Genetic diversity, population structure and AFLP markers associated with maize reaction to southern rust

Bragantia, Campinas, v. 78, n. 2, p.183-196, 2019 ABSTRACT: Maize is one of the species with greater genetic diversity among cereals and possibly the most diverse crop species known. Accessing this variability is essential for maize breeding, allowing breeders to achieve progress of yield increasing, overcome environmental challenges or deal with pests and diseases. Among the maize diseases, southern rust is one of the most important, causing significant losses in yield and presenting severe epidemics worldwide. In the present study, the AFLP technique was applied to analyze population structure and genetic diversity among 145 tropical maize inbred lines, and to test for preliminary evidence of association between AFLP markers and the reaction to southern rust. Disease severity was evaluated in two crop seasons and the accessions were genotyped through AFLP using four primer combinations. PLANT BREEDING Article

ABSTRACT: Maize is one of the species with greater genetic diversity among cereals and possibly the most diverse crop species known.Accessing this variability is essential for maize breeding, allowing breeders to achieve progress of yield increasing, overcome environmental challenges or deal with pests and diseases.Among the maize diseases, southern rust is one of the most important, causing significant losses in yield and presenting severe epidemics worldwide.In the present study, the AFLP technique was applied to analyze population structure and genetic diversity among 145 tropical maize inbred lines, and to test for preliminary evidence of association between AFLP markers and the reaction to southern rust.Disease severity was evaluated in two crop seasons and the accessions were genotyped through AFLP using four primer combinations.

PLANT BREEDING -Article
Genetic diversity, population structure and AFLP markers associated with maize reaction to southern rust INTRODUCTION Maize (Zea mays L.) is one of the species with greater genetic diversity among cereals.The huge diversity is shown by more than 350 races and multiple varieties, which carry many distinct characteristics (Vigouroux et al. 2008), allowing the worldwide crop growing in various environmental conditions (Ranum et al. 2014).
The conservation and identification of the genetic diversity remaining within the germplasm pools of crop species are crucial for the breeding process.Therefore, many efforts have been made over the recent decades in order to assess genetic variability of many species, such as maize, using morphological and molecular markers (Fu 2015).Despite the large genetic diversity of maize, the popcorn germplasm has a narrow genetic base (Ziegler 2001).This lack of diversity is a serious problem in breeding programs and can implicate the capacity of yield increasing, overcome environmental challenges or even in deal with pests or diseases.
One of the most important diseases that affect maize is the Southern rust, a fungal disease caused by Puccinia polysora Underw, which is considered the most destructive maize rust, historically presenting worldwide yield losses (Dudienas et al. 2013;Raid et al. 1988;Rhind et al. 1952).The ideal conditions for P. polysora development involve temperatures around 23 °C and long leaf wetness (Godoy et al. 1999).Thus the disease has been reported mostly in tropical and subtropical areas (Abadassi 2014).
The high variability and the great spreading of P. polysora are confirmed by the documentation of at least ten physiological races.Most of the resistance cases to P. polysora are race-specific and controlled by genes nominated Rpp (Brewbaker et al. 2011).Although the race-specific genes are more practical to work in breeding programs and usually confer a significant level of resistance, the horizontal resistance commonly is more durable and provides protection against greater range of races.Therefore, in tropical areas as Brazil, considering the high incidence and severity of southern rust, the horizontal resistance should be considered as a long-term strategy (Casela and Ferreira 2002).
Understanding the relationship between DNA polymorphism and variation in phenotypes is important for increasing the speed of selective breeding programs.Genome wide association studies (GWAS) can be used to identify significant association between molecular markers across the genome and phenotypic traits in panel of unrelated genotypes.This approach has been successfully applied to identify QTLs and genomic regions related to maize resistance to many important diseases, including southern rust (Zhou et al. 2018).
Although high-throughput genotyping technologies are becoming increasingly cheaper and can provide a vast number of SNPs, according to Zhang et al. (2014) the traditional markers, such as AFLP, can still act as easygoing approaches for many labs.For some studies, such as genetic diversity analysis, the resolution requirement for distinguishing the individuals can be reached by AFLP, and sequencing all genomes would be unnecessary and inflate the costs.Thus the AFLP technique has still been considerably applied in many investigations as genetic diversity access, population structure analysis and also association studies (Achleitner et al. 2008;Dadras et al. 2014;Ebrahimi et al. 2017;Saeed and Darvishzadeh 2017;Sharma et al. 2016) Therefore, the aim of the present study was to analyze the population structure and genetic diversity among 145 tropical maize inbred lines from the germplasm collections of Universidade Estadual de Maringá, and test for preliminary evidence of association between AFLP markers and the maize reaction to southern rust.

Field experiments
Field experiments were conducted in two crop seasons and evaluated in a randomized complete block design with two replicates.We tested 145 maize inbred lines derived from the germplasm collections of Universidade Estadual de Maringá among which 77 were previously classified as common maize and 68 as popcorn.The experiment performed in the first crop season was sown on November 30 th 2014, while the second crop season experiment was sown on January 19 th 2015, both by manual sown without any fungicide application.The experimental units consisted of two six-meter rows with 0.2 m × 0.9 m spacing between plants and rows, respectively.
Both experiments were performed at Iguatemi Research Station in Maringá, Paraná, Brazil, located at 23°25' S and 51°57' W. The region's climate is classified as Cfa following Köppen classification and the soil is categorized as Typical Red Dystrophic Argisol.The crop management and the fertilization were realized following the technical recommendation for maize crop.The climatological data of precipitation and average high and low temperatures were obtained by the weather station of the Iguatemi Research Station.
Regarding the high inoculum pressure of Puccinia polysora in the region, there was no necessity of artificial inoculation (Figure 1).Anyway, aiming to guarantee great inoculum pressure and high dispersion of the pathogen, the genotype IAC 112, previously classified as susceptible, was cultivated within and in the boarders of the experimental area.The southern rust severity evaluation was held by the infected leaf tissue area according to the referential scale proposed by Fantin 5 (1997).On average, five plants per treatment were scored when reached the VT stage.

DNA extraction and AFLP analysis
Around 30 days after emergency, leaves from five plants of each accession were collected in bulk and immediately frozen in liquid nitrogen until the storage at -80 °C.For DNA isolation, 300 mg of leaf tissue were powdered into liquid nitrogen and extracted using CTAB buffer according to Doyle and Doyle (1987) with minor modifications.DNA integrity and concentration was confirmed by electrophoresis in a 1% agarose gel and using the spectrophotometer NanoDrop 2000/2000c (Thermo Scientific, USA).
The AFLP reactions were performed following the protocol described by Vos et al. (1995), with some modifications.Approximately 700 ng of DNA from each accession was doubly digested with EcoRI and MseI enzymes (5 U each) in a 20 μL volume system for 18 h at 37 °C.The resulting DNA fragments were ligated with T4 DNA ligase (1 U) to EcoRI (0.5 μM) and MseI (5 μM) adapters in a mix containing 1X T4 DNA ligase buffer, NaCl (0.05 M), BSA (50 ng/μL), and DTT (0.25 mM) in a final volume of 10 μL.The reaction was carried in a thermocycler at 37 °C for 3 h, 17 °C for 30 min, and 70 °C for 10 min and then diluted (1:4) with ultrapure water.
Subsequently, the fragments were amplified using a pair of primers with one selective base in a step known as pre-amplification.This reaction was performed using 3.5 μL of GoTaq®Green Master Mix (Promega, USA), 0.58 μL of the pre-selective primer (4.75 μM), 3.0 μL of the restriction/binding dilution and ultrapure water for a total volume of 10 μL.The pre-amplification program was 2 min at 72 °C, 20 cycles of 1 s at 94 °C, 30 s at 56 °C, and 2 min at 72 °C, followed by a final stage of 30 min at 60 °C.The pre-amplified product was diluted in ultrapure water (1:4).
Finally, selective amplification was carried out using diluted pre-amplified product and four primer pairs with three selective nucleotides (E-ACG/M-CAG, E-AGC/M-CAG, E-ACT/M-CTG and E-AAG/M-CTG) wherein the primers for EcoRI were labeled with fluorophores: NED, PET, VIC and FAM, respectively.In this process, a 2.5 μL aliquot of the previous dilution was mixed with: 0.54 μL of each selective primer (5μM MseI and 1μM EcoRI), 3.5 μLGoTaq® Green Master Mix (Promega, USA) and ultrapure water, in a final volume of 10 μL.For the reactions the thermocycler was programmed as follows: one initial cycle of 2 min at 94 °C, 30 s at 64 °C and 2 min at 72 °C, followed by eight cycles of 1 s at 94 °C, 30 s at 64 °C (as touchdown with 1 °C lowering for each cycle) and 2 min at 72 °C, further 23 cycles of 1 s at 94 °C, 30 s at 56 °C and 2 min at 72 °C, and finally the last step of 30 min at 60 °C.
The products of selective reactions (0.2 μL of each) were subjected to capillary electrophoresis using the automated system ABI 3500 xL Genetic Analyzer (Applied Biosystems, USA) in a solution containing 0.2 μL of size standard 600-LIZ (GeneScan v2.0) and 8.8 μL of highly deionized (Hi-Di) formamide.The results of the electrophoresis were combined in a binary matrix, scored for presence (1) or absence (0) of polymorphic bands, using the software GeneMapper® v.4.1 (Applied Biosystems, USA)

Data analysis
When considered the phenotypic data, since the southern rust severity showed heteroscedasticity between classes according to Bartlett's test and non-normality of residuals according to Shapiro-Wilk's test, the data were transformed by the expression: y = √x + 0.5.
Analysis of variance (ANOVA) was conducted to determine the main effects and interactions using F test (p < 0.05).The following mixed model equation was used (Eq.1): where Y ijk is the severity measurement for the i th accession, in the j th crop season, on the k th block; μ is the general mean; G i is the random effect of the i th accession; E j is the fixed effect of the j th crop season; R/E jk is the random effect of the k th block within the j th crop season; GE ij is the random effect of the genotype by crop season interaction of the i th accession and the j th crop season; and ε ijk is the residual effect.The ANOVA, as well as the estimation of the genetic parameters of: broad-sense heritability (H 2 ), genotypic variance (σ 2 g ), genotype by environment variance (σ 2 ge ), environmental variance (σ 2 e ), intraclass correlation coefficient, genotypic coefficient of variation (CV g ), environmental and genotypic and environmental coefficient of variation ratio (CV g /CV e ) were computed using GENES software (Cruz 2006).
Analyses of the AFLP data were performed using the Jaccard genetic distance matrix and Ward's cluster analysis.These analyses were performed using the packages ade4 on R statistical computing environment (R Development Core Team 2008).For the population structure analysis, the program STRUCTURE v2.3.4 was used to identify K discrete subpopulations based on admixture model with correlated allele frequencies.Marker data were coded as 1 or 0 and individuals were treated as haploids to avoid any assumptions about dominance or heterozygotes.Values of K ranging from 1 to 11 were tested, with 11 independent interactions for each grouping.Each numerical solution was optimized setting burn-in of 50,000 repetitions and Markov Chain Monte Carlo (MCMC) simulations of 500,000 iterations.The number of K groups that best fit the data set was determined according to the ∆k value by Evanno method (Evanno et al. 2005).The membership in subpopulations was compared for reliability among replications and subsets.A representative solution was selected, and the membership of each accession in the K subpopulations was designated using a numerical index, establishing the Q matrix.
The pairwise relative kinship matrix (K matrix) of the association panel was obtained following Hardy (2003) by running SPAGeDi v1.4.All negative values between two individuals in the output from SPAGeDi were set to 0 and the matrix was formatted to a text file readable by TASSEL software v.5.2.
The association analysis between AFLP markers and maize reaction to P. polysora was performed in TASSEL v.5.2 using four different approaches.The first approach consisted in using a simple general linear model (GLM), known as naïve, which did not account for population structure or relative kinship as a potential cause of the genotype-phenotype relationship (GLM: G + P).In the second framework, GLM was tested taking population structure matrix (Q) as a cofactor (GLM: G + P+Q).The other two approaches consisted in utilizing mixed linear models (MLM) as suggested by Yu et al. (2006).Both have included the kinship matrix (K) among all accessions as random effect, initially not involving the population structure matrix (MLM: G + P + K) and then containing the kinship matrix plus the structure matrix (MLM: G + P+Q + K).The markers with minor allele frequency (MAF) < 0.05 were not taken into consideration for the analyses.
In this study, two thresholds were considered: Bonferroni test criterion assuming an alpha of 0.05, typically considered very strict; and the Bonferroni-corrected threshold (p = 1/n, where n = marker number), considered moderately conservative and thus widely adopted in the literature (Liu et al. 2016;Wang et al. 2012;Yang et al. 2014).

Southern rust severity
Maize southern rust severity was scored in the diversity panel of 145 inbred lines across two crop seasons, with two replications in each environment.Analysis of variance identified significant effect (p < 0.05) between accessions, crop seasons and for the interaction between accessions and crop seasons (Table 1).The P. polysora severity in the first crop season was lower than in the second (0.96% and 5.11%, respectively).If we consider the frequency distribution of the severity performing class scores as proposed by Fantin (1997), we can see the variability of the association panel for the resistance to southern rust.It is possible to note a great difference in the frequency distribution in each crop season.In the first crop, among the 145 inbred lines evaluated, disease incidence was absent in ten lines which received score 1, and only eight were scored with notes higher than 5.These results showed that in the first crop, despite the high incidence of P. polysora, the disease severity was not high (Figs.2a and 2b).In the second crop only lines A2560-65H23.3-176and A2560-62H23.2-167showed no signal or symptom of southern rust, while 86 had scores superior to 5 (Fig. 2b).
Higher severity of southern rust in the second crop was also reported by Kurosawa et al. (2016) in a field experiment evaluating 37 popcorn inbred under natural inoculation.The climatic conditions more favorable to the pathogen development and its previous presence in the experimental area may have contributed to this occurrence.Precipitation and monthly minimum and maximum temperatures in the first crop experiment (Dec-Feb) were higher than those of the second crop season (Jan-Apr) (data not shown).The high temperatures observed in the first crop might be inadequate to the P. polysora development, exceeding the 25 °C, described as optimum to the pathogen (Godoy et al. 1999;Raid et al. 1988).Furthermore, according to Godoy et al. (2003) late sowings commonly show higher inoculum presence which, associated to ideal weather conditions, can result in severe epidemics of southern rust.
The genetic parameters associated with the reaction of maize to southern rust were estimated to ascertain the genetic variability in the diversity panel and to predict genetic gains from selection.We detect the prevalence of genotypic variance (σ 2 g ) in relation to environmental variance, showing wide genetic differences between the accessions.This vast variability may be result of the existence of resistance genes in some genotypes, allowing the identification of ones with higher levels of southern rust resistance and thus the association studies.
The estimated broad sense heritability (H 2 ) was 83.17%.This result corroborates with other authors who observed H 2 values from 72% to 93% (Kurosawa et al. 2016;Wanlayaporn et al. 2013) and indicates that the resistance to southern rust can be mostly attributed to genetic rather than environmental factors.The high heritability value suggests the possibility of achieving great genetic gains, enabling the phenotypic selection of genotypes resistant to P. polysora.
Considering the genotypic coefficient of variation (CV g ), we observed a value of 33.8%.For southern rust severity, similar values were found by Kurosawa et al. (2016), who asserted that this parameter is important because it allows the comparison of the genetic variability for different traits and helps to more precisely define the breeding strategies.For the best interpretation of the results, CV g might be analyzed in conjunction with CV e by means of the ratio CV g /CV e , to accurately assess the condition of the trait for genetic breeding.In this study we observed a ratio CV g /CV e of 1.11, indicating the presence of wide genetic variability and suggesting good genetic gain with selection.

Genetic diversity and population structure
The four primer pairs used for AFLP analysis resulted in 1008 bands, of which 975 were polymorphic, representing 97% polymorphism.The combinations E-AGC/M-CAG, AAG/M-CTG, E-ACG/M-CAG, E-ACT/M-CTG produced 321, 214, 232, and 241 bands, respectively, what shows the high efficiency of these primers in detecting genetic polymorphisms in maize.
The Jaccard similarity coefficient used to calculate the genetic distance among the 145 maize accessions ranged from 0.51 to 0.84, with a mean distance of 0.74.By Ward's hierarchical clustering analysis, two groups were well defined; clusters I and II, composed predominantly of common maize and popcorn, respectively.Cluster I consisted of 69 accessions of which 68 were previously classified as common maize (Table 2).Cluster II comprised 76 accessions of which 67 were previously defined as popcorn (Table 3).Within the Cluster II was observed the formation of two sub-clusters (Fig. 3).The first sub-cluster involved 47 accessions, all described as popcorn, while the second consisted of 29 accessions, amongst which 20 of popcorn and nine of common maize.This result suggests that the inbred lines belonging to the last sub-cluster share genetic similarities of common maize and popcorn.
The clustering of the popcorn showed that the accessions were not grouped according to their origin.As reported by Kantety et al. (1995) a relatively small number of popcorn inbred lines were developed from flint corn germplasm and selected to popping expansion and quality.Thus, the association pattern of the accessions was possibly a result of the narrow genetic base of the popcorn germplasm (Ziegler 2001).
Based on the simulations performed with the software STRUCTURE and the Δk value methodology proposed  by Evanno et al. (2005), all 145 analyzed genotypes could be split into two groups (Figs.4a and 4b).The percentage of membership for the K clusters for each individual was obtained and utilized to assign the accessions to each group (Fig. 4c).
A genotype was assigned to a cluster when its percentage of membership was higher than 0.7, otherwise it was defined as admixed (Bitocchi et al. 2012;Dadras et al. 2014;Silva et al. 2015).Figure 4 shows that among the 145 accessions, 76 were assigned to group I, being the vast majority previously classified as common maize (Table 2).The group II was composed by 41 genotypes, most of them classified as popcorn (Table 3).Other 28 genotypes were classified as admixture.In this study, the clusters obtained by the Ward's hierarchical clustering analysis and those achieved by structure analysis had high concordance to each other and were capable of establishing two big clusters, one of them predominantly of common maize and another of popcorn.Therefore, the apparent structure might be used to associate with other methods of clustering and showed been fairly consistent.Access the population structure has been frequently applied in many genetic studies as those which aim: to understand the genetic diversity among maize genotypes (Bracco et al. 2016), to determine heterotic groups of maize germplasm lines (Larièpe et al. 2017) and to control false positive associations between marker loci and phenotypic traits (Achleitner et al. 2008;Dadras et al. 2014;Kang et al. 2008;Saeed and Darvishzadeh 2017).

Association Analysis
Significant associations between AFLP markers (MAF > 0.05) and maize response to southern rust for the four tested models are reported in Table 4.The models detected differences in relation to the markers associated to the maize response and to the number of significant associations.Due to the relatively low sample size, which may result in limited statistical power, two thresholds were considered: the usual Bonferroni test (p < 8.82 × 10 -5 ) and a cutoff value with a moderately stringent threshold (p < 1.76 × 10 -3 ).
The independent analysis to each crop season also resulted in different associations.In the second crop season, the number of markers associated to the accessions response to the disease was higher than in the first crop (Table 1).A possible explanation for this phenomenon is the greater levels of severity verified in the second crop, what may be resulted in better representation of the genotypes reaction to southern rust.
Regarding all the models used in this study, the highest rate of significant markers were found by the model GLM: G + P, known as naïve, which did not account for population structure or relative kinship, as a potential cause of the genotype-phenotype relationship.When using this model in the first crop season, we identified two AFLP markers (EactMctg187, EaagMctg36), while in the second crop this number was much higher, and 15 markers were found (EaagMctg4,EagcMcag129,EactMctg18,EactMctg186,EaagMctg90,EactMctg207,EagcMcag117,EaagMctg162,EaagMctg131,EactMctg81,EaagMctg201,EactMctg158,EaagMctg94,EaagMctg148,EaagMctg140) (Table 4).Although the naïve model do not consider the presence of associations caused by population structure and genetic relatedness, it may lead to a considerable number of false positives (Yu et al. 2006).
In the model GLM: G + P+Q, the structure coefficient membership (Q) was incorporated as a cofactor.This model narrowed the number of significant associations when compared to the naïve model, showing the importance of the population structure control in reducing false associations.Using this model, for the first crop season the same two markers (EactMctg187, EaagMctg36) were identified as occurred with the naïve model.Although, for the second crop experiment the number of significant associations were cut down to five (EaagMctg117, EactMctg205, EactMctg18, EaagMctg4, EactMctg169), among which two (EactMctg205, EactMctg169) were not observed when applying the naïve model.
When considering the mixed linear model MLM: G + P+ K, which did not contemplate the population structure information, but contained the information of genetic relatedness by means of kinship matrix (K), a stronger impact was verified over the number of associations than that produced by GLM: G + P+Q.In the model MLM: G + P+ K, for the first crop experiment there were no significant associations, while for the second crop, three markers were found (EactMctg18, EactMctg205, EactMctg169).This result shows that when the kinship matrix was incorporated to the model, it was capable of reducing the inflation of p-values.
Finally, we tested the model MLM: G + P+Q + K, which incorporated both kinship matrix and population structure.This model is considered the most complete and usually results in better fit with the data (Yu et al. 2006).When applying the MLM: G + P+Q + K to the first crop season data, as well as observed in the MLM: G + P+ K, no significant association was found.For the second crop data, the significant markers (EactMctg18, EactMctg205, EactMctg169) were the same verified by MLM: G + P+ K.
If we consider all the models applied in this study, the number of significant marker associations under the moderately stringent threshold was 19, of which two were significant for the first crop season and 17 were significant for the second crop.The more complete model MLM: G + P+Q + K, which fits both population structure and kinship matrix, narrowed the number of significant associations to three, all in the second crop data.The results of association analysis demonstrated that considering both factors, population structure and genetic relatedness, arising from common kinship, is important to control false positive associations between markers and phenotype.This result is in agreement with other association studies using AFLP markers (Achleitner et al. 2008;Dadras et al. 2014;Ebrahimi et al. 2017;Saeed and Darvishzadeh 2017).
Among the 19 significant markers, the majority (13) was associated to maize susceptibility to P. polysora.Between these associations, the markers EactMctg205 (328 bp) and EactMctg169 (293 bp) were highlighted, once they were identified by means of the model MLM: G + P+Q + K and presented the highest effects of susceptibility, with a percentage of phenotypic variance (r 2 ) of 8.4% and 7.8%, respectively (Table 4).
Concerning all the markers associated to maize response to P. polysora, only six (EactMctg18, EactMctg186, EagcMcag117, EaagMctg162, EaagMctg201, EactMctg158) were associated to maize resistance.Among these markers, EactMctg18 (78bp) should be highlighted, once it was associated by the most complete model MLM: G + P+Q + K and had r 2 of 8.2% (Table 4).
Although further validation tests are required, the results of this study provided at least three (EactMctg18, EactMctg205, EactMctg169) interesting candidates for further investigations, considering its inclusion in plant breeding strategies as marker assisted selection.Regarding this application, one of the issues that would need to be determined is if the linkage between the markers and QTLs is small enough and permits the employment of these markers for reliably selection for southern rust resistance.
The main reason for the few (GLM) or the absence (MLM) of significant association for the first crop season data is attributed to the lower severity of southern rust, consequence of the weather conditions that were not favorable to the disease aggressiveness.For the second crop season, the weather conditions were more beneficial to the disease expression resulting in a higher number of significant associations.
It is important to emphasize that for the effective use of AFLP markers in assisted selection, these markers should be converted into sequence characterized amplified region (SCAR).SCAR markers, which are detected by single loci, are less laborious, can be identified in agarose gel, and are more reproducible, which makes these markers suitable for large-scale screening (Wei et al. 2009).

CONCLUSION
Th e AFLP markers are effi cient to access the genetic diversity of maize accessions from the Universidade Estadual de Maringá germplasm, allowing the identifi cation of two clusters composed of common maize and popcorn.Th e mixed linear model, MLM: G + P+Q + K, which fi ts both population structure and kinship, narrowed the number of signifi cant associations, reducing the chance of obtaining false associations.Th ree AFLP markers: EactMctg18, EactMctg169, and EactMctg205 are associated with the maize response to southern rust and are considered promising for further studies, for genetic validation and for carrying out markerassisted selection.

Figure 1 .
Figure 1.Pustules of Puccinia polysora on the upper surface of a maize leaf infected under natural inoculation.

Figure 2 .
Figure 2. Histogram showing the frequency distribution of southern rust severity score across the maize diversity panel for (a) the first and (b) the second crop seasons.

Figure 3 .Figure 4 .
Figure 3. Ward's hierarchical clustering analysis of 145 maize inbred lines, based on Jaccard distances generated by the polymorphism of 975 AFLP markers.

Table 1 .
Analysis of variance for the severity of southern rust and estimation of the genetic parameters of 145 maize inbred lines and two crop seasons.

Table 2 .
Identification, name and origin of 77 common maize accessions from Universidade Estadual de Maringá germplasm collection.

Table 3 .
Identification, name and origin of 68 popcorn accessions from Universidade Estadual de Maringá germplasm collection.

Table 4 .
Portion of phenotypic variance (r 2 ), p-values (p) and size of AFLP markers significantly associated with maize reaction to southern rust based on four statistical models in two crop seasons.
Hardy (2003)l linear model; G = Genotyping data matrix based on AFLP; P = Phenotyping data of southern rust severity; Q = Population structure inferred by STRUCTURE software; MLM = Mixed linear model; K = Kinship matrix estimated according toHardy (2003).** p-value significant for Bonferroni test; * p-value significant for the moderately conservative test.