A Genome-Wide Association Study for Fumonisin Contamination Resistance in Tropical Maize


 Native genetic resistance has been considered an effective and environmentally safe alternative to control fungal infections and to reduce mycotoxins accumulation in maize grains. Marker-assisted breeding can be used to accelerate the genetic gains for fumonisin contamination resistance. Thus, the objective of this study was to identify quantitative trait loci (QTLs) and candidate genes for resistance to fumonisin contamination in an Embrapa’s tropical maize panel. Two-hundred and five inbreed lines were evaluated in three field trials Brazil in order to quantify fumonisin contamination in maize grains. The lines were genotyped-by-sequencing (GBS), generating 385,654 high-quality polymorphic SNPs. Mycotoxins in the grain samples were isolated using commercial immunoaffinity columns and its concentrations were evaluated by fluorometric technique. Nine quantitative trait loci (QTL) were found associated with fumonisin concentration resistance in maize. Seven candidate genes with annotated functions associated with biosynthetic pathways of pathogen resistance and four genes have not been previously described as related to fumonisins contamination resistance. These findings will be important to better understand the fumonisin contamination resistance and to support the development of SNP markers to accelerate the selection process in tropical maize breeding programs.


Introduction
Maize (Zea mays L.) is an important cereal crop grown worldwide for food, feed and for processed industrial products, being the third most consumed cereal after wheat and rice (FAOSTAT 2018). Maize production faces several leaf, ear and kernel diseases that cause grain contamination by hazardous toxins to animal and human health, in addition to grain yield losses. These diseases are caused by fungi species that attack and invade developing ears and kernels. For example, Fusarium ear rot caused by Fusarium verticillioides (Sacc.) Nirenberg (syn. F. moniliforme Sheldon). This pathogen is the major producer of fumonisins, including fumonisin B1, affecting the grain quality and marketability, besides reducing grain yield by 10  Fumonisins have proven toxicity on animals and possible carcinogenic effects to humans, according to the International Agency for Research on Cancer (IARC 2002). Fumonisin B1, for example, has carcinogenic properties (Gelderblom et al. 1996) and is associated to neural tube birth problems in humans (Missme et al. 1997). In general, tropical countries have more severe infections caused by Fusarium spp., due to its wet and warm weather, even before harvest. Because of the fungi spores can occur on the silks. Native genetic resistance has been successfully used to control several diseases in maize. Thus, in order to understand the genetic architecture of fumonisin incidence in maize grains, some QTL studies were performed for the resistance to Fusarium and fumonisin contamination ( Giomi et al. 2016). For example,  identi ed 15 QTLs for Fusarium contamination, 17 QTLs for fumonisin B1 contamination and candidate genes that could accelerate the development of inbred lines showing reduced disease severity and low fumonisin contamination. A genome-wide association study (GWAS) for fumonisin contamination, based on 256 maize inbred lines and 990,000 SNP markers, found 17 QTLs associated to fumonisin contamination resistance (Bolduan et al. 2019). There are currently few results about genomic regions and genes associated with fumonisin contamination resistance, due to the complex genetic architecture of resistance to fumonisin accumulation that appears to be controlled by many quantitative trait loci of small effect (Bolduan et al. 2019). Thus, in the present study, a GWAS was performed to identify genomic regions associated to fumonisin contamination resistance in tropical maize germplasm. Additionally, inbred lines showing low levels of fumonisin contamination were selected as resistance sources for the Embrapa's tropical maize breeding program.

Genotypes and experimental design
The maize diversity panel was comprised two hundred and ve maize inbred lines, from the breeding program of Embrapa (Brazilian Agricultural Research Corporation) Maize and Sorghum, located in Sete Lagoas, state of Minas Gerais, Brazil (Silva et al. 2019). During the 2014/2015 crop season, the 205 lines were evaluated in three eld trials conducted in the experimental station of Embrapa Mid-North, located in Teresina, state of Piauí, Brazil. Teresina presents the geographic coordinates of 05°05'S latitude and 42°48'W longitude. The climate, characterized as dry sub-humid, mega-thermal, with moderate water surplus in the summer, is located in a semi-arid area. The soil of the experimental area is a sandy loam textured Dystrophic Yellow Argisol. The lines per se were characterized for the resistance to fumonisin concentration, under natural infection, considering a 9 x 9 lattice design with 81 treatments per trial (75 treatments and 6 common checks), with three replicates. Plots were represented by a 4-m row with 0.8 m spacing between rows and an average density of 60,000 plants per hectare. Fertilization was applied at sowing, with 500 kg/ha of the formulated 08-28-16(N-P-K) 0.3 % of Zn, and in the cover was used 200 kg/ha of 44-0-0 (Urea). Weed, pest control and other agronomical practices were performed as recommendedfor maize crop.

Fumonisin determination
The evaluated trait was the fumonisin concentration in the grain (measured in parts per million, ppm). A sample of 500g of maize kernels was nely grounded and a subsample of 10 g was used to quantify the fumonisin concentration in parts per million (ppm). The analyses were performed in the Laboratory of Food Safety at Embrapa Maize and Sorghum. From this subsample, fumonisins were extracted in a solution of 100 mL of water/methanol mixture (20/80) and 5g of NaCl in a blender for 1 min. Afterwards, it was ltered through Whatman paper and an aliquot of 10 mL of ltered extract was diluted with 40 mL of 0.1% phosphate Tween-20 solution (phosphate buffer). Then, the solution was ltered again with a 1.0 mm micro ber lter, and 10 mL of this solution was passed through the FumoniTest column. The column was washed with 10 mL of phosphate buffer solution, followed by a second ow of 10 mL of phosphate buffer. The column content was eluted with 1.0 mL of methanol (HPLC grade), collected and mixed with 1 mL of developer. The fumonisin concentration in the grain was quanti ed in the samples by the FumonitestTM using the Fluorometer VICAM according to the manufacturer's protocols (VICAM, 2015).

Genome-Wide Association Study
Four genome-wide association models were examined for the Embrapa´s tropical maize panel, comprising 205 inbred lines: (i) the "naïve" model using the general linear model (GLM), with no control for population structure and relatedness; (ii) the GLM corrected for population structure, incorporating the scores for rst principal component; (iii) the GLM corrected for population structure, including the scores for the rst and the second principal components; (iv) the mixed linear model (MLM) with kinship (K) matrix. GLM and MLM models were tted using the software TASSEL v5.2.10 (Bradbury et al. 2007). Then, quantile-quantile plots (Q-Q plots) were used to select the best genome-wide association model for controling the detection of false-positive associations. Manhattan plots were drawn using the -log 10 (Pvalue) of the marker-trait associations and the marker physical positions in Megabasis (Mb). A moderate stringency threshold of -log 10 (P-value) equal to 4.0 was adopted to consider SNPs signi cantly associated to fumonisin contamination in maize.

Linkage Disequilibrium and candidate gene selection
The linkage disequilibrium (LD) decay was estimated using the software TASSEL v. 5 Based on the clustering analysis of the kinship matrix, six genetic diversity groups (represented by the colored bars bellow the dendrogram in Fig. 1a) were detected for the 205 Embrapa's maize inbred lines. The lines exhibiting low genetic relationship with the genetic diversity groups were not clustered and represented by the black bar bellow the dendrogram in Fig. 1a. Most of the IBS estimates showed values from 0 to 0.5 in a scale of 0 to 2, in which 2 is the maximum value of genetic similarity (Fig. 1a). The biplot of the rst two principal components (PC) is presented in Fig. 1b, in which the colored points represent the maize lines clustered according to the six genetic diversity groups identi ed by the clustering analysis of the kinship matrix (Fig. 1a). The rst (PC1) and the second (PC2) principal components explained 9% and 3% of the genetic variability, respectively.
The selection of the best GWAS model was based on the Q-Q Plot, representing the distribution of the observed by the expected -log 10 (P-values) under the null hypothesis (Fig. 2). The models corrected by population structure or kinship performed similarly to the naïve model in order to minimize false positives. Thus, the naïve model was selected and used in the association analysis of fumonisin contamination resistance.
Nine quantitative trait loci (QTL) with fumonisin contamination resistance were obtained, considering the threshold of -log10(P-value) equals to 4.0 (Fig. 3). QTLs were found in the bins 2.05, 2.08, 3.06, 4.05, 4.06, 5.01, 5.05, 10.03 and 10.04 (Table 2). It should be noted that the QTL in bin 5.01 looks promising due to the signi cant association, in addition, it has not been previously identi ed in other studies. The supporting intervals for the QTLs ranged from thousands to millions of bp and were located according to the B73 genome v3. Therefore, to investigate candidate genes next to associated QTLs with the fumonisin resistance, we considered the genome-wide LD of the Embrapa's tropical maize panel that tended to decay rapidly to r 2 = 0.1 within 10 kb (Online Resource 1). Were considered QTL when located in the same LD block (Table 2). Based on the LD blocks of each QTL, seven candidate genes were identi ed, showing annotated functions likely related to pathogen resistance ( Table 2, Online Resource 3).
Next, we explored the genetic constitution of inbreed lines selected for fumonisin resistance using a 10% selection pressure for lines more resistant (top) and the 10% of lines more susceptible bottom (Fig. 4a). This analysis was undertaken with the frequencies of alleles related to fumonisin resistance estimated for nine QTLs (Online Resource 4). The highest frequency of favorable alleles (RR -alleles that increase the resistance) were placed in top lines and, the highest frequency of unfavorable alleles (SS) in bottom lines (Fig. 4b).  2008) showed that the selection for Fusarium ear rot resistance is not always successful to reduce fumonisin contamination, requiring more QTL studies to better understand its genetic basis.
It was not possible to conduct out the study of Genotype x Environment (GxE). Giomi et al. (2016) looked at Genotype x Year (G xY) for maize inbreds Fusarium ear rot in two years trials and found G x Y to be small. The GxY variance components were minor compared to those of principal effects (results not shown) indicating that the ranking of genotypes for disease, severity tended to be stable across years and fungal species. Because of that, the three-eld trials conducted were considered similar in the Giomi et al. (Giomi et al. 2016) study. According to Samayoa et al. (2015) in study genome-wide association analysis for fumonisin content in maize kernels the phenotypic mean across environments would nely correspond to genotype performance because genotype x environment signi cant effects have been rather attributed to the heterogeneity of genotypic variances than to the lack of correlation of genotype performance in different environments. Hence, in a condition such as that, as our phenotypic trait was assessed with reasonable precisions based on our heritability estimates, we do not expect dramatic impacts of additional trials, particularly for QTL.  (Table 2) i.e. have not been previously described as related to fumonisins contamination resistance (Fig. 3). Some of these candidate genes colocalized with QTLs shown in Table 2 .03 to contain a large QTL conditioning resistance to several maize diseases. Therefore is important for resistance since common rust resistance genes rp1 and rp5 were found in this bin (Chen et al. 2016, Coan et al. 2018. Several SNPs associated with the candidate genes presented protein domains that have high similarity to the pathogenesis-related proteins and were reported to improve disease resistance. The gene GRMZM2G022213 (208 Mbp, bin 2.08 - Table 2) annotated as zinc nger protein MAGPIE, regulates tissue boundaries cell division and asymmetric cell division (Welch et al. 2007). GRMZM2G051270 gene located in the bin 5.05, 7 Mbp, correspond to a sulfate adenylyltransferase cysteine (Table 2). ATP-S could be involved in plant-tolerance to several abiotic stresses via different Scompounds pathogen responses (Álvarez et al. 2012). S-containing compounds is directly or indirectly modulated/regulated by ATP-S and are involved in plant tolerance to both biotic and abiotic stresses (Anjum et al. 2015). There is a high correlation between fumonisin contamination, linoleic acid content and masking action in maize hybrids with higher oleic to linoleic ratio (Dall'Asta et al. 2012). This masking phenomenon consists of the formation of covalent bonds between the tricarballylic groups of fumonisins and sulfhydryl groups of the side chains of amino acids in proteins. The gene GRMZM2G051270 present the sulfate groups and might be related to the increase of fatty acid composition on fumonisin contamination and the occurrence of hidden fumonisins in maize.
The SNPs linked to candidate genes signi cantly associated with fumonisin resistance could be used as molecular markers to select resistent genotypes and decrease mycotoxin contamination. The unknown genes or not directly involved genes have the potential to be investigated, since they might be involved in resistance, as biochemical and genetic pathways leading to resistance to fumonisin are complex and, for the most part, unknown (Zila et al. 2014). Thus, the QTLs signi cative associated with fumonisin resistance in this study are promising to be useful for whole-genome selection in tropical maize.
The Lines (410399_19_1, 371056_1, 211_0587_5, 552697_F, 2841, L724, L_228_3x45611_x228_3__2_4_x228_3__1_1, 3821095_5) had a higher frequency of favorable alleles for the resistance to fumonisin concentration (70% RR -alleles that increase the resistance) and on average provided the best resistance (Online Resource 4), also suggesting a predominance of the dominance effects in the resistance to fumonisin contamination in tropical maize.
The complex nature of resistance challenged maize breeders to effectively incorporate novel resistance alleles into adapted breeding pools; as a result, most commercial maize hybrids have lower levels of resistance than desired (Bush et al. 2004). Therefore, inbred lines that presented a higher frequency of favorable alleles and lower fumonisin content could be used in future crosses for the generation of resistant hybrids, supporting advances in plant breeding.

Conclusions
Fumonisin contamination in grains of the Embrapa's tropical maize panel showed substantial additive polygenic variation. GWAS identi ed 9 QTLs associated to the fumonisin contamination resistance. Seven candidate genes with annotated functions associated with biosynthetic pathways of pathogen resistance were identi ed within the LD blocks of these QTLs. Furthermore, four genes have not been previously reported in other studies for fumonisin contamination resistance were colocalized with signi cant QTLs. SNP markers located within these candidate genes should be validated and potentially used for introgressing favorable alleles in Embrapa's tropical maize elite lines. Quantile-quantile plots (Q-Q plots) for fumonisin association analysis using the "naïve", PC1, PC1 + PC2, and kinship models. The black line is the expected distribution under the nullity hypothesis. Assuming there are few true associations is expected that the observed P values will almost follow the expected P value