Host Determinants of Fungal Species Composition and Symptom Manifestation in the Sorghum Grain Mold Disease Complex

Sorghum (Sorghum bicolor) is an important cereal crop worldwide, with diverse culinary and industrial uses (Castro et al. 2017; Han et al. 2012; Rogerson 2019; Xiong et al. 2019). Given its nutritional value and hardiness even in marginal environments, sorghum plays a vital role in food system resilience in semiarid regions (Belton and Taylor 2004; Hadebe et al. 2017). Sorghum grain mold (SGM) is an important multifungal disease complex that compromises quality and safety of sorghum crops worldwide (Das et al. 2012). SGM-associated fungi can contaminate sorghum with fumonisin, a mycotoxin implicated in adverse health outcomes (Sharma et al. 2011; Shephard et al. 2007). The disease complex takes a substantial toll on the productivity and profitability of sorghum cropping systems (Das and Patil 2013). Resistance breeding has to date been insufficient in controlling SGM, owing to the marked diversity and spatiotemporal dynamics of the fungi implicated in disease outcomes. To more effectively prevent disease and mitigate the associated economic and public health risks, it is necessary to consider not only the genes associated with resistance but also the genetic underpinnings that shape the ecology and composition of the fungal disease complex.

Warm, humid environments during the growing season (and particularly during grain development) promote establishment of the disease (Waniska et al. 2001). The deployment of resistant cultivars is cited as the most effective means to control SGM (Rodriguez-Herrera et al. 2006), but complex host-pathogen-environment interactions have limited the success of resistance breeding efforts (Das et al. 2012). There are several morphological and bio chemical traits known to confer resistance against SGM. Traits involved in resistance include panicle openness, testa presence/pigmentation, glume coverage, pericarp structure, and kernel density (Bandyopadhyay et al. 2000). Biochemically, phenolic compounds present in the testa and caryopsis have been associated with SGM resistance (Little and Magill 2003; Menkir et al. 1996).

One of the major fungal contributors to the SGM disease burden is Fusarium verticillioides, a ubiquitous ascomycete fungus that is known to colonize sorghum, maize, and other important crop species in asymptomatic/biotrophic and necrotrophic relationships (Deepa and Sreenivasa 2017; Marín et al. 2004; Oren et al. 2003). F. verticillioides is a prolific producer of fumonisins, and it is known that SGM is associated with fumonisin contamination in several production contexts (Bhat et al. 2000; Waliyar et al. 2008), but the myriad interactions among sorghum, F. verticillioides, competing microbes, and the production environment limit the ability to effectively execute disease control and mycotoxin management.

F. verticillioides competes with numerous other fungal species that also play roles in the grain mold complex. The fungi most commonly associated with SGM include Fusarium spp., Curvularia lunata, Phoma sorghina, Bipolaris australiensis, Alternaria alternata, Colletotrichum graminicola, and others (Thakur et al. 2006). Despite the diversity in fungal contributors to the disease complex, their respective roles and relationships with SGM outcomes have not been thoroughly explored. This gap in understanding is in part attributable to the limited potential of conventional field phenotyping strategies, which are generally not conducive for the type of species- or symptom-specific investigations necessary to elucidate the range of symptom manifestations produced by the SGM disease complex. Generally, grain mold is appraised by visual scoring of disease severity, incidence, or damage (Butler et al. 2000) without specific attention to the community composition of fungi involved in the complex.

The fungi associated with SGM have nonuniform modes of infection and colonization, suggesting that the host may be protected by diverse resistance factors depending on the composition of the disease complex (Little 2000; Melake-Berhan et al. 1996). Moreover, spatiotemporal dynamics of host-pathogen-environment interactions can influence the morphophysiological outcomes of fusariosis disease in grain crops (Morales et al. 2018; Mutiga et al. 2014). We therefore hypothesized that to better understand the genetics underpinning grain mold resistance in sorghum, it would be important to dissect the classical mold severity phenotype and investigate whether there are genetic features that affect species composition and symptom manifestation of the disease complex.

In this study, we sought to understand the extent to which host genetics determine the symptom outcomes that manifest in the grain mold disease complex and whether there are genetic underpinnings that result in mold assemblages that favor infection by F. verticillioides. Specifically, the objectives of our study were to (i) identify and characterize the diversity of symptom manifestations associated with the SGM disease complex, (ii) describe a typology of distinct symptom manifestations that may be used to classify genotypes, (iii) assess the relative influence of F. verticillioides in the SGM disease complex across diverse sorghum germplasm, and (iv) understand the host genetic determinants of grain mold symptom outcomes.

Materials and Methods

Sorghum germplasm and field trial design

A panel of 390 diverse sorghum accessions was planted in 2017 and 2019 at Clemson University’s Pee Dee Research and Education Center in Florence, South Carolina, U.S.A. The diversity panel included 332 accessions from the original United States sorghum association panel (SAP) described by Casa et al. (Casa et al. 2008) and an additional 58 accessions based on unique characteristics (Boyles et al. 2016). The fields were planted in a randomized complete block design, with height and days to anthesis used as blocking factors as previously described (Sapkota et al. 2020). Resistant and susceptible controls (P850029 and Tx2911, respectively) were included in every block. Genotypes had two plot replicates per year, totaling four plot replicates per genotype across the 2 years of the study. Each plot-replicate consisted of two rows 6.1 m in length, with 60 seeds planted per row, for a total field density of 130,000 plants/ha assuming 75% plant establishment rate. Nitrogen fertilizer was applied to the field at 89.7 kg/ha as a side dress application 40 days after planting. Charger Max Atz was applied as pre-emergence weed control at the rate of 4.7 liter/ha. Sivanto Prime (0.5 liter/ha) and Prevathon (1.5 liter/ha) were applied postheading for control of sugarcane aphids and corn earworms, respectively. Fields were irrigated on an as-needed basis during the growing season to prevent bias in mold outcomes due to differential drought tolerance among lines in the diversity panel.

F. verticillioides isolates and field inoculation

The intent of this study was to ascertain SGM disease outcomes under disease pressure from endemic fungal strains. All F. verticillioides isolates used in this study were isolated from sorghum at the Pee Dee Research and Education Center. In the first year of the trial, five local F. verticillioides strains were isolated from overwintered sorghum detritus collected from the field in Spring 2017. Because it was not possible to conduct pathogenicity assays prior to the field season, these five isolates were mixed at equal concentrations. The field location is a conducive environment for SGM and is subject to high disease pressure even without inoculation. For the second year of the trial, one local pathogenic F. verticillioides strain was isolated from a naturally occurring SGM disease complex at the field site in Summer 2018; this isolate was the sole isolate used in the 2019 inoculation. All fungal isolates in the 2017 cocktail were identified at the species level via colony and spore morphology. The 2019 isolate was first identified morphologically in culture, and the the species identity was confirmed using F. verticillioides-specific ‘Fusq’ primers (Rodriguez Estrada et al. 2011). These isolates were intended to represent fungi endemic to the field environment and were not accessioned nor intended for further mycological study.

Single spore-derived fungal subcultures were plated onto Petri dishes containing potato dextrose agar and incubated for 7 to 10 days, until prolific sporulation was confirmed via compound microscopy. Conidia were manually harvested from each plate by scraping the mycelium with a bacterial loop, incorporating the tissue mass into 2 to 5 ml of sterile diH₂O, and decanting the Petri dishes into sterile 50-ml conical tubes. Spore slurries were filtered through three layers of cheesecloth, and spore concentrations were adjusted to 1 × 10⁶ spores/ml with sterile diH₂O. Surfactant Tween 20 was added to the inoculum immediately before application at concentration 0.2 ml/liter.

Plots were inoculated by spraying liquid spore suspension into panicles at 50% anthesis, as previously described (Bandyopadhyay and Mughogho 1988; Forbes 1986; Little and Magill 2003). We used an inoculation procedure adapted from Prom and Erpelding (2009) and Clements et al. (2003). Once the necessary volume of spore suspension had been diluted to the proper concentration as noted above, a cone-nozzle-equipped applicator (either backpack or spray bottle) was used to thoroughly wet the entire panicle from all angles until runoff (∼50 ml/plot). We used a two-tier buffering approach to prevent border effects: The primary panicles of 15 central plants in the odd row of each plot-replicate were inoculated, and the middle five panicles in the inoculated range were tagged for phenotyping. Tagged panicles were harvested by hand at physiological maturity, placed individually into fresh pollination bags, and subsequently dried to a stable moisture content to prevent further fungal growth.

Grain mold phenotypes

The four SGM phenotypes used in this study are summarized in Fig. 1. We used a classical panicle grain mold severity rating (PGMSR) to assess the magnitude of disease as described previously (Menkir et al. 1996). Each panicle was rated on a scale of 1 to 5, where 1 = ∼0% infected grains, 2 = 1 to 10% infected grains, 3 = 11 to 25% infected grains, 4 = 26 to 50% infected grains, and 5 = >50% infected grains. In this rating system, scores were assigned based on total mold presence irrespective of symptom typology.

Download as PowerPoint

While useful as a measure of disease magnitude, the conventional PGMSR does not provide information about the spread of disease symptoms across a panicle. Based on our hypothesis that understanding the distribution of symptoms can yield useful insights into the pathogenesis and phenomenology of disease, we developed an intrapanicle disease localization index (IPDLI) to represent the spatial distribution of SGM symptoms. Disease presence (Y/N) was recorded for each quintile of the panicle (tip, above midsection, midsection, below midsection, and base). The index ranged from –3 to 3 and was calculated by summing quintile values across all quintiles where disease was present, where tip = 2, below tip = 1, midsection = 0, below midsection = –1, and base = –2. Panicles with more positive values had disease localized toward the tip, and those with more negative values had disease localized toward the base. Panicles with indices close to zero had no discernible localization patterns.

As a simple measure of the breadth of species composition in the SGM disease complex observed on each panicle, the number of distinct intrapanicle symptom types (IPST) was counted. Distinct symptom typologies were identified morphologically and recorded according to coloration of visible mycelium in the following categories: White/pink (which was exclusively applied to morphologically identified Fusarium spp. molds), black, gray, red, green, or orange. Morphological identification of Fusarium grain mold on infected panicles was conducted using the symptom criteria described by Das et al. (2012). Taxonomic identity was additionally verified for a random subset of plots (∼2%) with “white/pink” symptoms by examining spore morphology with a compound microscope. Molds not morphologically classified as Fusarium spp. were presumed to correspond to other non-Fusarium taxa implicated in the complex for phenotypic analysis. Panicles with no observable disease symptoms were ascribed a count value of zero.

An objective in this study was to understand the relative contributions of F. verticillioides to the SGM disease complex in diverse host genetic backgrounds. Thus, we developed a Fusarium symptom dominance index (FSDI) that reflects the influence of this taxon in plot-level disease outcomes. Each panicle was scored either “1” for Fusarium dominance or “0” for Fusarium nondominance in the SGM manifestation, and FSDI was computed as the sum of all panicle-level binary (0/1) values averaged over all five panicles in each plot. The index was represented as a proportion ranging from 0 to 1, where 0 = complete nondominance, and 1 = complete dominance of Fusarium symptoms in the SGM disease complex.

Genotyping and single nucleotide polymorphism calling

The sorghum association panel was genotyped by sequencing as previously described (Boyles et al. 2016, 2017; Morris et al. 2013). Restriction enzyme ApeK1 was used for digestion of genomic DNA. Sequencing reads were aligned to the sorghum reference genome (v3.1.1, https://phytozome-next.jgi.doe.gov/) and filtered using the TASSEL 5.0 GBS pipeline (Glaubitz et al. 2014). Missing single nucleotide polymorphisms (SNPs) were imputed in TASSEL using the FILLIN method (Swarts et al. 2014), resulting in 268,896 total SNPs.

Trait analysis

Means and variances were computed for PGMSR, IPDLI, and IPST for each plot-replicate in both years of the study. Mean FSDI phenotypes were calculated at the plot-level as described. Values were fitted to linear mixed models using the lme4 R package (Bates et al. 2015), following the equation

Where y_ijk is the phenotypic value for the ith genotype in year j, and replicate k; G_i, Y_j, R_k, G_iY_j, and G_iR_k are the random effects of genotype, year, replicate, genotype-by-year interaction, and genotype-by-replicate interaction, respectively, and ε_ijk is the random effect of residuals following the distribution N(0, ${\sigma }_{\epsilon }^2$). Variance components (σ²) of random effects were calculated for each trait using the lmer() function in R. Variance estimates were used to compute broad-sense heritability (H²) of each trait on a line mean basis, with replicate used in place of location as previously described (Boyles et al. 2016) using the equation

Best linear unbiased predictors (BLUPs) were estimated based on the random effect of each genotype from the fitted models. The BLUP estimates were subsequently used instead of unadjusted phenotype values for characterizing the SGM traits and delineating a typology of SGM symptom manifestations.

Bivariate Pearson correlation coefficients and statistical significance levels were evaluated for each SGM trait pair based on BLUPs using the rcorr() function in the Hmisc R package (Harrell 2021). In addition to pairwise correlations between SGM traits, correlations were also examined among these traits and other panicle characteristics with known implications for disease outcomes, including the variables panicle openness (compact = low, semicompact = moderately low, semiopen = moderately high, and open = high), pericarp pigmentation (white, yellow, and red), glume pigmentation (straw = low, gray = moderately low, red = moderately high, and black = high), and glume coverage (0 to 25% = low, 26 to 50% = moderately low, 51 to 75% = moderately high, and >75% = high). The factor levels of panicle traits were codified as numerical variables for correlation analysis: Openness (1 to 4 scale corresponding to categories), pericarp pigmentation (white = 1, yellow = 2, and red = 3), glume pigmentation (1 to 4 scale corresponding to categories), and glume coverage (mean % coverage). Pearson’s correlations were examined among panicle traits and plot-level mean values for all four SGM traits.

Distributions of SGM trait BLUPs were examined with respect to accessions’ botanical race. Botanical race designations included each of the five major races (bicolor, caudatum, durra, guinea, and kafir) and combinations thereof. Designations for most accessions were derived from a prior study (Morris et al. 2013); the remaining race designations were determined based on population structure in the program STRUCTURE as previously described (Boyles et al. 2016). Analysis of variance (ANOVA) was used to test for significant differences in SGM traits across botanical races. Post hoc pairwise comparisons were examined using Tukey’s tests of honestly significant difference (HSD) implemented using the TukeyHSD() function in R. P values for all post hoc tests were adjusted using a Bonferroni correction for multiple comparisons.

A principal component analysis (PCA) was performed and visualized for all grain mold traits using the PCA() and fviz_pca_biplot() functions in the FactoMineR R package, respectively (Lê et al. 2008). Eigenvalues and variable loadings were examined to assess PCA quality, and the percent of variance was explained by principal components. A k-means clustering approach was implemented to delineate a typology of SGM manifestation by partitioning individuals into four groups. Clustering was executed using the kmeans() R function and visualized using the fviz_cluster() function in the FactoMineR R package. ANOVA and post hoc Tukey’s HSD tests were performed to test for cluster-wise differences in SGM trait outcomes. A radar plot of trait means was created using the ggradar R package (Bion 2022) to visualize the final typology.

Genome-wide association studies

A total of 297 and 334 genotyped individuals with phenotype data collected in 2017 and 2019, respectively, were included in the GWAS. The number of accessions included varied due to inability to obtain phenotype data from some plots in the field. The genomic dataset was filtered by minor allele frequency >0.05 and comprised 268,896 total SNPs. To ascertain the effect of genotype-by-environment interaction (GxE) in the data, significance of the GxE random effects from the models described above was determined using ANOVA-like tests of random effect terms, which was implemented with the ranova() function in the lmerTest R package (Kuznetsova et al. 2017). The GxE effect was highly significant for all traits, potentially obscuring signal in GWAS and leading to spurious observations. Thus, BLUP estimates for use in the association studies were based on the random effect of the GxE interaction for each genotype in each year. GWAS was executed separately for each year based on these BLUP estimates.

Fixed and random model circulating probability unification (FarmCPU) modeling was used for GWAS, which produces fewer false positives than conventional mixed modeling or stepwise regression approaches and eliminates confounding between relatedness and associated traits (Kusmec and Schnable 2018). FarmCPU models with parameters to control for relatedness and population structure were fit using BLUPs for each trait.

Association analyses were performed using the Genome Association and Prediction Integrated Tool (Lipka et al. 2012). The kinship matrix (K) used in the models was calculated in GAPIT using the VanRaden method (VanRaden 2008). Three principal components (PCs) generated from principal components analyses were included in the models to control for population structure and relatedness among individuals (Wei et al. 2017). To strengthen associations for these highly quantitative traits, pigmented testa presence (0/1) was included in the models as a covariate to control for confounding effect of tannin-associated Tan1 locus, which has been established as an important resistance factor (Cuevas et al. 2019). To account for any temporal differences in the infection biology of mold-associated fungi, flowering time was also included as a phenotypic covariate in the models. Quantile-quantile (Q-Q) plots of association results were used to confirm that the models effectively controlled for false positives and relatedness among lines (Supplementary Fig. S1).

Bonferroni-corrected significance thresholds for multiple comparisons can be overly stringent, as it assumes true independence between tests, and some SNPs are correlated and therefore not truly independent (Zhang et al. 2015). To determine an empirical significance threshold for evaluating associations, we performed Bonferroni-like multiple testing correction that estimated the number of tests using the extent of linkage disequilibrium (LD) across the genome (Matthies et al. 2014). Pairwise SNP LD (r²) was computed in TASSEL, and LD decay was plotted using the geom_spline() function in the ggplot2 R package (Wickham 2011). In this dataset, LD decayed to background levels (r² = 0.1) within 25 kb distance (Supplementary Fig. S2), which is within the range previously reported (Boyles et al. 2017; Hamblin et al. 2004; Hu et al. 2019). A Bonferroni-like multiple test correction was implemented using the effective number of independent tests (genome size/average extent of LD decay = 730 Mb/25 kb = 29,200) as previously described (Zhang et al. 2015). Given the alpha level 0.05, an empirical significance threshold for SNP associations was estimated at P = 10⁻⁶ based on this multiple test correction. Manhattan plots and Q-Q plots were generated using the ggplot2 package in R.

Functional genes in LD (within the estimated 25 kb window) with significant SNP markers were identified by mapping SNPs to the Sorghum bicolor reference genome v3.1.1 (McCormick et al. 2018) in Phytozome (https://phytozome-next.jgi.doe.gov/). The possible functionalities of identified genes were explored in the context of grain mold symptom manifestation and Fusarium spp. pathogenicity in sorghum and related crop species.

Results

Grain mold trait characteristics

The SAP accessions exhibited substantial variability in the four SGM traits evaluated here (Fig. 2A, B, C, and D; Table 1). The mean PGMSR rating was 2.76 (± 0.63), indicating moderate disease severity overall across the population under high disease pressure (i.e., artificial inoculation). The IPDLI scores on average were only slightly positive (mean = 0.73 ± 0.10) on a scale of –3 to 3, indicating that symptom manifestation was not consistently localized toward the panicle tip nor the panicle base. The mean IPST score was 1.34 (± 0.30), indicating that more than one distinct symptom type was present on evaluated panicles on average. This finding confirms that the SGM outcomes observed at the study site were attributable to a complex of diverse fungi even when saturated with F. verticillioides inoculum. Accordingly, Fusarium dominance in the disease complex was moderate, with a mean FSDI score of 0.39 (± 0.09). This suggests that there were a number of fungi in the study environment that were viable competitors with F. verticillioides, which had implications for the manifestation of SGM outcomes.

Download as PowerPoint

Table 1. Summary statistics of sorghum grain mold (SGM) traits evaluated^a

	SGM Trait
	PGMSR	IPDLI	IPST	FSDI
Description	Overall severity of disease	Extent of symptom localization or dispersion on panicle	Number of symptom types in complex	Whether Fusarium symptom types dominate
Parameter
Mean (SD)	2.76 (0.63)	0.73 (0.10)	1.34 (0.30)	0.39 (0.09)
Median (Q1, Q3)	2.75 (2.30, 3.20)	0.73 (0.66, 0.80)	1.36 (1.17, 1.54)	0.39 (0.32, 0.45)
Range	1.53–4.24	0.38–1.03	0.57–1.99	0.24–0.60
Variance partitioning and broad-sense heritability
G	0.6	0.06	0.15	0.02
Rep	0.05	0.00	0.01	0.00
Year	0.18	0.21	0.03	0.02
G:Rep	0	0	0	0
G:Year	0.25	0.21	0.07	0.03
Residual	0.52	0.6	0.17	0.07
Genotype contribution to total variance (%)	57.1%	5.6%	34.9%	14.3%
H2	0.76	0.28	0.71	0.45

^a PGMSR = panicle grain mold severity rating; IPDLI = intrapanicle disease localization index; IPST = intrapanicle symptom typology; and FSDI = Fusarium symptom dominance index. Summary statistics reported are based on best linear unbiased predictors.

View as image HTML

All pairwise SGM trait Pearson correlations were statistically significant at the 0.05 alpha level. PGMSR and IPST, the two SGM traits directly related to disease severity, were strongly positively correlated (Fig. 2E). PGMSR and IPST also exhibited moderate positive correlations with FSDI, indicating that more severe infections tended to be more dominated by Fusarium symptom types and that Fusarium tended to coinfect with assemblages of other fungal pathogens in the disease complex. IPDLI had moderate negative correlations with all other variables, which has a reasonable biological explanation: More severe infections, or those where a fungus may have a competitive advantage, are more likely to cover more of the panicle space and thereby are not “localized.”

Examination of variance partitioning in the data set revealed that the extent of phenotypic variation explained by genotypes is variable across SGM traits (Table 1). PGMSR had the highest genetic contribution to variance of any trait (57.1%) and also yielded the highest estimate of broad-sense heritability (0.76). This is consistent with previous studies that estimated H² between 0.49 and 0.85 (Diatta et al. 2019; Rodriguez-Herrera et al. 2000). While the heritability of IPST was relatively high (0.71), the genetic contribution to the observed variance was moderate (34.9%), indicating that this trait is likely to be affected by the growing environment. Much smaller genetic contributions (<15%) were observed for IPDLI and FSDI, and, accordingly, heritability of these traits was relatively modest (0.28 and 0.45 for IPDLI and FSDI, respectively). We can therefore conclude that mold severity and the number of species involved in the disease complex have relatively strong genetic contributions, but the relative importance of specific fungi in disease complex outcomes is strongly influenced by environmental determinants.

A significant effect of botanical race was observed for all SGM traits examined in ANOVA analysis, but there was marked variability even within races (Fig. 3). In post hoc Tukey analysis of pairwise comparisons by botanical race, only a few tests yielded significant differences after applying a P value adjustment for multiple comparisons (Supplementary Table S1). Most significant pairwise comparisons involved durra accessions, which had comparatively high values in disease severity traits (e.g., PGMSR and IPST) and comparatively low values in IPDLI on average. Only one significant pairwise comparison was observed for IPST (between durra-bicolor and durra-caudatum accessions), illustrating that the species richness of the SGM disease complex is comparably diverse across botanical races.

Download as PowerPoint

Several significant correlations were observed among grain mold phenotypes and panicle- or seed-related characteristics (Table 2). Glume coverage had moderate negative correlations with PGMSR and IPST but no statistically significant correlation with FSDI or IPDLI. Panicle openness was negatively correlated with PGMSR and IPST but had no observable relationship with FSDI or IPDLI; this suggests that open panicles are (i) less susceptible to grain mold and (ii) have fewer diverse fungal assemblages. The level of pigmentation in both the glume and the pericarp was negatively correlated with PGMSR and IPST, indicating a clear relationship between pigmentation and SGM symptom manifestation. While glume pigmentation exhibited no relationship with IPDLI or FSDI, pigmentation in the pericarp had significant weak to moderate correlations with these traits. It is possible that the weak correlations observed with glume pigmentation across traits may be partially attributable to visual scoring bias, as panicle glume pigmentation can be somewhat ambiguous or difficult to observe.

Table 2. Pearson’s correlation matrix among panicle characteristics and grain mold phenotypes^a

	PGMSR	IPDLI	FSDI	IPST
Panicle structure
Glume coverage (%)	–0.33***	0.09	–0.05	–0.38***
Panicle openness	–0.34***	0.09	–0.07	–0.35***
Panicle pigmentation
Glume pigmentation	–0.17**	0.10	–0.02	–0.18***
Pericarp pigmentation	–0.32***	0.17**	0.41***	–0.30***

^a PGMSR = panicle grain mold severity rating; IPDLI = intrapanicle disease localization index; FSDI = Fusarium symptom dominance index; and IPST = intrapanicle symptom typology. Significance values: * P < 0.05; ** P < 0.01; *** P < 0.001.

View as image HTML

Principal components analysis (PCA) was undertaken to characterize the range of distinct manifestations of SGM disease outcomes among SAP accessions. The first three PCs in the analysis had Eigenvalues of 2.5, 0.8, and 0.6, respectively, and cumulatively accounted for 97.0% of variance in the data (Table 3). The first PC is a linear combination of PGMSR, IPST, and IPDLI, indicating that this dimension represents the severity and distribution of disease in panicle space. Comparatively, FSDI was the only variable with strong loadings on PC2, suggesting that this dimension reflects the species composition in the SGM disease complex, and in this case, the role of Fusarium in the manifestation of disease outcomes. Individual genotypes generally did not exhibit well-defined clustering patterns, with many individuals localized at the center of the PCA factor map (Fig. 4A). No qualitative clustering by botanical race or geographic origin was observed. A typology of SGM disease manifestation was identified by partitioning the accessions into k = 4 groups by k-means clustering (Fig. 4B).

Table 3. Summary of sorghum grain mold trait principal component analysis Eigenvalues and contributions to explained variance

PCa	Eigenvalue	Variance percent	Cumulative variance percent
1	2.52	62.9	62.9
2	0.81	20.3	83.2
3	0.55	13.8	97.0
4	0.12	3.0	100

^a PC = principal component.

View as image HTML

Download as PowerPoint

The four SGM disease manifestation clusters (hereafter “clusters”) in the identified typology yielded statistically significant differences for all traits in ANOVA and significant post hoc pairwise cluster comparisons in nearly all cases (Supplementary Fig. S3). The clusters’ trait outcomes in the typology are summarized in Figure 5. Cluster 1 tended to be the most resistant to SGM, exhibiting low values for severity traits (PGMSR and IPST) and moderate dominance of Fusarium in the disease complex, suggesting a relatively diverse assemblage of fungal associates. Unlike the other clusters, Cluster 1 was the only cluster with substantial symptom localization in the panicle space, with more positive (i.e., tip-localized) values than the other clusters. Cluster 2 exhibited moderate mean values for all SGM traits on average, with a level of Fusarium dominance nearly identical to the modest level observed in Cluster 1. Cluster 3 exhibited the highest severity outcomes on average, minimal symptom localization in the panicle, and relatively high Fusarium dominance in the disease complex. This cluster, therefore, is likely to have the highest affinity for Fusarium infection and a high likelihood of severe SGM symptom outcomes and probable mycotoxin accumulation. Cluster 4 also has relatively high Fusarium dominance but exhibited less severe disease outcomes than Cluster 3.

Download as PowerPoint

Genome-wide associations and candidate gene identification

The GWAS led to the identification of 19 SNPs that met the empirically determined threshold (P ≤ 1 × 10^–6) for significant association with the four SGM traits (Fig. 6). At least one significant SNP was located in each linkage group, with the highest densities observed in chromosomes 3 (four SNPs), 4 (three SNPs), and 6 (three SNPs). The traits with the highest number of identified marker-trait associations were PGMSR and IPST (eight SNPs each), which is expected given that these traits are the most complex and likely modulated by many loci in tandem (Magill 2013; Upadhyaya et al. 2013). Notably, these traits also exhibited the highest heritability. Two significant SNPs were identified in association with FSDI, while only one significant SNP was identified in association with IPDLI. All significant marker-trait associations were observed in the 2017 dataset, illustrative of the strong GxE effect on SGM outcomes. In total, the significant SNPs were in LD with 86 gene models (Table 4), constituting a trove of novel exploratory evidence for the genetic underpinnings of SGM disease outcomes.

Download as PowerPoint

Table 4. Summary of significant single nucleotide polymorphisms (SNPs) implicated in marker-trait associations with sorghum grain mold traits

Trait	Year	SNP	Chra	P value	MAFb	Effect	Colocalized gene model(s)
PGMSR	2017	S1_72537074	1	8.11E–10	0.24	0.08	Sobic.001G448000 Sobic.001G448200 Sobic.001G447700 Sobic.001G447800 Sobic.001G447900 Sobic.001G448100
IPST	2017	S8_53116463	8	1.88E–09	0.24	0.03	Sobic.008G118700 Sobic.008G118800 Sobic.008G118900 Sobic.008G119000 Sobic.008G119100
IPST	2017	S6_50581647	6	2.77E–08	0.26	0.03	Sobic.006G143200 Sobic.006G143300 Sobic.006G143400 Sobic.006G143500 Sobic.006G143600 Sobic.006G143700 Sobic.006G143750 Sobic.006G143800 Sobic.006G143900
IPST	2017	S8_18771955	8	4.03E–08	0.14	–0.03	Sobic.008G088200
IPST	2017	S3_56148512	3	5.02E–08	0.39	0.03	Sobic.003G225300 Sobic.003G225500 Sobic.003G225600
PGMSR	2017	S4_6006350	4	6.96E–08	0.14	0.08	Sobic.004G073800 Sobic.004G074100 Sobic.004G074200 Sobic.004G074300
PGMSR	2017	S9_10724870	9	7.64E–08	0.05	–0.12	Sobic.009G079400 Sobic.009G079500 Sobic.009G079600 Sobic.009G079700 Sobic.009G079801 Sobic.009G079900 Sobic.009G079925 Sobic.009G079950 Sobic.009G079975 Sobic.009G080000 Sobic.009G080100
IPST	2017	S6_57172226	6	2.47E–07	0.08	0.05	Sobic.006G226200 Sobic.006G226300 Sobic.006G226400 Sobic.006G226500 Sobic.006G226601 Sobic.006G226700 Sobic.006G226800 Sobic.006G226900 Sobic.006G227000 Sobic.006G227050
PGMSR	2017	S6_50802793	6	4.01E–07	0.12	0.08	Sobic.006G145901 Sobic.006G146000 Sobic.006G146100 Sobic.006G146150 Sobic.006G146200 Sobic.006G146300 Sobic.006G146400
PGMSR	2017	S5_69698481	5	4.33E–07	0.17	0.07	Sobic.005G210200
PGMSR	2017	S10_52495841	10	5.15E–07	0.29	–0.07	Sobic.010G185100
FSDI	2017	S3_42491734	3	5.90E–07	0.38	0.04	Sobic.003G175000
FSDI	2017	S3_42488444	3	6.04E–07	0.35	0.04	Sobic.003G175000
IPDLI	2017	S4_64038624	4	6.33E–07	0.05	–0.15	Sobic.004G301250 Sobic.004G301300 Sobic.004G301400 Sobic.004G301500 Sobic.004G301600 Sobic.004G301650 Sobic.004G301700 Sobic.004G301800 Sobic.004G301900 Sobic.004G302000
IPST	2017	S3_30306837	3	6.60E–07	0.25	–0.03	Sobic.003G170100 Sobic.003G170200
PGMSR	2017	S2_36700052	2	6.85E–07	0.06	–0.12	Sobic.002G147466
PGMSR	2017	S7_26458985	7	7.10E–07	0.32	0.07	–
IPST	2017	S10_12645405	10	7.35E–07	0.06	–0.05	Sobic.010G116200 Sobic.010G116300 Sobic.010G116400
IPST	2017	S4_57556761	4	8.27E–07	0.05	0.06	Sobic.004G225000 Sobic.004G225100 Sobic.004G225200 Sobic.004G225300 Sobic.004G225400 Sobic.004G225450 Sobic.004G225500 Sobic.004G225550 Sobic.004G225600 Sobic.004G225700

^a Chr = chromosome.

^b MAF = minor allele frequency.

View as image HTML

The significant SNPs associated with PGMSR were in LD with 31 candidate gene models in total, including several loci that have been previously linked to disease- or stress-resistance functions. For example, there were 11 gene models colocalized with a significant SNP on chromosome 9 in a region of the genome that has previously been associated with other panicle-related traits in sorghum. The significant SNP is situated within a known QTL for panicle amino acid content (Kimani et al. 2020) and downstream of a QTL for panicle branch ratio (Kong et al. 2014).

Several gene models were also identified in association with eight significant SNPs for the IPST trait. One gene in this region, Sobic.006G143900, has noted orthology to genes in the sugar transporter family in Saccharum spp. (Zhang et al. 2021). A second significant SNP identified within this interval on chromosome 6 was associated with several annotated gene models, including five genes characterized as putative anthocyanidin reductases (Sobic.006G226400, Sobic.006G226700, Sobic.006G226800, Sobic.006G226900, and Sobic.006G227000). At least one of these putative anthocyanidin reductase genes (Sobic.006G226800) has been implicated in biosynthesis of 3-deoxyanthocyanidins after infection with fungal pathogens (Pfeiffer 2017).

The single significant SNP (S4_64038624) associated with IPDLI was located in a genomic region on chromosome 4 that is relatively gene-rich; situated upstream of the important Tan1 locus (Sobic.004G280800), which is involved in pericarp color (Nida et al. 2019); and has known implications for SGM disease outcomes (Nida et al. 2021). Nine annotated gene models colocalized with the significant SNP, including three stress- or disease resistance-related genes also associated with this SNP, including Sobic.004G301700 (similar to putative stress-induced protein sti1), Sobic.004G302000 (PF01657, salt stress response/antifungal), and Sobic.004G301900 (similar to Potyvirus VPg interacting protein, putative, expressed/PROTEIN OBERON 1-RELATED). Gene Sobic.004G301700 has been identified as a hop2 homolog (Tu and Li 2020); this protein is known to play a key part in homologous chromosome pairing and DNA repair in plants (Uanschou et al. 2014).

Marker-trait associations for FSDI were only observed at one locus: A 3.2kb-wide genomic region on chromosome 3 that contained two significant SNPs. Both SNPs colocalized with the same gene construct (Sobic.003G175000) annotated as a respiratory burst oxidase homolog protein (Rboh) (H-related). This gene has previously been identified as a homolog to an Arabidopsis gene encoding an Rboh (Calzado 2019). It has been demonstrated in wheat (Triticum aestivum) that Rboh genes play important roles in stress response and that they are expressed upon infection with important pathogens, including Fusarium spp. (Navathe et al. 2019).

Discussion

This study aimed not only to complement existing literature on the genetic architecture of SGM resistance but also to closely examine the utility of novel phenotypes for understanding host genetic determinants of specific assemblages of fungi in the disease complex. We demonstrated that traits related to aspects of SGM symptom manifestation were variable across diverse sorghum accessions and were significantly associated with genomic markers in GWAS. Our study was conducted over two growing seasons in Florence, South Carolina, U.S.A., where endogenous populations of the mycotoxigenic fungus F. verticillioides contribute substantially to the SGM disease burden and may result in excess mycotoxin accumulation in the sorghum value chain. While SGM is a complex pathosystem and complete resistance is unrealistic to pursue (Das et al. 2012), this study provides exploratory evidence suggesting that it may be possible to manipulate species composition of the disease complex through host genetic improvement, thus limiting the crop’s propensity to be colonized by mycotoxigenic strains.

We observed significant year-to-year variability in each of the mold phenotypes, illustrating the strong influence of environmental effects on symptom outcomes in the SGM disease complex. High temperatures, humidity, sunshine, and wind speeds favor Fusarium and Curvularia molds in the SGM disease complex, but these factors are negatively correlated with infection by other implicated species including Alternaria, Phoma, and Aspergillus (Magar and Kurundkar 2005). Average daily temperatures during the inoculation period in 2019 were around 2°C cooler than in 2017, while humidity was 10% higher in 2019 on average (https://www.wunderground.com/). It is possible that these factors may have influenced the taxonomic composition and differential dominance of Fusarium symptom types across study years, speaking to the complex nature of this pathosystem. Despite year-to-year differences, SGM was prevalent in both seasons and is known to be an important disease in the region.

All traits exhibited broad variability within the sorghum botanical races, and there were few significant pairwise differences between races, indicating that each race comprises accessions across the spectrum of susceptibility and resistance. This is consistent with previous reports that have also found high variability of mold outcomes within botanical races (Bandyopadhyay et al. 1988; Menkir et al. 1996). Across races, caudatum had the lowest PGMSR. While there is substantial morphological diversity within this race, it is noted for being well adapted to stressful environments (Venkateswaran et al. 2019), and the caudatum race has been the primary race utilized for modern breeding efforts that may play a role in the race’s mean resistance level. Durra-type lines had the highest levels of Fusarium symptom dominance. Durra sorghums generally have compact panicles and white seeds and are therefore adapted to environments with low mold pressure (Mann et al. 1983). Our finding that these lines have the highest levels of Fusarium dominance in the disease complex suggests that Fusarium spp. are highly competitive pathogens in susceptible varieties but not necessarily in varieties with more morphophysiological resistance mechanisms.

There was a strong positive correlation between the number of symptom types and the overall severity of mold disease. A plausible alternative would be rapid and total (symptomatic) colonization of a panicle by a single fungus. In this study, we have quantitatively demonstrated that diverse assemblages of molds are not only present in the study environment, but that the coinfection of multiple species is associated with more severe disease outcomes. Investigation of the relationships among panicle and seed characteristics and the range of SGM disease phenotypes yielded important insights into the morphophysiological determinants of symptom manifestation while also affirming the multigenic, quantitative nature of SGM resistance that is observable across botanical races. This study corroborates earlier evidence supporting the roles of panicle architecture and pigmentation in disease outcomes, but it should be noted that pigmentation and panicle architecture also have important trade-offs with respect to marketability and consumer preference, especially presence of pigmented testa/tannins and tenacious glumes (Bandyopadhyay et al. 2008).

Our focus on the role of mycotoxigenic Fusarium in the disease complex has also shed light on the interactions between this taxon and others in SGM. There was a positive correlation between Fusarium symptom dominance and the number of symptom types, suggesting that as the number of species in the assemblage increases, Fusarium is more likely to be the dominant contributor to disease outcomes. This is consistent with what has been observed of grain molds in maize, where F. verticillioides outcompeted Aspergillus flavus when coinoculated and in natural infection (Zorzete et al. 2008). It has also been shown that colonization of maize grain by F. verticillioides and F. proliferatum reduced the presence of A. flavus and Penicillium spp. (Marín et al. 1998). We concluded from these findings that Fusarium spp. is a competitive member of the SGM disease complex in the study environment. A limitation of this study is that the strains of F. verticillioides utilized are not characterized accessions. Thus, the extent to which these findings can be generalized to other production contexts with different fungal populations is not certain. Further investigations shedding light on the interplay between host genetics and fungal genetics may contribute deeper insights and are recommended.

The PGMSR has been used for decades to evaluate sorghum resistance to fungal disease (Menkir et al. 1996; Rodriguez-Herrera et al. 2000; Upadhyaya et al. 2013). Studies based on this phenotype have elucidated the major genetic and morphophysiological characteristics that confer resistance to SGM. For example, Esele et al. (1993) associated grain mold resistance with host genes controlling testa pigmentation and pericarp color. The same PGMSR phenotyping approach was used to identify strong association between mold resistance and kernel traits such as hardness and the concentration of phenolic compounds (Menkir et al. 1996). Resistance to SGM is quantitative and polygenic, owing to complex host-pathogen-environment interactions (Audilakshmi et al. 2005). This complexity is reflected in the genetic architecture of resistance, which is hitherto not completely understood. It is thought that deployment of pathogen-specific R genes may not be effective due to the diversity of causal fungi (Upadhyaya et al. 2013).

Despite the utility of PGMSR as an indicator of SGM disease severity, the interpretive value of PGMSR phenotype data is inherently limited by a lack of contextual information about the composition and distribution of the disease symptoms across panicle space. It has been reported that fungi in the SGM elicit distinct active defense responses in the sorghum host (Little and Magill 2003), suggesting that consideration of specific fungal assemblages in the disease complex could allow for selection based on pathogen-specific outcomes. By enriching the phenotypic perspective on the SGM disease complex with a more nuanced understanding of the distinct manifestations of disease outcomes, there is increased potential for crop improvement efforts to be responsive to the challenges posed by environment-specific fungal assemblages. Further applications of disease complex-sensitive mold phenotyping as described here could shed light on the extent to which distinct fungal ecologies elicit differential defense responses in the sorghum host.

Acknowledgments

We would like to thank field staff and technicians at the Pee Dee Research and Education Center for assisting in field management, harvesting, and providing logistical and technical support. We are also grateful for the assistance with mold phenotyping provided by M. Cardoso and L. Cramer at Cornell University.