ORIGINAL RESEARCH

published: 03 July 2018

doi: 10.3389/fpls.2018.00815

Linkage Disequilibrium and

Population Structure in Wild and
Cultivated Populations of Rubber
Tree (Hevea brasiliensis)
Livia M. de Souza 1† , Luciano H. B. dos Santos 1† , João R. B. F. Rosa 2,3 , Carla C. da Silva 1 ,
Camila C. Mantello 1,4 , André R. O. Conson 1 , Erivaldo J. Scaloppi Jr. 5 ,
Josefino de F. Fialho 6 , Mario Luiz T. de Moraes 7 , Paulo de S. Gonçalves 5 ,
Gabriel R. A. Margarido 2 , Antonio A. F. Garcia 2 , Vincent Le Guen 8 and
Anete P. de Souza 1,9*
1
Center of Molecular Biology and Genetic Engineering Center, University of Campinas, Campinas, Brazil, 2 Departamento de
Genética, Escola Superior de Agricultura “Luiz de Queiroz” Universidade de São Paulo, Piracicaba, Brazil, 3 FTS Sementes
S.A., Research and Development Center, Ponta Grossa, Brazil, 4 The John Bingham Laboratory, National Institute of
Edited by: Agricultural Botany, Cambridge, United Kingdom, 5 Center of Rubber Tree and Agroforestry Systems, Agronomic Institute of
Petr Smýkal, Campinas, Votuporanga, Brazil, 6 Centro de Pesquisa Agropecuária dos Cerrados (Cpac), Embrapa Cerrados, Empresa
Palacký University, Olomouc, Czechia Brazileira de Pesquisa Agropecuária, Planaltina, Brazil, 7 Departamento de Fitotecnia, Faculdade de Engenharia de Ilha
Solteira, Universidade Estadual Paulista, Ilha Solteira, Brazil, 8 Centre de Coopération Internationale en Recherche
Reviewed by:
Agronomique pour le Développement, UMR AGAP, Montpellier, France, 9 Department of Plant Biology, Biology Institute,
Hamid Khazaei,
University of Campinas, Campinas, Brazil
University of Saskatchewan, Canada
Marinus J.M. Smulders,
Wageningen University & Research, Among rubber tree species, which belong to the Hevea genus of the Euphorbiaceae
Netherlands
family, Hevea brasiliensis (Willd. ex Adr.de Juss.) Muell. Arg. is the main commercial
*Correspondence:
Anete P. de Souza source of natural rubber production worldwide. Knowledge of the population
anete@unicamp.br structure and linkage disequilibrium (LD) of this species is essential for the efficient
† These authors have contributed organization and exploitation of genetic resources. Here, we obtained single-nucleotide
equally to this work.
polymorphisms (SNPs) using a genotyping-by-sequencing (GBS) approach and then
Specialty section: employed the SNPs for the following objectives: (i) to identify the positions of SNPs
This article was submitted to on a genetic map of a segregating mapping population, (ii) to evaluate the population
Plant Breeding,
a section of the journal
structure of a germplasm collection, and (iii) to detect patterns of LD decay among
Frontiers in Plant Science chromosomes for future genetic association studies in rubber tree. A total of 626
Received: 18 December 2017 genotypes, including both germplasm accessions (368) and individuals from a genetic
Accepted: 28 May 2018
mapping population (254), were genotyped. A total of 77,660 and 21,283 SNPs
Published: 03 July 2018
were detected by GBS in the germplasm and mapping populations, respectively.
Citation:
de Souza LM, dos Santos LHB, The mapping population, which was previously mapped, was constructed with 1,062
Rosa JRBF, da Silva CC, markers, among which only 576 SNPs came from GBS, reducing the average interval
Mantello CC, Conson ARO,
Scaloppi EJ Jr., Fialho JdF,
between two adjacent markers to 4.4 cM. SNPs from GBS genotyping were used
de Moraes MLT, Gonçalves PdS, for the analysis of genetic structure and LD estimation in the germplasm accessions.
Margarido GRA, Garcia AAF,
Two groups, which largely corresponded to the cultivated and wild populations,
Le Guen V and de Souza AP (2018)
Linkage Disequilibrium and Population were detected using STRUCTURE and via principal coordinate analysis. LD analysis,
Structure in Wild and Cultivated also using the mapped SNPs, revealed that non-random associations varied along
Populations of Rubber Tree (Hevea
brasiliensis). Front. Plant Sci. 9:815.
chromosomes, with regions of high LD interspersed with regions of low LD. Considering
doi: 10.3389/fpls.2018.00815 the length of the genetic map (4,693 cM) and the mean LD (0.49 for cultivated and

Frontiers in Plant Science | www.frontiersin.org 1 July 2018 | Volume 9 | Article 815

de Souza et al. LD and Genetic Linkage of Hevea

0.02 for wild populations), a large number of evenly spaced SNPs would be needed to
perform genome-wide association studies in rubber tree, and the wilder the genotypes
used, the more difficult the mapping saturation.
Keywords: linkage mapping, genetic diversity, population structure, non-random association, genome-wide
association study, Hevea

INTRODUCTION Evaluation of the molecular diversity encompassed in rubber

tree genetic resources is a prerequisite for their efficient
Hevea brasiliensis, or the rubber tree, is an important crop species exploitation in breeding and the development of conservation
that produces a high-quality natural rubber in commercially strategies of genetic diversity. de Souza et al. (2015) analyzed
viable quantity, accounting for more than 98% of the total natural approximately 1,000 cultivated rubber tree accessions originating
rubber production worldwide (Priyadarshan and Goncalves, from various geographic areas in a Brazilian germplasm
2003). A native species of the Amazon rainforest, H. brasiliensis collection. These accessions were genotyped with 13 SSR markers
is a diploid (2n = 36, n = 18), perennial, and cross-pollinated tree distributed across the chromosomes of the species, and a total
species with an estimated haploid genome size of 1.47 Gb (Tang of 408 alleles were identified, 319 of which were shared between
et al., 2016). This species belongs to the Euphorbiaceae family, groups, while 89 alleles were specific to different groups of
comprising 11 inter-crossable species, of which H. brasiliensis accessions.
is the most economically important (Gonçalves et al., 1990). LD is the non-random association of alleles at distinct loci
The rubber tree has a heterozygous nature, with a long growing in the genome of a sampled population (Weir, 1979) and
cycle that includes 5 years before latex collection. Like most is the basis for association mapping approaches. LD can be
forest trees, the rubber tree has a long generation time, which used for many purposes in plant genomics research and has
explains the slow progress of breeding this species and elucidating received considerable attention as a tool for the study of
the genetic architecture of its complex traits using traditional marker-trait associations due to physical linkage, followed by
approaches. Genetic breeding programs are challenged by a low marker-assisted selection (MAS). Another important application
seed yield per pollination (an average of ten seeds obtained for of LD is the study of genetic diversity in natural populations
100 pollinated flowers) and inbreeding depression, making it and germplasm collections, where it can be employed for the
difficult to develop the appropriate progeny for classical genetic evaluation of population genetics and in crop improvement
studies (Lespinasse et al., 2000). Hence, relatively little is known programs, respectively (Gupta et al., 2005).
about genome-wide models of recombination, allele frequency With the rise of sequence-based genotyping, precise and
variation, and linkage disequilibrium (LD) in this important accurate estimates of population structure and the LD across
plant. the genome are now attainable for the rubber tree. Our goals
Over the last 15 years, many genetic maps of the rubber tree in this study were to characterize the genome distribution
have been constructed. The first rubber tree marker-based genetic of single-nucleotide polymorphisms (SNPs) in a rubber tree
maps were built with restriction fragment length polymorphisms mapping population using GBS technology, to examine
(RFLPs) and amplified fragment length polymorphisms (AFLPs) population structure, to investigate how LD breakdown relates
(Lespinasse et al., 2000), and dense genetic maps were to chromosomes, and to compare the LD between cultivated and
subsequently constructed using simple sequence repeats (SSRs) wild populations. For this study, we used accessions selected
(Le Guen et al., 2011; Triwitayakorn et al., 2011; Souza et al., from the germplasm collection previously analyzed by de Souza
2013). Saturated genetic linkage maps are important for the et al. (2015) and a mapping population (PR255 × PB217)
identification of genomic regions containing major genes and previously saturated with SSR markers described by Souza et al.
quantitative trait loci (QTLs) controlling agronomic traits, and (2013) and Rosa et al. (unpublished).
such maps are important for further breeding programs.
In recent years, advances in next-generation sequencing
technology (NGS) have lowered the cost of DNA sequencing MATERIALS AND METHODS
to the point that genotyping-by-sequencing (GBS) (Elshire
et al., 2011) is now feasible for high-diversity, large-genome Plant Materials and DNA Extraction
species, and a genetic map has been developed using the GBS Two sets of samples were selected for this study, and a total
approach (Pootakham et al., 2015). GBS utilizes restriction of 626 samples were sequenced. One set consisted of 368
enzymes to capture a reduced representation of the target H. brasiliensis accessions, composed of both wild germplasm
genome, and with DNA-barcoded adapters, it is possible to and cultivated genotypes. Details of the plant materials can
sequence multiple samples in parallel in a single run using be found in de Souza et al. (2015) and Supplementary
an NGS platform. GBS has recently been applied to the large Table S1. The other set is an important mapping population
barley (Hordeum vulgare L.) and wheat (Triticum aestivum) of the rubber tree consisting of 252 F1 hybrids, derived
genomes and has been shown to be an effective tool for from a cross between PR255 × PB217 and comprising three
developing molecular markers for these species (Poland et al., replicates of each parental genotype, which were mapped with
2012). 505 markers (SSRs, expressed sequence tag-SSRs, and SNPs)

Frontiers in Plant Science | www.frontiersin.org 2 July 2018 | Volume 9 | Article 815

de Souza et al. LD and Genetic Linkage of Hevea

prior to this publication (Souza et al., 2013; Rosa et al., Population Structure and Genetic
unpublished). Genomic DNA was extracted from leaves using Diversity
the DNeasy Plant Mini Kit (QIAGEN, Germany) according
R

The population structure was investigated in 368 genotypes

to the procedures described by the manufacturer. DNA quality
from the germplasm collection with data from SNPs anchored
parameters and concentrations were measured using a UV-Vis
in a certain LG using two different methods: STRUCTURE
spectrophotometer (NanoDrop, Thermo Scientific, Wilmington,
analysis and principal coordinate analysis (PCoA). Initially, the
DE, United States) and agarose gels.
structure was analyzed with the software STRUCTURE 2.3.4
(Pritchard et al., 2000). Ten replications were run for each of
SNP Discovery via GBS the subpopulation numbers (K), ranging from 1 to 10. Each
GBS library preparation and sequencing were performed at the
run included 500,000 Markov chain Monte Carlo (MCMC)
Institute of Genomic Diversity (Cornell University, Ithaca, NY,
iterations, among which the first 100,000 iterations (used to
United States) as described by Elshire et al. (2011). Genome
monitor whether a chain reached stationarity) were discarded as
complexity was reduced by digesting individual genomic
burn-in. The delta K method was used to identify the number
DNA samples with EcoT22I, a methylation-sensitive restriction
of subgroups in the dataset (Evanno et al., 2005). Based on the
enzyme. The resultant fragments from each sample were directly
posterior probability of membership (Q) of a given accession, the
ligated to a pair of enzyme-specific adapters and combined
accession was classified as admixed in clusters with a membership
into pools. PCR amplification was carried out to generate the
of Q < 0.70. Subsequently, genetic distances between pairs of
GBS libraries, which were sequenced on the Illumina HiSeq
accessions were calculated, and PCoA was performed for the
2500 platform (Illumina Inc., United States). The raw data
SNPs using the GenAlEx program (version 6.5) (Peakall and
were processed, and SNP calling was performed using TASSEL
Smouse, 2012).
5.0 (Glaubitz et al., 2014). Initially, the FASTQ files were
demultiplexed according to the assigned barcode. The reads from
each sample were trimmed, and the tags were identified using the Analysis of LD
following parameters: Kmer length of 64 bp, minimum quality LD was measured by calculating the squared correlation
score within the barcode and read length of 20, minimum Kmer coefficient (r2 ) between each pair of SNPs with the R software
length of 20. All sequence tags from each sample were aligned and GGT 2.0 (van Berloo, 2008), using data from SNPs anchored
to the reference rubber tree genome (Tang et al., 2016) with in the genetic map. We selected markers that were positioned on
Bowtie 2 (Langmead and Salzberg, 2012) using the very-sensitive the linkage map and were present in the germplasm to calculate
option. the LD in each LG separately, considering the subgroup inferred
To perform the analysis, the data were divided into with STRUCTURE. The decay of LD over genetic distance was
the mapping population and germplasms. SNP calling was investigated by plotting pair-wise r2 values against the distance
performed using the TASSEL 5 GBSv2 pipeline (Glaubitz et al., (cM) between markers on the same chromosome using the
2014) and filtered using VCFtools (Danecek et al., 2011) with following model: y = a + b−c/x (Ranc et al., 2012), where x
the following criteria: (1) missing data of 20%, (2) minor allele and y represent the genetic distance in cM and the estimated
frequency (MAF) greater than or equal to 5% (MAF 0.05), and r2 , respectively. The critical r2 for LD decay was determined by
(3) biallelic SNPs only. values of 0.1, which is considered the minimum threshold for a
significant association between pairs of loci and to describe the
Genetic Linkage Map maximum genetic or physical distance at which LD is significant
All linkage analyses were performed using OneMap software (Zhu et al., 2008).
(Margarido et al., 2007), version 2.0-1, employing a previously
constructed genetic map (Souza et al., 2013; Rosa et al.,
unpublished) as a basis for the inclusion of GBS-based SNPs with RESULTS
a minimum logarithm of odds (LOD) score of 8.21 (according
to the function in the R package Onemap “suggest_lod”) and a SNP Discovery and Evaluation
maximum recombination fraction of 0.35. The analysis was performed separately for the mapping
The map construction utilized only markers with 0.05 population and germplasm to produce a total of 1.785 million
missing data and tested the pattern of allelic segregation for reads of sequence data, of which 89% (1,586 million reads)
χ2 goodness of fit to the expected Mendelian segregation consisted of good barcoded reads. Of these reads, 69,408 (23.42%)
ratios, and markers with significant segregation distortion were were aligned to the mapping population exactly one time
excluded from further analysis. GBS-based SNPs were added and 177,143 (59.78%) more than one time, corresponding to
to the previous genetic map using the “try.seq” function in an 83.20% (246,551) overall alignment rate. The total rate of
OneMap, which determines the best position for a given alignment to the germplasm was 89.97% (818,807), of which
unpositioned GBS marker in a specific linkage group (LG). 22.59% of the tags (205,576) were aligned to the rubber tree
Finally, the fractions of recombination were converted to reference genome exactly one time (Tang et al., 2016) and 67.38%
centimorgans (cM) using the Kosambi map function (Kosambi, (613,231) more than one time.
1943), and the map was drawn in Mapchart, version 2.3 A total of 386,180 SNPs were identified in the germplasm,
(Voorrips, 2002). and 76,191 SNPs were detected in the mapping population, of

Frontiers in Plant Science | www.frontiersin.org 3 July 2018 | Volume 9 | Article 815

de Souza et al. LD and Genetic Linkage of Hevea

which 350,965 SNPs (germplasm) and 66,453 SNPs (mapping by a drastic decline starting at K = 3, suggesting that the
population) were biallelic. After excluding markers showing (1) optimal K value was 2. The assignment results for K = 2 showed
more than 20% missing data or (2) a MAF ≤ 0.05, the whole that some of the sampled individuals exhibited admixtures
dataset was reduced to 77,660 and 21,283 SNPs in the germplasm from two gene pools (Figure 2A): group 1 (red bars) mainly
and mapping populations, respectively. The SNP frequencies consisted of accessions originating from the Mato Grosso and
were one biallelic SNP every 20.7 kb for the mapping population cultivated genotypes, and group 2 (green bars) consisted entirely
and every 3.9 kb for the germplasm. of wild accessions from Amazonas, Rondônia, Para, and Acre
Sequence data are deposited under EMBL-EBI accession (identified as IS – Ilha Solteira). Twenty-one accessions showing
PRJEB26962. a membership probability (Q value) below 0.70 were defined as
admixed and were removed from subsequent analyses.
Saturation of the Linkage Map With SNPs As a second analysis of differentiation, we employed PCoA
From 21,283 SNPs, a total of 14,852 markers were selected based on a similarity matrix that explained 22.8 and 16.7%
after applying the chi-square test (P ≤ 0.05), which revealed of the genetic variation with the first and second PCoA
segregation ratios of 1:2:1 and 1:1. Markers exhibiting statistically axes, respectively. Plotting the two first PCoA axes separated
significant segregation distortion were excluded from further the germplasm into two clusters, though some overlap was
analysis to obtain accurate genetic linkage maps. The data present between the wild germplasms and cultivated genotypes
analyses were performed using SNPs with a maximum of 5% (Figure 2). On the first axis, most of the breeding genotypes were
missing data, resulting in a linkage map with 1,062 markers separated from the other genotypes. On the second axis, the wild
presenting 348 SSR markers (Souza et al., 2013; Mantello, 2014), germplasm samples were clustered together, but some accessions
576 SNPs from GBS (Supplementary Table S2) and 138 SNPs from the Amazon were isolated in their own subdivision.
markers genotyped using the Sequenom MassARRAY platform R

(AgenaBio, San Diego, CA, United States) and the Fluidigm R Evaluation of LD
platform (South San Francisco, CA, United States), developed Based on population genetic structure, accessions could be
from de novo transcriptome assemblies (Mantello et al., 2014; divided into two distinct groups (cultivated and wild group)
Salgado et al., 2014) and from EST full-length libraries (Silva et al., (Figure 2), and pairwise LD estimates were performed within the
2014). gene pool of each of these groups. To visualize LD throughout
The genetic map was organized according to the numbers the genome, heat maps were produced based on pairwise r2
obtained from the map previously developed by Rosa et al. estimates, and corresponding p-values were calculated using
(unpublished). Eleven markers were removed after data diagnosis permutations for all marker pairs (Supplementary Figure S1).
using heat map graphs, thereby permitting the visualization of These heat maps were employed to identify variations in LD
the recombination fraction and LOD scores from markers, to between the cultivated and wild rubber tree germplasm groups.
group the SNPs in the LGs without changing the order of the base In 16,025 pairwise combinations, we identified 186 (1.2%) and
map. Only LG18 was further divided into subgroups “A” and “B” 592 (3.7%) statistically significant associations that were in LD
(Figure 1). (P < 0.05) in the cultivated and wild germplasms, respectively. Of
Thus, a genetic map was generated spanning a cumulative these significant associations, 78 and 64 were intrachromosomal
length of 4,693 cM (Figure 1), distributed among all the in the cultivated and wild germplasms, respectively, accounting
chromosomes. The LGs ranged from 23.8 cM (LG18B) for 0.5 and 0.4% of the total possible intrachromosomal
(Supplementary Table S3) with 14 markers to the largest group correlations. Among the unlinked loci, the proportions of LD
with 404.6 cM (LG10) with 85 markers (of which 78 were from were 0.7 and 3.3% for the cultivated and wild germplasms,
GBS), and the average interval size between two adjacent markers respectively. In both clusters, an uneven distribution of LD
was 4.4 cM. among the 18 chromosomes was observed.
The maximum gap size was 36.4 cM (LG11), which was The strength of LD (P < 0.05) was very different between
maintained from the previous map. Furthermore, some regions two clusters, as reflected by the mean r2 values of 0.49 and 0.02
could not be sampled using the selected restriction enzyme; in obtained for the cultivated and wild germplasms, respectively.
LG16, for example, only two GBS-based markers were added. The 346 SNPs that were localized in the integrated map were
used for the estimation of LD decay among the different LGs.
LD decay was estimated across the germplasms and was found
Genetic Relationships Among to be more pronounced in the wild germplasms, with the range
Populations being dependent on the chromosome group (Supplementary
The germplasm collection was selected in a prior work by de Figure S2). Using a fixed baseline r2 value of 0.1 for the wild
Souza et al. (2015) examining the population structure with SSR germplasm, LD decay ranged from 2.3 cM (LG9) to 11.4 cM
markers. To confirm the population structure of the sample of (LG17). In contrast, in the cultivated accessions, this decay was
selected individuals, we performed a new analysis as performed slower, ranging from 4.3 cM (LG8) to 50 cM (LG17), with
with the SNPs from GBS genotyping. Only mapped SNPs that average values of 5.6 cM and 19.6 cM for the wild and cultivated
were common in the germplasm collection were used. Clustering germplasms, respectively (Figure 3). The patterns of LD can also
inference performed with K values ranging from 1 to 10 showed be visualized across the genome from the diagonal of the heat
that the model likelihood increased steeply at K = 2, followed maps (Supplementary Figure S1).

Frontiers in Plant Science | www.frontiersin.org 4 July 2018 | Volume 9 | Article 815

de Souza et al. LD and Genetic Linkage of Hevea

FIGURE 1 | Linkage map of the rubber tree. Markers in blue represent SNPs obtained using the genotyping-by-sequencing technique, and markers in black were
obtained from the previous map (Rosa et al., unpublished).

DISCUSSION An important breakthrough of the GBS approach is that

a reference genome is not necessary for SNP genotyping.
We performed a genetic mapping study with a bi-parental However, the availability of a reference genome offers additional
population (252 segregating hybrids between PR255 × PB217) benefits, as it allows proper alignment and ordering of the
and analyzed the population structure and LD with two different sequenced tags (Poland et al., 2012). The first draft of the
populations of rubber tree, formed from 47 accessions from genome sequence (Rahman et al., 2013) provided a source of
a breeding program and 300 accessions from a germplasm genomic information, after which three more genomes were
collection. For this study, we employed SNPs obtained via the published (Lau et al., 2016; Tang et al., 2016; Pootakham
GBS approach, which enabled the detection of polymorphisms et al., 2017). The assembly is highly fragmented, containing
distributed across the genome. more than 1 million contigs. The assembly of a complex

Frontiers in Plant Science | www.frontiersin.org 5 July 2018 | Volume 9 | Article 815

de Souza et al. LD and Genetic Linkage of Hevea

FIGURE 2 | Estimated genetic structure of the wild population and breeding population of rubber tree based on PCoA (A) and STRUCTURE analysis
(B). All analyses are based on the genetic variability of 438 SNP loci.

FIGURE 3 | Decay of LD (r 2 ) as a function of genetic distance (cM) between pairs of loci on all chromosomes. Only r 2 values with P < 0.05 are shown.

genome is challenging, owing in part to the presence of highly Tang et al. (2016) to map the SNPs. The SNPs positioned on the
repetitive DNA sequences, which introduce ambiguity during map are named in reference to the location of alignment with the
genome reconstruction. We used the genome published by genome.

Frontiers in Plant Science | www.frontiersin.org 6 July 2018 | Volume 9 | Article 815

de Souza et al. LD and Genetic Linkage of Hevea

Repetitive regions account for 71% of the Hevea genome and differences in mapping strategies, the mapping software and
(Tang et al., 2016), posing a major challenge for the de novo the ratio between the number of markers and the population size
assembly, particularly when exclusively short-read data are used. (Knox and Ellis, 2002).
This phenomenon might explain the prevalence of tags that Genetic mapping of the population (PR255 × PB217)
aligned to more than one unique region (59.78% to the mapping was initiated in the first publication by Souza et al. (2013)
population and 67.38% to the germplasm) (Tang et al., 2016). using microsatellite markers. These first LGs were organized
Based on the properties of the reference genome (characterized according to the numbers obtained from the map previously
by a great accumulation of repetitive sequences, primarily in developed by Lespinasse et al. (2000), and information from
heterochromatic regions) (Rahman et al., 2013), the restriction other maps (unpublished) consisting of microsatellites in
enzyme EcoT22I was selected because it is partially sensitive to common was used to identify syntenic markers. Subsequently,
methylation and rarely cuts retrotransposons. the PR255 ×PB217 mapping population was saturated with
The frequency of nucleotide substitutions was five times SSR markers and 243 SNPs obtained from platforms such as
higher in the wild (3.9 kb) than in the cultivated (20.7 kb) Sequenom MassARRAY iPLEX technology and KASP genotyping
germplasm sequences. One explanation for this finding is that (Rosa et al., unpublished). The SNPs from GBS used to construct
selection in cultivated breeding programs acts to reduce diversity the new genetic linkage map contributed to reducing the average
and alter allele frequencies in the DNA sequence. Depending interval between two adjacent markers (4.4 cM versus 7.4 cM).
on how LD is increased surrounding these loci; the effects of However, the marker density remained lower than that of the
such a selection may not extend sufficiently far to affect the first rubber tree map (Lespinasse et al., 2000), which presented
overall genome diversity. Rubber tree breeders have had to one marker every 3 or 0.89 cM according to Pootakham et al.
develop cultivars that are appropriate for the specific temperature (2015). LG11 displayed the maximum gap size (36.4 cM), which
and humidity conditions encountered in different cultivation may have been caused by sections of the genome that were
areas, along with various biotic and abiotic stress resistance identical among the parental genotypes and thus an absence
factors. High-density SNPs are also common in non-genetic of polymorphisms, or by recombination hotspots. Large gaps
regions where there is no selection pressure, and the abundance exhibiting a low degree of polymorphisms have also been
of these SNPs will be very useful for future assessments of reported by Shearman et al. (2015).
breeding populations. The high diversity of SNPs is characteristic Genetic maps are important tools not only for QTL mapping
of outbreeding trees; in other studies examining rubber tree, but also for anchoring genome assembly scaffolds into pseudo-
Pootakham et al. (2011) previously identified a frequency of one chromosomes. The draft genome of H. brasiliensis has been
SNP every 1.5 kb, while Pootakham et al. (2015) observed a reported to be highly heterozygous, with 71% of the genome
frequency of one SNP every 308 nucleotides. length comprising repeats (Tang et al., 2016). A total of 21,283
GBS does not require complete genome sequencing; only biallelic SNPs were identified, potentially aligning with 1,926
a targeted sequencing approach is necessary. Due to its high- scaffolds, corresponding to 26% of the total scaffolds of the
throughput efficiency, GBS has been used for SNP identification genome used. However, only 3% of these scaffolds were anchored
and mapping in many plant species (Poland et al., 2012; Rabbi to LGs due to the relatively fragmented scaffold sequences and
et al., 2014; Bekele et al., 2018; Rimbert et al., 2018). In rubber the usage of an insufficient genetic map to anchor the scaffolds.
tree, most of the genetic linkage maps constructed to date Thus, it is important to develop higher-quality genetic maps to
have employed molecular markers such as RFLPs and AFLPs help improve the scaffold anchoring ratio of future rubber tree
(Lespinasse et al., 2000) or microsatellites (SSRs) (Le Guen et al., genome assemblies.
2011; Triwitayakorn et al., 2011; Souza et al., 2013). However, Progress in the development of new molecular markers and
GBS was employed for linkage map construction in rubber tree genetic linkage maps is important for genetic improvement
very recently (Pootakham et al., 2015; Shearman et al., 2015). in rubber tree breeding programs. Despite the economic and
In the present study, we utilized the GBS platform to sequence ecological importance of this crop, restricted genomic resources
a PR255 × PB217 mapping population that was previously are available for rubber tree. Sequencing of the studied family
saturated with SSR markers and SNPs obtained from other allowed the identification and genotyping of many markers in
specific platforms (Souza et al., 2013; Rosa et al., unpublished). an efficient and cost-effective way. The linkage mapping analysis
The linkage map constructed in the present study exhibits resulted in a number of LGs corresponding to the rubber tree
a regular marker distribution. However, the cumulative genetic haploid chromosome number (n = 18) (Ong, 1975; Lespinasse
map is significantly longer (4,693 cM) than would be expected et al., 2000), and the development of GBS methods and genetic
for maps derived from other mapping populations that have maps represents an important advancement of the genomics tools
presented a lower resolution; for example, a map of 2,052 cM available for these crops, which currently lack a good reference
(Pootakham et al., 2015) was obtained using only SNPs, and genome sequence.
a map of 2,441 cM (Le Guen et al., 2011) was obtained using Diversity analyses were performed in a previous work that
both SSRs and AFLPs. Such expansions of the rubber tree included the entire collection of rubber germplasm, analyzing
genetic linkage map have also reported previously (4,160 cM, 1,117 genotypes with 13 microsatellite markers (de Souza et al.,
Shearman et al., 2015). Several factors may be responsible for 2015), and showed a mean observed heterozygosity (Ho) of
this phenomenon, including the numbers and types of mapped 0.64 and higher genetic diversity (He) than Ho in all cases.
loci, the genetic constitution of different mapping populations Wright’s fixation index (F) values were positive, with a mean of

Frontiers in Plant Science | www.frontiersin.org 7 July 2018 | Volume 9 | Article 815

de Souza et al. LD and Genetic Linkage of Hevea

0.16 obtained for the accessions overall. Among a total of 408 wild germplasm (0.02), corroborating the results of Chanroj
observed alleles, 89 represented unique alleles to different groups et al. (2017), who suggested that these results were caused
of accessions, demonstrating the high level of heterogeneity in by high gene flow in the wild Amazonian population. The
these genotypes. To confirm the population structure of the breeding accessions exhibited a notably higher level of LD,
sample of selected individuals based on the work of de Souza suggesting less genetic diversity in this subdivision, perhaps
et al. (2015), a new analysis of the population structure was because of the constraint of genetic variability employed
conducted with SNPs obtained from GBS genotyping. The 1K in breeding programs due to the recurrent process of
values obtained in this study indicated that the rubber tree selection. The low LD detected in the wild germplasm
germplasm could be divided into two groups and showed that in this study was expected because perennial outcrossing
some of the sampled individuals exhibited admixtures from two tree species display a high effective recombination rate,
gene pools, which was also confirmed by plotting the two first which leads to the rapid decay of LD (Krutovsky and Neale,
PCoA axes. 2005).
Recent work using SSR markers has revealed similar results The higher LD level detected in the cultivated group may
to those obtained using SNPs from GBS. For example, Le Guen have been influenced by the partly identical-by-descent of these
et al. (2009) demonstrated a separation between the Acre and genotypes from a limited number of founders of the breeding
Rondônia groups and the Mato Grosso group. One possible programs, with only a few generations between them. Thus,
reason for this separation is that the wild accessions are from many longer pieces of chromosomes have not had time to
geographically distant populations that are not connected by undergo disruptions. Domesticated crop cultivars necessarily
hydrographic networks (Le Guen et al., 2009; Chanroj et al., represent a subset of the genetic variation found in their wild
2017), whereas most of the Wickham clones were collected from ancestors, and the process of crop domestication is responsible
regions that are geographically closer to the Mato Grosso and the for genetic bottlenecks (McCouch, 2004). Although rubber
Tapajós river, enabling rubber seeds to flow from one region to tree breeding programs are very recent, the differences in
another via the river. de Souza et al. (2015) reported a difference LD patterns detected between cultivated and wild germplasms
in genotypes between the two groups, which are in different suggest that plant breeders may have selected for separate
river basins, thus separating the genotypes in the cultivated combinations of genes during the breeding process. Moreover,
and wild groups. Within the group of cultivated genotypes selection for high latex yields and the extensive use of
denoted as Mato Grosso and Wickhan, genotypes collected from particular clones as parents in rubber breeding programs
Mato Grosso were genetically close to genotypes used to initiate have further reduced the genetic diversity of commercial
the Asian breeding programs, denoted Wickhan genotypes in rubber germplasm (Priyadarshan, 2016), which may also
this article. These genotypes were collected by Henry Wickhan affect LD.
in 1976 in the same basin from which the genotypes from LD decay was estimated at 25.7 cM within cultivated
Mato Grosso were sampled (Gonçalves et al., 1990). The results and 5.7 cM within wild germplasm, and significant inter-
obtained by de Souza et al. (2015) corroborate the seminal study chromosomal LD was identified within cultivated in contrast
by Le Guen et al. (2009). to wild germplasm. The distances of LD decay between the
Population genetic structure is a principal factor influencing two groups were most different in LGs 3, 5, 7, 10, and 11,
the generation of false positives and the success of the association suggesting that these chromosomes may carry more genes
or LD mapping (Gupta et al., 2005). These 368 rubber tree related to agronomic traits that have been selected via organized
accessions could be divided into two distinct groups based on breeding of this crop. In previous studies involving Amazonian
population genetic structure: cultivated and wild germplasm. accessions of rubber tree, Chanroj et al. (2017) revealed an
The investigation of LD decay versus genetic distance based on LD decay of more than 0.5–6 cM and suggested that LD
markers is not possible without prior information regarding the estimates were significantly influenced by physical distance,
positions of markers in the genome. Since the information on the with LD decay greater than 2 kbp being observed in the
rubber tree genome is still inaccurate regarding these positions, widespread Amazonian population. These authors showed that
we employed the existing genetic map of the PR255 × PB217 LD decay over genetic distance was different for the 18 different
population to identify the positions of some SNPs and, thus, chromosomes, possibly because of the different recombination
enable a more accurate study of LD decay. rates of the 18 chromosomes. Rapid LD decay has been reported
LD analysis with the mapped SNPs revealed that LD for many other outcrossing tree species, such as Populus nigra, in
varied along the chromosomes, with regions of high LD being which a decay of r2 with distance in the CAD4 gene was observed
interspersed with regions of low LD (Supplementary Figure S1). at approximately 16 bp (Marroni et al., 2011). In Eucalyptus
LD estimation is possible without the positions of molecular globulus, candidate genes for wood quality were analyzed using
markers along the genome and was indeed performed in this SNPs, and LD was estimated to decay rapidly (Thavamanikumar
study. However, these positions are crucial if one is interested et al., 2011). Most LD estimation studies conducted in tree
in determining the decay or extension of LD in relation to the species are based on candidate genes (Krutovsky and Neale,
genetic distance, regardless of by chromosome. 2005).
The average r2 values were found to be very different when LD mapping relies on germplasm samples and, as such,
the two detected groups were compared. This measurement does not require the development of experimental crosses with
was higher in the breeding germplasm (0.49) than in the specific genetic backgrounds, and thus ease of use is an obvious

Frontiers in Plant Science | www.frontiersin.org 8 July 2018 | Volume 9 | Article 815

de Souza et al. LD and Genetic Linkage of Hevea

benefit in studies of perennial species with long life cycles. FUNDING

LD is a key factor in determining the number of markers
needed for genome-wide association studies (GWAS) and The authors gratefully acknowledge the Fundação de Amparo
genomic selection (GS). Genomes with low LD require a high a Pesquisa do Estado de São Paulo (FAPESP) (2007/50392-1;
marker density for GWAS or GS; therefore, our SNPs may 2012/50491-8) for financial support and for graduate
be valuable for GWAS in rubber tree breeding. Considering scholarships to CM (2011/50188-0 and 2014/18755-0) and CdS
the length of the genetic map (4,693 cM) and the mean LD (2009/52975-0) and LHBdS (2014/11807-5 and 2017/07908-9)
observed (0.49 in breeding and 0.02 in wild populations), and a post-doctoral fellowship to LMdS (2012/05473-1); the
many evenly spaced SNPs would be necessary to perform Conselho Nacional de Desenvolvimento Científico e Tecnológico
GWAS in the rubber tree, and the wilder the genotypes (CNPq) for financial support (478701/2012-8; 402954/2012-2),
that are used, the more difficult is the saturation of the a post-doctoral fellowship to CdS, and research fellowships
mapping. However, to obtain a sufficient SNP density throughout to AG, AdS, and PG and, a Ph.D. fellowship to AC; and the
the genome and to account for variation in LD along Coordenação de Aperfeiçoamento do Pessoal de Nível Superior
the chromosomes more markers must be genotyped. Our (CAPES) for financial support (Computational Biology Program
study results provide a valuable resource for further genetic and CAPES-Agropolis Program) and post-doctoral fellowships
studies involving linkage or association mapping, marker- to LMdS and AC.
assisted breeding and Hevea sequence assembly and comparative
mapping.
Furthermore, GWAS of wild germplasm accessions in the SUPPLEMENTARY MATERIAL
future will provide a substantial contribution to dealing with
new challenging situations that will arise as a consequence of The Supplementary Material for this article can be found online
global climatic changes. Useful QTLs and genes to face these new at: https://www.frontiersin.org/articles/10.3389/fpls.2018.00815/
situations are currently unknown, but the best way to identify full#supplementary-material
them would be through analyses relying on GWAS of large
panels of wild genotypes. A precise assessment of the LD pattern FIGURE S1 | Plots of LD heat maps for (A) with 47 breeding accessions and (B)
with 300 accessions from wild germplasms. The rubber tree LGs are represented
across the genome of Hevea is necessary for such endeavors,
by a diagonal bar. Markers were ordered on the x- and y-axes based on genomic
and the present study supplies a major contribution to this location; therefore, each cell of the heat map represents a single marker pair. The
goal. r 2 values for each marker pair are presented in the bottom half of the heat map
and are represented by shades of red increasing in intensity in equal increments of
0.1 from 0.0 (white) to 1.0 (red).
AUTHOR CONTRIBUTIONS FIGURE S2 | Decay of LD (r 2 ) as a function of genetic distance (cM) between
pairs of loci in individual LGs. Only r 2 values with P < 0.05 are shown.
LMdS, LHBdS, VLG, and AdS designed the study and performed
TABLE S1 | Origin of germplasm genotypes and population structure results.
the experiments. LMdS, LHBdS, AC, CdS, CM, ES, JF, and MdM
performed the experiments. LMdS, LHBdS, JR, GM, CM, VLG, TABLE S2 | SNPs from GBS and their genome information. The name of markers
are in agreement with the sequences of the reference genome (Tang et al., 2016).
MdM, PG, and AG analyzed the data. LMdS, LHBdS, and AdS
wrote the manuscript. TABLE S3 | Marker information for the genetic map.

REFERENCES Evanno, G., Regnaut, S., and Goudet, J. (2005). Detecting the number of clusters of
individuals using the software STRUCTURE: a simulation study. Mol. Ecol. 14,
Bekele, W. A., Wight, C. P., Chao, S., Howarth, C. J., and Tinker, N. A. (2018). 2611–2620. doi: 10.1111/j.1365-294X.2005.02553.x
Haplotype based genotyping-by-sequencing in oat genome research. Plant Glaubitz, J. C., Casstevens, T. M., Lu, F., Harriman, J., Elshire, R. J., Sun, Q.,
Biotechnol. J. doi: 10.1111/pbi.12888 [Epub ahead of print]. et al. (2014). TASSEL-GBS: a high capacity genotyping by sequencing analysis
Chanroj, V., Rattanawong, R., Phumichai, T., Tangphatsornruang, S., and pipeline. PLoS One 9:e90346. doi: 10.1371/journal.pone.0090346
Ukoskit, K. (2017). Genome-wide association mapping of latex yield Gonçalves, P. S., Cardoso, M., and Ortolani, A. A. (1990). Origem, variabilidade e
and girth in Amazonian accessions of Hevea brasiliensis grown in a domesticação da Hevea: uma revisão. Pesqui. Agropecu. Bras. 25, 135–156.
suboptimal climate zone. Genomics 109, 475–484. doi: 10.1016/j.ygeno.2017. Gupta, P. K., Rustgi, S., and Kulwal, P. L. (2005). Linkage disequilibrium and
1007.1005 association studies in higher plants: present status and future prospects. Plant
Danecek, P., Auton, A., Abecasis, G., Albers, C. A., Banks, E., DePristo, M. A., et al. Mol. Biol. 57, 461–485. doi: 10.1007/s11103-005-0257-z
(2011). The variant call format and VCFtools. Bioinformatics 27, 2156–2158. Knox, M. R., and Ellis, T. H. N. (2002). Excess heterozygosity contributes to genetic
doi: 10.1093/bioinformatics/btr330 1 map expansion in pea recombinant inbred populations. Genetics 162, 861–873.
de Souza, L. M., Le Guen, V., Cerqueira-Silva, C. B. M., Silva, C. C., Mantello, Kosambi, D. D. (1943). The estimation of map distances from recombination
C. C., Conson, A. R. O., et al. (2015). Genetic diversity strategy for the values. Ann. Eugen. 12, 172–175. doi: 10.1111/j.1469-1809.1943.tb02
management and use of rubber genetic resources: more than 1,000 wild and 321.x
cultivated accessions in a 100-genotype core collection. PLoS One 10:e0134607. Krutovsky, K. V., and Neale, D. B. (2005). Nucleotide diversity and linkage
doi: 10.1371/journal.pone.0134607 disequilibrium in cold-hardiness- and wood quality-related candidate
Elshire, R. J., Glaubitz, J. C., Sun, Q., Poland, J. A., Kawamoto, K., Buckler, genes in Douglas fir. Genetics 171, 2029–2041. doi: 10.1534/genetics.105.04
E. S., et al. (2011). A robust, simple genotyping-by-sequencing (GBS) approach 4420
for high diversity species. PLoS One 6:e19379. doi: 10.1371/journal.pone. Langmead, B., and Salzberg, S. L. (2012). Fast gapped-read alignment with Bowtie
0019379 2. Nat. Methods 9, 357–359. doi: 10.1038/nmeth.1923

Frontiers in Plant Science | www.frontiersin.org 9 July 2018 | Volume 9 | Article 815

de Souza et al. LD and Genetic Linkage of Hevea

Lau, N.-S., Makita, Y., Kawashima, M., Taylor, T. D., Kondo, S., Othman, A. S., et al. Rahman, A. Y., Usharraj, A. O., Misra, B. B., Thottathil, G. P., Jayasekaran, K.,
(2016). The rubber tree genome shows expansion of gene family associated with Feng, Y., et al. (2013). Draft genome sequence of the rubber tree Hevea
rubber biosynthesis. Sci. Rep. 6:28594. doi: 10.1038/srep28594 brasiliensis. BMC Genomics 14:75. doi: 10.1186/1471-2164-14-75
Le Guen, V., Doaré, F., Weber, C., and Seguin, M. (2009). Genetic structure Ranc, N., Muños, S., Xu, J., Le Paslier, M. C., Chauveau, A., Bounon, R., et al.
of Amazonian populations of Hevea brasiliensis is shaped by hydrographical (2012). Genome-wide association mapping in tomato (Solanum lycopersicum)
network and isolation by distance. Tree Genet. Genomes 5, 673–683. is possible using genome admixture of Solanum lycopersicum var. cerasiforme.
doi: 10.1007/s11295-009-0218-9 G3 2, 853–864. doi: 10.1534/g3.112.002667
Le Guen, V., Garcia, D., Doaré, F., Mattos, C. R. R., Condina, V., Couturier, C., et al. Rimbert, H., Darrier, B., Navarro, J., Kitt, J., Choulet, F., Leveugle, M., et al. (2018).
(2011). A rubber tree’s durable resistance to Microcyclus ulei is conferred by a High throughput SNP discovery and genotyping in hexaploid wheat. PLoS One
qualitative gene and a major quantitative resistance factor. Tree Genet. Genomes 13:e0186329. doi: 10.1371/journal.pone.0186329
7, 877–889. doi: 10.1007/s11295-011-0381-7 Rosa, J. R. B., Mantello, C. C., Garcia, D., Souza, L. M., Silva, C. C., Gazaffi, R.,
Lespinasse, D., Rodier-Goud, M., Grivet, L., Leconte, A., Legnate, H., and et al. (Submitted). QTL detection for growth and latex production in a full-sib
Seguin, M. (2000). A saturated genetic linkage map of rubber tree (Hevea spp.) population of rubber tree cultivated under suboptimal climate conditions.
based on RFLP, AFLP, microsatellite, and isozyme markers. Theor. Appl. Genet. Salgado, L., Koop, D., Pinheiro, D., Rivallan, R., Le Guen, V., Nicolas, M., et al.
100, 127–138. doi: 10.1007/s001220050018 (2014). De novo transcriptome analysis of Hevea brasiliensis tissues by RNA-
Mantello, C. C. (2014). Mapeamento Genético Molecular em Hevea brasiliensis. seq and screening for molecular markers. BMC Genomics 15:236. doi: 10.1186/
Ph.D. thesis, Universidade Estadual de Campinas, Campinas. 1471-2164-15-236
Mantello, C. C., Cardoso-Silva, C. B., da Silva, C. C., de Souza, L. M., Scaloppi Shearman, J. R., Sangsrakru, D., Jomchai, N., Ruang-Areerate, P., Sonthirod, C.,
Junior, E. J., de Souza, G. P., et al. (2014). De novo assembly and transcriptome Naktang, C., et al. (2015). SNP identification from RNA sequencing and linkage
analysis of the rubber tree (Hevea brasiliensis) and SNP markers development map construction of rubber tree for anchoring the draft genome. PLoS One
for rubber biosynthesis pathways. PLoS One 9:e102665. doi: 10.1371/journal. 10:e0121961. doi: 10.1371/journal.pone.0121961
pone.0102665 Silva, C. C., Mantello, C. C., Campos, T., Souza, L. M., Gonçalves, P. S.,
Margarido, G. R., Souza, A. P., and Garcia, A. A. (2007). OneMap: software for and Souza, A. P. (2014). Leaf-, panel- and latex-expressed sequenced tags
genetic mapping in outcrossing species. Hereditas 144, 78–79. doi: 10.1111/j. from the rubber tree (Hevea brasiliensis) under cold-stressed and suboptimal
2007.0018-0661.02000.x growing conditions: the development of gene-targeted functional markers
Marroni, F., Pinosio, S., Zaina, G., Fogolari, F., Felice, N., Cattonaro, F., for stress response. Mol. Breed. 34, 1035–1053. doi: 10.1007/s11032-014-
et al. (2011). Nucleotide diversity and linkage disequilibrium in Populus 0095-2
nigra cinnamyl alcohol dehydrogenase CAD4) gene. Tree Genet. Genomes 7, Souza, L. M., Gazaffi, R., Mantello, C. C., Silva, C. C., Garcia, D., Le Guen, V.,
1011–1023. doi: 10.1007/s11295-011-0391-5 et al. (2013). QTL mapping of growth-related traits in a full-sib family of rubber
McCouch, S. (2004). Diversifying selection in plant breeding. PLoS Biol. 2:e347. tree (Hevea brasiliensis) evaluated in a sub-tropical climate. PLoS One 8:e61238.
doi: 10.1371/journal.pbio.0020347 doi: 10.1371/journal.pone.0061238
Ong, S. H. (1975). “Chromosome morphology at the pachytene stage in Hevea Tang, C., Yang, M., Fang, Y., Luo, Y., Gao, S., Xiao, X., et al. (2016). The rubber tree
brasiliensis – a preliminary report,” in Proceedings of the International Rubber genome reveals new insights into rubber production and species adaptation.
Conference, Vol. 2 (Kuala Lumpur: IRRDB), 3–12. Nat. Plants 2:16073. doi: 10.1038/nplants.2016.73
Peakall, R., and Smouse, P. E. (2012). GenAlEx 6.5: genetic analysis in Thavamanikumar, S., McManus, L. J., Tibbits, J. F. G., and Bossinger, G. (2011). The
Excel. Population genetic software for teaching and research—an update. significance of single nucleotide polymorphisms (SNPs) in Eucalyptus globulus
Bioinformatics 28, 2537–2539. doi: 10.1093/bioinformatics/bts460 breeding programs. Aust. For. 74, 23–29. doi: 10.1080/00049158.2011.106
Poland, J. A., Brown, P. J., Sorrells, M. E., and Jannink, J. L. (2012). Development 76342
of high-density genetic maps for barley and wheat using a novel two-enzyme Triwitayakorn, K., Chatkulkawin, P., Kanjanawattanawong, S., Sraphet, S.,
genotyping-by-sequencing approach. PLoS One 7:e32253. doi: 10.1371/journal. Yoocha, T., Sangsrakru, D., et al. (2011). Transcriptome sequencing of
pone.0032253 Hevea brasiliensis for development of microsatellite markers and construction
Pootakham, W., Chanprasert, J., Jomchai, N., Sangsrakru, D., Yoocha, T., of a genetic linkage map. DNA Res. 18, 471–482. doi: 10.1093/dnares/d
Therawattanasuk, K., et al. (2011). Single nucleotide polymorphism marker sr034
development in the rubber tree, Hevea brasiliensis (Euphorbiaceae). Am. J. Bot. van Berloo, R. (2008). GGT 2.0: versatile software for visualization and analysis of
98, e337–e338. doi: 10.3732/ajb.1100228 genetic data. J. Hered. 99, 232–236. doi: 10.1093/jhered/esm109
Pootakham, W., Ruang-Areerate, P., Jomchai, N., Sonthirod, C., Sangsrakru, D., Voorrips, R. E. (2002). MapChart: software for the graphical presentation of linkage
Yoocha, T., et al. (2015). Construction of a high-density integrated genetic maps and QTLs. J. Hered. 93, 77–78. doi: 10.1093/jhered/93.1.77
linkage map of rubber tree (Hevea brasiliensis) using genotyping-by-sequencing Weir, B. S. (1979). Inferences about linkage disequilibrium. Biometrics 35, 235–254.
(GBS). Front. Plant Sci. 6:367. doi: 10.3389/fpls.2015.00367 doi: 10.2307/2529947
Pootakham, W., Sonthirod, C., Naktang, C., Ruang-Areerate, P., Yoocha, T., Zhu, C., Gore, M., Buckler, E. S., and Yu, J. (2008). Status and prospects
Sangsrakru, D., et al. (2017). De novo hybrid assembly of the rubber tree of association mapping in plants. Plant Genome 1, 5–20. doi: 10.3835/
genome reveals evidence of paleotetraploidy in Hevea species. Sci. Rep. 7:41457. plantgenome2008.02.0089
doi: 10.1038/srep41457
Pritchard, J. K., Stephens, M., and Donnelly, P. (2000). Inference of population Conflict of Interest Statement: The authors declare that the research was
structure using multilocus genotype data. Genetics 155, 945–959. conducted in the absence of any commercial or financial relationships that could
Priyadarshan, P. M. (2016). “Genetic diversity and erosion in Hevea rubber,” in be construed as a potential conflict of interest.
Genetic Diversity and Erosion in Plants: Case Histories, eds M. R. Ahuja and
S. M. Jain (Cham: Springer International Publishing), 233–267. doi: 10.1007/ Copyright © 2018 de Souza, dos Santos, Rosa, da Silva, Mantello, Conson, Scaloppi,
978-3-319-25954-3_6 Fialho, de Moraes, Gonçalves, Margarido, Garcia, Le Guen and de Souza. This is an
Priyadarshan, P. M., and Goncalves, P. S. (2003). Hevea gene pool for breeding. open-access article distributed under the terms of the Creative Commons Attribution
Genet. Resour. Crop Evol. 50, 101–114. doi: 10.1023/A:1022972320696 License (CC BY). The use, distribution or reproduction in other forums is permitted,
Rabbi, I. Y., Hamblin, M. T., Kumar, P. L., Gedil, M. A., Ikpan, A. S., Jannink, provided the original author(s) and the copyright owner(s) are credited and that the
J. L., et al. (2014). High-resolution mapping of resistance to cassava mosaic original publication in this journal is cited, in accordance with accepted academic
geminiviruses in cassava using genotyping-by-sequencing and its implications practice. No use, distribution or reproduction is permitted which does not comply
for breeding. Virus Res. 186, 87–96. doi: 10.1016/j.virusres.2013.12.028 with these terms.

Frontiers in Plant Science | www.frontiersin.org 10 July 2018 | Volume 9 | Article 815