Assessment of a New Strategy for Selective Phenotyping Applied to Complex Traits in Brassica Napus

Open Journal of Genetics, 2012, 2, 190-201 OJGen http://dx.doi.org/10.4236/ojgen.2012.24025 Published Online December 2012 (http://www.SciRP.org/journal/ojgen/)

Assessment of a new strategy for selective phenotyping applied to complex traits in Brassica napus

Christophe Jestin1,2, Patrick Vallée1, Claude Domin1, Maria J. Manzanares-Dauleux3,4, Régine Delourme1

1Institut National de la Recherche Agronomique, UMR1349, Institut de Génétique, Environnement et Protection des Plantes, Le Rheu, France 2Centre Technique Interprofessionnel des Oléagineux et du Chanvre, Centre de Grignon, Thiverval-Grignon, France 3Agrocampus Ouest, UMR1349, Institut de Génétique, Environnement et Protection des Plantes, Rennes, France 4Université Européenne de Bretagne, Rennes, France Email: [email protected]

Received 25 September 2012; revised 24 October 2012; accepted 22 November 2012

ABSTRACT include several hundred genes [1,2]. These large support intervals are a major limitation to the use of QTL in The accurate mapping of quantitative trait loci (QTL) breeding programs through marker-assisted selection depends notably on the number of recombination (MAS). Indeed, the larger the support interval, the events occurring in the segregating population. The greater the risk that the QTL will be lost or undesirable cost of phenotyping often limits the sample size used genes will be introgressed. The number of recombination in QTL mapping. To get round this problem, we as- events is a determining factor for mapping QTL accur- sessed a selective phenotyping method, called qtlRec ately. However, the type and restricted size of popula- sampling. In order to improve the accuracy of QTL tions (F2, doubled haploids…) commonly used to detect mapping, a subset of individuals was selected to and map QTL in plants limit these events. One solution is to maximize the number of recombination events at pu- use populations with much larger numbers of individuals tative QTL positions; the usefulness of this subset was but then it is time consuming and costly, both financially compared to a selected sample built to maximize the and in terms of human means, to genotype markers and/or recombination rate over the whole genome. We as- phenotype traits and this can limit this approach. sessed this method on the quantitative oil content trait To get round these problems, different methodologies in Brassica napus. We showed that the qtlRec strategy were developed to either improve the experimental could allow increasing accuracy (both support interval design or use more efficient statistical methods. Among and position) of QTL location while it maintained a the different possibilities, the use of phenotypically (se- similar power of detection. We then applied this ap- lective genotyping), or genotypically (selective pheno- proach to the B. napus—Leptosphaeria maculans patho- typing) selected samples can improve QTL mapping system for which resistance QTL with minor effect precision in comparison to random samples of the same were previously identified. This allowed the validation size. Selective genotyping (SG) [3,4] consists in select- of the QTL in six genomic regions. The qtlRec method ing the most phenotypically informative progeny out of is an attractive strategy for validating QTL in multiple the whole population: only the individuals from the ex- year and/or location trials for a trait which requires tremes of the phenotypic distribution are genotyped. costly and time-consuming phenotyping. Several authors refer to the success of this method applied with classical QTL detection methods (e.g. [5]) Keywords: Selective Phenotyping; QTL; Brassica or combined with methods of QTL detection based on Napus; Leptosphaeria maculans linkage disequilibrium, so-called association mapping (e.g. [6]). However, SG offers less benefit for a 1. INTRODUCTION quantitative trait which is mediated by a large number of QTL with small effects than for a trait controlled by a Many traits in plants as well as animals and humans are few QTL with large effects when selection is made on complex and controlled by quantitative trait loci (QTL). rather small populations (<500 individuals) [7]. Selective The genetic analysis of these complex traits showed that phenotyping (SP), used when trait phenotyping is costly, most reported QTL correspond to large genomic regions consists of selecting a subset of the most informative covering 10 to 30 centiMorgans (cM), which usually individuals based solely on genotypic data from a large

OPEN ACCESS C. Jestin et al. / Open Journal of Genetics 2 (2012) 190-201 191 population. Brown and Vision [8] developed the MapPop which maximizes the number of recombination events in software which proposes to select a reduced subset of the nine previously detected QTL regions [18]. An appro- most informative individuals, based on the number and ximate selection rate of 50% was applied since Jin et al. position of crossover sites detected from the genotype [14] showed that, for a trait showing high heritability, data. This selection should optimize the distribution of selectively phenotyping of 50% of the entire progeny recombination points all over the genome. The effect of retains most of the information needed for QTL detection. this selection procedure on QTL detection was evaluated In order to evaluate the impact and relevance of our in simulation [9] and empirical studies in different plants sampling strategy, we compared the power and accuracy such as Arabidopsis [10], barley [11], maize [12] or of QTL detection in the selected and whole populations pepper [13]. With the same objective, other authors pro- as well as in a sample selected using the method imple- posed, through simulation studies, methods to select 1) mented in MapPop software. individuals which maximize their genotypic dissimilarity Since we showed that this methodology could improve using markers across the entire genome, markers on the QTL accuracy, we then applied it to oilseed rape quanti- chromosome that contained a known QTL or a single tative stem canker resistance. Actually, phenotyping of marker near the QTL [14]; 2) the individuals with a this trait entails substantial financial costs and human maximal number of recombination events considering means. The stem canker disease caused by the fungus (uniRec) or not (maxRec) the uniformity of their distri- Leptosphaeria maculans is an internationally important bution across the genome [15]. Methods based on genetic disease of oilseed rape causing serious losses in Europe, dissimilarity are intended to improve the power of QTL Australia and North America [19,20]. Both qualitative detection whereas methods based on recombination rate and quantitative resistance were identified in B. napus or are intended to improve the accuracy of QTL detection. in related species and are reviewed in [21,22]. In pre- In all cases, the sample size, selected or not, affects QTL vious studies, our search for quantitative resistance fac- detection. As previously shown, the larger the population, tors has focused on one source of resistance, the variety the more accurate the QTL detection [16]. However, a “Darmor”, for which resistance QTL were detected in selected population brings more power and precision to two genetic backgrounds [23,24]. From the above whole the detection process of QTL with small effects in com- DYDH population, 152 DH lines were originally used parison to an unselected sample of the same size, and it for stem canker resistance QTL detection [24]. We applied can reduce the number of false positive QTL [9]. our sampling strategy based on eight stem canker QTL Until now, SP methods were applied to refine the posi- regions identified in this previous study and we also tion of major QTL (e.g. [17]) or using a selection based compared the power and accuracy of QTL detection with on the entire genome without prior information on puta- this second complex trait. tive QTL positions. The previous simulation studies using prior information on QTL position for selection were 2. MATERIAL AND METHODS rather simplistic (one QTL) or involved ideal situations 2.1. Plant Material which are rarely representative of reality: No missing data, a small genome, few QTL identified… In this con- The segregating DH population used in this study is text, we investigated the potential interest of a SP me- derived from the “Darmor-bzh” x “Yudal” cross and con- thod using a sampling strategy based solely on QTL mar- sists of 442 DH lines including 225 tall and 217 dwarf kers, in a case where many QTL were identified. This lines, available at INRA (Le Rheu). This material was strategy intends to improve the accuracy of QTL mapp- obtained as described in [25]. “Darmor-bzh” is a dwarf ing (position and support interval) by choosing the indi- isogenic line resulting from the introduction of the dwarf viduals that maximize recombination at QTL markers bzh gene in the French winter cultivar “Darmor”. “Yu- and could thus be used for the efficient validation of dal” is a spring Korean line that behaves as an early- QTL in different environments for a trait that is pheno- flowering winter type in temperate climates. “Darmor- typing-costly. The method would then consist in select- bzh” and “Yudalv” are resistant and susceptible to L. ma- ing a sample of individuals from a large population culans, respectively. which has been totally genotyped and phenotyped in a single environment for a first QTL detection. In order to 2.2. Selection of Subpopulations test this strategy, we used an oilseed rape (Brassica For oil content, a total of nine QTL regions on eight napus L.) segregating population derived from the cross linkage groups (A1, A3, A6 (two regions), A10, C2, C3, “Darmor-bzh” x “Yudal” (DY). It included 442 doubled C5 and C6) were chosen from a previous study [18]. A haploid (DH) lines that were genotyped and phenotyped subset of individuals which maximized recombination at in three experiments and used for seed oil content QTL QTL positions was selected from the 442 DYDH identification [18]. We selected a subset of individuals population. For this, each DH line was recorded as

Copyright © 2012 SciRes. OPEN ACCESS 192 C. Jestin et al. / Open Journal of Genetics 2 (2012) 190-201 recombined (1) or not (0) at each QTL position. At each the different samples was compared using a student test position, a DH line was recorded as recombined if a applied on the differences between the lengths of each recombination occurred between the markers located in LG (α = 0.01). the confidence interval of the QTL. The sum of the number of recombined regions per DH line was used to select 2.4. Field Experiment for Stem Canker the DH lines with the maximum recombination rate over A field experiment with the 279 DH lines of the DY all the QTL regions. We refer to this methodology as population was conducted at one location (INRA Experi- qtlRec, in reference to the uniRec or maxRec methods of mental unit, Le Rheu, France) using an design with three Jannink [15]. The subpopulation consisted of 200 DH (called “qtlRec sample”). In order to compare our sampl- replicates in 2006-07. The 129 dwarf and 150 tall lines ing strategy with that based on the number of recombi- were arranged in separate trials. Control lines as well as nation events throughout the genome, another subpopu- the parental lines were included in both trials. The con- lation of 200 DH (called “MapPop sample”) was chosen trols were winter-type B. napus cultivars showing dif- out of the same full population of 442 DH, using the ferent levels of L. maculans resistance: “Jet Neuf” (resis- MapPop 1.0 software [8]. The selection criterion applied tant), “Darmor” (resistant), “Falcon”(partially resistant), to identify the most informative lines was the expected “Eurol” (moderately susceptible). Infected rapeseed stub- maximum bin length (eMBL), i.e. the expected maximum ble collected from the previous year’s trial was scattered distance between two recombination points. eMBL for through the field to increase inoculum pressure. the MapPop population was 8.95 cm compared to 6.83 The stem canker severity was evaluated for each line cm for the whole population. For stem canker resistance, as in [24]. Forty plants per plot were uprooted and crown a total of nine regions on eight linkage groups (A2, A6, canker was assessed on a 1 - 6 scale as follows: 1 = no A7, A8, A9, C2, C4 (two regions) and C8) were chosen disease, 2 = 1% - 5%, 3= 6% - 50%, 4 = 51% - 75%, 5 = from previous results [24]. Two “qtlRec” samples were 76% - 100% of crown section cankered. An additional selected from the same 442 DH population: A 150 DH disease score category of 6 was used to indicate plants “qtlRec” sample (“150Q”) which was compared to the broken at the crown from severe canker. All crown 150 DH previously studied sample [24] (“150R”) and a canker data were transformed to a standardized 1 - 9 200 DH “qtlRec” sample (“200Q”) to get the same disease severity scale using the formula: G2 index = [(N1 selection rate as for oil content. As 118 DH lines were * 0) + (N2 * 1) + (N3 * 3) + (N4 * 5) + (N5 * 7) + (N6 * present in all sub-populations, the whole population used 9)]/Nt where N1, 2,…6 = the number of plants with a in this study for stem canker resistance consisted of 279 canker score of 1, 2,…6, respectively, and Nt = the total DH. Based on the results obtained with the qtlRec me- number of plants assessed. thod (see Results section) on oil content, no MapPop sample was used for stem canker trait. 2.5. Statistical Analyses for Stem Canker Trial For each dwarf and tall DH trial, the analysis of variance 2.3. Genetic Markers and Maps (ANOVA; proc GLM of statistical Analysis System, For the DY population, the published maps [23,26,27], SAS, [31]) partitioned total variation into line, replicate that were recently updated [28], were used as a starting and error effects (Pij = µ + Li + Rj + eij where Pij is the point to choose the markers and build the DY genetic G2 disease index of the ith line located in the jth map. In addition, physical functional markers (prefixed replicate, µ the mean of all data, Li the line i effect, Rj “CZ”), obtained through a Genoplante project in collabo- the replicate j effect and eij the residual). The trial effect ration with INRA-Evry France (coll. H. Belcram and B. was also tested from data on the control varieties. Herita- 2 Chalhoub, unpublished data; primers are available upon bility (h ) was estimated for the whole DH population 2 2 2 2 2 request to Genoplante) were used. PCR assays were con- with the formula: h = g /[g + (e /n)] with g the 2 ducted essentially as described in [18]. In all, a set of 549 genetic variance, e the environmental variance and n markers was chosen according to the map coverage and the number of replicates. the number of missing genotyping data. The linkage groups (LGs) were built from the whole 2.6. QTL Mapping population using the 549 chosen markers and a LOD The mean seed oil content in three experiments (RE01: threshold of 5.0 with Mapmaker/Exp 3.0 software [29]. Rennes 2001; RE02: Rennes 2002; SE02: Lille 2002) as Genetic distances expressed in centiMorgan (cM) bet- calculated by Delourme et al. [18] and the mean disease ween markers were estimated with the Haldane function index for stem canker calculated over the three replicates [30]. This map was used for QTL detection in all sub- of this study were used for QTL mapping. QTL detection populations. However in order to evaluate the effect of was performed for each population and trait using com- selective sampling, the size of the LGs determined with posite interval mapping (CIM) implemented in Windows

QTL Cartographer 2.5 [32]. A forward-backward step- was detected in the whole population but not in a sub- wise regression analysis was used with Pin/out = 0.05 population, it was considered as a false negative (FN) in and QTL were detected using CIM procedure, with 10 the subpopulation; in contrast, if a QTL was not detected cofactors and a 10 cm window size. The graphic of each in the whole population but was detected in the sub- LG carrying QTL was generated with MAPCHART 2.2 population, it was considered as a false positive QTL software [33]. Only the genetic map obtained with the (FP). According to these postulates, we defined the spe- whole population was drawn and the QTL identified in cificity and sensitivity criteria closely related to the de- all populations were projected on this same map. We tection power, as previously proposed by many authors used the LG nomenclature proposed by the Multinational (e.g. [14]). The specificity was defined as the proportion Brassica Genome Project Steering Committee. The QTL of true QTL among all QTL detected within each were named according to their location on each LG [27] subpopulation. The sensitivity is the proportion of true for stem canker resistance, i.e. QLmA9 for QTL of QTL detected in a subpopulation compared to the total resistance to L. maculans located on the LG A9, and true QTL detected in the whole population. The spe- according to their location and the trial for seed oil cificity (Sp) and the sensitivity (Sn) were calculated as content, i.e. OilRE01A9 for QTL located on the LG A9 follows: and detected in RE01 trial. Sp TP TP FP Sn TP TP FN 2.7. Comparison of QTL Analyses between Sub- and Whole Populations 3. RESULTS To estimate the effect of our qtlRec selective sampling, 3.1. Genetic Maps Used for Oil Content Analyses we compared different parameters between the popula- The size and the average space between markers on each tions: The QTL detection power estimated with the sen- LG for the qtlRec, MapPop and whole populations are sitivity and specificity criteria and the support interval of shown in Table 1. The size of the genetic map obtained the QTL. We adopted the following conventions: A QTL peak in the MapPop sample was significantly (P < 0.01) larger was defined at the maximum LOD value and the support (3914.2 cm) than the one obtained on the qtlRec subpo- interval was defined as the interval where the LOD score pulation (Total map size = 3525 cm), which is consistent decreased of one LOD unit on both sides of the maxi- with our strategy of selecting individuals that maximize mum. We assumed that QTL detected in the whole po- recombination on the whole genome or at previously pulation were true QTL. If the support interval of a QTL identified QTL to built these two populations. The size mapped in the subpopulations overlapped with a true of the genetic map in the whole population was lower QTL, it was considered as a true positive QTL (TP); if (3292.3 cm) than in the two subpopulations (qtlRec and not, it was considered as a false positive (FP). If a QTL MapPop samples), as expected.

Table 1. Length and average space (in Haldane cM) between markers on each linkage group (LG) for each sub-population and the whole population derived from the cross “Darmor-bzh” x “Yudal”. LGs used for selecting the qtlRec sample are indicated by an asterisk.

Sample selected using the MapPop software Sample selected using the qtlRec strategy Whole population Linkage groups (LG) Length Average space between markers Length Average space between markers Length Average space between markers A1* 223.6 5.9 226.9 6.0 198.9 5.2 A2 170.0 7.7 57.2 7.1 147.8 6.7 A3* 305.3 6.0 269.0 5.3 253.8 5.0 A4 107.2 6.0 89.0 4.9 97.5 5.4 A5 153.8 8.1 138.0 7.3 133.8 7.0 A6* 178.8 5.6 193.9 6.1 150.4 4.7 A7 173.4 7.9 141.5 6.4 143.2 6.5 A8 205.2 7.9 152.6 5.9 158.6 6.1 A9 236.4 5.5 206.3 4.8 198.7 4.6 A10* 211.4 6.6 223.0 7.0 170.7 5.3 C1 220.6 7.6 192.6 6.6 191.6 6.6 C2* 275.5 8.6 223.9 7.0 219.9 6.9 C3* 349.5 9.4 296.1 8.0 299.9 8.1 C4 208.7 6.5 187.1 5.8 175.4 5.5 C5* 213.4 8.2 245.1 9.4 173.8 6.7 C6* 169.9 10.0 141.3 8.3 135.3 8.0 C7 184.9 7.7 169.5 7.1 160.1 6.7 C8 196.6 9.8 154.5 7.7 167.0 8.4 C9 131.0 4.5 117.5 4.1 115.9 4.0 Total 3915.2 7.1 3525.0 6.4 3292.3 6.0

3.2. Additive QTL Detection in the Whole and 500 permutation tests. However, QTL at LOD 2.5 were Sub-DH Populations for Oil Content also considered since most of them colocalized with QTL detected at other LOD thresholds in the other popu- We observed the same continuous distribution pattern in the whole, MapPop and qtlRec samples. No significant lations or for the other variables. QTL mapping per- difference was found between the sub-population means formed on the 442 DH lines globally revealed 35 QTL or between the sub-populations and the whole population for the three variables (OilRE01, OilRE02 and OilSE02), means. This confirms that the selection based on the distributed on 14 LGs. Of these, 25 QTL were located in genotypic data had no effect on the phenotypic distribu- the regions that were used for qtlRec sampling. In the tion. MapPop and qtlRec populations, 20 and 19 QTL were The results of the CIM analyses for the whole and identified on 12 and 10 LGs, respectively. The overall- sub-DH populations are summarized in Table 2. LOD explained phenotypic variation was estimated at 46.5%, thresholds of 3.0 (for the whole and qtlRec populations) 51.9% and 44.4% in the whole, the MapPop and the and 3.1 (for the MapPop population) were obtained after qtlRec populations, respectively.

Table 2. The oil content QTL detected on the linkage groups of the “Darmor-bzh” x “Yudal” DH populations: LOD score, peak position and support interval (SI) in cM for each QTL.

Whole population qtlRec sample MapPop sample QTL LOD Position (SI) LOD Position (SI) LOD Position (SI) OilRE01A1 17.9 128.8 (17.6) 5.4 118.7 (18) 7.7 108.3 (9.5) OilRE02A1 14.8 128.8 (8.2) 3 126.8 (15.6) 5.1 106.3 (27.6) OilSE02A1 23.3 126.8 (5.0) 9.7 120.7 (6.1) 5.9 106.3 (11.2) OilRE02A2.1 3.6 53.5 (14.9) OilRE01A2.2 7.7 130.2 (24.5) 3.7 111.7 (16.0) 8.2 136.2 (15.5) OilRE02A2.2 3 117.7 (21.5) OilRE01A3 3.1 241.3 (22.8) 5.3 234.9 (16.8) OilRE02A3 2.9 239.7 (36.7) OilSE02A3 4.8 234.5 (13.4) 4.4 213.0 (12.4) 3.1 218.6 (25.6) OilSE02A4 2.7 68.2 (24.0) 2.6 68.2 (39.4) OilRE01A6.1 8.4 27.6 (4.2) OilRE02A6.1 3.7 27.6 (15) OilSE02A6.1 7.3 34.4 (15.3) OilRE01A6.2 9.5 136.5 (11.2) 6.8 136.5 (11.2) 5.3 138.5 (18.2) OilRE02A6.2 14.6 138.5 (13.5) 5.9 138.5 (11.2) 7.7 136.5 (13.2) OilSE02A6.2 3.1 128.5 (18.4) 3.5 128.1 (18.4) OilRE01A7 4.3 90.2 (15.6) 4 39.3 (17.7) OilRE01A9 3.3 115.0 (26.2) 3.1 108.1 (13.4) OilRE01A10.1 3.6 0.0 (10) OilRE01A10.2 3.4 67.5 (19.4) OilRE02A10.2 2.6 56.8 (33.9) 3.2 56.8 (9.5) 2.6 56.8 (4.3) OilSE02A10.2 7.6 55.1 (18.1) 5.3 59.4 (9.1) OilRE01C2.1 3.6 46.8 (58.5) OilRE02C2.1 12.9 14.5 (14.5) 2.8 10.0 (47.7) 3.8 14.5 (18.5) OilSE02C2.1 3.2 0.0 (10) OilRE02 C2.2 3.3 171.3 (33.7) OilRE02C3.1 2.7 105.2 (20.6) OilRE01C3.2 4.8 258.9 (35) 3.8 265.2 (12) OilRE02C3.2 2.7 267.2 (22.1) OilSE02C3.2 3.2 274.9 (34) 4 264.9 (12.4) OilRE01C5 13.6 134.4 (11.5) 4 134.4 (15.5) 6.6 136.4 (19.5) OilRE02C5 14.6 132.9 (19.5) 4.4 132.9 (19.5) 10 136.4 (17.5) OilSE02C5 7.6 134.4 (15.5) 5.8 134.4 (11.5) 6.1 134.4 (17.5) OilRE02C6.1 3.4 30.4 (19.8) 4 67.4 (19.2) 4.8 8.0 (20) OilSE02C6.1 3 48.2 (49) OilRE01C6.2 2.8 95.6 (2.8) OilRE02C6.2 4.6 76.6 (11.3) 3.4 75.4 (17.3) OilRE02C7 4.4 107.9 (18.4) 3.4 70.7 (63.4) OilRE01C9 3.4 112.2 (11.2)

In order to compare the effect of our sampling strategy, 100 A we considered true QTL as those that were detected in 90 the whole population. Out of the 35 true QTL, 13 QTL 80 70 were detected at the same position as the true QTL in the 60 (cM) two sub-populations (on LGs A1, A2, A3, A6, A10, C5 50 Whole and C6). Thirteen true QTL were detected in none of the 40 qtlRec length 30 subpopulations. Four and five true QTL were detected MapPop either in qtlRec or in MapPop sub-populations. This was 20 Region 10 summarized by the two parameters (sensitivity and spe- 0 cificity) used to evaluate the significance and reliability 0510LG of the selected samples in comparison to the whole A1 A3 A6.2 A10.2 C2 C5 C6.1 sample. Sn was estimated on average at 0.48 and 0.51 (a) and Sp was estimated on average at 0.91 and 0.88 for the qtlRec and MapPop populations, respectively (Table 3). QTL support intervals were compared in seven regions that were used for selective sampling and where QTL were identified in the whole and the selected populations. In these selected regions, the recombination rate in the qtlRec sub-population was either equal or higher than in the MapPop one but on C2 and C6 LGs (Figure 1). As these regions were detected in one, two or three environments, they included 12 QTL in all. The support interval was lower in the qtlRec than in the MapPop subpopulation in six QTL (on LGs A1, A3, A6 and C5). In that case, it was equal to the support interval obtained in the whole population except for OilRE01A2 QTL (Figure 1). (b) In four cases (OilRE01A1, OilRE02A6.2, OilRE02C5 and Figure 1. (a) Representation of the recombination increase OilRE02C6.1), the same support interval was obtained in in seven selected QTL regions: The size of the regions was the MapPop, the qtlRec and the whole populations. For estimated (in cm) on the whole population and on the the two remaining regions, either qtlRec and MapPop MapPop and the qtlRec sub-populations after seletion. (b) support intervals were lower than whole population one Length of the support interval for the QTL detected in these selected regions on all the populations. (OilRE02A6.2) or qtlRec support interval was higher than MapPop and whole population ones (OilRE02C2.1). in two separate field trials, one for the dwarf and one for the tall DH lines. ANOVA within each trialshowed sig- 3.3. Phenotypic Evaluation for Stem Canker nificant (P < 0.001) phenotypic variation among lines Resistance and replicates. ANOVA performed on the control varie- The 279 DH lines of the “Darmor-bzh” x “Yudal” whole ties from the two trials showed no significant trial effect population, as well as the parental lines and control va- and the average disease indexes of each trial (4.52 and rieties, were evaluated for their resistance to L. maculans 4.74 for the dwarf DH trial and the tall DH trial, respectively) were not significantly different (P = 0.09). There- Table 3. Number of true positive (TP), false positive (FP) fore, the phenotypic data from the two trials were pooled and false negative (FN) QTL, criterion of sensitivity (Sn) and the mean of three replicates was used for the follow- 2 and specificity (Sp) in the qtlRec (Q) and MapPop (M) se- ing analyses. The heritability was very high: h = 0.93, as lected populations in comparison to the whole population estimated in the whole population. The parental lines with LOD threshold 2.5 “Darmor-bzh” and “Yudal” showed a mean G2 disease index of 1.46 +/– 0.29 and 7.72 +/– 0.45, respectively. OilRE01 OilSE01 OilSE02 The control varieties showed a level of resistance to L. Q M Q M Q M maculans, which was consistent with expected levels TP 6 6 7 7 4 5 (“Jet Neuf”: G2 = 0.99 +/– 0.20; “Darmor”: G2 = 1.34 +/– FN 7 7 5 5 6 5 0.15; “Falcon”: G2 = 2.71 +/– 0.56; “Eurol”: G2 = 4.43 +/– 0.75), illustrating the high inoculum pressure. Resistance FP 1 1 1 2 0 0 throughout the whole DH population showed a conti- Sn 0.46 0.46 0.58 0.58 0.4 0.5 nuous distribution pattern (Figure 2), confirming the Sp 0.86 0.86 0.88 0.78 1 1 quantitative and polygenic control of the resistance ob-

Copyright © 2012 SciRes. OPEN ACCESS 196 C. Jestin et al. / Open Journal of Genetics 2 (2012) 190-201 served in previous studies [23,24]. All the QTL detected were also detected in the 200Q We observed the same continuous distribution pattern qtlRec population at the same position, except on C4 and in the random and qtlRec samples (Figure 2). No signifi- A4 LGs. In this latter 200Q population, only one addi- cant difference was found between the sub-population tional QTL was detected on A2 LG. The estimated phe- means or between the sub-populations and the whole po- notypic variation explained by individual QTL varied pulation means. This confirms that the selection based on from 3.1% to 9.3% and the overall explained phenotypic the genotypic data had no effect on the phenotypic distri- variation was 65.6% and 56.5% for 150Q and 200Q bution. populations, respectively. The allele increasing stem canker resistance was derived from “Darmor-bzh” for all 3.4. Additive QTL Detection in the Whole and the QTL, except for QLmA3.2 and QlmC1 where “Yu- Sub-DH Populations for Stem Canker dal” contributed the resistance allele. Resistance The results of the CIM analyses for the whole (279 DH) and subpopulations (150R, 150Q and 200Q) are summarized in Table 4 and Figure 3. LOD thresholds of 2.8 (for the whole population) and 3.1 (for thesub-populations) were obtained after 500 permutation tests. As for oil content, a LOD threshold of 2.5 was retained for the comparisons. QTL mapping performed on the whole population revealed ten QTL over nine LGs. The estimated phenotypic variation explained by individual QTL varied from 2.5% to 15.5% and the overall explained phenotypic variation was 48.4 %. In the 150R random Figure 2. Frequency distribution for the adjusted mean G2 population, seven QTL were identified on seven LGs. disease index in the “Darmor-bzh” x “Yudal” populations. The arrows show the mean value of the parental lines. The The estimated phenotypic variation explained by indi- hatched, speckled, dark-grey and black bars correspond to vidual QTL varied from 4.5 to 15.2% and the overall ex- the population taken at random (150R), the qtlRec 150Q plained phenotypic variation was 55.4%. In the 150Q and 200Q selected populations and the whole population, qtlRec population, seven QTL were identified on six LGs. respectively.

Table 4. The stem canker QTL detected on the linkage groups of the “Darmor-bzh” x “Yudal” DH populations: LOD score, peak position and support interval (SI) in cM for each QTL.

Whole population 150R population 150Q population 200Q population LOD Position (SI) LOD Position (SI) LOD Position (SI) LOD Position (SI) QLmA1 4.23 114.7 (14.2) QLmA2.1 3.24 11.3 (22.6) 4.42 23.4 (15.6) 2.61 22.6 (27.8) QLmA2.2 3.55 103.5 (24.5) QLmA3.1 3.23 200.0 (12.0) 4.87 202.0 (8.2) QLmA3.2 2.84 230.5 (20.6) 7.35 219.3 (3.6) 7.39 219.3 (2.5) QLmA4.1 14.78 2.0 (4.0) 8.69 17.5 (19.1) 3.49 2.0 (6.4) QLmA4.2 3 31.9 (18.7) QLmA7 6.89 44.4 (12.9) 7.81 39.3 (15.7) 6.81 42.4 (7.3) 6.55 44.4 (7.6) QLmA8 4.48 49.3 (25.4) 3.02 56.0 (22.1) 3.74 62.0 (18.8) QLmA9.1 4.22 68.9 (19.4) QLmA9.2 3.66 134.8 (15.1) QLmC1 4.4 7.5 (14.1) QLmC2 2.9 11.1 (22.2) QLmC4.1 5.55 60.0 (14.1) 3.54 66.2 (14.1) QLmC4.2 3.37 113.5 (13.2) QLmC7 3.28 94.1 (29.1) QLmC8 2.86 119.5 (66.2) 6.14 84.8 (30.1) 4.71 82.8 (35.2)

A1 A2 A3 A4 0.0 CB10081 0.0 sR12095a 0.0 CB10388 0.0 CB10448 Qlm 6.4 sN11641a 11.0 sR94102a 15.4 F15.680 4.4 CB10347 Qlm Qlm Qlm 11.1 CZ0r227217 H09.570 Qlm 18.6 sS2368b sN0786a 19.8 7.5 ‐‐ ‐ Qlm R12137Ib T 25.0 150R 14.5 sN11707b 31.6 CZ5b703024b ‐ 11.4 sN0464b 200Q ‐ ‐ 150R 15.8 sNRE47a 47.5 CZ5r224888 T 31.0 CHS.4 sNRD71 sR12610a ‐ 17.6 CB10097 Bras037 sN3761b 43.0 I11.2000 12.1 sR0357z 150Q 29.0 C02.860 sR9481a 58.4 sR2103a 57.7 Bras029 13.5 W08.1620 32.5 Ol10D03bRa2G09 60.3 sR6293 62.4 Ra2E11 19.5 CZ5b696093 40.2 sR1377b 67.1 O15.1000 66.1 F15.1110 29.8 sN2025 sR9546a 49.1 CZ0b702196 68.4 At5_003 72.0 sN1915bsR0253a 40.2 FAD3.A 74.9 CZ0b699112 81.3 sN11531 77.9 Na14E02 42.0 P05.1420 79.9 Ol12F11b 85.6 Na12H09c Qlm 88.2 F01.1340 52.2 CB10196 sR12386b 88.6 At2_020b 90.4 87.6 CB10093 ‐ 68.2 sR12307I 95.7 Bras041a 94.1 At4_001 T 90.3 sN8841b 77.5 At3_015bY10.1300 99.7 Bras084 98.1 E02.1200 91.4 Na12E02 97.4 Bras024 100.1 RPS2.2 107.6 ScJ14 93.9 S2Nx062 100.4 Bras078 124.0 CB10540 100.6 sS1716b sR12015a 109.2 Bras013a Qlm 141.0 sNRE30c MKD152.2 108.8 111.6 Na12C06 143.4 Na12E03 109.4 W05.630

‐ 112.9 sN12574b 112.4 Bras026 T 147.6 CZ0b696769 A7 113.4 IGF2071e 129.3 CZ4a00790 114.8 sN2305a 137.5 A16.3250 0.0 V09.1300 122.2 G12.710 140.7 sS1867Ac 8.0 sR4047 124.8 sN3523Rb sN0983Fa 151.4 CB10144 8.7 H06.940sR0282R Bras043 152.4 sR6803b 11.8 Ra2G08a

132.3 Qlm

154.5 sN11722a Qlm 135.0 sN2321a A9 18.3 sS2367 Q Qlm l

159.6 CZ3a12080 22.9 Bras023 m 137.9 ELI3 ‐ 150R ‐ Na10F06b ‐ O15.1100 162.3 CB10450 150Q Bras067 0.0 39.4 200Q 151.1 ‐ 156.7 Bras074 7.4 A07.500 163.4 G14.1045 42.5 A08.2340 T 18.0 sR6806b 167.3 CZ3a12880 49.7 P07.CD1 Qlm

159.8 ACPlicor Qlm 165.2 CZ3b697214 28.6 W03.550 180.4 W08.2360 52.3 W11.1690

Bras010b ‐ sR9447 183.3 Qlm Ol09A06b

35.2 Qlm 66.7 ‐ sR12355b 150Q 167.8 200Q 168.7 sS1867Ab 42.7 N18.940 186.9 CB10415 70.9 CB10439 Qlm ‐ sN1988b 189.0 Bras056a ‐ sNRA59 43.6 200Q 75.1 198.0 Na14F11a 150Q 198.8 CZ5a24103 47.9 sR50163a 195.4 Na10E02c 84.1 A14.880 ‐ 57.4 Bras020 198.7 MKD156.3 T 90.7 C02.1375 68.7 CZ0b697380 200.0 MYJ243.1 102.5 Ol12E03 79.8 CB10311 202.6 CZ5b711213 110.3 sR12387b CZ1b705119 83.2 MR216c 206.9 sR0404a 111.4 T04.680 A8 86.9 W11.1890 216.6 E85M32a 114.6 CZ1r225888 0.0 CB10026b 97.4 sN2713c 219.2 V09.2100 143.2 O15.1360 Qlm E84M40a 14.1 CZ1a44936 105.9 CB10402 221.2 112.4 CB10029 222.8 IGF9014c

17.3 sR9480 ‐ 22.1 Na12G05a Na10E02a 112.5 Na10G06b T 224.4 P78M71a sR6068 sR9251Ja sN1839b 228.5 Na12C07 24.1 113.9 G03.960 25.8 CB10364 sR12296a ScL12 234.9 sORA84b sNRC18a 236.8 sN5108a C7 27.3 sR7178b 114.5 sORC38b Qlm 237.7 Bras002b 0.0 Bras047 Qlm 30.7 CZ4a21020 Qlm 32.9 L08.790 Qlm 115.8 Ol12F02b 241.3 P78M41b 13.7 Bras066 ‐

150R CZ0b714264 ‐ ‐ Ol10D02bRa2A11 253.8 21.9 sN0706 150Q 150R 34.3 sS1702 ‐ 116.4 35.7 Bras039 200Q Na10A08c ScNP004 22.8 Na12F03 35.8 sR7167a 117.4 ScNP046 23.5 Na12B07c 41.5 CB10013b 118.3 Na12E06B 26.7 sNRH63 56.2 A08.960 121.6 MR216b 29.2 Bras019 69.6 IGF3222b 134.2 W15.1470 55.3 CB10299 74.0 CZ0b700305 144.2 CB10106 62.8 PC1 77.3 sR3688 148.4 ScH09 C4 72.8 Bn92A1 A15.850 162.0 Na14B03 91.0 CB10268 81.4 0.0 Bras072b Qlm 100.9 Na12B05 168.4 sN11670a 94.2 Bras014 173.4 I11.1130 16.2 sS2277 96.8 sNRB93b CB10475 ‐ 105.8 28.4 S2No063 T 115.4 G11.510 174.6 A07.790 108.0 U13.2160 178.9 sR5795a 34.3 Bras063b 109.1 N12508QaN12508Xa 121.1 Na12A08b 44.1 sORF73a 122.2 Na14G02b 180.4 sR6755a 111.1 sS1949b 192.8 Y04.1510 46.5 sN12353a 120.7 sR0404b 138.6 CZ2a26750 57.0 MR119 158.8 CB10193 198.2 CZ1b693390b 123.8 MKD156.4 57.7 CB10107 127.5 CB10431 63.8 At2_020a 141.9 Na12A10 70.8 Bras123 142.9 R12302Ib

73.5 Na12B07b Qlm 148.6 b4NB12 C1 C2 75.6 Na12E6Aa Qlm 160.3 sR6715b CB10159a ‐ 0.0 Y15.2000 CZ0b686968 76.6 200Q 0.0 ‐ 7.5 sN11641b 11.1 A01.870 78.2 Bras061 T 14.1 CB10429 Qlm 12.4 sR12095b 79.2 Na14E11

82.2 W15.1560 Qlm

sN11707a ‐ 22.8 14.4 Fad8 Qlm 28.7 sNRE47b 150R 31.8 CZ5b694914 86.0 CB10501 C8 ‐

F01.540 150Q 35.1 CB10587 ‐ 90.6 34.7 L08.1190 150R 0.0 Bras031 49.2 CB10369 51.6 CZ5b703024a 105.3 Ol10D03a 3.2 CB10602 65.0 Ol12F11a 54.6 ScN01 109.1 P05.1375 9.9 sR7178a 69.5 O13.1490 58.3 W11.610 115.5 ScN13 25.2 RPS5a 71.5 sN9425 62.3 sR12779b 121.2 sN0539b 30.4 H12.1010 75.7 S4No025 63.9 Na14H11a 124.5 ScNP048 60.7 sN2189a 85.6 L02.940 64.5 MR216a 125.3 ScNP009 95.4 CB10449

133.8 A09.1000 Qlm 86.7 CB10206 70.5 Ol13G05a 104.8 CZ3b697541b Qlm sS1725a Y15.1880 80.5 Na12H09b 139.1 Na12D09 111.6 sR12386a

88.0 ‐ P09.1160At2_019 ‐

143.2 200Q sN1838a 85.1 CB10026a 113.4 CB10179 150Q 89.3 sNRE30d 90.5 Ol09A06a 146.5 At3_015a 119.8 H06.CD1 Qlm 92.4 Na12C08 CZ1b704751 97.6 sN3514Fa 170.8 CB10493a Bras060e

135.8 ‐ 93.3 Na12H01 118.5 CZ0b707602 173.0 S2No065 137.2 sR5795b T 98.0 CZ0b696150 124.7 ScNP007 175.0 S2No066 138.6 sR6755b 105.1 Bras044 134.4 ScP05 P05.CD1 147.4 C02.1460 124.9 CZ3a12840b 138.2 Bras075b 147.8 CB10092 133.7 sN2321b 140.8 Na12B07a 152.0 A09.700 134.8 Ol10A11Na10H03 150.1 CZ0b716882 154.9 sN0866a 137.4 CB10572 158.7 D20.760 156.8 sORB29A 154.3 sR12355a 167.1 Ol10H02 167.0 CB10139 191.4 Na14F11b 167.2 Bras102b 173.7 I20.760 184.7 sNRE30a 191.8 Ol09A03 199.3 Na12C03 219.7 CB10543a Figure 3. Additive QTL controlling resistance to L. maculans in the “Darmor-bzh” x “Yudal” whole and subset populations. Only linkage groups carrying resistance QTL are shown. Additive QTL detected in the whole population are indicated by black-colored bars (QLm-T), in the qtlRec selected population by speckled bars (QLm-150Q) and by grid bars (QLm-200Q) and in the random population by hatched bars (QLm-150R). The genetic map was calculated on the population taken at random. The bar length corresponds to the 1-LOD support interval. When more than one QTL affecting a trait was identified for one population and on the same linkage group, QTL are distinguished by different letters. The cumulated distance between two markers is expressed in centimorgans (Haldane).

In order to compare the effect of our sampling strat- 70 egy, we considered true QTL as those that were detected 60 in the whole population. Ten true QTL were detected, of which five were located in the regions used for sampling 50 the qtlRec population. The power of QTL detection was 40 (cM) higher in the whole population than in all selected sub- Whole populations. A single QTL was detected at the same po- 30 150R interval

150Q sition as the true QTL in all the sub-populations on A7 20 200Q LG. Four true QTL (QLmA1, QLmA2.2, QLmA9.1 and Support QLmC7) were detected in none of the sub-populations. 10

Only the 200Q sample allowed the increase of true posi- 0 tive and the decrease of false negative and false positive 02468 QTL as shown by the two parameters used to evaluate QTL QlmA2.1 the significance and reliability of our selected samples. QlmA3.2 QlmA4.1 QlmA7 QlmA8 QlmC4.1 QlmC8 Sn was estimated at 0.30, 0.30 and 0.60 and Sp was es- Figure 4. Length of QTL support interval in the whole and timated at 0.43, 0.43 and 0.75 for the random (150R) and the selected populations. Only the regions where QTL were qtlRec (150Q and 200Q) populations, respectively (Ta- detected in the whole or the 150R and in the 150Q or the 200Q populations are presented in order to perform com- ble 5). parison. QTL support intervals (SI) were compared in the regions where QTL were detected in the whole or 150R cost of phenotyping. We evaluated the qtlRec sampling populations and in the 150Q or 200Q populations (Fig- strategy on oil content in B. napus for which QTL were ure 4). The support intervals were lower in the qtlRec identified in a previous study on a large genotyped and populations than in the random or the whole populations phenotyped population. Our study showed that the for QlmA3.2, QlmA7, QlmA8 and QlmC8. They were qtlRec sampling strategy performed as well as the sam- similar to SI estimated in the whole population for pling strategy based on MapPop software for the power QlmA4.1 and QlmC4.1. For QlmA2.1, SI was higher in of QTL detection. It also showed that qtlRec sampling the qtlRec population than in the random or the whole strategy decreased the support interval of the QTL more populations. than the MapPop strategy. We then applied the qtlRec sampling strategy to the quantitative resistance to L. 4. DISCUSSION maculans in B. napus, which confirmed its interest with this second complex trait, in comparison to the random In this study, we present the assessment of a new sam- sample, and allowed the validation of the QTL in six pling strategy, referred to as qtlRec. It is based on the genomic regions through the phenotyping of a selected selection of individuals which maximize the recombine- sample. tion at targeted QTL regions in order to improve the accuracy of QTL location in comparison to a MapPop sample 4.1. Effect of Sampling Strategy on the Detection of the same size. This selective phenotyping strategy and Mapping of Oil Content QTL would be useful when QTL were identified in a preli- minary study from a large genotyped population and a Considering as true QTL those detected in the whole further phenotypic evaluation in multiyear/environment population, we identified 35 true QTL overall the three trials is required on a limited sample size due to the high variables. It was not possible to identify all true QTL in any of the sub-populations. Fifty percent of the true QTL Table 5. Number of true positive (TP), false positive (FP) were detected whatever the sampling strategy was. The and false negative (FN) QTL, criterion of sensitivity (Sn) sensitivity and specificity parameters showed that qtlRec and specificity (Sp) in the random (150R) and the two selective sampling led to a similar rate of true positive qtlRec (150Q and 200Q) selected populations in comparison QTL detection, compared to the MapPop sampling stra- to the whole population with LOD threshold 2.5. tegy. Overall, we observed that the detection of QTL 150R 150Q 200Q with low individual effects was particularly affected by TP 3 3 6 both sampling strategies. In both cases, very few false positive QTL were observed. Two of the four identified FN 7 7 4 false positive for OilRE02 variable were located at the FP 4 4 2 position of a true QTL for the other variables, indicating Sn 0.3 0.3 0.6 that true positive could also be missing on the whole population. The decrease in the number of true QTL Sp 0.43 0.43 0.75 detected was expected when the number of individuals

Copyright © 2012 SciRes. OPEN ACCESS C. Jestin et al. / Open Journal of Genetics 2 (2012) 190-201 199 was reduced as previously reported by several authors reasonable to predict that our methodology could have e.g. [16]. This was also observed by [10-13] who tested been more effective if the full population (442 DH) had a SP strategy with the sampling method implemented in been previously phenotyped in a single environment and MapPop software. The benefits obtained with a selected the qtlRec selection applied on the QTL regions iden- sample, especially when specific genetic regions are tified from this larger population. The increase in re- targeted, were reported by Jin et al. [14]. Even if the combination would have been more highly focused due authors used a sampling procedure based on genetic to an accurate position of the QTL obtained from this dissimilarity, they showed that a selected sample was large population. However, since we had two pheno- better than a random sample, particularly, when specific typing years available, we were able to choose the re- genetic regions were targeted. Thus the power and ac- gions to use for selection of recombinants and were able curacy of QTL analysis improves when the selection is to assess the sampling methodology from 2007 expe- more finely targeted using prior knowledge. Sen et al. riment, considering the 279 DH population as the re- [34] confirmed this but the efficiency of SP decreased as ference one for this comparison. Ten true QTL were de- the number of unlinked loci considered increased. Ac- tected. It was not possible to identify all true QTL in any cording to the authors, when only a small sample can be of the sub-populations. Nevertheless, an increase of sen- phenotyped, the efficiency of the SP is still higher com- sitivity and specificity was obtained with the 200Q pared to random sampling, even when more than ten loci qtlRec sample compare to the random and the 150Q are used for the selection. qtlRec samples. More true positive and less false Selective sampling based on recombination rates is positive were observed in this 200Q qtlRec sample. This intended to improve the precision of QTL location. The result is due to the combined effect of the sampling qtlRec sampling strategy reduced the support interval for strategy and of the increase of the size of the population. six QTL compared to only two for the MapPop sampling Some of the false positives observed in 150Q qtlRec strategy, and the two strategies were equivalent for four sample resulted from a lack of accuracy in the QTL QTL. Other studies [9,15] referred to the positive po- position, as shown in [37], where small population sizes tential of the SP method to increase the accuracy of QTL generated a shift or an increase in the support interval of mapping. Indeed, the support interval will be smaller the QTL position. Thus, the QTL considered as false in when there are a greater number of recombinant events the 150Q qtlRec population on the A4 and C4 LGs could in the QTL support interval. The six QTL with reduced correspond to the true QTL identified in the whole support interval in the qtlRec sample were located in population, but were poorly located on these LGs. regions used for selective sampling where the recom- Selective sampling based on recombination rates is bination rate was actually higher in the qtlRec than in the intended to improve the precision of QTL location. The MapPop sample. support intervals obtained with the 200Q qtlRec sample In our study, the global R² was slightly higher in the were either similar or smaller than those obtained with MapPop sample than that of the whole and the qtlRec the whole population. For the only QTL that had the 2 populations. Global R depends on the genetic map ac- exact same location on all the populations (QlmA7), the curacy, inter-marker distances [35] as well as the num- support interval was lower in the qtlRec samples com- ber and accuracy of detected QTL. An overestimation of pared to the random sample, as was desired. individual R2 values for each QTL was observed when the population size was reduced [11,13], owing to the 4.3. Consistency of Stem Canker Resistance Beavis effect [36]. The estimations of individual and QTL across the Years global R² might be less biased in the qtlRec sample than in the MapPop one due to the improvement of accuracy Our analysis confirmed that stem canker resistance is of some QTL. controlled polygenically, mainly by small effect QTL and is a trait with a very high heritability (0.93). These 4.2. Effect of Sampling Strategy on the Detection results are consistent with those reported previously. Pi- and Mapping of Stem Canker QTL let et al. [34] identified eight and five QTL and estimated the heritability at 0.89 and 0.88 in 1995 and We showed with the oil content trait that the use of a 1996, respectively, in a 150 DH population from the priori knowledge in targeted regions of interest to select same DY cross which corresponds to our 2007 random the individuals maximizing recombination at QTL posi- population. A second mapping study was then performed tions lead to a similar QTL power detection compare to on the “Darmor” x “Samourai” (DS) cross [23] and the MapPop sampling strategy but led to an increase in identified five and four QTL on a 134 DH population in the accuracy of small effect QTL location. Then, we 1998 and 1999, respectively, and four QTL on a 185 applied qtlRec sampling to another complex quantitative F2:3 population in 1998. Four regions were consistent trait, the stem canker resistance to L. maculans. It seems across the two crosses (DY and DS) and the four years

(1995, 1996, 1998 and 1999) on the A2, C2, C4 and C8 Métropolitains) and PROMOSOL. We thank the team of the INRA LGs [23]. Experimental Unit (Le Rheu) for performing the disease evaluation From our 2007 experiment, QTL were detected on all trials. Genotyping was performed on Biogenouest® platform. LGs where QTL were previously identified with the

1995 and 1996 data, except for two LGs. A strong effect QTL was detected on A4 in 2007, which was not REFERENCES identified before. On the A6 LG, a QTL with a strong [1] Kearsey, M.J. and Farquhar, A.G.L. (1998) QTL analysis effect was identified at the Bzh gene position in 1996 but in plants; where are we now? Heredity, 80, 137-142. not in 1995 [24] and was not detected in our 2007 study. doi:10.1046/j.1365-2540.1998.00500.x The dwarf trait seemed to affect the expression or the [2] Salvi, S. and Tuberosa, R. (2005) To clone or not to evaluation of resistance in 1996 but not under the 1995 clone plant QTLs: Present and future challenges. Plant or 2007 field conditions. This result may be related to a Science, 10, 297-304. lower level of disease in 1996 than in 1995 and 2007 as [3] Lander, E.S. and Botstein, D. (1989) Mapping mendelian hypothesized by [24]. These inter-year differences high- factors underlying quantitative traits using RFLP linkage light the interest of testing QTL x environment interac- maps. Genetics, 121, 185-199. tions in order to study the global genetic architecture of [4] Lebowitz, R.J., Soller, M. and Beckmann, J.S. (1987) quantitative stem canker resistance. Trait-based analyses for the detection of linkage between marker loci and quantitative trait loci in crosses between On the LGs where QTL were detected in 2007 and in inbred lines. Theoretical and Applied Genetics, 73, 556- 1995 or 1996, either the position of 1995/1996 QTL was 562. doi:10.1007/BF00289194 similar to the QTL considered as true QTL in this study [5] Navabi, A., Mather, D.E., Bernier, J., Spaner, D.M. and (on A2, A7, A8, and C8 LGs) or was similar to the Atlin, G.N. (2009) QTL detection with bidirectional and position of QTL detected in the sub-populations (on A9, unidirectional selective genotyping: Marker-based and C2 and C4 LGs). Thus, the QTL identified in 1995 and/ trait-based analyses. Theoretical and Applied Genetics, or 1996 on A9 and C4 LGs [24,27] could have been 118, 347-358. doi:10.1007/s00122-008-0904-2 poorly located as the ones detected in 2007 in the 150R [6] Kimura, T., Kobayashi, T., Munkhbat, B., Oyungerel, G., or 150Q sub-populations, due to the low population size. Bilegtsaikhan, G., Anar D., Jambaldorj, J., Munkhsaik- han, S., Munkhtuvshin, N., Hayashi, H., Oka, A., Inoue, I. All these QTL were identified in at least two years of and Inoko, H. (2008) Genome-wide association analysis experiments. Four were confirmed in the DS genetic with selective genotyping identifies candidate loci for background [23]. This number is probably underesti- adult height at 8q21.13 and 15q22.33-q23 in Mongolians. mated due to the small size and incomplete map of the Hu- man Genetics, 123, 655-660. DS population. The stability of these QTL across the doi:10.1007/s00439-008-0512-x years is an important factor to take into consideration in [7] Sun, Y., Wang J., Crouch, J.H. and Xu, Y. (2010) Efficiency breeding programs. The validation of the QTL across of selective phenotyping for genetic analysis of complex different environments and multiple genetic back- traits and potential applications in crop improvement. Molecular Breeding, 26, 493-511. grounds also strengthens their interest in Marker-Assist- doi:10.1007/s11032-010-9390-8 ed Selection (MAS). The detection of consistent QTL [8] Brown, D. and Vision T. (2000) Software for selective between our study and previous studies is valuable mapping and bin mapping. information for breeding programs for stem canker resis- http://www.bio.unc.edu/faculty/vision/lab/mappop tance in B. napus. [9] Xu, Z.L., Zou, F. and Vision, T.J. (2005) Improving quanti- In this study, we demonstrated that our sampling stra- tative trait loci mapping resolution in experimental crosses tegy, based on the choice of individuals which maximi- by the use of genotypically selected samples. Genetics, 170, zed recombination only at QTL markers, gives similar 401-408. doi:10.1534/genetics.104.033746 power and can lead to greater accuracy in QTL detection, [10] Simon, M., Loudet, O., Durand, S., Berard, A., Brunel, compared with sampling individuals that maximizes D., Sennesal, F.S., Durand-Tardif, M., Pelletier, G. and recombination over the whole genome. This strategy, Camilleri, C. (2008) Quantitative trait loci mapping in which is especially effective for a trait controlled by five new large recombinant inbred line populations of Arabi- dopsis thaliana genotyped with consensus single-nucleotide multiple small effect QTL, could be used for QTL vali- polymerphism markers. Genetics, 178, 2253-2264. dation in multiple years and/or locations of traits which doi:10.1534/genetics.107.083899 require costly and time-consuming phenotyping. [11] Vales, M.I., Schon, C.C., Capettini, F., Chen, X.M., Corey, A.E., Mather, D.E, Mundt, C.C., Richardson, K.L., 5. ACKNOWLEDGEMENTS Sandoval-Islas, J.S, Utz, H.F. and Hayes, P.M. (2005) Effect of population size on the estimation of QTL: A This work was supported by the French Institut National de la Recher- test using resistance to barley stripe rust. Theoretical and che Agronomique—Department of Génétique et Amélioration des Applied Genetics, 111, 1260-1270. Plantes, CETIOM (Centre Technique Interprofessionnel des Oléagineux doi:10.1007/s00122-005-0043-y

[12] Birolleau-Touchard, C., Hanocq, E., Bouchez, A., Bau- [24] Pilet, M.L., Delourme, R., Foisset, N. and Renard. M. land, C., Dourlen, I., Seret, J.P., Rabier, D., Hervet, S., (1998) Identification of loci contributing to quantitative Allienne, J.F., Lucas, P., Jaminon, O., Etienne, R., Baud- field resistance to blackleg disease, causal agent Leptos- huin, G. and Giauvret, C. (2007) The use of MapPop1.0 phaeria maculans (Desm.) Ces. et de Not., in Winter for choosing a QTL mapping sample from an advanced rapeseed (Brassica napus L.). Theoretical and Applied backcross population. Theoretical and Applied Genetics, Genetics, 96, 23-30. doi:10.1007/s001220050704 114, 1019-1028. doi:10.1007/s00122-006-0495-8 [25] Foisset, N., Delourme, R., Barret, P., Hubert, N., Landry, [13] Barchi, L., Lefebvre, V., Sage-Palloix, A.M., Lanteri, S. and B.S. and Renard, M. (1996) Molecular-mapping analysis Palloix, A. (2009) QTL analysis of plant development in Brassica napus using isozyme, RAPD and RFLP mar- and fruit traits in pepper and performance of selective kers on a doubled haploid progeny. Theoretical and Ap- phenotyping. Theoretical and Applied Genetics, 118, plied Genetics, 93, 1017-1025. 1157-1171. doi:10.1007/s00122-009-0970-0 doi:10.1007/BF00230119 [14] Jin, C., Lan, H., Attie, A.D., Churchill, A.G., Bulutuglo, [26] Lombard, V. and Delourme, R. (2001) A consensus link- D. and Yandell, B.S. (2004) Selective phenotyping for age map for rapeseed (Brassica napus L.): Construction increased efficiency in genetic mapping studies. Genetics, and integration of three individual maps from DH popu- 168, 2285-2293. doi:10.1534/genetics.104.027524 lations. Theoretical and Applied Genetics, 103, 491-507. doi:10.1007/s001220100560 [15] Jannink, J.L. (2005) Selective phenotyping to accurately map quantitative trait loci. Crop Science, 45, 901-908. [27] Delourme, R., Piel, N., Horvais, R., Pouilly, N., Domin, doi:10.2135/cropsci2004.0278 C., Vallée, P., Falentin, C., Manzanares-Dauleux, M.J. and Renard, M. (2008) Molecular and phenotypic charac- [16] Charcosset, A. and Gallais, A. (1996) Estimation of the terization of near isogenic lines at QTL for quantitative contribution of quantitative trait loci (QTL) to the vari- resistance to Leptosphaeria maculans in oilseed rape (Bras- ance of a quantitative trait by means of genetic markers. sica napus L.). Theoretical and Applied Genetics, 117, Theoretical and Applied Genetics, 93, 1193-1201. 1055-1067. doi:10.1007/s00122-008-0844-x doi:10.1007/BF00223450 [28] Wang J., Lydiate, D.J., Parkin, I.A.P., Falentin, C., De- [17] Orgogozo, V., Broman, K.W. and Stern, D.L. (2006) lourme, R., Carionand, P.W.C. and King, G.J. (2011) In- High- resolution quantitative trait locus mapping reveals tegration of linkage maps for the amphidiploid Brassica sign epistasis controlling ovariole number between two napus and comparative mapping with Arabidopsis and dro- sophila species. Genetics, 173, 197-205. Brassica rapa. BMC Genomics, 12, 101. doi:10.1534/genetics.105.054098 doi:10.1186/1471-2164-12-101 [18] Delourme, R., Falentin, C., Huteau, V., Clouet, V., Hor- [29] Lincoln, S., Daly, M. and Lander, E. (1992) Constructing vais, R., Gandon, B., Specel, S., Hanneton, L., Dheu, J.E, genetic linkage maps with Mapmaker/Exp 3.0: A tutorial Deschamps, M., Margale, E., Vincourt, P. and Renard, M. and reference manual. 3rd Edition, Whitehead Institute (2006) Genetic control of oil content in oilseed rape Technical Report. (Brassica napus L.). Theoretical and Applied Genetics, 113, 1331-1345. doi:10.1007/s00122-006-0386-z [30] Haldane, J.B.S. (1919) The combination of linkage values, and the calculation of distances between the loci of [19] Fitt, B.D.L., Brun, H., Barbetti, M.J. and Rimmer, S.R. linked factors. Journal of Genetics, 8, 299-309. (2006) World-wide importance of phoma stem canker (Lep- tosphaeria maculans and L. biglobosa) on oilseed rape [31] SAS II. (1989) SAS/STAT users guide, version 6.0. 4th (Brassica napus). European Journal of Plant Pathology, Edition, SAS institute Inc., Cary. 114, 3-15. doi:10.1007/s10658-005-2233-5 [32] Wang, S., Basten, C.J. and Zeng, Z.B. (2007) Windows [20] West, J.S., Kharbanda, P.D., Barbetti, M.J. and Fitt, B.D.L. QTL cartographer 2.5. North Carolina State University, (2001) Epidemiology and management of Leptosphaeria Raleigh. http://statgen.ncsu.edu/qtlcart/WQTLCart.htm maculans (phoma stem canker) on oilseed rape in Austra- [33] Voorrips, R.E. (2002) MapChart: Software for the gra- lia, Canada and Europe. Plant Pathology, 50, 10-27. phical presentation of linkage maps and QTLs. Journal [21] Delourme, R., Chèvre, A.M., Brun, H., Rouxel, T., Bales- of Heredity, 93, 77-78. doi:10.1093/jhered/93.1.77 dent, M.H., Dias, J.S., Salisbury, P., Renard, M. and Rim- [34] Sen, S., Johannes, F. and Broman, K.W. (2009) Selective mer, S.R. (2006) Major gene and polygenic resistance to genotyping and phenotyping strategies in a complex trait Leptosphaeria maculans in oilseed rape (Brassica napus). context. Genetics, 181, 1613-1626. European Journal of Plant Pathology, 114, 41-52. doi:10.1534/genetics.108.094607 doi:10.1007/s10658-005-2108-9 [35] Jansen, R.C., Vanooijen, J.W., Stam, P., Lister, C. and Dean, [22] Rimmer, S.R. (2006) Resistance genes to Leptosphaeria C. (1995) Genotype-by-environment interaction in genetic- maculans in Brassica napus. Canadian Journal of Plant mapping of multiple quantitative trait loci. Theoretical and Pathology, 28, S288-S297. Applied Genetics, 91, 33-37. doi:10.1007/BF00220855 doi:10.1080/07060660609507386 [36] Beavis, W.B. (1998) QTL analyses: Power, precision, and [23] Pilet, M., Duplan, G., Archipiano, M., Barret, P., Baron, accuracy. In: Molecular dissection of complex traits, edited C., Horvais, R., Tanguy, X., Lucas, M.O., Renard, M. by Patterson AH. CRC Press, Boca Raton, 145-162. and Delourme. R. (2001) Stability of QTL for field re- [37] Hyne, V., Kearsey, M.J., Pike, D.J. and Snape. J.W. (1995) sistance to blackleg across two genetic backgrounds in QTL analysis: Unreliability and bias in estimation proce- oilseed rape. Crop Science, 41, 197-205. dures. Molecular Breeding, 1, 273-282. doi:10.2135/cropsci2001.411197x doi:10.1007/BF02277427