Maize is the most important cereal crop in sub-Saharan Africa (SSA), growing over 35 million hectares, with an average production of over 70 million metric tons of grain (FAOSTAT, 2021). The crop is mainly grown by several million smallholder farmers for their food security, income, and livelihoods across SSA (Prasanna et al., 2020, 2021). The average maize yield in SSA is very low (∼1.7 tons/hectare) compared to the world average (∼5 tons/hectare) due to various abiotic and biotic stresses (Erenstein et al., 2022). Maize in eastern and southern Africa, specifically in highlands, is affected by many fungal diseases, mainly Turcicum leaf blight (TLB), gray leaf spot (GLS), and Fusarium ear rot (Shi et al., 2007; Kibe et al., 2020a; Nyaga et al., 2020; Omondi et al., 2023; Ndlovu et al., 2024). GLS caused by Cercospora zea maydis (Ward et al., 1999) and C. zeina, maydis (Crous et al., 2006; Liu and Xu, 2013) are major threats to maize production in the world (Ward et al., 1999; Katwal et al., 2013; Omondi et al., 2023);. During the 1960s and ‘70s, the disease was first reported in the USA and later spread worldwide and became a key concern for maize production (Liu et al., 2016) with significant economic yield loss. In SSA, the average yield loss exceeds 70% implicating a significant economic and food security concerns (Vivek et al., 2010; Kinyua et al., 2011; Yigrem and Yohannes, 2019).

Pathogens causing GLS are favored by environmental conditions such as high humidity, moderately high temperature, and extended leaf wetness. Initial symptoms of GLS appear on the lower leaves and are progressively observed in the upper leaves later during the season. The characteristic symptoms of mature GLS lesions are gray to tan in color, sharply rectangular, long and narrow, and run parallel to the leaf veins (Latterell, 1983; Katwal et al., 2013). The GLS damage has been linked to loss of photosynthetic capability and premature plant death (Latterell, 1983; Dodd and McGee, 1989). Due to the widespread impact of the disease in eastern and southern Africa and globally, there is a critical need to adopt effective disease management strategies. The main disease management approaches include the application of fungicides, cultural practices, and, most importantly, host resistance mechanisms (Katwal et al., 2013). Many maize breeding improvement programs in eastern and southern Africa resorted to develop and deploy GLS resistant maize germplasm, as other methods such as the application of fungicides are neither affordable and economical nor environment friendly (Menkir and Ayodele, 2005; Nzuve, 2013; Shi et al., 2014). Hence, breeding for resistant germplasm through conventional methods and integrating advanced molecular tools is the most effective method to control diseases and ensure maize-based food security in SSA.

Finding the resistance source of germplasm and understanding the genetic basis of resistance is critical in managing plant diseases. However, this process is still in progress for GLS resistance. Earlier studies indicated that GLS resistance is a complex trait controlled by multiple genes with small to moderate additive effects (Berger et al., 2014; Benson et al., 2015; Kibe et al., 2020a; Omondi et al., 2023) and strongly influenced by the environment (Clements et al., 2000). Quantitative trait loci (QTL) mapping is an effective tool to understand the genetic basis of complex traits like GLS resistance. Previous QTL mapping studies identified several genomic regions that confer resistance to GLS, which helped to understand its genetic architecture (Zhang et al., 2012, 2017; Berger et al., 2014; Xu et al., 2014; Mammadov et al., 2015; Du et al., 2020; Qiu et al., 2021). QTLs on chromosome bin 5.04 and 5.06–07 were consistently detected in different mapping studies and are of interest to improve GLS resistance (Martins et al., 2019). Meta-QTL analysis revealed that bin 8.08 on chromosome 8 possesses a cluster of QTLs and significant consensus QTLs for GLS, TLB, and southern leaf blight (SLB) with less than 5 cM of confidence interval and also found to be associated with two nucleotide-binding site (NBS) family of R genes (Ali et al., 2013). QTL studies with different populations also revealed QTLs on chromosome bin 8.05/8.06 has been detected for GLS as well as for other foliar diseases like common rust and common smut (Bubeck et al., 1993; Kerns et al., 1999). The consistent genomic regions identified in these studies are needed to focus on improving GLS resistance. However, not many studies were done on eastern and southern African-adapted germplasm.

However, QTL mapping has its limitations, like low mapping resolution due to few recombinations in population development, and each mapping population represents only two alleles. In addition, low relatedness between mapping populations and breeding populations also hampers the translation of the identified QTL into breeding targets. On the contrary, the linkage disequilibrium-based genome-wide association study (GWAS) in a set of mapping panels that represents a broad diversity of the germplasm in breeding programs is a powerful tool for dissecting oligogenic and polygenic traits. GWAS panels with genetically unrelated individuals are expected to accumulate a large number of historical recombination events from the past, which can help to overcome the problems related to the lack of recombination events (Mammadov et al., 2015). In maize, GWAS has been widely used to identify the allelic variants that contribute to improve resistance to many of the maize diseases like maize lethal necrosis (Gowda et al., 2015; Sitonik et al., 2019; Nyaga et al., 2020), GLS (Shi et al., 2014; Kuki et al., 2018; Kibe et al., 2020a; Omondi et al., 2023), sugarcane mosaic virus (Tao et al., 2013), maize streak virus (Nair et al., 2015), common rust (Zheng et al., 2018; Kibe et al., 2020b, Nyaga et al., 2020), Tar spot complex (Mahuku et al., 2016), Fusarium ear rot (Liu et al., 2021), head smut (Wang et al., 2012), and TLB (Poland et al., 2011; Rashid et al., 2020; Ndlovu et al., 2024). Even though several research groups reported GWAS on GLS resistance, the studies on screening a large number of lines to identify the best donor lines and conducting GWAS on lines adapted to SSA are seldom. Therefore, the present research was designed to screen a large number of locally adapted, widely used elite lines and conduct GWAS for GLS resistance in SSA.

GWAS is widely used to find trait-linked markers. The application of trait-linked markers in breeding is limited to large effect QTLs or markers. On the contrary, genomic selection (GS) uses genome-wide markers and captures variations explained by both large-effect and small-effect QTLs or markers, which is effective for complex traits like grain yield and moderately complex traits like GLS and TLB (Kibe et al., 2020a). In GS, the training population has genotypes with both phenotypic and marker data and is used to establish prediction models. From the marker effects estimated from the training population, the genomic-estimated breeding values (GEBVs) are predicted in a testing population that has only marker data but no phenotypic data (Meuwissen et al., 2001). By doing this, we can phenotype only selected lines under controlled environments to produce reliable data. Empirical research has shown the advantage of GS for accelerating the genetic gains per unit of time over phenotypic selection. In maize, GS has been widely applied for many traits for inbred line prediction (Zhao et al., 2012; Technow et al., 2013; Beyene et al., 2019, 2021; Ndlovu et al., 2022, 2024; Sadessa et al., 2022) and hybrid performance prediction (Guo et al., 2019; Schrag et al., 2019; Li et al., 2020). These findings demonstrate the potential of GS to help in the selection of elite lines for disease resistance. The objectives of the present study are (i) to evaluate 427 elite lines for 4 years under artificially inoculated conditions and identify the best GLS resistance donor lines, (ii) to identify the genomic regions associated with GLS resistance through GWAS, and (iii) to examine the potential of GS in predicting GLS resistance and other agronomical traits.

A collection of 427 elite inbred lines widely used in eastern and southern Africa subtropical maize breeding programs was assembled. These inbred lines have potential resistance sources to various foliar diseases, including GLS. The information on the pedigree and other details is provided in Supplementary Table S1 . All these lines have been planted to test for their response to GLS under field conditions with artificial inoculation at the KALRO (Kenya Agricultural and Livestock Research Organization) Research Station, Kakamega, Kenya (0°16′N and 34°49′E, 1406 masl) for 4 years (2016, 2017, 2018, and 2019). The monthly average rainfall and temperature (minimum and maximum) in this disease screening period are provided in Figure 1 . The disease screening location is a hot spot for foliar diseases, including GLS. However, climate change has affected temperature and rainfall patterns in the region, leading to inconsistent disease infection and expression. To have accurate disease severity data, each trial was artificially inoculated to ensure uniform disease expression.

Four hundred and twenty-seven (427) entries were planted in a 5 × 86 alpha (α)–lattice design, randomized and replicated two times each year, by using CIMMYT’s field book (Vivek et al., 2010). These inbred lines were planted in one row of four-meter plots with 15 plants per row. The trial was conducted in the main rainy season (April–September). For these trials, two seeds were planted per hill and later thinned to a single plant per hill three weeks after emergence. This was done to ensure a uniform plant density. All standard agronomic practices were applied during the disease screening period.

GLS is one of the serious foliar diseases in maize (Kibe et al., 2020a; Omondi et al., 2023) and is caused by Cercospora zeae-maydis and C. zeina, in Africa C. zeina is more prevalent (Crous et al., 2006; Meisel et al., 2009; Liu and Xu, 2013). With the changing climate and increase in cropping the same crop over a larger area, GLS becomes a serious threat to maize production, particularly in smallholder farmers of SSA. To understand the genetics of GLS resistance, in this study, we selected a set of 427 elite lines from tropical and sub-tropical breeding pipelines adapted to eastern and southern Africa and screened them over 4 years in Kakamega under artificial inoculation of C. zeina.

Among the 427 lines screened for GLS, 14 were identified as resistant lines with a score of <4 on a 1–9 scale ( Table 2 ). These lines represent both the intermediate and late maturity groups, which occupies the major market share in eastern and southern Africa. Among the 14 lines, CKDHL142989 was the best-performing line with a disease severity score of 2.6, followed by CKL14500, which showed a score of 2.9, which is promising to be used as a donor line in GLS resistance breeding. Interestingly, these lines are also tolerant to TLB, which makes them multiple disease-resistant lines. Among the selected 14 lines, CML536, CKDHL120423, CKL14500, CKL14529, and CML540 also showed resistance to TLB (data not shown), suggesting the possibility of using them as donors for resistance to both diseases. CML 536 is also known to be tolerant to drought and low soil nitrogen stress. CML 540 is resistant to MSV, TLB, common rust, early maturing, and tolerance to drought, which provide additional benefits to use these as elite lines in breeding programs. Furthermore, some of these lines may be useful as parents in abiotic stress-tolerant hybrids. For example, line CKL14500, known for GLS resistance, also carries favorable alleles for drought tolerance; it is evident as it derived from two known drought-tolerant lines (CML444 and CML395). The single cross tester (CML444 x CML395) is frequently used as one of the parents for many commercially released drought-tolerant hybrids. Although line CKL14500 has some good trait combinations, it may not be a suitable parent in bi-parental crosses because of its susceptibility to common rust (Sserumaga et al., 2020). Another good line for GLS resistance (CML559) was derived from source germplasm Population 500 (P500), which is resistant to common rust, MSV, and stem borers as well as tolerant to TLB. Another line, CML566 is tolerant to drought with moderate resistance to TLB and MSV. CML574 is a yellow line and is tolerant to MLN, tar spot complex, and fall armyworm and can be used as a donor for multiple diseases and pests. Overall, the identified GLS-resistant lines not only contribute to disease resistance but also carry useful alleles for several economically important traits, including drought tolerance, which makes them multi-trait elite donors.

GLS resistance breeding is influenced by several factors, including genotype, environment, and their interactions. Across environment, analyses revealed significant genotypic and genotype x environment interaction effects for GLS disease severity and AUDPC values ( Table 1 ). The magnitude of variance components for genotype x environment interaction variance was approximately 1.5 times of genotypic variance which indicates the role of both additive and non-additive effects. On the contrary, the magnitude of genotypic variance is more than twice that of genotype x environment interaction variance for flowering time traits like AD and SD, and for PH and EH suggesting the major role of additive effects over non-additive effects. A high magnitude of genotype x environment interaction effect was also observed in an earlier study with the IMAS association panel (Kibe et al., 2020a). The heritability estimates for these traits are also on similar expectations with moderate heritability for GLS disease traits and high heritability for flowering traits. Kibe et al. (2020a) also reported moderate heritability for GLS disease severity traits in biparental populations. GLS disease severity and AUDPC values are negatively and significantly correlated with AD, SD, PH, and EH ( Figure 3 ). This indicates the selection of lines with early flowering and low to medium height has better resistance to GLS over lines with tall and late flowering plants. These correlations are consistent with earlier studies in association panels evaluated in SSA (Kibe et al., 2020a; Omondi et al., 2023).

GWAS results are influenced by population structure, as it influences marker-trait associations, including the false positives in an association mapping panel. We observed low to moderate population structure with PC1 and PC2 explaining 6.53% and 5.56% of variation, respectively. The association panel lines are broadly grouped into two clusters. This was also supported by phylogenetic tree grouping, though three groups were formed, a closer look shows groups 1 and 2 form one major cluster ( Figure 6A ). Similar findings were also observed by Kibe et al. (2020a) in an association panel and Zhao et al. (2012) in a large set of DH lines which comprised tropical and subtropical lines. Further to understand the structure of the panel used in this study, STRUCTURE software was used, where the ad hoc statistics ΔK were used to determine the optimum number of subgroups based on the output log likelihood of data [LnP (D)]. The peaks of the line plot ( Figure 5A ) suggest that the GWAS panel could be broadly divided into two groups or five subgroups. The kinship matrix also suggests moderate structure among the lines used for GWAS ( Supplementary Figure S3 ). The moderate structure observed in the panel with no clear differentiation of major adaptation groups is in anticipation of earlier studies where CIMMYT maize germplasm was not found to have a strong population structure (Wu et al., 2016; Rashid et al., 2020). Several researchers have also reported moderate structure in the tropical maize germplasm (Kibe et al., 2020a; Nyaga et al., 2020; Sadessa et al., 2022). CIMMYT’s germplasm pools and populations are known for high genetic diversity and serve as the source of germplasm for many breeding lines in the tropical and subtropical regions (George et al., 2004; Warburton et al., 2005). This is one of the possible reasons we did not observe any well-defined population structure in this study.

The rate of LD decay indicates the presence of diversity in the selected germplasm or panels. Fast LD decay suggests higher diversity at the nucleotide base level, which might have resulted from the historic recombination events. In temperate maize germplasm, LD decay distance (10–100 kb) is several times higher than that of tropical maize germplasm (5–10 kb, Lu et al., 2011). Romay et al. (2013) found that LD decays much more rapidly in the tropical germplasm to about 1 kb at r ² = 0.2. A rapid LD decay was observed in each chromosome and across the genome in the panel (1.44 Kb at r² = 0.2). The LD decay observed in this study corroborates earlier studies (Rashid et al., 2020; Kibe et al., 2020a), which suggests the presence of sufficient diversity in the selected set of lines for GWAS.

GWAS revealed 13 markers significantly associated with GLS disease severity and 11 markers with AUDPC value ( Table 3 ). Interestingly, though GLS disease severity was positively and significantly correlated with AUDPC values, no common SNPs were detected across GLS traits in GWAS analyses. Nevertheless, there are few regions overlapped in terms of their bin locations ( Table 3 ). Many studies reported QTLs for GLS resistance are distributed in all 10 chromosomes, where most of them explained small to moderate effects, except for a very few that had a major effect of >10% and were used for further fine mapping studies (Du et al., 2020; Zhu et al., 2021; Kibe et al., 2020a; Omondi et al., 2023). For three foliar diseases including, GLS, TLB, and SLB revealed 147 multiple disease resistance mQTLs through meta-QTL analyses and identified bins 3.04–08, 5.04–07, and 8.05–06 as significant regions for resistance to these diseases (Ali et al., 2013). Summarizing the earlier QTL studies revealed five major clusters or hot spots for GLS resistance, namely in chromosome 1 at bin’s 1.05–1.06 (Saghai Maroof et al., 1996; Lehmensiek et al., 2001; Balint-Kurti et al., 2008; Pozar et al., 2009; Xu et al., 2014), on chromosome 2 at bin’s 2.03–2.05 (Bubeck et al., 1993; Saghai Maroof et al., 1996; Zwonitzer et al., 2010; Zhang et al., 2012; Lennon et al., 2016), in chromosome 4 at bin’s 4.05–4.08 (Bubeck et al., 1993; Saghai Maroof et al., 1996; Clements et al., 2000; Balint-Kurti et al., 2008; Zwonitzer et al., 2010; Zhang et al., 2012; Benson et al., 2015; Lennon et al., 2016), in chromosome 5 at bin’s 5.03–5.06 (Bubeck et al., 1993; Clements et al., 2000; Lehmensiek et al., 2001; Zhang et al., 2012; Lennon et al., 2016) and in chromosome 7 at bin’s 7.02–7.03 (Bubeck et al., 1993; Pozar et al., 2009; Zwonitzer et al., 2010; Berger et al., 2014; Benson et al., 2015; Mammadov et al., 2015). Comparison of these hot spots with our results revealed two markers on chromosome 2 (DT2_55324276; DT2_153752700), three markers on chromosome 5 (DT5_77120507; DT5_80046482; DT5_24851058), one marker on chromosome 7 (DT7_100370211) were co-located within these regions. This supports the earlier findings and indicates their consistent association with GLS resistance in different genetic backgrounds. These regions are of potential interest to identify possible potential candidate genes and use them to improve GLS resistance.

The other SNPs, though not fallen into these hotspot regions, are overlapped with a few reported QTLs from other studies (Du et al., 2020; Kibe et al., 2020a; Chen et al., 2021; Qiu et al., 2021; Zhu et al., 2021). For instance, the most significant SNP associated with GLS disease severity in this study was DT3_18370709 at the physical position 18.37 Mb on chromosome 3 (P = 1.37 × 10⁻⁸) was overlapped with “consensus QTL” on bin 3.04 in the IBM2005 map (Shi et al., 2007), and the QTL (qYCM-DS3-1) reported in the RIL population (Chen et al., 2021). Another SNP (DT3_129473727) at bin 3.04 is overlapped with an earlier reported SNP through GWAS (Kuki et al., 2018) and QTL mapping (Du et al., 2020). All three SNPs detected in this study were also overlapped with an earlier study by (Kibe et al., 2020) in the GWAS panel and biparental populations. The remaining markers, especially on chromosomes 9 and 10, appeared to be specific for the current study and new additional sources for GLS resistance.

Flowering traits, both AD and SD, also play crucial roles in selecting GLS resistance. SNP (DT5_80046482) at chromosome 5 is significantly associated with AD and also showed a strong association with GLS disease severity ( Tables 3 , 4 ). Further, we also observed eight bins, namely, bins 3.03, 3.04, 3.05, 4.09, 5.03, 6.04, 7.02, and 9.03 shared SNPs for both GLS disease severity and flowering time traits ( Tables 3 , 4 ). Other agronomic traits like PH and EH are also significantly and negatively correlated with GLS resistance. A comparison of markers detected for GLS disease severity and PH and EH revealed three bins, bins 6.08, 9.05, and 10.04 shared the markers. These findings suggest strong linkage or clustering of markers in certain regions of the genome and selecting early flowering time, and medium height also indirectly helps in improving GLS resistance.

Among the eight candidate genes identified in this study for GLS resistance, one on chromosome 5 (Zm00001d015224) associate with both GLS and AD encodes for Salicylate/benzoate carboxyl methyltransferase ( Table 5 ). Salicylate carboxyl methyltransferase responsible for formation of methyl salicylate which plays important role in signaling for local defense and systemic acquired resistance of plants against pathogens invasion (Koo et al., 2007). Another candidate gene associated with AUDPC value on chromosome 5 (Zm00001d013920) encodes for Phosphatidylinositol 4-phosphate 5-kinase which plays a role in plant defense and cellular function (Zarreen et al., 2023). Another candidate gene encodes for receptor homology region transmembrane domain- and RING domain-containing protein involved in transport of storage proteins to protein storage vacuoles. Overall, the identified candidate genes in this study are involved in plant defense and development.

GLS resistance-linked QTLs are reported across all chromosomes. Some of the major effects of QTLs on chromosomes 1, 2, 5, and 8 were also fine-mapped (Liu et al., 2016; Zhang et al., 2017; Du et al., 2020; Qiu et al., 2021). As the number of QTL needed to be considered to achieve maximum resistance to GLS increases, it complicates the success of marker-assisted selection. Unlike GWAS, which identifies the markers linked to the trait of interest, GS calculates the estimated breeding values of the trait/s for genotypes in practical breeding which is used to select superior-performing candidates. GS has a high predictive power since it uses genome-wide markers to predict the breeding value of individuals in the testing population (Meuwissen et al., 2001). The rapid innovation in next-generation sequencing technology able to produce millions of markers with reduced cost of genotyping makes GS a critical method in breeding programs. Currently in maize, GS is used for complex traits like grain yield and drought tolerance (Zhao et al., 2012; Beyene et al., 2019, 2021b). Compared to abiotic stress traits and grain yield, diseases like GLS are relatively less complex, following this expectation, we observed a high prediction correlation for GLS disease severity and a moderate correlation for AUDPC values ( Figure 8 ). These correlations are higher than observed correlations in IMAS association panel for GLS disease severity but on par with correlations observed in DH populations (Kibe et al., 2020a; Omondi et al., 2023). GS on several disease traits showed promising results with a prediction accuracy of as high as 0.70 for TLB (Technow et al., 2013) 0.86 for MLN (Sitonik et al., 2019), and moderate accuracy of 0.46 for Fusarium ear rot resistance (Holland et al., 2020; Kuki et al., 2020) and common rust resistance (Kibe et al., 2020b; Nyaga, et al., 2020). By using a large, related, and improved training population, the prediction accuracy could be greatly elevated as shown for GLS in the previous study, GS accuracy was low-to-moderate with a range of 0.29–0.56 for GLS resistance with a small training population, which was elevated to 0.77 when increasing the diversity and size of the training set (Kibe et al., 2020). Therefore, complementing GS with phenotypic selection is promising to achieve high genetic gain for GLS resistance with optimal resources.

In this study, 427 diverse tropical maize inbred lines were evaluated in 4 years under artificial inoculation of GLS (Cercospora zeina) in Kakamega, Kenya. Wide variation was observed for GLS disease severity and AUDPC values across years, but significantly and negatively correlated with agronomic traits such as flowering time and plant height. We identified fourteen GLS-resistant lines that can be used as either donors or parents in a resistance hybrid breeding program. In SSA, three-way cross hybrids are the final commercial products. Therefore, using lines with moderate resistance to GLS can combine the desirable alleles from all three parents, which also adds up to higher resistant hybrids. From 427 lines a set of 140 lines were genotyped with DART GBS genotyping. Population structure analyses revealed moderate structure in the panel. GWAS analyses revealed 24 SNPs significantly associated with GLS traits. Most of the detected SNPs were also co-located with earlier studies for GLS resistance. This indicates the consistency in the detection of the markers across genetic backgrounds. GS prediction correlations are moderate to high, which opens new avenues to improve breeding for GLS disease resistance with optimum allocation of resources.