes et al., 2016). The Stg loci influence several aspects of sorghum development, including canopy architecture, water use, and grain yield (Borrell et al., 2014b). Stg alleles were identified in a temperate-adapted breeding line BTx642 (formerly B35) that is derived from a tropically-adapted Ethiopian durra landrace (IS12555). However, the prevalence of the Stg alleles in sub-Saharan Africa or their role in drought adaptation (if any) is not known. Understanding the genetic basis of drought adaptation in sorghum could elucidate the process of environmental adaptation and facilitate breeding of drought-tolerant varieties.
Local varieties have been under natural and farmers selection for adaptation to environmental conditions and farming systems. Local varieties of sorghum have adapted to various environmental conditions since their domestication (Harlan & De Wet, 1972; Wendorf et al., 1992). Consequently, positive pleiotropic loci for combined pre- and post-flowering drought tolerance might exist in locally-adapted varieties. West African sorghum is extremely diverse and there have been few cycles of selection in breeding programs (Mauboussin et al., 1977; Leiser et al., 2014). The West African sorghum association panel (WASAP), including landraces and breeding lines that consist of working collections of breeding programs, was assembled and genotyped using genotyping-by-sequencing technology. However, the genetic architecture underlying grain yield and its components under various drought scenarios remains largely unknown in the germplasm. We hypothesized that positively pleiotropic QTLs confer combined pre- and post-flowering drought tolerance in the West African sorghum.
Genome-wide association studies (GWAS) contribute to the identification of natural variants, taking advantage of historical recombinations within diversity panels (Yu & Buckler, 2006; McCouch et al., 2016; Yano et al., 2016; Zhao et al., 2019). A grass species such as sorghum is suitable to identify natural variants underlying complex agronomic traits partly due to its small genome size and moderate LD (Paterson et al., 2009; Mace et al., 2013; McCormick et al., 2018). Disentangling positive pleiotropic effects of drought-yield QTLs through GWAS can contribute to detect and characterize the natural allelic variation existing within locally-adapted populations. In this study, we performed GWAS on 756 sorghum accessions of the WASAP under ten different environments using the previous GBS SNP dataset. We (i) characterize the genetic variation of yield components under various water stress environments; (ii) identify genetic variants at known and novel drought tolerance loci with high productivity under pre- and post-flowering water stress; (iii) investigate the pleiotropic effect of drought tolerance QTLs associated with STI and reduction of yield components under various drought scenarios; and (iv) determine signatures of selection overlapping with identified drought tolerance QTLs. The present study provides knowledge of the genetic architecture of yield components under various drought scenarios.
MATERIALS AND METHODS
Plant materials
The West African Sorghum Association Panel (WASAP) consists of N = 756 genotyped accessions from the four West African countries of Senegal (118 accessions), Mali (123), Togo (156), and Niger (359) (Faye et al., 2021) (Fig. 1a). The panel includes predominantly landraces along with some local breeding lines and local improved varieties. Five local breeding lines were used as checks for use in augmented design: T1 (IRAT 204/CE151-262), T2 (CE145-266), T3 (ISRA-621B/Faourou), T4 (CE180-33), and T5 (53-49). Two international drought-response reference lines, Tx7000 (pre-flowering drought tolerant, post-flowering drought susceptible) and BTx642 (pre-flowering drought susceptible, post-flowering drought tolerant), were used as controls (Burke et al., 2013; Borrell et al., 2014a).
Field trials
Field experiments were performed over four years (2014–2017) in Senegal at the Bambey Research Station, CNRA–Centre National de Recherche Agronomique (14.42°N, 16.28°W) in the Soudano-Sahelian zone (Fig. 1a). The average annual precipitation is ∼600 m, which occurs strictly in the rainy season (“hivernage”) of July to October, with maximum monthly precipitation typically occurring in August (Fig. 1b). In total, ten experiments were performed in an incomplete randomized block design (augmented block design) across the four years (Table 1; Fig. S1a-f). The experimental set-up followed a column–row field layout with 30 blocks for 2014 experiments or 25 blocks for 2015-2017 experiments, with 19 genotypes and the 5 local check varieties (present in each block for spatial variation analysis) within each block. Each entry was sown in a 3 m row with 0.6 m space between rows and 0.2 m space between plants (or hills) within a row. Each entry was surrounded by one row of fill material (IRAT 2014). Ten days after planting, plants were thinned to keep only one plant per hill, for a density of about 84,000 plants ha-1. Two experiments were carried out under rainfed conditions (RF) during the rainy season in 2014 with one-month planting date interval: RF1 (planted in August) and RF2 (planted in September). Managed drought stress experiments were conducted in the off-season to take advantage of the complete lack of precipitation during the Sahelian dry season (Fig. 1b).
Managed drought stress
Well-watered (WW) and pre-flowering water stress (WS1) experiments were planted during the hot off-season in 2015 (March to August). Three experiments, under WW, WS1, and post-flowering water stress (WS2), were planted during the cool off-season in 2015–2016 (October 2015 to March 2016; note as “2016” experiments) and 2016-2017 (October 2016 to March 2017; noted as “2017” experiments). During the rainy season of 2014, the cumulative rainfall recorded was 395 mm. The average daily temperature ranges between 22.4 and 35 °C and average relative humidity between 42 and 89%. For WW, irrigation was applied twice a week (30 mm each time) until physiological maturity. For WS1, water limitation was applied 30 DAP, to mimic a one-month pre-flowering drought, and irrigation was restarted 60 DAP until physiological maturity. For the WS2, water limitation was applied when 75% of plants in a maturity group flowered and maintained until physiological maturity. Three maturity groups were defined based on accession phenology characterized during 2014 experiments for water deficit application in WS2. The fraction of transpirable soil water (FTSW) in different managed drought stress experiments was determined using a DIVINER 2000 (Sentek Pty Ltd, Adelaide, SA).
Phenotypic measurements
In each environment, phenological, physiological, and yield component traits were measured. Days to 50% flowering (DFLo) of plants in a plot (one row), above-ground dry biomass (DBM), plant height (PH), and yield components including grain weight per panicle (GrW), panicle weight (PW), grain number per plant (GrN), and thousand-grain weight (TGrW) were measured and used for association mapping studies. For each trait except for DFLo and TGrW, three plants from the middle row of each plot were used for measurements. The drought stress tolerance index (STI) (Li et al., 2018a; Yuan et al., 2019) for grain weight was calculated from the GrW under WW and WS1 or WS2 as follows:
Where Yww and Yws is the grain weight of a given genotype in WW and WS environments, respectively, and Ym.ww is the mean value of GrW in the WW environment. For the STI, the higher the value, the more tolerant the genotype to the stress. The drought reduction of each yield component relative to the control environment was calculated as follow:
Where Ri is the drought response of a genotype for trait i, Yww and Yws are the performance of the genotype in the control environment and water-stressed environment, respectively.
Statistical analysis of phenotypes
Each year-treatment combination is considered an environment. Statistical analysis was performed using the R program (R Core Team, 2016). Spatial variation within each environment was analyzed based on the check varieties in each block using the SpATS package (Rodríguez-Álvarez et al., 2018) to obtain genotype-adjusted means. The variance components were estimated by fitting the mixed linear model with random effects for all genotypes (G), water regimes (WR), years (Y), and GxY interaction effects using the lme4 package (Bates et 2010). Broad sense heritability (H2) was calculated based on variance components derived from the mixed effect model. H2 was estimated for each trait across environments based on the genotypic variance and the total phenotypic variance. Phenotypic correlations among traits were calculated using the Pearson correlation coefficient of the PerformanceAnalytics package (Peterson et al., 2014). Tukey’s Honestly Significant Difference (TukeyHSD) test in the Agricolae package (Mendiburu, 2009) was used to test the difference of genotype performance between environments or botanical types. The BLUP values of the phenotypes were calculated by combining data for a given water regime across years or across all environments. The phenotypic BLUPs and genotype-adjusted means were used for the genome-wide association analysis across environments.
Genome-wide association studies
To identify drought-yield QTLs, GWAS was performed using the general linear model (GLM) with principal component (PC) eigenvalues and mixed linear model (MLM) in the GAPIT package (Lipka et al., 2012). These two GWAS models were used as complementary because the GLM may identify false-positive associations while MLM may lead to false-negative associations when controlling for false-positive significant associations. The SNP dataset was filtered for MAF > 0.02, which corresponds to >15 observations of the minor allele within the panel of N = 756 genotyped accessions. The first five PCs and the kinship matrix were used to account for population structure and genetic relatedness effects, respectively for the MLM. The significance level of GWAS associations were defined based on Bonferroni-corrected p-value 0.05 for the GLM with PC (referred to as GLM+Q hereafter) or at least top five SNPs above p < 10-5 cutoff for the MLM. The most highly-associated SNP (“lead SNP”) within a 150 kb genomic region defined based on average linkage disequilibrium (LD) decay in global sorghum germplasm (Morris et al., 2013) was chosen to represent the association. A list of a priori candidate genes of cloned cereal flowering times from a previous study (Faye et al., 2019) was used for colocalization analysis between lead SNP and candidate genes.
Locus-specific analyses
LD heatmaps were constructed using the R package LD heatmap 0.99-4 (Shin et al., 2006). BLUP values of phenotypes across water stress environments were used for the estimation of the proportion of phenotypic variance explained (PVE) by lead SNPs from the GWAS. The PVE was estimated using linear models with fractions of ancestry inferred by ADMIXTURE (Alexander et al., 2009) used as fixed covariates. Statistical enrichment analysis for colocalization between GWAS lead SNPs and all Stg QTLs from the sorghum QTL Atlas (Mace et al., 2019) was performed based on 1000 permutation tests. Statistical significance was assessed with a two-sample t-test with α = 0.05. Geographic distribution of the associated lead SNP alleles with DFLo or putative drought tolerance was determined using an existing set of georeferenced global sorghum landraces (Lasky et al., 2015). Lead associations within Stg1-3 QTLs were selected based on their association with drought tolerance variables, LD with other lead associations within a locus, contribution to the phenotypic variation, and availability in the GBS data for global sorghum landraces.
Genome-wide selection scans
For selection scans, we included 550 worldwide sorghum accessions including wild relative sorghum accessions with available sequencing data (Morris et al., 2013). Genome-wide selection scans were performed based on 100 kb sliding windows using the vcftools program (Danecek et al., 2011). Decreased genome-wide nucleotide diversity (π) in durra-caudatum, durra, and guinea cultivars relative to wild relatives was performed to assess domestication and diversification selections for drought responses to dry (in durra-caudatum and durra genome) versus humid (in guinea genome) regions. Statistical enrichment analysis for colocalization between π outlier regions and Stg1–4 loci was performed based on 1000 permutation tests. Statistical significance of mean differences were based on two-sided two-sample t-tests with α = 0.05.
RESULTS
Phenotypic variation for drought tolerance related traits
A total of 590 WASAP accessions were evaluated for phenological, physiological, and yield component traits under ten environments across four years in Senegal (Fig. 1a,b; Fig. S1). To assess the level of drought stress applied, we estimated the fraction of transpirable soil water (FTSW) in the WW, WS1, and WS2 (Fig. 1c-f). FTSW was estimated to be 0.6 in both WW and stressed treatment before water deficit treatment, then dropped to ∼0.2 and 0.3 in WS1 and WS2 environments, respectively. To assess the effect of each water condition, we characterized the grain yield components and days to flowering of genotypes. A non-significant cross-over genotype-environment interaction (p < 0.08) was observed between the two drought tolerance reference lines, BTx642 and Tx7000 in WS1 and WS2 (Fig. 1g). As expected, the average grain weight and number of genotypes was significantly reduced in WS1 and WS2 relative to WW treatment (Fig. S1a,b). Overall, DFLo was significantly delayed in 2015 hot off-season environments, whereas it was reduced in cool off-seasons of 2016 and 2017 relative to rainfed conditions (Fig. S1c). DFLo was delayed in WS1, whereas it was not different in WS2 relative to the WW controls. DBM was significantly reduced in all stressed environments, except in WS1 of 2015 relative to RF (Fig. S1d). Average grain weight was not significantly different between RF and WS2 (Fig. S1e). The average GrN was significantly lower in WS1 than in WS2 (Fig. S1f).
Genetic variation in drought response
Broad-sense heritability (H2) estimates varied from moderate to high with values ranging from 33% for GrN to 92% for PH in the whole WASAP (Table S1). The average grain weight was not significantly different between caudatum accessions and durra and guinea accessions within each water regime in terms of production under drought stress (Fig. 2a). The durra-caudatum intermediates had significantly higher average grain weight than caudatum (13%, p < 0.05) and guinea (16%, p < 0.05) accessions, but not with durra (7%, p < 0.1) accessions. The average GrN was not significantly different between botanical types (Fig. 2b). Significant correlations were observed among yield components, including GrW, DBM, and STI for grain weight, across WS1 and WS2 regimes (Fig. S2a). High positive correlation was observed between BLUP of GrW, PW, DBM, and GrN, while TGrW was negatively correlated with grain number (Fig. S2b). Significant correlations were observed between DBM in WS1 and WS2 and other yield components, GrW, GrN, DFLo, and PH across RF conditions (Fig. S2c,d). Overall, genetic differences contributed to the phenotypic variation in managed water stress conditions.
Genome-wide association studies of flowering time
To identify loci potentially underlying quantitative trait variation in West African sorghum, we carried out GWAS using 130,709 SNP markers. First, we considered DFLo under WW off-season environments of 2015, 2016, and 2017 and BLUPs across all off-season environments to map known flowering time candidate genes using GLM+Q. No significant peak above the Bonferroni-corrected p-value of 0.05 was identified for DFLo of the 2015 data, but significant associations were identified for DFLo of the 2016 and 2017 data (Fig. S3). Two SNPs, S6_55280640 and S3_62811196, were significantly associated with DFLo in both years, and co-localized with a priori candidate flowering time genes SbZfl1 (Sobic.006G201600; 9 kb away) and SbCN12 (Sobic.003G295300; 61 kb away), respectively. In both 2016 and 2017, S6_55280640 was the lead SNP (p < 10-10 in 2016; p < 10-10 in 2017) of the associated region on chromosome 6. A third SNP, S2_67812515, was significantly associated with DFLo in 2017 data and colocalized with the a priori candidate gene Maturity2 (Sobic.002G302700; 70 kb away). Significant associations were not identified above the Bonferroni threshold (p > 10-5) when the MLM with PCA and kinship matrix were used to account for both population structure and genetic relatedness effects (Fig. S3).
The same associated SNPs near SbZfl1 and SbCN12, noted above, were observed for flowering time BLUPs across all off-season environments (Fig. 3a; File S1). Lead SNP S6_55280640 was located one gene away from SbZfl1 (Fig. 3c). The T allele of S6_55280640, associated with shorter flowering times under RF conditions (Fig. 3d), had a wide geographic distribution and was found at high frequency in accessions of the Sahel, Ethiopia, and west India (Fig. 3e). Lead SNP S3_62811196 was the top association near SbCN12 (Fig. 3f). The T allele of S3_62811196, associated with short flowering times under RF conditions (Fig. 3g), is globally-rare, found mostly in accessions from Niger and northern Nigeria (Fig. 3h).
Genome-wide association studies for drought tolerance
To generate hypotheses on the loci that underlie drought tolerance variation in sorghum, we performed GWAS for grain weight STI and the reduction of PW (RPW), DBM (RDBM), GrN (RGrN), PH (RPH), TGrW (RTGrW) in water-stressed environments. We considered water-stress scenarios separately (WW vs. WS1, WW vs. WS2) and together (WW vs. WS1 and WS2). In total, 222 and 214 associations were identified by the GLM+Q and MLM, respectively for drought response variables and STI in all drought stress environments (File S2; Fig. S4). Among the associations, 134 were commonly identified by both GWAS methods.
To determine QTLs that have positive pleiotropic effect on pre- and post-flowering drought tolerance among the associations above, we looked for common associations across different water-stressed environments. We defined a pleiotropic QTL as one lead SNP or locus being mapped in both pre- and post-flowering drought scenarios, or associated with several drought response variables. Among the associations, 16 putative pleiotropic associations for drought responses were observed across water stress environments (Table S2). For example, the SNP S4_67777846 was associated with STI under WS1 of 2016 and 2017 and WS2 of 2017 using both GLM+Q and MLM. SNPs S3_13763609 and S1_74186408 were associated with RPW in WS1 and WS2 of 2017 using both GLM+Q and MLM. The identified pleiotropic lead SNPs showed significant allelic effect and significantly (p < 10-8) explained 11–25% of STI for grain weight across water deficit treatments (Table S2). Of the 16 putative pleiotropic associations, 6 associations (S4_67777846, S2_18195896, S9_57781496, S10_1402513, S6_55048997) overlapped with associations identified for the STI BLUPs across water-stressed environments (Fig. 4; File S3).
Drought response associations colocalizing with stay-green loci
To test the hypothesis that Stg loci identified from Ethiopia, we characterized the colocalization of GWAS peak SNPs with previously defined Stg QTL intervals as summarized in the Sorghum QTL Atlas. The interval of Stg3 (Stg3a and Stg3b) was defined based on the introgressed region by the ICRISAT breeding program (Vadez et al., 2013). Of the total lead SNPs associated with STI for grain weight and drought response variables, 78 overlapped with 54 QTLs of the published Stg QTLs (File S4, Fig.4a,b), which represents a significant enrichment (p < 10-16). Thirty lead SNPs colocalized with known Stg1–4 loci (Table S3). The lead SNPs colocalizing with each locus could explain up to 16% (p < 10-10, Stg1), 20% (p < 10-13, Stg2), 19% (p < 10-13, Stg3a), 27% (p < 10-16, Stg3b) and 21% (p < 10-15, Stg4) of the phenotypic variation across WS1 and WS2 over years based on STI BLUPs. At Stg2, SNP S3_56094063 was the top association (p-GLM < 10-19, p-MLM <10-13) for STI in WS2 and WS1. At Stg3b, S2_62095163 was the top association (p-GLM <10-18, p-MLM <10-13) with high effect for STI in WS2. The remaining lead SNP associations did not colocalize with Stg loci.
Putative pleiotropic drought response associations colocalizing with stay-green loci
At each of the Stg1–4 loci there were several associations observed across two or more drought scenarios or drought response variables (Table S3). The Stg3a and Stg3b (which are next to each other) region covered associations for STI in WS1 of 2015 and 2016, STI in WS2 of 2016 and 2017, RPW in WS1 of 2017, and RDBM in WS1 of 2016. There was a strong LD among the lead SNPs within Stg3b but no LD among lead SNPs within Stg3a (Fig. 4c). The SNP S2_62095163 was in strong LD with other lead SNPs in Stg3b but not in LD with lead SNPs in Stg3a. Stg2 colocalized with putative pleiotropic associations for STI in WS1 of 2015 and 2017, WS2 of 2017, RGrN in WS1 of 2017, and RDBM in WS1 of 2016. The Stg1 locus covered associations for RPW in WS1 and WS2 of 2017 and associations for STI in WS1 of 2017. At both Stg1 and Stg2 there was strong LD among several lead SNPs (Fig. 4d). At the Stg4 locus there were associations for RPW in WS1 of 2017 and for STI in WS1 of 2015 and in WS2 of 2017, and moderate LD among lead SNPs (Fig. S4f).
Evolutionary signals around drought tolerance loci
To investigate the possibility of positive selection for drought tolerance at Stg loci, we conducted a genome scan of pairwise nucleotide diversity (π) ratios for West African sorghums relative to wild relatives (i.e. outliers with high πsorghum/πwild ratio), considering 95th and 99th percentile outliers (Fig. 4e,f; Fig. S5). Twelve of the lead SNPs associated with RPW and grain weight STI overlapped with π ratio outliers in durra-caudatums and durras, but not in guineas (Table 2). Colocalizations of π ratio outliers with Stg1–4 loci were significantly enriched (p < 10-16). In durra-caudatums and durras, but not in guineas, some 99th percentile π ratio outliers were localized within Stg1 (Fig. 4e,f; Fig. S5). We characterized the geographic distribution of two selected lead associations within each Stg locus to determine whether the Stg alleles are rare and only involved in local adaptation or are broadly involved in adaptation across sorghum landraces, beyond known sources in Ethiopia sorghums (Fig. 5a,b). The rare alleles associated with increased STI at a few selected lead SNPs within Stg1-3 were broadly geographically distributed in sorghum landraces (Fig. 5c-h). (Stg4 was excluded due to its large interval). However, the increased STI-associated allele at lead SNPs that overlapped with strong selection outliers were found mostly in WA sorghums (Fig. 5h), except for S3_66366589 (Fig. 5g).
DISCUSSION
How well do managed stress trials reveal the genetics of drought tolerance in sorghum?
In this study we sought to better understand genetics of drought adaptation in sorghum, a crop that is well known for drought tolerance, but whose mechanisms of drought adaptation are not yet understood (Choudhary et al., 2020). We characterized a diverse panel of West African sorghum germplasm in common-garden managed drought stress field trials with the aim of balancing experimental repeatability (via the use of irrigation in off-season) with biological and breeding relevance (via the use of a field environment) (Cooper et al., 2014). The usefulness of managed stress experiments to understand crop evolution and improve crop resilience depends on several criteria we consider in turn:
Was the intended stress applied? Two lines of evidence support the contention that the intended drought stress was successfully imposed via irrigation management in the off-season. First, the measured soil moisture was consistently high in well-watered control treatment (FTSW ≈ 0.6), but dropped to ∼0.2 at the intended times in pre- or post-flowering drought stress treatments (Fig. 1f). The FTSW values achieved in WS1 and WS2 were similar to the critical values (∼0.2–0.5) for decreases in transpiration and leaf expansion in diverse sorghum lines (Choudhary et al., 2020), suggesting that a physiologically-relevant stress was experienced by the plants. Second, we observed a substantial (but not complete) reduction of yield components (∼50%; Fig. 2) under managed drought stress (WS1 and WS2 relative to WW), suggesting the stress was also agronomically relevant (Blum, 2010).
Was the stress comparable to previous stress experiments? To be able to address this question, we included two international drought tolerance check lines, which are the canonical post-flowering (BTx642) and pre-flowering (Tx700) drought tolerant genotypes based on many studies in the U.S. and Australia (Tuinstra et al., 1996; Burke et al., 2013; Borrell et al., 2014b). Consistent with the idea that our managed drought stress was comparable to natural and managed drought stress in the U.S. and Australia, a strong cross-over genotype-environment interaction for grain yield of Tx7000 vs. BTx642 under pre- vs. post-flowering drought in the expected direction (Fig. 1g).
Was the timing and severity of stress comparable to that in the TPE? Among the three criteria, this is the most difficult to assess. A formal envirotyping study, which quantifies the frequency of particular water deficits relative to crop phenology, would be necessary to address this question (Chenu et al., 2011; Cooper et al., 2014). One particular concern for off-season managed stress would be that differences in photoperiod regime relative to the TPE (i.e. the rainy season) could change in growth or developmental dynamics in a way that alters the drought response (Blum, 2010; Gano et al., 2021). However, the overall similarity of grain yield components in the rainy season (RF) and off-season experiments (Fig. S1a,e; Fig. S2c,d) suggest that the managed drought stress is broadly comparable to drought in the TPE. Ultimately, to rigorously test hypotheses on the similarity of off-season managed drought to the drought in the TPE, a comparison of phenotypes under managed stress to multi-environment field trials under natural drought stress will be necessary (Cooper et al., 1995).
Evidence for a role of SbZfl1 and SbCN12 in flowering time variation and for SbCN12 in drought adaptation
Flowering time is a critical component of geographic adaptation (Lasky et al., 2015; Castelletti et al., 2020) and a potential contributor to drought adaptation via early-flowering drought escape (Blum, 2010). Among the six canonical sorghum photoperiodic flowering genes (Maturity1–Maturity6) characterized in U.S. germplasm, (Murphy et al., 2011, 2014; Yang et al., 2014; Casto et al., 2019) we identified colocalization of associations only at Ma2 (Fig. 3a). Instead, the top QTL mapped two a priori flowering time candidate loci, SbZfl1 (chr6: 55.289–55.293 Mb) and SbCN12 (chr3: 62.747–62.749 Mb) that are not known to underlie genetic variation in U.S. germplasm (Fig. 3; Fig. S3).
SbZfl1 is the ortholog of maize ZFL1/2 and rice RLF, which induce early flowering by activating vegetative-to-reproductive transition (Bomblies & Doebley, 2006; Rao et al., 2008). While SbZfl1 variation has not been previously identified via linkage mapping (Mace et al., 2019), SbZfl1 was identified as a top candidate in a recent GWAS of photoperiodic flowering rating in a Senegal regional panel (Faye et al., 2019). The MAF of the SbZfl1 QTL was high (>0.4) in both WASAP and global georeferenced landraces (Fig. 3e), suggesting a common, moderate-effect variant exists at SbZfl1. Sorghum is a short day species, so under short days (e.g. the cool off-season in West Africa; Fig. 1b) it is expected to flower early, regardless of photoperiodism. Given SbZfl1 was the top flowering time association under short days, SbZfl1 may be a regulator of basic vegetative phase (BVP), the thermal time component of flowering regulation that acts independently of photoperiodic flowering regulation (Guitton et al., 2018). This hypothesis could explain the lack of a flowering time QTL at SbZfl1 in a previous GWAS under long days (rainy season) in the WASAP (Faye et al., 2021)—subtle BVP variation at SbZfl1 could have been masked by large-effect photoperiodic variants at Ma6, SbCN8, or other loci. However, this hypothesis would not explain the photoperiod flowering association at SbZfl1 previously observed in Senegalese germplasm (Faye et al., 2019). Given inherent limitations of association studies (Korte & Farlow, 2013) and the complexity of photothermal flowering (Li et al., 2018b), linkage mapping and ecophysiological modeling will be needed to test these hypotheses on the role of SbZfl1 in flowering time adaptation (Guitton et al., 2018; Li et al., 2018b).
SbCN12 (also known as SbFT8) is a co-ortholog of the canonical florigen Arabidopsis FT gene and ortholog of maize ZCN12 (Yang et al., 2014; Castelletti et al., 2020), which was identified as a likely sorghum florigen based on conserved sequence and expression dynamics (Yang et al., 2014; Wolabu et al., 2016). The current GWAS findings, along with previous finding that SbCN12 explained up to 12% of variation in global nested association mapping population (Bouchet et al., 2017; Hu et al., 2019), provide strong support for the hypothesis that functional allelic variation exists at SbCN12. Given the early-flowering associated allele near SbCN12 is globally rare (Fig. 3h), it may be a useful new allele for sorghum breeding programs targeting earlier flowering for stress escape. Molecular cloning of causative variants at SbCN12 and SbZfl1 could shed light on their role in flowering time evolution (Bomblies & Doebley, 2006; Castelletti et al., 2020) and facilitate development of molecular marker to recover locally-adaptive flowering time.
The evidence for a role of these flowering time genes in drought adaptation (e.g. via drought escape) is mixed. On one hand, SbCN12 colocalized with a drought response association (RPW for WS1 vs. WW; S3_62836558; 64 kb away; Table S3), so could plausibly underlie some variation for pre-flowering drought response of this yield component. Also the same SNP near SbCN12 was in a window of reduced π in guinea genotypes (Table 2), suggesting selection at this locus. (Note, this is not the same SNP as the rare flowering time associated variant S3_62811196, but a common variant 25 kb away). On the other hand, SbZfl1 did not colocalize with the drought response QTL (STI, RPW, etc.; the nearest association with STI, S6_55048997, was at ∼240 kb away) and there was no evidence of positive selection around SbZfl1 based on the π ratios (Fig. 4; Fig. S5). Given that causative variants at SbCN12 and SbZfl1 are not yet known, hypotheses on the role of these genes in drought adaptation remain speculative, but could be tested using near-isogenic lines (NILs).
Insights on the genetics of drought adaptation in sorghum
The botanical types of sorghum vary strikingly in their morphology and geographic distribution, and based on a phytogeographic adaptation model (Vavilov, 2009). It has long been hypothesized that they vary in their drought adaptedness (Harlan & De Wet, 1972; Lasky et al., 2015; Wang et al., 2020). For instance, durra sorghums, which predominate in arid regions, are thought to be the most drought tolerant (Harlan & De Wet, 1972), while guinea sorghums, which predominate in humid regions are thought to be adapted to high humidity (De Wet et al., 1972). However, previous studies of large sorghum diversity panels under managed drought stress have not directly tested this hypothesis, for instance, by comparison of drought response for yield among botanical types (Vadez et al., 2011; Lasky et al., 2015; Upadhyaya et al., 2017). Surprisingly, in this study we find no evidence of overall differences in drought tolerance among botanical types based on the drought response of yield components (Fig. 2). These findings could be explained by one of two competing hypotheses. First, it is possible that the drought scenarios we applied do not correspond to the drought scenarios in the TPE, such that true differences in drought tolerance among botanical types were not reflected in the phenotypes. Alternatively, it may be that the major botanical types in West Africa all harbor substantial drought tolerance, presumably because droughts are common even in the higher precipitation portions of the sorghum range (Traore et al., 2014). In either case, our findings suggest that long-held views on differential drought adaptation among botanical types in sorghum require more formal testing.
Theoretical considerations on water use tradeoffs (Tardieu, 2012) and the apparent lack of sorghum genotypes harboring both pre- and post-flowering drought tolerance (Burke et al., 2013) suggest that a tradeoff might exist between early versus late stage tolerance mechanisms. However, the moderate positive correlation of the grain yield estimates under pre- and post-flowering drought (e.g. for GrW or STI; Fig. S2a) suggest no major physiological tradeoff for tolerance to these contrasting drought scenarios, at least at this broad scale of diversity. Colocalization of MTA for drought tolerance related traits in WS1 and WS2 would provide further evidence for genetic factors that contribute pleiotropically to both pre- and post-flowering drought tolerance. Consistent with this hypothesis, sixteen distinct MTAs (Table S2) were detected for drought-related traits (mostly STI) under both the pre- and post-flowering drought treatments, including one (S3_56094063) that colocalized with Stg1–4 loci (Table S3).
Among the positive pleiotropic associations, the STI MTA at S4_67777846 may be the most interesting candidate for further study, given that it had the highest PVE estimate (25%) in both pre- and post-flowering water stress over two years. This MTA did not colocalize with Stg QTLs or other a priori candidate genes, and there were no obvious post hoc candidate genes near the SNP, so we have no hypothesis on the biological basis of this association at this point. If confirmed, positive pleiotropic drought tolerance QTLs, which could contribute to yield stability across drought scenarios, would be of great interest for breeding of broadly-adapted climate-resilient varieties and help elucidate mechanisms that circumvent potential tradeoffs (Tardieu, 2012).
Another question that motivated our study was whether canonical Stg alleles, which were originally discovered in Ethiopia-derived materials (BTx642) (Borrell et al., 2014a), are also present in West African landraces (Fig. 5a). The hypothesis that canonical Stg alleles have a broad role in drought adaptation across Africa is plausible since it is well established that durra sorghums diffused from Ethiopia across the Sahel to West Africa (Harlan & Stemler, 1976; Morris et al., 2013). Indeed, we observe a statistically significant enrichment of drought tolerance related MTA colocalized with canonical Stg QTL intervals, which provides preliminary support for the shared Stg hypothesis (File S4; Table S3). Most notable among these are the highly significant association for grain weight STI under post-flowering drought at Stg3 (S2_62095163) and Stg2 (S3_57614567). Further, the West Africa drought tolerance associated alleles in the Stg intervals are found in Ethiopia, as would be expected if they were shared across Africa. While these findings are suggestive, they are not sufficient to exclude the alternate hypothesis (Fig. 5b) that West African drought tolerance loci are unrelated to Ethiopian-origin Stg alleles. Testing this hypothesis conclusively would require positional cloning of the West Africa drought tolerance QTL and Stg alleles.
The final hypothesis we considered was that drought tolerance alleles underlie drought adaptation per se and were subject to positive selection. This finding was supported by significant enrichment for colocalization of selection outliers with Stg QTLs and common allele frequencies of lead SNPs overlapping with selection outliers in durra-caudatums and durras relative to guineas (Fig. 4e,f; Fig. S5; Table 2). As with the other findings, the development of NILs and the validation of major effect QTL in breeding populations (Borrell et al., 2014b,a) will be crucial to rigorously test the proposed role of QTL in these genomic regions for drought adaptation. Overall, our findings support the long-standing hypothesis that genetic variation for drought tolerance exists in West African sorghum, and provide preliminary evidence for a broad role of canonical Stg drought tolerance alleles across Africa.
Author Contributions
GPM, DF, NC, BS: design of the research; EAA, BS, CD: performance of the research; JMF, EAA, BS, GPM: data analysis, collection, or interpretation; JMF, GPM: writing the manuscript.
SUPPORTING INFORMATION
ACKNOWLEDGMENTS
This study is made possible by the support of the American People provided to the Feed the Future Innovation Lab for Collaborative Research on Sorghum and Millet through the United States Agency for International Development (USAID) under Cooperative Agreement No. AID- OAA-A-13-00047. The contents are the sole responsibility of the authors and do not necessarily reflect the views of USAID or the United States Government. Analyses were made possible by Beocat High-Performance Computing Cluster at Kansas State University.