Characterization of phenylpropanoid pathway genes within European maize (Zea mays L.) inbreds

Background Forage quality of maize is influenced by both the content and structure of lignins in the cell wall. Biosynthesis of monolignols, constituting the complex structure of lignins, is catalyzed by enzymes in the phenylpropanoid pathway. Results In the present study we have amplified partial genomic fragments of six putative phenylpropanoid pathway genes in a panel of elite European inbred lines of maize (Zea mays L.) contrasting in forage quality traits. Six loci, encoding C4H, 4CL1, 4CL2, C3H, F5H, and CAD, displayed different levels of nucleotide diversity and linkage disequilibrium (LD) possibly reflecting different levels of selection. Associations with forage quality traits were identified for several individual polymorphisms within the 4CL1, C3H, and F5H genomic fragments when controlling for both overall population structure and relative kinship. A 1-bp indel in 4CL1 was associated with in vitro digestibility of organic matter (IVDOM), a non-synonymous SNP in C3H was associated with IVDOM, and an intron SNP in F5H was associated with neutral detergent fiber. However, the C3H and F5H associations did not remain significant when controlling for multiple testing. Conclusion While the number of lines included in this study limit the power of the association analysis, our results imply that genetic variation for forage quality traits can be mined in phenylpropanoid pathway genes of elite breeding lines of maize.


Background
Maize (Zea mays L.) is widely used as a silage crop in European dairy agriculture. While breeding efforts in recent decades have substantially increased whole plant yield, there has been a decrease in cell wall digestibility, and consequently feeding value, of elite silage maize hybrids [1,2]. Digestibility of cell walls of forage crops is influenced by several factors, including the content and composition of lignins [3]. Lignins are complex phenolic polymers derived mainly from three hydroxycinnamyl alcohol monomers (monolignols): p-coumaryl-, coniferyl-, and sinapyl alcohol. p-hydroxyphenyl-(H), guaia-cyl-(G), and syringyl units (S), respectively, are derived from these alcohols and polymerize by oxidation to form lignins. In monocots, lignins are predominantly comprised of G and S units [4].
The four brown-midrib (bm) mutants of maize are characterized by a decreased lignin content, an altered cell wall composition, and a brown-reddish colour of leaf midribs. bm1 is caused by a severe decrease in CAD enzyme activity, possibly resulting from a decrease in CAD transcription [9,13], bm3 is caused by a knock-out mutation in the COMT gene [14,15], while the genes underlying the bm2 and bm4 mutations are unknown. Of the four known bm mutants, bm3 exhibits the strongest effect on plant phenotype, including a reduction in total lignin and an altered lignin composition [16]. A positive effect of the bm3 mutant has been observed on intake and digestibility of forage maize [3]. However, inferior agronomic performance such as lodging and lower biomass yield result from this mutation as well, restricting the use of bm3 mutants in maize breeding programs [17]. The bm1 mutant is also characterized by a reduction in total lignin and an altered lignin composition [16]. Characterization of genetic diversity associated with forage quality traits in genes of the phenylpropanoid pathway might facilitate identification of alleles more applicable to breeding programs.
Levels of nucleotide diversity and linkage disequilibrium (LD), and associations to forage quality traits have been reported for several genes involved in the phenylpropanoid pathway [18][19][20][21]. Due to population bottlenecks and selection, LD is generally higher among elite breeding lines than within distantly related germplasm [22]. In agreement with this, extended LD, spanning from hundreds of kb to tens of cM, has been reported among elite inbred lines [23][24][25][26]. Contrasting levels of LD have been observed between genes in the phenylpropanoid pathway. While LD decreased rapidly within few hundred bp at the COMT and CCoAOMT2 loci [20,21], LD persisted over thousands of bp at a PAL locus [18]. The extent of LD is relevant in the context of association (LD) mapping as it determines both the marker saturation necessary for association mapping as well as the possibility to discriminate between phenotypic effects of individual polymorphisms. The first candidate gene-based association mapping study in plants, associating individual dwarf8 polymorphisms with flowering time of maize [27], has been followed by numerous studies in maize [28] and other crop plants [29]. Associations between maize forage quality traits and individual polymorphisms have been reported for the PAL, CCoAOMT2, and COMT genes [18,20,30] as well as for the ZmPox3 maize peroxidase gene, putatively involved in the oxidative polymerization of monolignols [31,32]. Consequently, target sites within The phenyhlpropanoid pathway catalyzing the biosynthesis of monolignols in grasses (modified from Boerjan et al. 2003) Figure 1 The phenyhlpropanoid pathway catalyzing the biosynthesis of monolignols in grasses (modified from Boerjan et al. 2003). Enzymes are shown in bold.
phenylpropanoid pathway genes for functional marker development [33] for forage quality traits have been identified.
In the present study, partial genomic sequences of C4H, 4CL1, 4CL2, C3H, F5H, and CAD were obtained in a set of 40 European forage maize inbred lines. Since European elite material was included in this study, LD was expected to span whole genes. Therefore, sequencing efforts were directed towards obtaining partial sequences of several genes as compared to obtaining the full sequence(s) of one/few genes, the rationale being that this would increase the number of unlinked polymorphisms available for testing by subsequent association analysis in a broader range of materials. The objectives were to (1) examine nucleotide diversity within genes, (2) examine LD within and between genes, and (3) to test for associations between individual polymorphisms and three forage quality traits.

Phenotypic data
Analysis of variance and phenotypic correlations were published previously [18]. Mean phenotypic values for individual lines across five environments ranged from 50.33 to 63.03 for neutral detergent fiber (NDF), 67.23 to 77.98 for in vitro digestibility of organic matter (IVDOM), and 49.59 to 60.99 for digestibility of neutral detergent fiber (DNDF) ( Table 1). The least significant differences between lines were 3.71, 2.69, and 2.70 for NDF, IVDOM, and DNDF, respectively. Heritabilities were 86.5%, 89.5%, and 92.2% for NDF, IVDOM, and DNDF, respectively.

Nucleotide-and haplotype diversity and selection
Partial genomic fragments were amplified for six candidate genes (names in parenthesis refer to identical genes in the MAIZEWALL database [8]): C4H (C4H1), 4CL1 (4CL), 4CL2 (not identified), C3H (C3H), F5H (F5H1), and CAD (Y13733). The resulting alignments were from 461 bp (C4H) to 1,306 bp (4CL1) in length and were based on 16 (F5H) to 40 (C4H) lines ( Table 2). The exonintron structure at individual loci was estimated by alignments to the mRNA sequences from which primers were developed. GENSCAN estimations supported the structures predicted by the alignments and all amplified sequences were predicted to include both coding and non-coding regions. A total of 54 SNPs were identified out of which 25 were non-redundant for discrimination of haplotypes. Total nucleotide diversity (π) ranged from 0.00049 at the CAD locus to 0.01025 at the 4CL2 locus, and Tajima's D did not indicate selection at any of the six loci ( Table 2). The number of haplotypes defined by SNPs ranged from two at the CAD locus (where only one SNP was identified), four at the C4H and 4CL1 loci, five at the F5H locus, to seven at the 4CL2 and C3H loci (Tables 3, 4 , 5, 6, 7, 8).

Intra-and inter-locus linkage disequilibrium
Extended LD was identified at the 4CL1 locus at which all polymorphisms, with the exception of two 1-bp deletions, were in high LD (P > 0.001) across the entire amplified sequence (~1.3 kb; Figure 2). At the C4H, C3H, 4CL2, and F5H loci, breakdown of LD was observed within ~200 bp. Inter-locus LD was examined by estimating LD between SNP haplotypes of the six loci as well as PAL (32 lines), COMT (42 lines), CCoAOMT1 (40 lines) and CCoAOMT2 (34 lines) ( [18,21], unpublished results). This revealed that C4H were in high (P < 0.0001) LD with CCoAOMT2 and intermediate (P < 0.001) LD with CCoAOMT1 and CAD. Significant LD was not observed between any other pairs of loci ( Figure 3). Examining LD between individual SNPs at these three loci pinpointed that a single non-synonymous SNP, changing the 27 th amino acid of the C4H enzyme from Threonine to Serine, was in high LD with several SNPs at the CCoAOMT1 and CCoAOMT2 locus, respectively (data not shown).

Population structure and marker-trait associations
Within Structure we evaluated whether the 40 lines constitute one, two, three, or four subpopulations, respectively. Two subpopulations (K = 2) was the most likely scenario (results not shown). Most lines were estimated to be > 99% Flint or Dent, in agreement with pedigree information. Under the assumption of two subpopulations, four lines showed approximate 3:1(AS27 and AS29) or 1:3 (AS34 and AS39) ratios of genetic background of Dent:Flint.
The estimated population structure matrix was included in the association analysis, performed as GLM analysis in TASSEL. At the 4CL1 locus a 1-bp indel was associated with NDF and IVDOM ( Table 9). The insertion allele was present in only one line (AS18), which exhibits NDF = 61.43 compared to an overall mean of 56.25, and IVDOM = 67.95 compared to an overall mean of 73.30 (Table 1). At the C3H locus, a non-synonymous G/C SNP at position 294 of the alignment was associated with both IVDOM and DNDF. The C allele was present in two lines (AS14 and AS28). While IVDOM and DNDF values for AS14 are slightly below the overall means, AS28 exhibits the lowest overall values for both IVDOM and DNDF, 67.23 and 49.59, respectively (Table 1). At the F5H locus, two nonsynonymous SNPs, at positions 5 and 6 and in complete LD, were associated with NDF. The G and C allele, respectively, of these two C/G SNPs were present in lines AS20 to AS24. The mean NDF value of these five lines is 52.96 compared to an overall mean of 56.25. The line AS23 is differing from the other four lines in this haplotype as it exhibits an NDF value above the overall mean (Table 1).
In addition, two SNPs in the intron region of F5H were associated with DNDF (C/G SNP, position 610) and NDF (C/T SNP, position 817). At position 610 a singleton SNP was present in line AS24, exhibiting the highest overall DNDF value (Table 1). At position 817, the C allele was present in lines AS14, AS15, and AS20 to AS22, the mean of these lines being below the overall mean of NDF. It should be noted that for F5H, only 16 lines was included in the sample. In addition, the SNP at position 817 was genotyped for only 13 lines due to an indel polymorphism in this region. Consequently, this SNP was not included in the haplotype overview (Table 7). No associations with forage quality traits were detected for the 4CL2, C4H, and CAD gene fragments. The associations identified by GLM were validated by the MLM method, which in addition to overall population structure also corrects for finer scale relative kinship. By MLM, significant associations (P < 0.05) of the 4CL1 indel with IVDOM, the C3H SNP with IVDOM, and one F5H intron SNP with NDF were identified (Table 9). No association to DNDF was detected when correcting for both overall population structure and relative kinship. Controlling for multiple testing by the FDR method requires P < 0.005 to reject the hypothesis of no association. One association, identified by GLM analysis, satisfied this constraint: the association of the 1 bp frameshift indel in 4CL1 with IVDOM (P = 0.0017).

Nucleotide diversity and linkage disequilibrium in the phenylpropanoid pathway
In the present study, the partial genomic sequence of six genes putatively involved in the phenylpropanoid pathway has been obtained for 16 to 40 inbred lines of European maize. Population bottlenecks and selection are expected to decrease nucleotide diversity and increase LD at a given locus [22,23]. While selection was not indicated at any of the six loci (Table 2) nucleotide diversity (π) varied considerably between loci, ranging from 0.00049 at the CAD locus to 0.01025 at the 4CL2 locus. Comparable levels of nucleotide diversity have been reported for other genes of the phenylpropanoid pathway within a similar and overlapping set of lines [18,21] as well as within a more diverse set of lines [20]. Also, a comprehensive study of six genes of the starch pathway of maize revealed similar levels of diversity [34]. Nucleotide diversity at the CAD locus is exceptionally low as compared to other phenylpropanoid pathway genes, with only one SNP identified across 38 genotypes (Table 2). While the CAD sequence is relatively short (~0.5 kb), several SNPs were identified within fragments of similar length for other genes ( Table  2).
Levels of LD varied between loci, spanning the full 4CL1 sequence (~1.3 kb) while decaying within few hundred bps at the C4H, C3H, 4CL2, and F5H loci. Due to population bottlenecks and selection, LD can be expected to be higher among elite breeding lines as compared to more distantly related germplasm. In agreement with this, a rapid LD decay (r 2 < 0.1 within few hundred bps) has been reported for several loci in diverse sets of maize germplasm [35,36] while extended LD, up to tens of cM, has been reported among elite inbred lines [23][24][25][26]. However, extended LD was also observed at the sugary1 locus in a set       [20,21] while spanning more than 3.5 kb at the PAL locus [18], supporting that LD decay is differentiating more between loci than between samples of different origin, e.g., between elite breeding lines and more distantly related germplams. In agreement with this, LD decay at the COMT locus was similar between a diverse set of lines (r 2 = 0.2 within ~250 bp) [20] and a set of elite European breeding lines (r 2 = 0.2 within ~500 bp) [21].
Varying levels of nucleotide diversity and LD between loci could reflect different levels of constraints put on individual loci by selection. It might also be speculated that these parameters would be influenced by length of exons and introns contained in individual gene amplicons. However, levels of nucleotide diversity and LD did not seem to be correlated to exon:intron proportions of individual genes ( Table 2 and data not shown). Nucleotide diversity at the CAD locus, encoding the enzyme catalyzing the last step in monolignol biosynthesis, the reduction of phydroxycinnamaldehydes into their respective alcohols, is found to be exceptionally low, with only one SNP identified across the ~0.5 kb examined in this study. It should be noted that the level of nucleotide diversity identified here might not be indicative for the CAD locus as a whole.
A recent comprehensive study of gene expression in relation to cell wall biosynthesis in maize identified a total of seven CAD gene family members, of which the one examined here was highly expressed in internodes [8]. In addition, reduced CAD enzyme activity and altered lignin content and structure were observed in the bm1 mutant [9], most likely resulting from decreased expression of this and/or other CAD genes [9,13]. Thus, an important role in lignification is indicated for this CAD gene, suggesting selection against detrimental mutations at this locus.
While nucleotide diversity at the 4CL1 locus was found to be ~10 fold higher than for the CAD gene, all SNPs across the 4CL1 locus (~1.3 kb) were in high LD ( Figure 2). This is comparable to the situation at the PAL locus, at which all informative polymorphisms were in complete LD across ~2.5 kb within an overlapping sample of lines [18]. PAL is the first enzyme in several phenylpropanoid pathways, catalyzing the production of a number of phenylpropanoids, including monolignols, from phenylalanine. In Arabidopsis it has been observed that PAL mutants were affected not only in the monolignol pathway, but that also carbohydrate-and amino acid metabolisms were altered [37]. While the 4CL enzyme is further downstream in the phenylpropanoid pathway, it is before the branching of the pathway into monolignol-, flavonoid-and other biosynthetic pathways. The 4CL1 gene investigated in the present study is highly expressed in leaves and young stems of maize [8], indicating an important func-     tion of the enzyme in these tissues. Thus, functional constraints of the enzyme might restrict recombination rates at the gene, resulting in the extended LD observed at the 4CL1 locus.

polymorphisms (SNPs) in the p-coumarate 3-hydroxylase (C3H) gene of maize and average phenotypic values of lines included in individual haplotypes. Numbers denote bp position of individual
While LD decay was rapid within the C4H gene, a single non-synonymous SNP in C4H was in high LD with several SNPs in the CCoAOMT1 and CCoAOMT2 genes, respec-tively. While C4H is located on chromosome 8 (unpublished results), the CCoAOMT1 and CCoAOMT2 genes are located on chromosomes 6 and 9, respectively [20]. It could be speculated that C4H, CCoAOMT1, and CCoAOMT2 are epistatically interacting, i.e., particular allelic variants leading to altered C4H enzymes are dependent on specific allelic properties at the two CCoAOMT loci. An expression-QTL for cell wall biosyn-Linkage disequilibrium between haplotypes of individual genes in the phenylpropanoid pathway in maize Figure 3 Linkage disequilibrium between haplotypes of individual genes in the phenylpropanoid pathway in maize.
thesis genes has been identified at bin 9.04 [38] near CCoAOMT2 at bin 9.02 [20]. Thus, given the limited precision of QTL mapping experiments, it could be speculated that CCoAOMT2 is involved, directly or indirectly, in the regulation of several cell wall biosynthesis genes.

Association of genetic variation in the phenylpropanoid pathway and forage quality
Previous studies have identified associations between phenylpropanoid pathway genes and forage quality traits [18,20,30]. In the present study we have identified associations between polymorphisms in 4CL1, C3H, and F5H and NDF, IVDOM, and/or DNDF, of which DNDF is the most relevant trait in relation to forage quality. No associations were detected for 4CL2, C4H, and CAD polymorphisms. When correcting for multiple testing the association between 4CL1 and IVDOM remained significant. The 4CL1 gene investigated in the present study is homologous to the 4CL1 of Arabidopsis [39] and is highly expressed in leaves and young stems of maize [8]. In Arabidopsis, 4CL1 has been shown to be involved in the biosynthesis of lignin, antisense lines being depleted in G monolignol units [40]. In the present study, a 1-bp indel in 4CL1 was found to be associated with IVDOM by both GLM and MLM. The insertion allele of this indel is present in only one line, AS18, which exhibits the second lowest overall value of IVDOM. The insertion results in a frameshift in the first exon, introducing a premature stop codon four amino acids downstream the insertion. It is thus likely that this indel directly influences the function of the 4CL1 enzyme. In relation to association analysis the situation of a single phenotypically extreme individual is a potential problem. This individual might show numerous specific mutations, which consequently would show association to the phenotype if tested. In the present study, associations based on one/few phenotypically extreme individuals could explain why associations are identified to IVDOM and not to DNDF, and vice versa, in spite of these two traits being highly correlated [18]. Including population structure and kinship into association analysis reduces the number of false positive associations. However, associations based on single phenotypically extreme individuals should be considered with caution and validated in broader plant material.
While not significant when controlling for multiple testing, a non-synonymous SNP in the terminal exon of C3H was associated with IVDOM by both GLM and MLM. The C allele of this G/C SNP was identified in two lines of which AS28 exhibits the lowest overall values of both IVDOM and DNDF. In Arabidopsis [41] and maize [8], a single C3H gene has been identified, which in maize is expressed in relatively low levels in different tissues [8]. A reduced transcription of C3H has been shown to affect lignin content and composition in both Arabidopsis [42] and alfalfa [43]. Moreover, alfalfa lines down-regulated in C3H transcription exhibited increased in vitro dry matter digestibility (IVDMD) [44]. Given a similar function of the C3H enzyme in maize, allelic variation at this locus might directly affect lignin content and composition and in turn digestibility of the maize cell wall.
At the F5H gene, a SNP in the intron was associated with NDF by both GLM and MLM, but not when controlling for multiple testing. Two F5H genes have been identified in both Arabidopsis [45,46] and maize [8,11]. In Arabidopsis [47] and alfalfa [44], plants deficient in F5H transcription exhibit an altered composition of lignin. However, an effect on NDF or IVDMD was not observed in F5H down-regulated lines of alfalfa [44]. In maize, the F5H gene analyzed in this study is highly expressed in young stems and in leaves [8]. It can be questioned if a SNP in the intron region has a causative effect on phenotypic traits. However, this SNP might be in LD with causative variation in other regions of the ORF.
One of the factors greatly affecting the outcome of the association analysis is the choice of method for testing associations (Table 9) [28,48]. By correcting for overall population structure (Q) ten significant associations were identified, five of which were confirmed when including correction for relative kinship (K). No additional associations were identified when correcting for both Q and K as compared to when correcting only for Q. Not surprisingly, a greater number of associations were identified when no correction for population structure was made (data not  shown). It has been shown that the number of false positive associations is reduced when controlling Q and, to a greater extend, when controlling for both Q and K [48]. However, both trait values and causative polymorphisms might be confounded with population structure, e.g., between Flint and Dent lines. In consequence, true positive associations might not be identified by association analysis when considering population structure. To avoid this, it might be beneficial (if possible) to ensure an even distribution of trait values within and between subpopulations in the plant genotype sample employed for association analysis. Independent of the method, sample size is an important factor and, in the present study, a limiting factor in relation to association analyses. Consequently, the associations reported here should be considered indicative and validated in a larger sample of lines before applied in, e.g., breeding programs. In addition, due to the limited extent of LD at four out of six genes, full-length sequences of these genes would likely increase the number of unlinked polymorphisms to be tested for associations.

Deriving functional markers for forage quality traits
With the results presented here, genes encoding most of the enzymes in the phenylpropanoid pathway in maize have been tested for association with forage quality traits.
The PAL gene was investigated in a set of 32 European elite inbred lines, overlapping with the lines used in this study [18]. A 1-bp deletion in the second exon of PAL, introducing a premature stop codon, was associated with high IVDOM. Two CCoAOMT genes have been investigated in a set of 34 diverse lines used in both European and US breeding programs [20]. While no associations were detected for CCoAOMT1, an SSR-like insertion in the first exon of CCoAOMT2 was associated with an increase in cell wall digestibility. For the COMT gene, indel polymorphisms in the intron region have been associated with cell wall digestibility in two different sets of lines, one of which are overlapping with the set employed in the present study [20,30]. Specifically, a 1-bp deletion in a putative splice site recognition site was associated with high cell wall digestibility [20]. Likewise, a MITE insertion in the second exon of the ZmPox3 gene, encoding a peroxidase putatively involved in monolignol polymerization, was associated with high cell wall digestibility [32].
Combining previous results with the results reported in the present study, it seems likely that several genes of the phenylpropanoid pathway can be considered candidate genes for deriving functional markers for forage quality. In addition, it is indicated that useful variation in these genes can be identified even within elite breeding lines of maize, although alleles with larger effects on phenotype might be mined from a broader and larger sample of lines. Polymorphisms in genes encoding enzymes downstream in the phenylpropanoid pathway have been indicated to increase cell wall digestibility [20,30,32], while similar polymorphisms have been associated with increase and decrease in IVDOM for PAL [18] and 4CL1 (Table 9), respectively. It could be speculated that genes more downstream in the phenylpropanoid pathway ( Figure 1) would make more suitable targets for functional marker development in relation to digestibility of the cell wall. Such genes might be more specific to lignin biosynthesis as compared to genes acting earlier in the phenylpropanoid pathway, which could possibly affect several pathways as illustrated by PAL in Arabidopsis [37]. However, the recent identification of gene families in most genes of the phenylpropanoid pathway in maize [8] might suggest that specialization towards biosynthesis of lignin occur earlier in the phenylpropanoid pathway than previously assumed.

Plant materials and phenotypic analyses
A collection of 40 maize inbred lines consisting of 22 Flint and 18 Dent lines were included in this analysis. The line collection is identical to the one published previously [18], with the addition of eight lines. Thirty-five lines were from the current breeding program of KWS Saat AG and five lines were from the public domain (AS01, AS02, AS03, AS39, and AS40 identical to F7, F2, EP1, F288, and F4, respectively;

Analysis of sequence data
The exon-intron structure of the amplified genomic sequences was estimated by alignment to the mRNA sequences used for primer development ( Nucleotide diversity (π), the average number of nucleotide differences per site between two sequences was estimated by using DNASP Version 4.10 [55]. Indel polymorphisms were excluded from the estimates of π .
Tajima's D statistic [56] was also estimated by DNASP to test for selection at individual loci. LD between pairs of polymorphic sites (SNPs and indels, excluding singletons) within and between loci was estimated by the TAS-SEL software, version 1.9.0 [27,57]. Various measurements for LD have been developed [58] of which squared allele frequency correlations (r 2 ) [59] were chosen for our calculations. The significance of LD between sites was tested by Fisher's exact test. For the estimation of inter-locus LD, the alignments of PAL, COMT, CCoAOMT1, and CCoAOMT2 ( [18,21], unpublished results) were included.

Population structure and association analysis
Lines were genotyped with 101 simple sequence repeat markers (SSRs) providing an even coverage of the maize genome. The employed SSR markers are publicly available [60]. Population structure was inferred from the SSR data by the Structure 2.0 software [61,62]. Structure applies a Bayesian clustering approach to group individual lines in subpopulations based on marker profiles. A Q matrix is produced that lists the estimated membership coefficients for each individual in each subpopulation. A burn-in length of 50.000 followed by 50.000 iterations was used. The Admixture model was applied with independent allele frequencies. Data were defined as haploid.
Association analysis was carried out as a general linear model (GLM) analysis in TASSEL to test for associations   Table 1). The Q matrix produced by Structure was included as covariate in the analysis to control for populations structure. All polymorphisms (including singletons) were tested and the Pvalue for individual polymorphisms was estimated based on 10,000 permutations of the dataset. Associations were further tested by the unified mixed model method for association mapping (MLM) in TASSEL [48]. The MLM simultaneously accounts for overall population structure (Q) and finer scale relative kinship (K). Loiselle kinship coefficients [63] between lines (a K matrix) were estimated by the SPAGeDI software [64] based on the SSR data mentioned above. Negative values between two individuals in the K matrix were set to 0 as a negative value indicates that two individuals are less related that random individuals [64]. The diagonal of the K matrix were assigned the value 2. The False Discovery Rate (FDR) method [65] was applied to correct for multiple testing.