| |

Received: 31 October 2018    Revised: 25 January 2019    Accepted: 1 March 2019

DOI: 10.1111/pbr.12703

ORIGINAL ARTICLE

A fine‐scale genetic linkage map reveals genomic regions

associated with economic traits in walnut (Juglans regia)

Mallikarjuna K. Aradhya1  | Dianne Velasco2  | Ji‐Rui Wang2,3  |

2
Ramesh Ramasamy  | Frank M. You  | Chuck Leslie  | Abhaya Dandekar2
4 2
 |
2 2
Ming‐Cheng Luo  | Jan Dvorak

1
National Clonal Germplasm
Repository, USDA‐ARS, University of Abstract
California, Davis, California A genetic linkage map of walnut containing 2,220 single nucleotide polymorphisms
2
Department of Plant Sciences, University of
(SNPs) in 16 linkage groups (LGs) was constructed using an F1 mapping population
California, Davis, California
3
Triticeae Research Institute, Sichuan
from a cross between “Chandler” and “Idaho,” two contrasting heterozygous parents.
Agricultural University, Wenjiang, China Five quantitative yield traits, lateral fruitfulness, harvest date and three nut traits
4
Cereal Research Centre, Agriculture and (shell thickness, nut weight and kernel fill) were then mapped on to linkage groups. A
Agri‐Food Canada, Morden, Canada
significant quantitative trait locus (QTL) in LG 11 with negative additive effects sug‐
Correspondence gested heterozygote superiority in the expression of lateral bearing. A set of three
Mallikarjuna K. Aradhya, National Clonal
Germplasm Repository, USDA‐ARS, QTLs explaining ~10% of the variation in harvest date was located in LG 1. Shell
University of California, Davis, CA. thickness, nut weight and kernel fill were under the control of two to three linked
Email: [email protected]
pleiotropic QTLs in LG 1 segregating from “Idaho.” The marginal positive additive
Funding information effects of QTLs for harvest date, shell thickness and nut weight and small negative
California Walnut Board, Grant/Award
Number: 106-10162; UC Discovery, Grant/ additive effects for kernel fill suggested that the QTLs had a marginal effect on the
Award Number: IT106-10162 expression of these traits.
Communicated by: Henryk Flachowsky
KEYWORDS
interval mapping, marker‐assisted selection, multiple QTL mapping, quantitative trait loci
(QTLs), single nucleotide polymorphisms (SNPs)

1 |  I NTRO D U C TI O N parent permits for construction of female and male parental genetic
maps (Grattapaglia & Sederoff, 1994). Since trees are heterozygous
Perhaps the single most important application of molecular markers there are ample marker loci segregating within each of the parents to
in the genetic improvement of woody perennial crops is marker‐as‐ produce moderate to fine‐scale parental genetic linkage maps, which
sisted selection. Advances in sequencing technologies have enabled are then merged into a fine‐scale consensus map (Van Ooijen, 2011).
the development of essentially unlimited numbers of genetic mark‐ The genetic linkage‐based QTL mapping employs regression
ers rapidly and cost effectively for deciphering the genetic basis analyses and maximum likelihood estimations that provide statistical
of complex economic traits. Markers have been widely applied in tests for the presence of a QTL and associated parameters (Darvasi,
genome‐wide association studies, quantitative trait locus (QTL) Weinreb, Minke, Weller, & Soller, 1993; Haley & Knott, 1992; Jansen
mapping and genomic selection and marker‐assisted selection to im‐ & Stam, 1994; Jensen, 1989; Knapp, 1991; Lander & Botstein, 1989;
prove selection efficiency and accelerate development of improved Paterson et al., 1988). Linkage disequilibrium between molecular
cultivars of tree crops. markers and the large effect alleles of QTLs governing the additive
Genetic mapping in trees generally relies on F1 progeny of di‐ genetic variance of economic traits is the basis for genetic linkage
verse bi‐parental crosses where genetic recombination within each and QTL mapping. High‐density maps permit dissection of complex

Plant Breeding. 2019;138:635–646. © 2019 Blackwell Verlag GmbH |  635

wileyonlinelibrary.com/journal/pbr  
 |
636       ARADHYA et al.

QTLs to identify candidate genes of large effects within QTLs, and KNFL. The crosses were made using “Chandler” as the female parent
in most cases their effects are additive in woody perennials (Kearsey and “Idaho” as the male in 2003, 2005 and 2007 generating 425 F1
& Pooni, 1998). Markers statistically associated with one or more seedlings. The F1 plants were genotyped with a dozen microsatellite
large effect QTLs should serve as an indirect selection criterion in a markers to verify if they were true hybrids. The parents and 425 F1
marker‐assisted selection scheme. plants were grafted on to the clonal “Paradox” interspecific hybrid
The Persian walnut (Juglans regia L.) is a diploid with 2n=32 (J. hindsii × J. regia) rootstock and grafted plants were planted in the
chromosomes and genome size of ~606 Mbp (Horjales, Redondo, UC Davis teaching orchard (Davis, California, USA) during 2007–
Blanco, & Rodríguez, 2003) grown in the temperate regions of the 2008. Standard horticultural practices were followed to establish
world. The long reproductive cycle and extended juvenile period the mapping population. Phenotype data on LATRL, HARVJ, TTWT,
are limiting factors in walnut genetic improvement through conven‐ THICK and KNFL were recorded for 221 (5 and 6 years old) trees.
tional breeding. Deployment of molecular markers and marker‐as‐ The phenotypes were scored as per the modified IPGRI descriptors
sisted selection should accelerate progress in walnut breeding. To (McGranahan, Germain, Ramos, & Rigert, 1994) described as fol‐
implement this objective, extensive genomic resources have been lows: LATRL as the percentage of laterally borne nuts and HARVJ as
developed for walnut, such as bacterial artificial chromosome (BAC) Julian date; and nut traits: TTWT in grams, THICK in millimetres and
libraries, BAC‐end sequences (BESs), thousands of single nucleotide KNFL on an ordinal scale of 1–9, ranging from 1 for poorly filled to 9
polymorphism markers (SNPs) (Wu et al., 2012; You et al., 2012), a for well filled. The phenotypic data were averaged over both years,
SNP‐based genetic map (Luo et al., 2015) and a genome sequence and analysed for descriptive statistics. The data were checked for
(Martínez‐García et al., 2016). normality of distribution using the Ryan‐Joiner normality test and a
The application of molecular breeding strategies such as genetic Box–Cox transformation was applied to traits that deviated signifi‐
linkage analysis, QTL and association mapping approaches should cantly from normality.
improve selection efficiency in breeding and development of im‐
proved walnut cultivars. There is only a single QTL mapping study
2.2 | SNPs and genotyping
reported in walnut, which used a specific length amplified fragment
marker‐based genetic linkage map to identify a QTL for anthracnose The details of SNP development has been described elsewhere (You
resistance (Zhu et al., 2015). Here, we report a SNP‐based genetic et al., 2012). The mapping population comprising 425 F1 individu‐
linkage map of walnut utilizing segregation of SNP markers in an F1 als and the two parents were genotyped using a “Chandler”‐based
mapping population of 425 individuals from a bi‐parental cross be‐ Infinium II 6K SNP array (You et al., 2012). The array consisted
tween contrasting parents, “Chandler” and “Idaho.” The genetic map of 3,866 genic SNPs and 2,134 non‐genic SNPs. Data were ana‐
is then used to discover and map QTLs segregating for the lateral lysed using the Illumina Genome Studio v1.0 software (San Diego,
bearing (LATRL), harvest date (HARVJ) and nut characteristics, such California) genotyping module with a GenCall Threshold of 0.15. The
as nut weight (TTWT), shell thickness (THICK) and kernel fill (KNFL), following quality metrics were used: GenTrain score ≥0.50, minor al‐
in this population. We used a two‐step strategy to map major ef‐ lele frequency ≥0.15 and call rate >80%. Low‐quality SNPs were re‐
fect QTLs. First, we used a single QTL model, which employs interval moved manually from the data set. Highly distorted SNPs or those
mapping (IM; Lander & Botstein, 1989) to detect putative QTLs at 1‐ missing for one or both parents were removed in GenomeStudio. A
cM intervals across the genome. This step is then followed by the final data set comprising 2,879 SNPs across 425 F1 individuals and
application of a multiple QTL model (MQM; Jansen, 1993; Jansen & the parents was assembled for the linkage analysis.
Stam, 1994) which uses selected cofactors located close to the pu‐
tative QTLs in a series of iterations to locate the major effect QTLs,
2.3 | Linkage analysis
thus eliminating the variation from nearby QTLs while reducing re‐
sidual variation. The data set comprised 1,706 loci heterozygous in “Chandler” coded
as lm × ll; 175 heterozygous in “Idaho” coded as nn × np and 998 het‐
erozygous in both parents coded as hk × hk for cross pollination (CP)
2 |  M ATE R I A L S A N D M E TH O DS
type (outbreeding full‐sib family) mapping populations (Van Ooijen,
2009). Linkage analysis was performed with JoinMap® v4.1 (Kyazma
2.1 | Mapping population and phenotyping
B.V., Wageningen, Netherlands) software. Segregation distortion
An F1 mapping population derived from a bi‐parental cross be‐ among marker loci was checked using chi‐square goodness‐of‐fit
tween two contrasting clonally propagated heterozygous culti‐ (p < 0.05) for a 1:1 or 1:2:1 segregation ratios, depending on whether
vars “Chandler” and “Idaho” was used to map the walnut genome markers segregated in a test cross or co‐dominant fashion, respec‐
(Figure 1). “Chandler” is a laterally fruitful, late maturing cultivar tively. Loci with distorted segregation, missing data or monomorphic
bearing medium‐sized, thin‐shelled nuts with good KNFL, and were excluded from analysis as were individuals with missing data.
widely planted in California and around the world. “Idaho” is a ter‐ The mapping algorithm in JoinMap v4.1 for CP‐type populations
minal bearing selection from Parma, Idaho, maturing earlier than simultaneously estimates two parental maps under the constraint
“Chandler” by a week, bears large, thick‐shelled nuts with poor that the order of loci is the same in both maps. It accommodates
ARADHYA et al. |
      637

F I G U R E 1   Contrasting phenotypes
(a)
of the walnut cultivars “Chandler” and
“Idaho” used to produce the mapping
population: (a) lateral bearing habit of
“Chandler”; (b) terminal bearing habit
of “Idaho” and (c) nut characteristics in
“Chandler” (top) and “Idaho” (bottom)—
shell thickness and kernel fill. White
arrows in a and b indicate the location of
fruits showing the bearing habit [Colour
figure can be viewed at wileyonlinelibrary.
com]

(b)

(c)

markers with different segregation types and differences in recom‐ with markers segregating in both parents. Linkage groups were de‐
bination rates between parents. Finally, the maps are integrated termined by a regression mapping procedure with a threshold mini‐
into a consensus map by averaging lengths over anchored segments mum logarithm of odds difference (LOD) score of 4 and a maximum
 |
638       ARADHYA et al.

recombination fraction of 0.45. Recombination fractions were con‐ phenotypic variation in the trait was identified. For CP‐type popu‐
verted into map distance in cM using the Kosambi mapping function lations, MapQTL computes the additive effects associated with the
(Kosambi, 2016). The final integrated map consisting of 2,220 mark‐ marker locus as per the formula (muA – muH), where muA and muH
ers including the test cross‐type markers segregating in each of the are the means of the distributions of the trait associated with the “A;
parents and co‐dominant type markers segregating in both parents homozygous” and “H; heterozygous” genotypes of parent A, which
was drawn using MapChart v2.2 (Voorrips, 2002). is “Chandler” here, respectively.
Traits that deviated from normal distribution were also evaluated
by the non‐parametric Kruskal–Wallis (KW) rank sum test, which is
2.4 | QTL analysis
equivalent to the locus‐wise two‐sided Wilcox rank sum test for two
The linkage map was used to locate major effect QTLs for LATRL, genotype classes with no assumptions of the probability distribu‐
HARVJ and nut traits, TTWT, THICK and KNFL. Genome‐wide inter‐ tions of the metric traits (Lehmann & D'abrera, 1975). Generally,
val mapping (IM; Lander & Botstein, 1989; Van Ooijen, 1992) at 1‐cM both the IM and the MQM mapping procedures are quite robust
increments was performed to assess the likelihood of presence and against deviations from normality (Van Ooijen, 2009). However,
genotypic effects of segregating QTLs and residual variance irre‐ the data for the metric traits deviating from normality were trans‐
spective of marker positions. The genome‐wide (GW) LOD threshold formed using Box–Cox power transformation before performing IM
for a Type I error α < 0.05 value was obtained by a permutation test and MQM mapping. JoinMap input files and phenotypic data will
(Churchill & Doerge, 1994) with 1,000 independent permutations be available through Figshare (https​://figsh​are.com/s/e09aa​0 eb28​
for each trait. The regions of putative QTLs along the linkage groups 2d024​0688c​).
were detected based on the LOD exceeding the significant thresh‐
old value. However, the limitation of IM approach is that it does not
3 | R E S U LT S
account for interaction among QTLs, which affects the power of de‐
tection and locations of QTLs, and may even produce ghost QTLs
3.1 | Linkage map
(Lander & Botstein, 1989; Martinez & Curnow, 1994).
The multiple QTL model implemented in MQM mapping on the The final linkage map consisted of 2,220 markers mapped into 16
other hand investigates interactions among QTLs increasing the linkage groups corresponding to the haploid chromosome comple‐
efficiency and accuracy of mapping QTLs and estimation of asso‐ ment of walnut. Of the total number of markers, 1,384 (62.4%) and
ciated parameters (Jansen, 1993), thus reducing the residual vari‐ 107 (4.8%) segregated as dominant marker in 1:1 ratio in “Chandler”
ation. The MQM approach combines a multiple linear regression and “Idaho,” respectively, and 729 (32.8%) segregated as co‐domi‐
method with conventional IM to fit one QTL at a time simultaneously nant markers in 1:2:1 ratio in both parents. The overrepresentation
using the linked markers as cofactors to eliminate or to reduce ge‐ of SNPs segregating in “Chandler” compared to “Idaho” was because
netic background noise by partially absorbing QTL effects (Jansen, SNP discovery was based on “Chandler” sequences. Of the 2,879
1993; Jansen & Stam, 1994). In fact, in large backcross populations SNPs used in linkage analysis, 659 showed segregation distortions,
the linked markers, when used as regressors fully absorb effects of missing data, mapped to identical locations on the linkage map or
nearby minor QTLs (Stam, 1991). Mapping of QTLs and estimating failed to map. We used only markers that segregated in a Mendelian
the allelic effects in MQM mapping requires completion of missing fashion (1:1 or 1:2:1, p < 0.05) for linkage map construction. The
molecular marker data. The completion of missing data is accom‐ length of LGs ranged from 46.2 cM (70 markers) for LG 5 to 99.5 cM
plished by assigning weights to all potential variants at the locus. (202 markers) for LG 7 with a genome‐wide average of 69.6 cM (139
These weights are conditional probabilities based on the values of markers) and a total length of 1,114.1 cM (Table 1, Figure 2). The
neighbouring linked markers and the estimated recombination rate average distance between markers in LGs ranged from 0.35 in LG 3
by using a Monte Carlo Expectation Maximization (MCEM) algo‐ to 1.51 in LG 15. Large gaps were found in LG 15 (17 cM) and LG 10
rithm (Dempster, Laird, & Rubin, 1977; Jansen & Stam, 1994). The (13.9 cM) while the remaining 14 LGs had gaps ranging from 2.2 cM
completion of missing data equally applies to both putative QTLs in LGs 3 and 11 to 7.7 cM in LG 9, and a genome‐wide average gap of
and any missing marker genotype. The completed data are used 6.1 cM. Some gaps in the genetic map are probably due to regional
for the weighted regression of phenotype on genotype to estimate homozygosity in the “Chandler” genome (Luo et al., 2015).
model parameters and residuals (Jansen, 1996; Jansen & Stam,
1994). Therefore, to detect interactions among QTLs identified in
3.2 | Descriptive statistics and test of normality of
the IM, we applied the MQM approach as implemented in MapQTL
distribution of traits
6 (Kyazma B.V.; Van Ooijen, 2009). Marker loci displaying equal or
above LOD thresholds around the detected QTLs were selected as an The coefficient of variation in the mapping population ranged from
initial set of cofactors for the MQM scan, which accounts for greater 1.65% for HARVJ to 74.2% for LATRL. The LATRL was negatively
variation for nearby QTLs, thus reducing the residual variance. This skewed with bimodal distribution (p < 0.05) (Table S1, Figures S1
process was repeated by readjusting the cofactors until the QTL(s) & S2) suggesting a simple Mendelian or oligogenic control. HARVJ
with the highest LOD score(s) explaining the largest percentage of showed normal distribution. THICK, TTWT, and KNFL were
ARADHYA et al. |
      639

TA B L E 1   Salient features of the integrated genetic linkage map. of walnut (Juglans regia) derived from an F1 mapping population of a cross
between the two contrasting, heterozygous, clonally propagated walnut genotypes, “Chandler” and “Idaho.” The linkage groups (LG) are
pseudomolecules

Marker typea

LG Length (cM) # Markers Max gap (cM) Markers/cM lm × ll hk × hk nn × np

1 79.9 167 7.3 2.11 111 51 5

2 72.3 158 5.5 2.19 95 56 7
3 66.7 190 2.2 2.85 120 64 6
4 58.2 141 2.9 2.42 97 35 9
5 46.2 70 5.8 1.52 40 27 3
6 61.4 172 3.2 2.8 108 55 9
7 99.5 202 3.5 2.04 113 78 11
8 94 128 6.9 1.36 73 47 8
9 59 88 7.7 1.49 55 28 5
10 80.9 186 5.2 2.3 127 49 10
11 67.4 164 2.2 2.44 107 52 5
12 63.3 135 6.6 2.14 76 52 7
13 77.1 132 2.5 1.71 82 45 5
14 61 103 3.5 1.69 78 14 11
15 63.4 42 17 0.66 20 21 1
16 63.8 142 6.9 2.23 82 55 5
Total 1,114.1 2,220 97.6   1,384 729 107
Average 69.6 138.8 6.1 2.0 86.5 45.6 6.7
a
lm × ll = loci segregating in “Chandler”; nn × np = loci segregating in “Idaho”; and hk × hk = segregating in both parents.

positively skewed. All five traits exhibited platykurtic distribution. which is only about 3 cM from the peak‐marker identified in KW
The Box–Cox power transformation of data somewhat improved test confirming a strong linkage between these markers. The MQM
the normality of data as indicated by the Ryan‐Joiner normality test, scans with cofactors around the QTL with the highest LOD score
which assesses normality by calculating the statistical significance (JH073K04r_424), adjusted for each run, located a major QTL at
of correlation between the sample data and the normal scores of 48.367 cM (JM070A22r_143 from “Chandler”) on LG 11 with a
the data. LOD of 67.11 accounting for 59% of the total phenotypic variation
(Figure 3b). Interestingly, the strong negative additive effects asso‐
ciated with the QTL locus inferred heterozygote superiority in the
3.3 | QTL mapping of traits
expression of lateral bearing in walnut (Table 2).
Box–Cox transformed data were used for QTL analyses. The GW
LOD threshold (p < 0.05) was 3.2 for LATRL, HARVJ, THICK, TTWT
3.5 | Harvest date
and 3.5 for KNFL QTLs. The results of interval mapping by LGs are
presented in Figures S3–S7. QTL effects were considered significant Box–Cox transformation of HARVJ data brought the distribution
when LOD scores exceed the GW LOD threshold levels. near normality. The IM revealed three narrow regions between
41.9 cM and 60.65 cM on LG 1 with a number of marker loci sig‐
nificantly associated with HARVJ exceeding GW LOD threshold
3.4 | Lateral bearing
(p < 0.05) (Figure 4a). The first region was located between 41.93
Both non‐parametric KW test and IM revealed roughly similar re‐ and 45.13 cM with 6 loci, the second region between 52.94 cM and
sults suggesting a strong genetic basis for the trait. A large number 53.97 cM with 7 loci and the third region was between 57.97 cM
of loci originating from “Chandler” showed significant association and 61.65 cM with a tightly linked set of 3 loci and 4 other imputed
with LATRL (Figure 3a) spanning a large portion of LG 11, between loci. Most of the loci were from the “Chandler” background, while
13 cM and 67 cM at the end of the linkage group with LOD ex‐ three heterozygous loci segregated in both parents. However, loci
ceeding the GW LOD threshold level. In the KW test, statistic K* in all three regions possessed significant LOD scores accounting,
(p < 0.005) peaked at 44.962 cM (JM001H03r_540) (Table S2). In on an average, for about 8% of the total variation at each region in‐
IM, the LOD reached a peak of 70 at 48.065 cM (JH073K04r_424), dicating the presence of weak QTLs for this trait. The MQM scans
 |
640       ARADHYA et al.

F I G U R E 2   Integrated genetic linkage map based on single nucleotide polymorphism markers showing 16 linkage groups corresponding to
the haploid chromosome complement of walnut

using a set of linked loci as cofactors across the three regions S3). In addition, a locus at 19.216 cM on LG 11 showed significant
completely absorbed the nearby QTL effects from the second association (p < 0.005).
and third regions and identified a narrow region with two SNPs,
JM031P19r_51 from “Chandler” and JH003H22r_63 segregating
3.6 | Shell thickness, nut weight and kernel fill
in both parents, and an imputed locus as the putative QTLs for
HARVJ (Figure 4b). These QTLs showed significant LOD scores THICK attained normality upon transformation and exhibited a com‐
and explained an average of 10% of the total phenotypic varia‐ plex multifactorial genetic control in the IM analysis with significant
tion for HARVJ. The additive component of variation for the two QTLs on LGs 1, 5 and 11. Three loci including 2 imputed ones be‐
assayed loci and one imputed locus did show a weak, but positive tween 9.943 and 10.943 cM, on LG 1; 5 loci between 11.139 and
effect on harvest date suggesting a complex inheritance of this 13.895 cM, and 67 loci between 16.596 and 46.176 cM on LG5; and
trait (Table 2). 5 loci between 43.235 and 44.962 cM and 41 loci from 47.150 to
The non‐parametric KW test also showed a complex genetic con‐ 54.233 on LG11, exceeded the GW LOD threshold level. The MQM
trol with regions in LGs 1 (8.9 cM–79.2 cM with 53 loci), 12 (0 cM– scans with markers at 10‐cM intervals on LGs 1, 5 and 11 used as
17.8 cM with 28 loci) and 13 (between 27 cM and 33 cM with 10 cofactors completely absorbed the QTL effects on LGs 5 and 11,
loci) showing significant association (p < 0.001) with HARVJ (Table leaving the three weak but significant QTLs (JH016J20r_112 from
ARADHYA et al. |
      641

F I G U R E 3   Quantitative trait locus

(QTL) mapping of lateral bearing trait in
walnut. (a) Interval mapping (IM) showing
the single nucleotide polymorphism
loci significantly associated with the
trait on linkage group 11 along with the
percentage variation explained by the
QTLs. Genome‐wide logarithm of odds
difference (LOD) threshold based on
permutation test with 1,000 permutations
is 3.2 (p < 0.05); (b) Detection of a major
QTL (JM070A22r_143 at 48.367 cM)
by multiple QTL model scans using
neighbouring linked markers as cofactors
(LOD 67.1 explaining 59% of total
variation in lateral bearing). x‐axis = Map
position (cM) and corresponding marker
position; y‐axis = LOD (%) [Colour figure
can be viewed at wileyonlinelibrary.com]

“Idaho” and two imputed loci at 9.943 cM and 10.943 cM) on LG 1. (JM055N23r_171 from “Chandler”) significantly associated with
The KW test showed similar results as IM with LG 1 (21 loci between KNFL. The same locus showed a slightly elevated LOD in IM. Overall,
39.6 cM and 58.6 cM and 1 at 8.9 cM,), LG 5 (43 loci between 2 cM THICK, TTWT and KNFL appear to be under the control of the same
and 46.1 cM) and LG 11 (67 loci between 25 cM and 63.9 cM) in‐ set of linked pleiotropic QTLs with significant LOD scores (Figure 5).
volved in the genetic control of THICK (Table S4).
Both IM and MQM mapping of TTWT and KNFL produced iden‐
4 | D I S CU S S I O N
tical results and showed significant QTLs on LG 1, approximately
at the same interval as that of THICK (Figure 5a,b,c). Three SNPs
4.1 | Genetic linkage map
(JH016J20r_112 from “Idaho” and two other imputed loci at 1 cM
interval at 9.943 cM and 10.943 cM showed significant association The consensus linkage map consisting of 2,220 SNPs in 16 LGs con‐
with TTWT explaining 18.9, 16.8 and 7.7% of the total variation re‐ structed here is an improvement over an earlier female‐backcross
spectively. The same region on LG 1 controlled the KNFL, but in‐ map based on the same “Chandler” × “Idaho” population (Luo et al.,
volved only two QTLs (JH016J20r_112 at 9.943 cM and one imputed 2015) with 1,525 SNPs segregating only in “Chandler”. The cur‐
locus at 9.943 cM) accounting for 27.6% and 20.6% of the total phe‐ rent map included markers also segregating in “Idaho”. Both maps
notypic variation respectively. However, only a small proportion of showed high levels of synteny and collinearity within and among
the QTL effects was accounted for by the additive component pos‐ LGs. There was a marginal increase in the length of different LGs
itively affecting the TTWT, but negatively the KNFL (Table 2). The between the earlier and the current maps (1,049.5 vs. 1,114.1 cM
KW test matched the results of IM for both TTWT and KNFL, but a respectively) with a concomitant increase in the number of markers
single locus at 9.943 cM on LG 1 (JH016J20r_112) showed signifi‐ (1,525 vs. 2,220 respectively) and corresponding marker density
cant K* values of 39.8 and 59.4 for TTWT and KNFL respectively (1.45 markers/cM vs. 1.99 markers/cM respectively) contributed
(Table S3). Additionally, there was one locus on LG 1 at 32.018 cM by both parents. The patterns of distribution of markers and gaps
 |
642       ARADHYA et al.

TA B L E 2   Estimated genetic effects of quantitative trait loci (QTLs) detected for five yield traits in walnut

Logarithm of odds
Trait Linkage group Position Locus difference muA muH R2 Additive

LATRL LG11 48.367 JM070A22r_143 67.11 5.65 68.66 59.00 −63.01

HARVJ LG1 41.937 JM031P19r_51 5.10 280.28 277.08 10.10 3.20
LG1 42.691 JH003H22r_63 5.09 280.28 277.07 10.10 3.20
LG1 43.691   5.04 280.26 277.07 10.00 3.18
THICK LG1 8.943 JH016J20r_112 6.06 1.45 1.33 10.00 0.12
LG1 9.943   6.19 1.47 1.31 10.20 0.16
LG1 10.943   4.51 1.48 1.31 7.60 0.18
TTWT LG1 8.943 JH016J20r_112 10.46 20.43 16.06 18.60 4.37
LG1 9.943   9.32 20.91 15.47 16.70 5.44
LG1 10.943   4.09 20.22 16.17 7.70 4.06
KNFL LG1 8.943 JH016J20r_112 16.06 3.75 5.18 27.20 −1.42
LG1 9.943   11.26 3.68 5.28 20.00 −1.60

Note: muA = Mean of the distribution of quantitative trait associated with the homozygous genotype of “Chandler”; muH = Mean of the distribu‐
tion of quantitative trait associated with the heterozygous genotype of “Chandler”; R 2 = Percent variation explained by the QTL; Additive = Additive
genetic effects associated with the QTL.

within and among LGs between the earlier and current map were superiority at the candidate QTL controlling the trait. Unlike ag‐
consistent, but the total size of gaps reduced in the current map ronomic crops, tree crop breeders generally select for high levels
(129.2 cM vs. 97.6 cM respectively). The gaps were perhaps due to of heterozygosity as it confers greater adaptation, vigour and ho‐
regional homozygosity within the “Chandler” genome or increased meostasis to buffer temporal and spatial environmental fluctua‐
recombination percentage contributed by both parents. tions (Fridman, 2015; Seehausen, 2004). Such a selection strategy
in walnut should invariably include the heterotic locus for lateral
bearing that confers yield superiority along with other desirable
4.2 | Quantitative trait loci
traits associated with high levels of heterozygosity.
Walnut cultivars are identified based on bearing habit as lateral, ter‐ The mapping of HARVJ produced a more complex picture than
minal or intermediate, depending on whether the pistillate flowers LATRL. This is perhaps not surprising because HARVJ is a complex
are borne laterally or terminally at tips of current year twigs. Lateral trait in tree crops greatly influenced by a number of dormancy–growth
bearing walnut cultivars are preferred for their yield superiority cycle‐driven phenological stages, such as leafing, flowering and fruit
over terminal bearing cultivars. Deciphering the genetic architec‐ set, and development and maturity. Each of these developmental
ture of the lateral bearing trait is a top priority in walnut breeding events is presumably under independent and interactive genetic con‐
programmes. We used a QTL mapping approach to unravel the un‐ trol interacting with the growing environment and cultural practices
derlying genetic control of LATRL. Unlike most quantitative traits, contributing to the complex genetic makeup of HARVJ. Being under
which are unimodal, the distribution for LATRL in the mapping the control of two minor effect QTLs, one segregating in Chandle'r' and
population was bimodal suggesting that the trait is possibly under one segregating in both parents, along with an imputed locus explaining
the control of a major Mendelian gene. The IM detected a region 10% of total variation, suggests a multifactorial inheritance of the trait.
of significant QTLs associated with LATRL spanning most of LG 11 THICK exhibited complex genetic control involving three LGs,
with a large number of loci segregating from “Chandler,” roughly 1, 5 and 11. The QTLs on LG 11 overlapped with the region con‐
matching the results of the non‐parametric KW test. A noteworthy trolling the LATRL, which includes the centromere of chromosome
discovery is the detection of a single major QTL (JM070A22r_143 11. Walnut chromosomes show greatly depressed recombination in
from “Chandler”) in MQM mapping with a LOD score of 67.11 ac‐ the proximal regions (Luo et al., 2015), and co‐location of QTLs is
counting for 59% of the total variation in LATRL almost behaving very likely caused by co‐segregation of a large number of loci. All
like a Mendelian gene. Fine‐scale mapping around this QTL is cur‐ three nut traits, THICK, TTWT and KNFL, appeared to be under the
rently underway to locate the causal gene governing lateral bearing genetic control of two to three pleiotropic loci on LG 1. The addi‐
in walnut. The identification of the gene for LATRL should improve tive effect of QTLs was marginally positive for THICK and TTWT,
selection efficiency and permit for juvenile selection of productive whereas it was negative for KNFL. The major QTL (JH016J20r_112),
lateral bearing cultivars thus saving time and resources in walnut originating in the “Idaho” genetic background, marginally influencing
breeding. There is a significant component of additive effect af‐ THICK, TTWT positively and KNFL negatively is probably a com‐
fecting the trait negatively (muA – muH) indicating heterozygote pound QTL with interactive tightly linked loci.
ARADHYA et al. |
      643

F I G U R E 4   Quantitative trait loci

(QTLs) mapping of harvest date in walnut.
(a) Interval mapping (IM) showing the
single nucleotide polymorphism loci
significantly associated with the trait on
linkage group 1 along with the percentage
variation explained by the QTL. Genome‐
wide logarithm of odds difference
(LOD) threshold‐based permutation
test with 1,000 permutations is 3.2
(p < 0.05). (b) Detection of three major
QTLs (JM031P19r_51, JH003H22r_63
at 41.947, 42.691 cM, and a locus at
43.691 cM imputed by MCEM algorithm)
on linkage group 1 by multiple QTL
model scans using a subset of linked loci
as cofactors on linkage groups 1, 12 and
13 (average LOD 5 explaining ~10% of
total variation). x‐axis = Map position
(cM) and corresponding marker position;
y‐axis = LOD (%) [Colour figure can be
viewed at wileyonlinelibrary.com]

Understanding the genetic effects (additive, dominance and 5 | CO N C LU S I O N S

epistatic interactions) of QTLs controlling the target traits is valu‐
able for developing effective MAS (Jannink, Moreau, Charmet, & This study contributes to the understanding of genetic architecture
Charcosset, 2009). The large number of minor effect QTLs detected of five important yield attributes in walnut. Notable among the QTLs
across several LGs in the IM of HARVJ, THICK, TTWT and KNFL identified is the one for LATRL, which has the potential to be useful
probably represents polygenic loci with epistatic interactions, some as marker‐assisted juvenile selection criteria to select for productive
of which may contribute to a steady response to selection, without lateral bearing trees early in the seedling and sapling stages in wal‐
major leaps in improvement. The major effect QTL for LATRL could nut breeding programmes. We are currently focusing on locating the
represent a single heterozygous locus, where a dominance effect casual genetic locus governing the bearing habit of walnut through a
determines the bearing habit, while several interacting minor effect fine‐scale mapping within the genomic region where the major QTL
loci for other traits suggest that response to selection depends on (JM070A22r_143) was discovered in this study. The weak QTLs gov‐
the nature of epistatic interaction (Hansen, 2013). erning HARVJ and nut traits such as THICK, TTWT and KNFL may be
 |
644       ARADHYA et al.

F I G U R E 5   Quantitative trait loci

(QTLs) mapping of nut traits, shell
thickness, nut weight and kernel fill
in walnut. (a) QTL on linkage group 1
governing shell thickness in walnut
detected by multiple QTL model (MQM)
scans using automatic cofactor selection
of loci on linkage groups 1, 5 and 11
with significant association with the trait
discovered by interval mapping (average
logarithm of odds difference (LOD)
5 explaining ~10% of total variation).
Genome‐wide LOD threshold‐based
permutation test with 1,000 permutations
is 3.2 (p < 0.05). (b) QTL significantly
associated with nut weight on linkage
group 1 detected by interval mapping
and MQM scan produced similar results.
Genome‐wide LOD threshold‐based
permutation test with 1,000 permutations
is 3.2 (p < 0.05). (c) QTL significantly
associated with kernel fill on linkage
group 1 detected by interval mapping
and MQM scan produced similar results.
Genome‐wide LOD threshold‐based
permutation test with 1,000 permutations
is 3.5 (p < 0.05). x‐axis = Map position
(cM) and corresponding marker position;
y‐axis = LOD (%) [Colour figure can be
viewed at wileyonlinelibrary.com]
ARADHYA et al. |
      645

of limited application as selection criteria and will require further in‐ Hansen, T. F. (2013). Why epistasis is important for selection and ad‐
vestigation in diverse genetic backgrounds. Overall, this study forms aptation. Evolution, 67(12), 3501–3511. https​ ://doi.org/10.1111/
evo.12214​
the foundation for the future research on QTL analysis of economi‐
Horjales, M., Redondo, N., Blanco, A., & Rodríguez, M. A. (2003).
cally important traits to develop juvenile selection criteria for rapid Cantidades de DNA nuclear árboles y arbustos. Nova Acta Cientifica
development of improved cultivars of walnut. Compostelana (Bioloxía), 13, 23–33.
Jannink, J. L., Moreau, L., Charmet, G., & Charcosset, A. (2009).
Overview of QTL detection in plants and tests for synergistic epi‐
AC K N OW L E D G E M E N T S static interactions. Genetica, 136(2), 225. https​://doi.org/10.1007/
s10709-008-9306-2
The authors thank the editor and the anonymous reviewers for their Jansen, R. C. (1993). Interval mapping of multiple quantitative trait loci.
constructive comments and suggestions. This work was supported Genetics, 135(1), 205–211.
by the California Walnut Board (106‐10162) and the UC Discovery Jansen, R. C. (1996). A general Monte Carlo method for mapping multiple
quantitative trait loci. Genetics, 142(1), 305–311.
(IT106‐10162) Grants.
Jansen, R. C., & Stam, P. (1994). High resolution of quantitative traits into
multiple loci via interval mapping. Genetics, 136(4), 1447–1455.
Jensen, J. (1989). Estimation of recombination parameters between a
C O N FL I C T O F I N T E R E S T
quantitative trait locus (QTL) and two marker gene loci. Theoretical
and Applied Genetics, 78(5), 613–618. https​ ://doi.org/10.1007/
None.
BF002​62554​
Kearsey, M. J., & Pooni, H. S. (1998). The genetical analysis of quantitative
traits. Cheltenham, UK: Stanley: Thornes (Publishers) Ltd.
AU T H O R C O N T R I B U T I O N S
Knapp, S. J. (1991). Using molecular markers to map multiple quantitative
trait loci: Models for backcross, recombinant inbred, and doubled
MA, CL, AD, MCL and JD planned the study; FMY, MCL, JW, DV and
haploid progeny. Theoretical and Applied Genetics, 81(3), 333–338.
CL contributed to data development and assembly; CL developed https​://doi.org/10.1007/BF002​28673​
the mapping population and collected phenotypic data; MA, JW, RR Kosambi, D. D. (2016). The estimation of map distances from recombina‐
and DV analysed the data; MA drafted the manuscript; and MA, DV tion values. In R. Ramaswamy (Ed.), D. D. Kosambi (pp. 125–130). New
and JD revised manuscript. All authors read and approved the final Delhi, India: Springer.
Lander, E. S., & Botstein, D. (1989). Mapping mendelian factors under‐
draft of the manuscript.
lying quantitative traits using RFLP linkage maps. Genetics, 121(1),
185–199.
Lehmann, E. L., & D'abrera, H. J. (1975). Nonparametrics: Statistical meth‐
ORCID ods based on ranks. San Francisco, CA: Holden‐Day.
Luo, M.‐C., You, F. M., Li, P., Wang, J.‐R., Zhu, T., Dandekar, A. M., …
Mallikarjuna K. Aradhya  https://orcid.
Dvorak, J. (2015). Synteny analysis in Rosids with a walnut phys‐
org/0000-0002-2118-7885 ical map reveals slow genome evolution in long‐lived woody pe‐
Dianne Velasco  https://orcid.org/0000-0003-1432-0630 rennials. BMC Genomics, 16(1), 707. https​ ://doi.org/10.1186/
s12864-015-1906-5
Ji‐Rui Wang  https://orcid.org/0000-0002-5019-6844 Martinez, O., & Curnow, R. N. (1994). Missing markers when estimat‐
Ramesh Ramasamy  https://orcid.org/0000-0002-6771-4177 ing quantitative trait loci using regression mapping. Heredity, 73(2),
198–206. https​://doi.org/10.1038/hdy.1994.120
Abhaya Dandekar  https://orcid.org/0000-0001-7925-4086 Martínez‐García, P. J., Crepeau, M. W., Puiu, D., Gonzalez‐Ibeas, D.,
Whalen, J., Stevens, K. A., … Chakraborty, S. (2016). The walnut
Ming‐Cheng Luo  https://orcid.org/0000-0002-9744-5887
(Juglans regia) genome sequence reveals diversity in genes coding
for the biosynthesis of non‐structural polyphenols. The Plant Journal,
REFERENCES 87(5), 507–532.
McGranahan, G. H., Germain, E., Ramos, D. E., & Rigert, K. (1994).
Churchill, G. A., & Doerge, R. W. (1994). Empirical threshold values for Descriptors for walnut (Juglans spp.). Rome, Italy: International Plant
quantitative trait mapping. Genetics, 138(3), 963–971. Genetic Resource Institute.
Darvasi, A., Weinreb, A., Minke, V., Weller, J. I., & Soller, M. (1993). Paterson, A. H., Lander, E. S., Hewitt, J. D., Peterson, S., Lincoln, S.
Detecting marker‐QTL linkage and estimating QTL gene effect and E., & Tanksley, S. D. (1988). Resolution of quantitative traits into
map location using a saturated genetic map. Genetics, 134(3), 943–951. Mendelian factors by using a complete linkage map of restriction
Dempster, A. P., Laird, N. M., & Rubin, D. B. (1977). Maximum likelihood fragment length polymorphisms. Nature, 335(6192), 721–726.
from incomplete data via the EM algorithm. Journal of the Royal Seehausen, O. (2004). Hybridization and adaptive radiation. Trends
Statistical Society. Series B (Methodological), 39(1), 1–38. in Ecology & Evolution, 19(4), 198–207. https​ ://doi.org/10.1016/j.
Fridman, E. (2015). Consequences of hybridization and heterozygosity tree.2004.01.003
on plant vigor and phenotypic stability. Plant Science, 232, 35–40. Stam, P. (1991). Some aspects of QTL analysis. In J. Pešek, M.
https​://doi.org/10.1016/j.plant​sci.2014.11.014 Herman, J. Hartmann (Eds.), Proceedings of the 8th Meeting of the
Grattapaglia, D., & Sederoff, R. (1994). Genetic linkage maps of Eucalyptus EUCARPIA Section Biometrics on Plant Breeding, July 1–6, 1991, Brno,
grandis and Eucalyptus urophylla using a pseudo‐testcross: Mapping Czechoslovakia (pp. 23–32). Hrušovany, Czechoslovakia: Research
strategy and RAPD markers. Genetics, 137(4), 1121–1137. Institute of Agroecology and Soil Management.
Haley, C. S., & Knott, S. A. (1992). A simple regression method for map‐ Van Ooijen, J. W. (1992). Accuracy of mapping quantitative trait loci in
ping quantitative trait loci in line crosses using flanking markers. autogamous species. Theoretical and Applied Genetics, 84(7–8), 803–
Heredity, 69(4), 315–324. https​://doi.org/10.1038/hdy.1992.131 811. https​://doi.org/10.1007/BF002​27388​
 |
646       ARADHYA et al.

Van Ooijen, J. W. (2009). MapQTL® 6, Software for the mapping of quan‐ locus for anthracnose resistance in walnut (Juglans regia L.). BMC
titative trait in experiment populations of diploid species. Wageningen, Genomics, 16(1), 614. https​://doi.org/10.1186/s12864-015-1822-8
Netherlands: Kyazma BV.
Van Ooijen, J. W. (2011). Multipoint maximum likelihood mapping in a
full‐sib family of an outbreeding species. Genetics Research, 93(5),
343–349. https​://doi.org/10.1017/S0016​67231​1000279 S U P P O R T I N G I N FO R M AT I O N
Voorrips, R. E. (2002). MapChart: Software for the graphical presenta‐
tion of linkage maps and QTLs. Journal of Heredity, 93(1), 77–78. https​ Additional supporting information may be found online in the
://doi.org/10.1093/jhere​d/93.1.77 Supporting Information section at the end of the article.     
Wu, J., Gu, Y. Q., Hu, Y., You, F. M., Dandekar, A. M., Leslie, C. A., … Luo,
M.‐C. (2012). Characterizing the walnut genome through analyses of
BAC end sequences. Plant Molecular Biology, 78(1–2), 95–107. https​://
How to cite this article: Aradhya MK, Velasco D, Wang J‐R, et
doi.org/10.1007/s11103-011-9849-y
You, F. M., Deal, K. R., Wang, J., Britton, M. T., Fass, J. N., Lin, D., … Dvorak, al. A fine‐scale genetic linkage map reveals genomic regions
J. (2012). Genome‐wide SNP discovery in walnut with an AGSNP associated with economic traits in walnut (Juglans regia). Plant
pipeline updated for SNP discovery in allogamous organisms. BMC Breed. 2019;138:635–646. https​://doi.org/10.1111/pbr.12703​
Genomics, 13(1), 354. https​://doi.org/10.1186/1471-2164-13-354
Zhu, Y., Yin, Y., Yang, K., Li, J., Sang, Y., Huang, L., & Fan, S. (2015).
Construction of a high‐density genetic map using specific length
amplified fragment markers and identification of a quantitative trait