Introduction

Commonly found in the global temperate regions, Gibberella ear rot (GER) is a disease of maize that is caused by the fungus Fusarium graminearum. The economic importance of GER is mainly due to the associated mycotoxins (Munkvold 2003; Mesterhazy et al. 2012). F. graminearum produces the two common mycotoxins, deoxynivalenol, and zearalenone, as secondary metabolites during the infection and disease development process; deoxynivalenol can contribute to disease development in maize (Harris et al. 1999). The mycotoxins have been reported to cause sickness when contaminated grains are consumed (Pestka et Pestka and Smolinski 2005; Wu 2007; Sabrova et al. 2010). The use of disease resistant genotypes is, therefore, key to avoid risk of food and feed contamination (Munkvold 2003).

The existence of genetic variability for GER resistance has been previously reported (Reid et al. 1994; Mesterhazy et al. 2012). To maximize the exploitation of this genetic variability, it is necessary to understand the genetic architecture of the trait. There are a number of reports that suggest the presence of different resistance mechanisms based on initial point of disease entry (Reid et al. 1994; Chungu et al. 1996a; Miller et al. 2003). Generally, the fungus enters into the ear through the silk channel or when the husk is damaged by birds or insects, and then disease directly establishes on the kernels (Sutton 1982). Thus, the mechanism by which the plant retards the development of the disease on silk or kernel tissue could be different. Reid et al. (1994) showed differences in resistance levels between genotypes based only on silk characteristics. A more recent study by Kebede et al. (2015) demonstrated differences in silk and kernel resistance based on difference in kernel drydown rate, and they proposed fast kernel drydown rate as one resistance mechanism that can bring selection gains towards development of GER resistance hybrids. These findings underline the need to characterize the resistance for both silk and kernel traits.

Previous efforts to characterize GER resistance indicated that the trait is quantitatively inherited for which the majority of the genetic variation is due to additive gene effects in kernel inoculated experiments (Chungu et al. 1996b); however, some levels of dominance and partial dominance were also reported for silk-related resistance traits (Reid et al. 1994). Heritability estimates range from moderate to high, but, in most cases, resistance is highly affected by the environment and thus testing in multiple environments is required (Reid et al. 1994; Bolduan et al. 2009; Löffler et al. 2010). Despite the high heritability estimate that ensures effective phenotypic selection, the requirement of multi-environment screening as well as artificial inoculation and scoring are resource intensive and is challenging to breeders.

An alternative method to better understand the genetic architecture and improve efficiency in selecting resistant genotypes would be to study the resistance mechanism at the molecular level. Methods presently available for dissecting a trait at the molecular level include, but are not limited to, the use of molecular markers. Currently, the most preferred choice of marker is the Single-Nucleotide Polymorphism (SNP) (Rafalski 2002). Quantitative trait loci (QTL) mapping studies were performed in the past using relatively lower density SNPs generated from array-based genotyping platforms, but recent advances in maize genome sequencing enabled the use of high-density SNP markers (Davey et al. 2011). One such method is known as genotyping-by-sequencing (GBS, Elshire et al. 2011). This method utilizes methylation sensitive restriction enzymes and the next-generation sequencing platform to reduce genome complexity and decrease cost of generating high-density genetic maps (Elshire et al. 2011; Poland and Rife 2012). Not only does GBS provide high-density genetic maps but also the exact chromosomal location of SNPs when sequence data are aligned with a reference genome; therefore, ascertainment bias is decreased, unlike array-based genotyping platforms (Ganal et al. 2012; Frascaroli et al. 2013). Recent utilization of GBS in QTL mapping in maize has been reported by Chen et al. (2014). Additionally, QTLs identified from GBS can potentially be used for further genetic studies, such as to identify candidate quantitative trait genes (QTGs) by comparison of the QTL regions with differentially expressed gene products and thus improve our understanding of the host–pathogen interaction (Ye et al. 2013).

In addition to understanding the mechanism of GER resistance, the use of SNPs linked with the desired trait can be a useful tool in marker-assisted selection (MAS). MAS can potentially reduce the time and cost required to develop resistant genotypes and its effectiveness highly depends on accurate QTL marker prediction. Larger population size and higher marker densities provide better genetic map resolution and tighter linkage between marker and QTL, as well as more accurate estimate of effect sizes for small to medium effect QTLs (Mackay 2001; Holland 2007; Li et al. 2010; Stange et al. 2013). Past efforts to identify QTLs that are involved in GER resistance were able to identify some chromosomal regions associated with the trait (Ali et al. 2005; Martin et al. 2011, 2012; Stange et al. 2013). However, these studies were based on lower population sizes, ranging from 101 to 227 recombinant inbred lines (RILs). Furthermore, the molecular markers utilized to identify the QTLs were not densely populated enough to result in map resolutions that can maximize the effectiveness of MAS.

In the current study, we report detection of chromosomal regions associated with both silk and kernel GER resistance using high-density SNPs from a novel genotyping-by-sequencing method and a relatively large population size (more than 400 RILs) in an effort to identify QTLs with more accurate effect size estimates and relatively smaller SNP marker support intervals. We also assessed the extent of difference between silk and kernel resistance and investigated the relationship between disease resistance and some associated agronomic traits at the phenotypic and molecular levels, further characterizing the mechanism of disease resistance.

Materials and methods

Plant material and field experimental condition

The plant material consisted of 410 RILs that were obtained from a bi-parental cross between inbred lines B73 and CO441. The inbred B73 is a publicly available inbred that has been genome sequenced (Schnable et al. 2009) and is highly susceptible to GER, while CO441 is an GER resistant inbred developed at the Eastern Cereal and Oilseed Research Center (ECORC), Agriculture and Agri-Food Canada, Ottawa, Canada (Reid et al. 2003). The RILs were developed from a single F1 ear of the cross B73 × CO441, starting with 507 plants at the F1 generation, and 410 RILs were subsequently advanced to the sixth generation (F6) using the method of single-seed descent.

The RILs, along with their two parental inbred lines, were evaluated for silk and kernel GER resistance for 3 years 2010, 2011, and 2014, at the ECORC Central Experimental Farm in Ottawa. ECORC is located at 45º23′N, 75º43′W and has an altitude of 93 m above sea level. Annual mean maximum and minimum temperatures for the growing period, i.e., summer season, are 28 and 15 °C, respectively. Average annual rainfall is 940 mm (accessed at http://climate.weather.gc.ca/index_e.html).

In each experimental year, there were two separate experiments for each of the inoculation techniques, silk channel and kernel, that were planted adjacent to one another. One separate experiment was conducted in 2014 with no disease inoculations for agronomic trait assessment. All seven experiments were laid out in an alpha lattice design (Patterson and Williams 1976) with two replications. The plots were 3.8 m in length with spacing of 0.76 m between rows and 0.20 m between plants within a row.

Disease inoculation

We used disease inoculum from two 15-ADON-producing F. graminearum isolates (DAOM 180378 and DAOM 194276; Canadian Collection of Fungal Cultures, Ottawa, ON) that were known for aggressiveness and were able to differentiate between genotypes for resistance to GER (Reid et al. 1993). We prepared the inoculum following the protocol developed by Reid et al. (1993) in which the two fungal isolates were cultured in a modified Bilay’s culture liquid medium under controlled laboratory conditions and spores filtered and quantified to 2.5 × 105 macroconidia/ml. The two fungal isolates were amplified in separate medium and were mixed at the time of inoculation. For silk channel inoculations, we injected 2 ml of the inoculum through the husk into the midpoint of the silk channel above the tip of the primary ear. For kernel inoculations, we applied the same amount of inoculum by piercing the husk and injecting inoculum into kernels in the middle of the ear (Reid et al. 2009). We inoculated 10 uniform plants in the middle of each row at the prescribed stage of silk development, i.e., 6 days post silk emergence (for silk channel inoculations) and 15 days post silk emergence (for kernel inoculations) to inflict disease levels that would maximize genotypic differences (Reid et al. 2002).

Data scoring

We scored the disease severity visually by rating the proportion of visible infection on each ear using a percentage-based rating scale, 0–100 %. Disease level was recorded from each plot at the time of harvest by removing the husk from each ear and recording data while the ear was still attached to the plant. All disease scoring was determined by a single individual for consistency. We measured days to silking, the duration of time from planting until 50 % of plants exhibit silk emergence, in each plot in both disease and non-disease inoculated experiments. Duration of days to silking lasted for 3 weeks, from the third week of July to the first week of August in each experimental year.

In the non-disease inoculated experiment, we measured husk coverage and ear moisture content at three different stages of kernel development. Husk coverage was recorded just before harvest time using a 1–5 scale according to Betrán and Isakeit (2004), where 1 = tight husks covering the whole ear and extending 2 cm beyond the ear tip, 2 = husks cover the ear tip and extend up to 2 cm, 3 = husks cover up to ear tip only, 4 = ear tip exposed by 1 cm, and 5 = more than 1 cm of the ear tip exposed. Ear moisture content was measured at 5, 8, and 11 weeks after silk emergence directly from the ear using a hand-held moisture meter (Reid et al. 2010). These moisture readings were then used to calculate the area under the drydown curve (AUDDC) using the formula by Yang et al. (2010):

$$AUDDC = \sum\limits_{i}^{n - 1} {\left[ {\left( {MC_{i} + MC_{i + 1} } \right)/2} \right]\left( {t_{i + 1} - t_{i} } \right)},$$

where n was the number of moisture measurements, MC was the moisture content at i and i + 1 recording dates, and t was the date of the measurement. A higher value of AUDDC signifies slower kernel drydown rate and a lower value indicates a faster rate.

DNA sample preparation

Genomic DNA samples were prepared from the 410 RILs and the two parental inbred lines. To extract the DNA, 3.5 cm above ground tissue samples were cut from 9-day-old seedlings grown in a sterilized environment in petri dishes. These samples were approximately 12 mg dry weight after lyophilization. High-quality pure DNA was isolated using the DNeasy Plant mini Kit (Qiagen, Mississaugua, ON). The amount of DNA was quantified using a QuantiFluor® dsDNA system (Promega, Madison,WI). Quality of the extracted DNA was assessed by agarose gel electrophoresis. Each sample of 1 μg DNA was prepared at a concentration of 100 ng/μl and was sent to Cornell University Life Sciences Core Laboratories Center for genotyping-by-sequencing.

Genotyping-by-sequencing and linkage map construction

DNA sequencing was performed using the genotyping-by-sequencing (GBS) technique described in detail by Elshire et al. (2011). Briefly, genomic DNA was fragmented using a type II methylation sensitive restriction endonuclease enzyme, ApeKI that recognizes the cut site CWGC. The fragmented DNA was ligated with barcoded adapters and was amplified using appropriate primers in a polymerase chain reaction (PCR). The amplified fragments were then sequenced on an Illumina Hiseq 2000 instrument (Illumina Inc., San Diego, CA). Raw sequence data were then filtered for quality and aligned with the reference genome (B73 RefGen v2, Glaubitz et al. 2014) using the Burrows-Wheeler Alignment tool (BWA) integrated in the Tassel-GBS pipeline (Glaubitz et al. 2014). Subsequently, SNPs were called and technically missing SNPs were imputed using Fast Inbred Line Library ImputatioN (FILLIN) method described by Swarts et al. (2014).

For the current QTL mapping study, polymorphic SNPs were filtered from the GBS data through three stages of filtering. First-stage filtering was obtained by selecting SNPs which have missing values in less than 10 % of the RILs and a minor allele frequency of 0.4. From a total of 955,690 sequenced sites, 92882 SNPs were filtered using these two criteria in Tassel V5.0 software. In the second stage, the 92882 SNPs were further reduced to 13967 by selecting those SNPs found only within functionally annotated genes in the B73 genome. We selected the SNPs within functional genes only to reduce genotyping error due to insertion and deletion of SNPs in non-genic regions. The third-stage filtering required the construction of haplotype blocks using the 13967 SNPs and then identifying haplotype tagging SNPs. Haplotype blocks were constructed according to Gabriel et al. (2002) where pairwise linkage disequilibrium (LD) or D prime estimates of SNPs within 500 Kbp physical genetic map distance and 95 % confidence bound were used to define blocks. To avoid overlapping, all possible blocks were sorted starting from the largest block and the next block was added if it did not overlap with a block that was already declared. Haplotype tagging htSNPs were then generated according to Bakker et al. (2005) using mean r 2 ≥ 0.8. A total of 1004 htSNPs captured 100 % of the 13967 functional gene SNPs. The SNPs were named after their physical chromosomal location on the B73 reference sequence version 2.

Segregation distortion from the 1:1 ratio was tested for each marker, and 73 htSNPs were found to be significantly distorted and were removed from further analysis. Subsequently, a genetic linkage map was constructed from the data of 410 RILs and 931 filtered htSNPs using IciMapping V 4.0 (Li et al. 2007). To construct the linkage map, htSNPs were grouped based on their physical chromosomal locations on the B73 reference genome V2 (Schnable et al. 2009) and ordered using REcombination Counting and ORDering (RECORD) and COUNT algorithm (Van Os et al. 2005). Genetic map distance was estimated in centi-Morgan (cM) and was based on Haldane mapping function (Haldane 1919). A graphical presentation of the genetic map was done using MapChart (Voorrips 2002).

Field experiment data analysis

To reduce heterogeneity of variance between the different years of testing and meet the normality assumption, disease severity data were transformed using arcsine square root transformation. By considering year of testing as environment, analysis of variance for each year experiment as well as combined over the 3 years was performed using the REML option of the PROC MIXED procedure of SAS 9.3 (SAS Institute, Cary NC). All main and interaction effects were considered random. Linear model followed was

$$Y_{ijkl} = \mu + G_{i} + E_{j} + GE_{ij} + R(E)_{k(j)} + B(RE)_{l(jk)} + \varepsilon_{ijkl},$$

where \(\mu\) was the overall mean, \(G_{i}\) was the effect of line i, \(GE_{ij}\) was the effect of interaction between line i and year j, \(R(E)_{k(j)}\) was the effect of replication k nested within year j, \(B(RE)_{l(jk)}\) was the effect of block l nested within replication k and year j, and \(\varepsilon_{ijkl}\) was the plot error.

Broad-sense heritability (H 2) for E number environments and R number of replicates was estimated according to Hallauer et al. (2010):

$$H^{2} = \frac{{\sigma_{g}^{2} }}{{\sigma_{g}^{2} + \frac{{\sigma_{ge}^{2} }}{E} + \frac{{\sigma_{\varepsilon }^{2} }}{ER}}},$$

where \(\sigma_{\text{g}}^{2}\), \(\sigma_{\text{ge}}^{2}\), and \(\sigma_{\varepsilon }^{2}\) were the genetic, genotype × environment interaction, and error variance component estimates, respectively.

Phenotypic (r p) and genotypic (r g) correlations between the different traits were estimated according to Mode and Robinson (1959) and were based on line means that were calculated by considering lines as fixed effect in the linear model.

QTL analysis

Quantitative trait loci (QTL) analysis was performed using across 3 years entry mean estimates for agronomic traits and back transformed entry means for GER disease severity. To detect the QTLs, we employed the Inclusive Composite Interval Mapping (ICIM), a mapping function that considers all markers simultaneously to compute stepwise regression of the markers (Li et al. 2007). To declare an additive QTL as significant, the logarithm of odds (LOD) score was estimated through 1000 permutation tests (Churchill and Doerge 1994) and a type 1 error rate of α ≤ 0.05. The threshold LOD value we estimated using the above criterion was 3.3. For epistatic QTLs, a fixed threshold value of 5.0 was used. Since we did not detect any epistatic QTLs in our study, the subject is not discussed further in the results. The genotypic variation explained by each QTL was calculated from the ratio: R 2/H 2 where R 2 was the percent phenotypic variation explained by the QTL and H 2 was the broad-sense heritability estimate for the given trait.

Results

Field performance

A population of 410 RILs descended from a single bi-parental cross was evaluated for GER disease resistance for three experimental field seasons conducted in 2010, 2011, and 2014. Although silk emergence was spread over a 3 week interval, climatic conditions during the silking and inoculation period were rather stable enough not to unduly compromise our disease severity data (accessed at http://climate.weather.gc.ca/index_e.html). Additionally, we applied sprinkler irrigation during the inoculation period whenever there was limited rain to create a more uniform and favorable environment for initial disease infection and development. Nonetheless, disease severity levels varied based on the year of testing, and the lowest severities for both silk channel (GERsilk) and kernel (GERkernel) inoculations were recorded in 2014. As temperatures in 2014 were the lowest recorded over the evaluation years, this may have contributed to a less favorable environment for disease growth and development (Table S1).

The two parental inbred lines that were used to develop the RILs showed contrasting values for all traits recorded except AUDDC (Table 1; Fig. 1). As expected, the resistant parent CO441 showed lower disease severity and earlier silking than the susceptible parent B73. The RILs showed disease severity levels ranging from 8.2 to 91.8 % for GERsilk and 12.2 to 82.0 % for GERkernel with means (\(\bar{X}_{RIL}\)) 38.3 and 42.9 %, respectively. The population means for the RILs were statistically similar with the parental inbred means (\(\bar{X}_{\text{Parent}}\)) for disease severity levels but were slightly closer to the parent CO441 for days to silk and husk coverage. In contrast, RILs had AUDDC means outside the range of the two parental inbreds, suggesting trait over-dominance and the presence of transgressive segregation. The difference between \(\bar{X}_{\text{RIL}}\) and \(\bar{X}_{\text{Parent}}\) for days to silk and AUDDC might have been caused by natural selection towards earlier maturity since RILs were developed in a short season location in Ottawa, ON. Almost one-fifth of the population was lost during the advancement from the F1 to F6 generation mostly due to sterility or the lack of male and female flower nicking. The portion of the RIL population that was lost at least partially account for the difference between parental and progeny mean differences. Significant (P < 0.01) genotypic variances (\(\sigma_{\text{g}}^{2}\)) were observed for all traits recorded while genotype-by-environment interaction variance (\(\sigma_{\text{ge}}^{2}\)) was significant (P < 0.01) for disease severity only. Heritability estimates were moderately high for disease severity and AUDDC and were very high for days to silk and husk coverage, indicating the higher contribution of genetic factors to trait expression than the environment.

Table 1 Means, variance components, and broad-sense heritability estimates for Gibberella ear rot (GER) disease severity (percentage of surface ear area with visible infection) in silk channel and kernel inoculated experiments, days to silking, husk coverage, and area under the drydown curve (AUDDC)
Full size table
Fig. 1
figure 1

Frequency distribution for 410 RIL population for GER disease severity and agronomic traits

Full size image

Correlations between disease resistance and agronomic traits

Disease severity for both GERsilk and GERkernel were significantly (P < 0.01, 0.001) but weakly correlated to agronomic traits: days to silking, husk coverage, and AUDDC, except between GERsilk and days to silking (Table 2). GERkernel had relatively higher correlations with agronomic traits than GERsilk. Irrespective of inoculation technique, disease severity had positive correlation with AUDDC, days to silk (only in kernel inoculated experiment), and negative correlation with husk coverage. This implies higher disease severity was observed in those RILs that were slow to drydown, late to silk, and had tight and full husk coverage. Moderately high and positive phenotypic (r p = 0.68) and genotypic (r g = 0.87) correlations were observed for disease severity between GERsilk and GERkernel.

Table 2 Phenotypic (r p , below diagonal) and genotypic (r g , above diagonal) correlations between disease severity in silk channel (GERsilk) and kernel (GERkernel) inoculations, days to silking, husk coverage, and area under the drydown curve (AUDDC)
Full size table

Chromosomal regions associated with disease resistance and agronomic traits

The genetic linkage map constructed from the 931 htSNPs was 1720.8 cM in length and had 1.9 cM average marker-to-marker distance (Table S3). Chromosome 1 had the highest SNP density (0.62 SNPs/cM) and chromosome 3 the lowest (0.32 SNPs/cM). The largest marker-to-marker distance (41.8 cM) was observed on chromosome 3. This resulted in a QTL identified at the same location for GERkernel appear to cover a very large region on the chromosome (Fig. 2) which in turn may reduce the practical applicability of that QTL in a future marker-assisted breeding efforts.

Fig. 2
figure 2

Genetic linkage map of Quantitative trait loci (QTL) associated with Gibberella ear rot disease severity for silk channel (green) and kernel (pink) inoculations, days to silk (red), husk coverage (yellow), and area under the drydown curve (blue). QTL regions are depicted using the 1-LOD confidence interval to the right of each chromosome

Full size image

In our study, QTLs affecting GER resistance were identified on chromosomes 1, 2, 3, 5, 8, and 9 for GERsilk and chromosomes 1, 2, 3, 5, 8, and 10 for GERkernel (Table 3; Fig. 2). These QTLs explained a total of 45.2 and 37.6 % of the phenotypic and 59.4 and 47.0 % of the genotypic variation we observed in silk channel and kernel inoculations, respectively. Favorable alleles for disease resistance were contributed by both parental inbred lines but higher proportions of the phenotypic and genotypic variations were explained by QTLs coming from the resistant parent, CO441. Four QTLs on chromosomes 1, 2, and 8 co-localized at the same region for the two inoculation techniques, while one QTL on chromosome 3 had an overlapping QTL confidence interval although the two QTLs were flanked by different left and right markers.

Table 3 Quantitative trait loci associated with disease severity and agronomic traits in disease inoculated and non-inoculated experiments
Full size table

For the agronomic traits, QTLs were identified in various regions of the maize genome either co-localized with disease resistance QTLs or in their own unique locations. We identified 11 QTLs affecting Days to silking on chromosomes 1, 3, 4, 5, 7, and 8 where two of these QTL regions overlapped with QTLs for GERkernel on chromosome 1 and GERsilk on chromosome 3 (Table 3; Fig. 2). All QTLs for earliness in days to silking except one QTL on chromosome 5 were contributed by only one parent, CO441. Similarly for husk coverage, only one parent, B73 in this case, contributed five of the six QTLs detected for tight and full coverage on chromosomes 1, 4, 6, 7, and 9. Despite the significant correlations between husk coverage and all other traits evaluated, we did not detect any overlapping QTL region for this trait. For AUDDC, six QTLs were detected on chromosomes 1, 3, 6, and 8. From all other traits in the current study, the QTLs detected for AUDDC accounted for the lowest proportion of the phenotypic (28.5 %) and genotypic (35.6 %) variation. A QTL on chromosome 1 shared the same region with GERsilk and GERkernel and one QTL on chromosome 8 shared an overlapping confidence interval with GERsilk only.

Discussion

Resistance to Gibberella ear rot (GER) in maize is a complex trait that is affected by multiple factors. These factors are involved in retarding initial fungal infection and the subsequent spread to healthy tissue. Obviously, the presence of favorable environmental conditions like rain or high humidity coupled with warm temperatures (26–28 °C) during the flowering season are prerequisites for initial disease infection (Munkvold 2003). However, the successful spread of the disease from infected to healthy tissue is also determined by additional factors that are governed by the genetic architecture of the plant. In this paper, we identified QTLs for those genetic factors that contribute to lower disease severity and tried to understand the disease resistance mechanism from two perspectives. The first perspective was assessing the difference between silk and kernel resistance and determining whether resistance was actually due to the involvement of different genes or chromosomal regions in each tissue. The second perspective was investigating the extent of some agronomic traits contribution towards promoting or deterring disease establishment and spread.

Comparison between silk and kernel resistance

The current study identified 10 QTLs for each of the two GER inoculation techniques and four co-localized in the same region on chromosomes 1, 2, and 8. The co-localized QTL detected on chromosome 2, which is the highest effect QTL for GERsilk and second highest for GERkernel, indicated a potential strong effect QTL that can be exploited in future marker-assisted selection efforts. Unlike the above mentioned co-localized GER resistance QTL, one QTL found on the short arm of chromosome 8 was obtained from the susceptible parent B73 suggesting the possibility of combining GER resistance QTLs from both resistant and susceptible inbreds. Similar findings from previous efforts (Ali et al. 2005; Martin et al. 2011, 2012) on favorable alleles found from both susceptible and resistant parent corroborate our results.

In each inoculation technique, out of ten QTLs detected for GER resistance, six were uniquely localized suggesting that GER resistance could be tissue specific. It should be noted that all these unique QTLs may not necessarily be different. Since the unique QTLs on chromosomes 1, 3, and 8 were ≤11 cM apart, there is a chance that these QTLs are the same but localized to different regions due to unforeseen phenotyping errors. Unique QTLs on chromosome 9 for GERsilk and QTL on chromosome 10 for GERkernel are stronger advocates for tissue-specific resistance. Particularly, QTL on chromosome 9 was detected in all three year evaluations even if all three were not exactly in the same location (Fig. S1). The QTLs identified on chromosome 9 might therefore harbor the genes responsible for silk tissue related resistance. This finding underscored what had been reported earlier in the literature (Reid et al. 1994; Chungu et al. 1996a; Munkvold 2003) and validated the existence of tissue-specific disease resistance mechanisms.

Five of the QTLs we identified were detected in different genetic backgrounds in previous studies by Martin et al. (2011, 2012). Two were co-localized QTLs for GERsilk and GERkernel and were found on chromosomes 2 and 8, while the remaining three were uniquely detected on chromosomes 2, 9, and 10 (Fig. 3). This suggested that these QTLs were stable in multiple genetic backgrounds. It is to be noted that the comparison was simply made using the physical location of the flanking SNP markers from our study with the physical and/or genetic location from IBM2 2008 neighbors map of the flanking SSR markers of the aforementioned earlier studies and may not necessarily be precise enough in estimating the exact position of the QTLs. Interestingly, our QTLs were flanked by much narrower chromosomal regions relative to what had been reported earlier showing the benefit of using higher marker density. Presently, research aimed at identifying candidate genes responsible for GER resistance that are located within QTL regions is under way by identifying differential gene expression between disease inoculated and non-inoculated genotypes.

Fig. 3
figure 3

QTLs identified in common with previous studies of four different RIL populations (Martin et al. 2011, 2012). Physical chromosomal locations covered by flanking markers of present GERsilk (turquoise) and GERkernel (green) and earlier studies (dark blue) on B73 reference genome version 2

Full size image

Effect of agronomic traits on disease development

In our study, QTLs identified for days to silking overlapped with QTLs for GERkernel on chromosomes 1 and 3. Similar findings were reported previously where QTL for GER resistance and days to silk co-localized in similar chromosomal regions (Martin et al. 2012). Generally, QTL region overlap may indicate linkage (when two genes affecting different traits co-segregate) or pleiotropic gene effect (when a single gene affects the expression of multiple traits), depending on whether the QTL is contributed by the same or a different parent in the two compared traits. Thus, the QTL overlap on chromosome 1 was contributed by the same parent and may indicate pleiotropic gene effect, whereas on chromosome 3, it was by different parents, indicating linkage. Although the confidence interval overlapped on chromosome 1, GER resistance and days to silking might be different QTLs but still very closely linked to each other since the two traits were flanked with different SNPs. This indicates days to silking could influence selection for the trait but may not necessarily control it. Genotypic correlation estimated in our study was weak with GERkernel and appeared to be non-existent with GERsilk. The difference in correlation between days to silk and disease severity in the two inoculation techniques was attributed to the indirect effect of husk coverage. Silk channel length is directly proportional to husk coverage, and hence, RILs with low husk coverage would have shorter silk channel and appear to have higher disease severity in silk than kernel inoculated experiments. Since days to silking had a negative correlation with husk coverage, the correlation between days to silk and GERsilk would seem non-existent when there was actually a positive correlation between days to silk and disease severity. The positive correlation was supported by the occurrence of an overlapped QTL region for days to silk and disease resistance on chromosome 1. Earlier reports indicated that response to GER resistance is influenced by silk channel length, moisture content, and silk age (Reid et al. 1992, 2002). Further studies on genes responsible for silk channel length and other silk-related traits such as the source and sink dynamics could in part elucidate the mechanism behind GER resistance.

Kernel drydown rate which we measured as AUDDC was reported to have a high correlation with GERkernel and had been recommended as a factor contributing to GER resistance (Kebede et al. 2015). Despite the weak correlation we observed for AUDDC and GER resistance in the current study, we detected common QTL regions on chromosomes 1 and 8. These suggested the involvement of pleiotropic genes affecting the two traits and the potential utilization of these regions in marker-assisted selection for improving both traits simultaneously. The current study might have under estimated the strength of correlation between the two traits when compared with an earlier report by Kebede et al. (2015). This could have been due to the natural selection caused by the location of RIL population development. The short season nature of this location (2750–2800 CHU) may have favored genotypes that mature early and thus shifted the allele frequency towards lower AUDDC, and consequently affected the correlation with GER resistance.

The practical advantage of assessing strength of correlation between traits would be the use of an easily measured trait as an indirect selection criterion for a trait that is difficult to measure. This difficulty could be caused either by low heritability or high resource requirement. Despite the high heritability, GER resistance breeding is highly resource intensive and requires artificial disease inoculation and multiple location and/or year testing. AUDDC on the other hand is a trait measured in routine yield testing trials, particularly in short season regions of northeastern US and Canada and easier to measure by comparison. The use of AUDDC, as an indirect selection criterion or perhaps in an index with major-effect QTLs for GER and phenotypic data from AUDDC, could therefore improve GER resistance breeding by reducing the cost required for setting up independent disease screening experiments.

Our results demonstrated the advantages of using a very large set of SNP markers from genotyping-by-sequencing and a relatively large RIL population in QTL mapping of GER, a highly complex trait that is controlled by many genes with minor effects and highly influenced by environmental and agronomic factors. The advantage of high marker density was observed in lower marker-to-marker distance estimated with average distance <2 cM, the shortest ever reported in GER mapping experiments. Detected QTLs were therefore localized within very small chromosomal regions except for one QTL for GERkernel on chromosome 3 (Fig. 2). Additionally, QTLs identified for GER resistance, which mostly tended to occur in the same region with QTLs for days to silk in previous reports (Martin et al. 2012), were distinctly separated although there were some overlap in their confidence intervals. This showed the power of high marker density resolving the effect of closely linked QTLs (Stange et al. 2013). The size of the mapping population was also an important factor because of higher number of recombination events leading to the identification of more QTLs per RIL population.

Conclusion

The current study reported relationships between GERsilk, GERkernel and agronomic traits, and identified QTLs common and unique for each trait. Moderately high correlation was observed between the two inoculation methods and weak but significant correlation was observed between GER resistance and agronomic traits. Chromosomes 1, 2, 3, 5, 8, 9, and 10 harbored multiple QTLs for GER resistance that could be exploited using MAS in future GER resistance breeding for both GERsilk and GERkernel. Two AUDDC QTLs found on chromosomes 1 and 8 co-localized in the same region with GER resistance, showing the role of this agronomic trait on disease development. Days to silk on the other hand showed some influence on GER resistance because of the occurrence of closely linked QTLs only. Tissue-specific QTLs were discovered for each of the inoculation methods emphasizing the importance of screening GER resistance for silk and kernel resistance separately. QTL mapping using high-density SNP markers and a large mapping population screening for multiple associated traits could therefore greatly improve our understanding of highly complex traits and most importantly optimize breeding efforts.

Author contribution statement

AZK was involved in field evaluation, data analysis, and writing of the manuscript. TW assisted in collection of relevant field data and performed the disease scoring. LJH developed the RIL population. LMR and LJH conceived, designed, and directed the experiments. All the authors read, edited, and approved the manuscript.