Brazilian cultivated pastures are the basis for the production of beef cattle, and they cover extensive areas, most of which are populated by grasses of the genus Urochloa (syn. Brachiaria) (Jank et al., 2011; Associação Brasileira das Indústrias Exportadoras de Carne [ABIEC], 2016). This genus belongs to the Poaceae family and comprises species of plants distributed in tropical and subtropical regions, mainly in the African continent (Renvoize et al., 1996; Valle et al., 2009).

One of the most widely cultivated species of genus is Urochloa decumbens (Stapf) R. D. Webster, also known as signalgrass. This forage grass has relevant agronomic attributes, including exceptional adaptation to poor and acidic soils that are typical of the tropics, leading to good animal performance (Valle et al., 2010). However, the species is susceptible to several types of spittlebug, including Notozulia entreriana Berg (Hemiptera: Cercopidae), which is the most damaging pest for tropical pastures (Gusmão et al., 2016).

Spittlebugs reduce biomass production, decrease palatability, and reduce the carrying capacity of pastures (Valério and Nakano, 1988). N. entreriana nymphs constantly suck the sap of roots, causing yellowing of the plant, and the saliva of the adults induces phytotoxemia that causes plant death (Valério and Nakano, 1992; Gusmão et al., 2016). Therefore, breeding programs aim to develop new cultivars with reasonable or high resistance to spittlebugs to reduce forage loss. To achieve this advance, it is essential to understand the genetic mechanisms involved in the spittlebug resistance response in forage grass and to identify valuable markers for selection in breeding schemes. Currently, spittlebug resistance in Urochloa grasses is thought to be under relatively simple genetic control and is highly heritable (Miles et al., 1995; Miles et al., 2006). Therefore, despite being a trait with little genetic information, spittlebug resistance can probably be easily manipulated in a plant breeding program.

Genetic breeding of U. decumbens is recent and has proven challenging because this grass is predominantly tetraploid (2n = 4x = 36) and apomictic, reproducing mainly by facultative apospory (Naumova et al., 1999; Valle et al., 2008). In this scenario, the number of currently available molecular genetic resources is limited. Previous molecular studies have increased knowledge of U. decumbens genetics, including the development of sets of microsatellite markers (Ferreira et al., 2016; Souza et al., 2018) and the first transcriptome (Salgado et al., 2017) of the species as well as linkage maps from the interspecific progeny of U. decumbens × U. ruziziensis (Worthington et al., 2016). Other studies with molecular markers have analyzed the genetic relationships of this species to other species in the Urochloa genus (Almeida et al., 2011; Pessoa-Filho et al., 2017; Triviño et al., 2017).

Until recently, it was not possible to perform intraspecific crossings due to the ploidy barriers between apomictic (polyploid) and sexual (diploid) accessions, but this restriction changed when diploid accessions were artificially tetraploidized using colchicine (Simioni and Valle, 2009). This advance allowed the generation of a base population and the exploration of a genetic variability previously conserved by apomixis, favoring new molecular studies, including building genetic linkage maps, which provide useful information about the species.

Genetic maps are essential tools for genetics and genomics research, and they are becoming increasingly common in polyploids species, mainly due to the development of next-generation-sequencing (NGS) technologies and advances in genetic and statistical methods (Garcia et al., 2013). In this study, we utilized the genotyping-by-sequencing (GBS) technique, which enables the discovery of 1000s of SNP at low cost (Elshire et al., 2011; Poland et al., 2012). In addition, robust software tools have been developed for polyploid organisms, including programs that can estimate the allele dosage of SNPs and provide more genetic information to generate linkage maps with high marker densities (Serang et al., 2012; Garcia et al., 2013; Mollinari and Serang, 2015; Bourke et al., 2016, 2018; Gerard et al., 2018).

The mapping process is more complex in polyploid species, mainly because of the larger number of possible genotypes (Ripol et al., 1999; Garcia et al., 2013; Vigna et al., 2016). For example, in autotetraploid species, five possible dosage classes exist: nulliplex (aaaa), simplex (Aaaa), duplex (AAaa), triplex (AAAa), and quadruplex (AAAA). Most polyploid genetic maps have been limited to the use of single-dose markers (SDMs). In this case, either each marker is considered as a single-allele copy from only one of the parents of the cross, with a 1:1 segregation ratio, or the SDMs in both parents segregate in a 3:1 ratio (Wu et al., 1992). This method has been developed and successfully adopted for various forage grass species, such as Panicum maximum (Ebina et al., 2005), Paspalum notatum (Stein et al., 2007), Panicum virgatum (Okada et al., 2010), and Urochloa humidicola (Vigna et al., 2016). However, SDMs represent only a partial sample of loci available in the genome, and not all configurations that can be present in the progeny. All possible offspring genotype configurations in an autotetraploid cross were detailed by Hackett et al. (2013) and have been considered in this work. Knowledge of the dosage of an SNP is essential for genetic studies in polyploid species and can significantly increase the information imparted by each locus (Voorrips et al., 2011; Garcia et al., 2013; Bourke et al., 2016, 2018).

High-resolution genetic linkage mapping for polyploid species can identify beneficial trait loci and allow genomics-based breeding programs (Shirasawa et al., 2017). Molecular markers have been widely used to locate quantitative trait loci (QTLs) associated with quantitative resistance to insects in many crop plants (Samayoa et al., 2015; Oki et al., 2017; Vosman et al., 2018); however, genomic loci related to resistance to spittlebugs have not been determined in any plant species. Particularly in U. decumbens, the identification of loci involved in this trait is a promising tool for characterizing the genetic architecture and can assist in the design of strategies to be introduced into breeding programs in order to increase the efficiency of the selection processes and accelerate the release of new cultivars (Valle et al., 2009).

This study reports the development and application of GBS for mapping studies in an intraspecific progeny of U. decumbens. To the best of our knowledge, there have been no reports of QTL mapping in signalgrass. Our goals were to (i) build a GBS-based integrated genetic map using autotetraploid allele dosage information in a bi-parental progeny, (ii) identify QTLs related to spittlebug resistance on the integrated genetic map, and (iii) search for putative candidate genes that may be involved in spittlebug resistance.

The development of GBS technology enables the identification of thousands of SNP markers across the genome of a given species at a relatively low cost. This advance represents an evolution in genetic studies of polyploid species, since a larger part of the genome can be represented, favoring approaches such as the construction of genetic maps. The map reported here was developed using an intraspecific progeny of U. decumbens of Embrapa Beef Cattle and includes GBS-based markers with autotetraploid allele dosages, an approach never previously employed in a forage grass species. Although not completely elucidated, some evidence suggests that U. decumbens is a segmental allopolyploid (Mendes-Bonato et al., 2001; Worthington et al., 2016), considered to have arisen from hybridization between very closely related species (Stebbins, 1947).

Our molecular markers fit a tetrasomic segregation model (Supplementary Figure S6), probably because autotetraploids, as well as segmental allotetraploids, do not behave as diploids with respect to the mode of inheritance (Bourke et al., 2018). Another explanation is the fact that one of the parents of the mapping population was tetraploidized artificially and therefore behaves as autotetraploid. Therefore, we chose to use all possible SNP configurations to add even more genetic information to the genetic map. Our map provides important genomic information, such as the detection of QTLs involved in spittlebug (N. entreriana Berg) resistance, a relevant trait for breeding programs in the Urochloa genus.

We used the genome of S. viridis, a member of the Poaceae, or grass, family, to align our reads and discover GBS-based markers. S. viridis is a species widely used as a model in genetic studies of grasses of the subfamily Panicoideae (Li and Brutnell, 2011), such as U. decumbens. S. viridis is a phylogenetically related species to U. decumbens that currently has a sequenced genome with a better assembly for use in genetic studies such as ours. This alignment was sufficient to identify a relevant number (4,240) of high-quality SNPs with allele dosage estimated. In previous analyses, we tested other reference genomes for our read alignment, but the S. viridis genome alignment produced the best results in terms of the number of markers identified (Supplementary Table S3), probably due to genetic proximity with U. decumbens.

Worthington et al. (2016) detected 3,912 SNP markers after filtering using the UNEAK (Universal Network-Enabled Analysis Kit) pipeline (Lu et al., 2013), which is similar to the number of markers that we detected using the S. viridis genome as reference. In a previous analysis, we tested the UNEAK pipeline for SNP identification in our intraspecific progeny, but this method detected fewer markers (1,210) compared with alignment to the genome reference (4,240). Most likely, the cross between two different species (U. decumbens × U. ruziziensis) favored the discovery of more SNPs through the UNEAK pipeline by Worthington et al. (2016), in addition to the different methodology used for GBS library construction.

The sequences obtained herein by the GBS libraries are dynamic, allowing the raw data to be reanalyzed as bioinformatics methods evolve (Poland and Rife, 2012), as well as when some tetraploid Urochloa spp. genome is sequenced and assembled. We reinforce that our map can help in this process. Therefore, our sequences may be reanalyzed in the future to obtain a higher alignment rate, as well as to identify a larger number of markers.

Previous polyploid genetic studies have been developed with markers scored as presence/absence of alleles, but this method does not make use of all the information available in the SNP at multiple doses. The GBS data obtained here were sufficient in terms of read depth to call allele dosage in 9,345 markers from the S. viridis genome using SuperMASSA software (Serang et al., 2012; Pereira et al., 2018) (Supplementary Table S3). This methodology provided the distribution of alleles in the progeny and the relative intensities of each allele, increasing the amount of genetic information obtained (Serang et al., 2012; Garcia et al., 2013; Bargary et al., 2014; Pereira et al., 2018).

After analysis using SuperMASSA software (Serang et al., 2012) and subsequent filtering steps, 4,240 SNPs were obtained representing all possible configurations for an autotetraploid species. Approximately 55% (2,345) of the SNP markers identified from the S. viridis genome alignment followed the parental genotype configurations of simplex (AAAA × AAAB/ABBB × BBBB) and double-simplex (AAAB × AAAB/ABBB × ABBB). Slightly less than 1% of the identified markers followed the nulliplex (AAAA × BBBB) and simplex-duplex (AAAB × AABB) configurations (Table 2). These results were satisfactory and relevant for genetic studies with a tetraploid species, but a GBS library construction methodology with more than one restriction enzyme might provide a larger read depth and, accordingly, increase the number of markers detected.

The 4,240 high-quality SNP markers were inserted in TetraploidSNPMap software (Hackett et al., 2017), and 2,725 were identified with distortion from the expected Mendelian segregation ratio. Although numerous studies have shown that markers with segregation distortion can increase the density of the map and help improve QTL detection (Hu et al., 2018), the mapping software used does not allow the inclusion of such markers. Segregation distortion is a common phenomenon that can occur due to non-biological factors such as sampling and genotyping error, or due to biological factors such as gametophyte and/or zygotic selection, hybrid incompatibility, deleterious alleles or chromosomal rearrangements (Manrique-Carpintero et al., 2016; Edae et al., 2017). In our study, crossing of a natural segmental allotetraploid male parent with an artificially autotetraploidized female parent may have caused uneven transmission of the alternative alleles due to reduction of pairing affinity and meiotic abnormalities, causing the appearance of a large number of distorted markers. However, more detailed studies are needed to clarify this hypothesis.

The remaining 1,515 SNPs showed significance in the χ² goodness-of-fit statistic, and 1,000 were placed in the linkage map (Table 1). These 1,000 SNP markers were distributed across nine homology groups, with a total length of 1,335.09 cM and an average map density of 1.33 cM (Figure 2, 3 and Table 1), reflecting the molecular source and mapping methodology.

In Worthington et al. (2016) study, genetic maps for an interspecific progeny resulting from a cross between U. decumbens × U. ruziziensis were built with GBS-based markers, and 1,916 SDA (single-dose allele) markers were distributed across the parental linkage maps. The different mapping method used in our study may have influenced the number of markers linked in the map. However, although our map has a lower marker density, it contains SNP markers with different allele dosages, adding more genetic information to the map and enabling more robust genetic analyses (Garcia et al., 2013). In addition, our genetic map has a higher density of markers than other Urochloa genetic maps (Thaikua et al., 2016; Vigna et al., 2016).

In our study, the density of markers per HG on the consensus map ranged from 63 (HG6) to 187 (HG9) (Figure 2, 3 and Table 1). In relation to the map of the parents, the density of polymorphic markers per HG ranged from 20 (HG6) to 78 (HG9) on the female parental map (D24/27) and from 47 (HG8) to 163 (HG9) on the male parental map (D62) (Supplementary Figures S1, S2 and S3). The greater density of markers for certain HGs may correspond to a greater recombination frequency. However, less saturated groups may have fewer SNP in these regions and/or correspond to highly homozygous regions that have lower recombination frequencies (Bai et al., 2016). We also observed a significant difference in the total number of polymorphic markers on the map of each parental: 377 markers for the female parental genetic map and 856 markers for the male parental genetic map. The lower polymorphic marker density in the map of the female parental (D24/27) is not surprising given the artificial tetraploidization process that this genotype was submitted, which could have a directly influence in the genetic variability on this parent.

The largest distance between adjacent markers on the consensus map, 12.4 cM, was observed in homology group eight (Figure 3 and Table 1), one of the groups with the fewest mapped markers. Compared with other linkage map studies, our map has low average marker intervals in all HGs (Figure 2, 3). Nevertheless, substantial intervals are common and even expected using the GBS technique, mainly due to centromeric regions, which are not reached with this methodology (Conson et al., 2018). Additionally, it is possible that the number of hybrids in the mapping population was not sufficient to observe recombination in these intervals (Ma et al., 2014), that these intervals represent regions that were not captured with the methodology used or that the intervals represent distorted portions of the genome. One possible method to fill these distances is to use different combinations of enzymes to enlarge the sequencing pools and thereby enable the capture of markers in other genomic regions. Furthermore, we mapped only markers identified by aligning with the S. viridis genome. Therefore, these observed intervals may contain markers that are exclusive to the U. decumbens genome and thus cannot be represented with the methodology used. Therefore, advances in the assembly of complex polyploid genomes are expected to enable the use of the full signalgrass genome as a reference for future studies (Margarido and Heckerman, 2015).

Of the 1000 mapped SNPs, 59% (591) followed the simplex parental genotype configuration (Figure 2, 3 and Table 2). This configuration was also the most frequent in other genetic mapping studies (Hackett et al., 2013; da Silva et al., 2017), probably because it is easily scored and analyzed. The least common configuration was X-double-simplex (XSS), followed by double-duplex (DD) (Table 2), which is the most complete configuration because it allows the representation of all types of segregation for autotetraploid species. Overall, the map displays most of the possible SNP configurations, and these data significantly increase the information about each locus and provide several advantages for genetic analysis (Garcia et al., 2013). When building genetic maps for polyploid species, the use of markers with allele dosage information allows the development of robust genetic maps that allow the detection of QTLs (Hackett et al., 2013; da Silva et al., 2017).

Until recently, no intraspecific progeny have been available for U. decumbens; therefore, no QTL mapping studies of this species have been performed to date. The identification of genes/loci underlying spittlebug (N. entreriana) resistance is critical for forage grass molecular breeding programs because these pests severely damage pastures (Valle et al., 2009). In our study, QTL mapping for spittlebug resistance was performed by applying interval mapping for the tetraploid progeny (Hackett et al., 2017). Although more accurate QTL models are available (such as composite and multiple interval mapping), these models have not been extended to polyploidy and therefore do not use molecular markers with allele dosage information.

For the multi-allelic QTL analysis, we performed a permutation test to obtain the threshold for declaring significant QTLs. This analysis allowed the identification of three significant QTLs which explained between 4.66 and 6.24% of the phenotypic variation (Table 5). The identification of a small number of QTLs with small effects reflects the median heritability found for spittlebug resistance (H_C² = 0.37). Spittlebug resistance in forage grasses is probably not genetically complex but likely involves more than a single major resistance gene (Miles et al., 1995). Therefore, values of phenotypic variation and the number of QTLs detected in our analysis may be mainly due the low genetic variability for spittlebug resistance in the analyzed progeny, as the parents do not differ greatly for this trait and are both considered susceptible. Moreover, the methodology used for QTL detection and/or map saturation may have influenced these results.

Although U. decumbens is considered a susceptible species to the pasture spittlebug, our results showed that the segregation of different alleles in each tetraploid parent contributed to the observed polymorphisms in the progeny. Therefore, our results can be useful for breeders to identify the alleles from the male parent that contribute to increased values of this phenotypic trait, consequently allowing genotypes that are more resistant to the development and survival of spittlebugs to be found more rapidly.

Quantitative trait loci with greater effects on spittlebug resistance are more efficient for use in breeding programs, but our results represent a great advance since it is the only study about genetic architecture using an U. decumbens intraspecific progeny. Therefore, the detection of these QTLs is the first step in the advancement of genomic studies involving spittlebug resistance. The annotation of sequences that originated markers with mapped QTLs is important to identify candidate genes involved in spittlebug resistance in U. decumbens. Thus, the regions that gave rise to these QTLs should be more thoroughly evaluated to identify more genetic information for future applications in the marker-assisted selection in signalgrass breeding programs.

To respond to mechanical damage caused by insect attacks, plants possess multiple molecular defense mechanisms (War et al., 2012). Based on an investigative and comparative analysis, the QTL regions of this study showed similarity to genes containing protein domains reported to be involved in defense responses against biotic stresses in plants (Supplementary Table S2).

For spittlebug resistance trait, we can highlight the similarity of the region of the QTL1 with pathogenesis-related gene 1 (PR1). PR proteins are induced by the plants as a defense response system in stress conditions such as insect attacks, as the PR1 gene is involved in biological processes related to defense responses and systemic acquired resistance (Metzler et al., 1991; Van Loon and Van Strien, 1999; Van loon et al., 2006; Sinha et al., 2014). This region also showed similarity to the WRKY11 gene, which plays a crucial role in cellular defense responses and defense-related gene transcriptional regulation (Jiang et al., 2016) (Supplementary Table S2).

For QTL3, we can highlight the similarity of the region to the adjacent markers with WRKY DNA binding domains. These transcription factors are involved in plant defense responses to biotic stresses (Du and Chen, 2000; Souza et al., 2017). Moreover, F-box and WRKY DNA binding domains have orthologs in A. thaliana, with genes that develop a function in plant defense. This QTL region also showed similarity to Thaumatin-like proteins (TLPs), which play a variety of roles, including plant protection against biotic stress (Liu et al., 2010; Mukherjee et al., 2010; Misra et al., 2016).

The regions of the markers adjacent to the mapped QTLs 2 and 3 showed similarity to NBS-LRR protein domains, which are widely distributed in plants and thought to respond to pathogen attacks, including viruses, bacteria, fungi, and even insects (Song et al., 2017; Souza et al., 2017). NBS-LRRs are the most prevalent class of resistance proteins (Gupta et al., 2012). These regions also showed similarity with F-box proteins that control many important biological functions such as pathogen resistance (Lechner et al., 2006), and the PPR superfamily proteins that have been implicated in plant biotic stress responses (Xing et al., 2018). Moreover, the NB-ARC protein domain is a functional ATPase domain containing a nucleotide-binding site that is proposed to regulate plant disease resistance (Van ooijen et al., 2008).

Although we used the S. viridis genome as a reference to investigate the marker regions adjacent to the identified QTLs, we selected only genes that were also found in the U. decumbens transcriptome (Salgado et al., 2017) to lend greater reliability to our results. Therefore, our results provide a starting point, indicating possible candidate genes involved in spittlebug resistance in signalgrass. However further studies should be conducted to validate these genomic regions as well as the role of the candidate genes in signalgrass and their effects on phenotypic expression.

We can conclude that this recent opportunity to analyze a U. decumbens full-sib progeny has opened new genetic molecular perspectives for this and related species. The genetic map presented herein includes important approaches such as the estimation of tetraploid dosage of each molecular SNP and greatly increases genetic information and represents an important evolution for polyploid studies. In addition, the map allowed the identification of genomic regions related to spittlebug (N. entreriana) resistance, the main insect that attacks forage grasses, providing new insights about this trait. Furthermore, the molecular data developed here are dynamic and can be applied in future studies, such as genome assembly and other QTL analyses in U. decumbens.

LC, SB, CdV, JV, FT, and AdS conceived and designed the experiments. RF, LC, SB, CdV, JV, and FT conducted the experiments and collected data. RF, LL, and AG analyzed the data and interpreted the results. RF, LL, and AdS wrote the manuscript. All authors read and approved the manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.