The initial mapping population comprised 189 F_{2 : 3} families that originated from a cross between inbred T32 (a highly resistant genotype) and J51 (a highly susceptible genotype). F₂ individuals were developed by planting F₁ seeds in the winter of 2014 in Hainan, the southernmost province of China. Next, an F_{2 : 3} segregation population developed from selfing each F₂ individual was planted in the spring of 2015 in Jiangdong (JD), Mengga (MG), and Mengwen (MW), in Yunnan, a southwest province of China. The F₂ and the F_{2 : 3} families were used to conduct a QTL mapping analysis, which resulted in identifying an anti-disease position (named QTL-qGLS8) on chromosome 8 of maize (Yang, 2016).

Fine QTL mapping was performed by developing 20 BC₃F₂ lines, in which the recombination occurred within the region containing the target QTL. The original cross T32 (donor parent) × J51 (recurrent parent) was made in Hainan during the winter of 2013. Randomly selected F₁ ears of the cross were pollinated by the recurrent parent J51 in the spring of 2014 to develop a backcross population (BC₁F₁). BC₁F₁ individuals were continuously backcrossed with the recurrent parent J51 using MAS until the generation of the BC₃F₁ population in the spring of 2015. Twenty individuals were selected from the BC₃F₁ population using simple sequence repeat (SSR) markers, denoted as D27, D128, D187, D221, D352, D355, D405, D428, D441, D493, D544, D550, D552, D600, D640, D647, D661, D707, D752, and D1123. The selected individuals formed a NIL population (named BC₃F₂-DQT) of 1468 plants, with BC₃F₂-D752 used to develop a NIL micro-population (named BC₃F₂-Micro) containing 180 plants. The NIL population was planted for phenotypic evaluation of GLS resistance in the spring of 2016 in JD, where a GLS epidemic was extreme (Figure 1).

A total of 189 F_{2 : 3} families were evaluated for GLS resistance under natural conditions in two replicates at each of three locations, JD, MG, and MW. A scoring system with five levels (1, 3, 5, 7, and 9) was used to rank GLS resistance. A completely randomized block design with two replications and 20 plants in each plot was established in each of the three sites. All plants were self-pollinated, and 13 plants from the middle of each plot were sampled at 2- and 4-week post-pollination and ranked for GLS resistance (Table 1) (Zhang et al., 2012). The phenotypic evaluation of the BC₃F₂ population was the same as was used for the F_{2 : 3} population.

Statistical analysis of the phenotypic data was performed using IBM SPSS Statistical 20.0 software. The variances for genotype, environment, and interaction between genotype and environment were estimated using Microsoft Office Excel based on a random model. The broad-sense heritability (H²) was calculated as follows: H2=δ2G/(δ2G+δ2GEn+δ2E/nr), where δG2 represents the genetic variance, δGE2 represents the genotype and environment interaction, δE2 represents the error variance, and n and r are the numbers environments and replications, respectively (Knapp et al., 1985).

Genomic DNA was extracted from samples of young leaves of each plant using the cationic detergent cetyl-trimethyl ammonium bromide (CTAB) method (Chen and Ronald, 1999). The genotype of an individual was identified using a 10% non-denaturing polyacrylamide gel electrophoresis. A total of 286 polymorphic markers between T32 and J51 were used to develop the genetic map in JoinMap version 4 (Jacobs et al., 1995) based on publicly available SSR markers retrieved from the Maize Genetics and Genomics Database (http://www.maizegdb.org) (Portwood et al., 2019). The linkage map was constructed from 181 markers across 10 chromosomes. The recombination frequency between linked loci was transformed into the genetic distance (centimorgans, cM) through Kosambi's function.

For the QTL mapping of the BC₃F₂-micro population, QTL-qGLS8 was detected in a CI between SSR markers CSU329 and umc1034 with a physical distance of 1,102 kb according to the maize B73 whole-genome physical map (http://www.maizesequence.org/). Bacterial artificial chromosome sequences within the qGLS8 CI were downloaded from the website. These sequences were used to design new markers between the T32 and the J51 parent lines. The sequences were first scanned using SSRHunter1.3 software (Li and Wan, 2005) to identify SSRs. Primers were then designed using Primer Premier 5.0 software with the following criteria: primers should be ~20 nucleotides in length with a 40–60% guanine-cytosine content, there should be no consecutive tracts of a single nucleotide and no secondary structure. The primer sequences, mapping position, and the amplified length of the eight new markers developed in this study are listed in Table 2.

Inclusive composite interval mapping was performed using QTL IciMapping software (Li et al., 2007) to identify the QTL and estimate their effect. Parameters for forward regression analysis were set with window size and walking speed of 10 and 1 cM, respectively. The significance threshold for QTL detection included 1,000 random permutations of the phenotypic data at the 5% level. The gene action mode of each significant QTL was estimated in accordance with the rate (D/A) of additive (A) and dominant (D) effects and classified as additive (A = 0–0.20), partial dominant (PD = 0.21–0.80), dominant (D = 0.81–1.20), and over-dominant (OD > 1.20) (Edwards et al., 1987; Tuberosa et al., 1998). QTLs were named using a standardized approach, for example, in JDGLS4a, “JD” represents the environment (site) in which the QTL was identified (JD, MG, and MW are abbreviations for the Jiangdong, Mengga, and Mengwen planting sites, respectively); “GLS” represents the name of the trait; the number “4” represents the serial number of the chromosome on which the QTL is located; and “a” represents the serial alpha code of the detected QTL.

A total of 1,468 BC₃F₂ plants were genotyped using 10 SSR markers to construct a small-scale linkage map. Eight of the SSR markers were newly developed. The small-scale genetic mapping of the qGLS8 genes was performed using JoinMap version 4 software to combine the genotypic data of molecular markers with QTL-qGLS8 genes.

The t-tests of the differences in GLS resistance in the two parents within a site were highly significant (P < 0.01; Table 3). GLS resistance in the T32 parent was significantly higher than it was in the J51 parent. The scores for GLS resistance exhibited continuous and approximately normal distributions across the three sites with low skewness and kurtosis, indicating that GLS resistance is a typical quantitative trait (Table 3; Supplementary Figure 1). A transgressive segregation of the resistance score was observed in the populations across the three sites. In addition, variance analysis indicated that the effect of genotype (F_{2 : 3} families), environment, and a genotype/environment interaction on GLS resistance was highly significant (P < 0.01), and that heritability was >80% (Table 4).

The length of the QTL map constructed using the 181 polymorphic SSR markers (Figure 2) was 1153.497 cM, and the average distance between markers was 6.37 cM. The average number of markers per chromosome was 18.1, ranging from 16 on chromosome 2 to 22 on chromosome 6. A few differences in marker order on a chromosome were compared to the IBM 2008 Neighbors Frame 6 (Supplementary Figure 2). These variations in distance and order may be due to the different materials used in the population.

Fourteen QTLs for GLS resistance were detected in F_{2 : 3} families, which comprised five QTLs at JD, four QTLs at MG, and five QTLs at MW (Table 5). Two of the loci were detected in different environments. JDGLS4a, MGGLS4a, and MWGLS4a were located in a common genetic region on chromosome 4 within the Phi076–bnlg1890 marker interval (bin4.11) and explained 5.86, 6.75, and 6.20% of the phenotypic variation in GLS resistance at the JD, MG, and MW sites, respectively. Additive effects of three QTLs on increased GLS resistance were contributed by T32, whose effect values were −0.34, −0.32, and −0.36, respectively. JDGLS8a, MGGLS8a, and MWGLS8a were also located in a common genetic region on chromosome 8 in the umc1130–umc2354 marker interval (bin8.03) and explained 7.89, 14.71, and 11.51% of the phenotypic variation in GLS resistance at the JD, MG, and MW sites, respectively. Additive effects of three QTL on increased GLS resistance were also contributed by T32, whose effect values were −0.22, −0.61, and −0.46 respectively. Compared to qGLS8, the effect value of qGLS4 was lower. Therefore, qGLS8 was used as the main locus for further study. The additive effects of the T32 allele in 86% of the QTL increased the GLS resistance level by 0.22–0.61, compared to the additive effect of the J51 allele. The modes of gene action were A, PD, and OD.

Twelve QTLs were detected for GLS resistance in the joint analysis combining data from all three sites (Table 6). The two common genetic regions detected in the single environment (site) analysis were also detected in the joint analysis within the same marker intervals with the relatively higher log of odd (LOD) values of 8.25 and 10.11. The two QTLs detected in only one environment and the one QTL detected only in two environments (sites) in the single environment analysis were also detected in the joint analysis within the same marker intervals. One common genetic regions detected in the MG and MW sites in the single environment analysis was also detected in the joint analysis, located near the genetic region of QTL-JDGLS7a, which was detected only at the JD site with the same marker (umc1944). Some QTLs detected in the joint analysis across all three environments (sites) were not detected in the single environment analysis. These QTLs probably have a minor effect and are detected when the analysis includes a greater number of environments.

Randomly selected F₁ ears of the original T32 (donor parent) × J51 (recurrent parent) cross were pollinated by the recurrent parent J51 to develop a backcross population (BC₁F₁). BC₁F₁ individuals were continuously backcrossed with the recurrent parent J51 using MAS until the BC₃F₁ generation was obtained. Twenty individuals were selected from the BC₃F₁ population using SSR markers (designated as umc1130 and umc2354). The BC₃F₂ population, developed from selfing each BC₃F₁ individual, was used to identify recombinants for fine mapping. In 20 individuals, the T32 chromosomal segment harboring qGLS8 was present in the genetic background of J51. A total of 180 BC₃F₂-Micro individuals were genotyped using 12 SSR markers located in the target region and evaluated for GLS resistance. Fine mapping confirmed a QTL peak between markers CSU329 and umc1034 with an LOD score of 4.55 and an R² of 11.11% (Table 7). This peak differed from the results of the original F_{2 : 3} mapping (Tables 5, 6), which may have been due to the increased SSR marker density used in the fine mapping and the elimination of interference from the genetic background.

The phenotypic means of BC₃F₂-Micro individuals were compared among the three genotypic classes defined by the allele constitution of markers in the CSU329–umc1034 interval to confirm the gene action of the QTL-qGLS8 identified in this study. The difference in the level of GLS resistance between plants having a homozygous T32 introgression in the target region and those having a homozygous introgression of J51 DNA was 0.61 (a difference of 8.24%). These results confirmed the identification of the primary QTL in the F_{2 : 3} population and demonstrated that the transgressive segregant could be identified in subsequent generations. The mean value of GLS resistance in the homozygous T32 class (7.40) was not significantly different from that of the heterozygous class (7.37); however, both were significantly different from that of the homozygous J51 class (8.01, P < 0.01). These results indicate that the T32 allele was completely dominant over the J51 allele. This analysis implied that the homozygous T32 class and the heterozygotes could be combined for use in the phenotypic analysis during fine mapping, which provided a larger sample size and greater statistical power.

Four newly developed polymorphic SSR markers were added to the existing SSR markers to genotype BC₃F₂-Micro individuals to precisely define the location of the critical breakpoints. A high-resolution genetic linkage map was constructed in the target region. QTL analysis confirmed the highly significant peak between markers ym20 and ym31 with an LOD score of 4.34 and an R² value of 10.51% (Table 7).

In addition to the QTL analysis, an extreme sampling strategy was used to clarify on which side of ym20–ym31 the qGLS8 locus was located. A total of 180 BC₃F₂-Micro individuals were compared genotypically using six of the markers in the target region. Based on these data, 35 recombinants were identified (Figure 3A). The mean phenotypic performance of individuals with and without the T32 introgression was compared, and results indicated that groups B and C were significantly different (P < 0.05). Thus, the ym20–ym31 interval represents a critical region of recombination, with the critical breakpoint falling between ym20 and ym31. This finding further supported the results of the QTL linkage analysis.

A total of 1,468 BC₃F₂ plants were genotyped for the fine mapping of qGLS8 and included the use of four new markers developed to further subdivide the target interval. The QTL analysis confirmed the presence of an extremely significant peak between markers ym20 and ym51 with LOD and R² values of 58.61 and 17.46%, respectively (Table 7). A total of 28 recombinants were identified and grouped into six genotypic classes consisting of 5, 6, 3, 2, 4, and 8 individuals. When the phenotypic means for each class were compared with the T32 and J51 parents (Figure 3B), the most informative comparison was between classes B and C, where the phenotypes (between T32 and J51) were significantly different (P < 0.05). Based on these results, the location of qGLS8 was concluded to be in the recombination area between ym20 and ym51. This conclusion was further supported by the comparison between classes B and E. Consequently, the location of qGLS8 was narrowed down to a 124 kb fragment defined by ym20 and ym51 markers on chromosome 8.

The eight newly developed markers were authentically mapped on bin 8.03 with two of the markers (ym20 and ym51) within and six of the markers located outside of the resistance qGLS8 region. The newly developed markers will greatly facilitate both map-based cloning and MAS breeding for GLS resistance in maize (Figure 4). The 124-kb region harbors seven genes based on the B73 Reference-Gramene_v4 (Tello-Ruiz et al., 2018) and the Maize GDB, including five genes and two lincRNAs (Table 8).