The genetic control of germination and hypocotyl heterotrophic growth was studied at sub-optimal temperatures using a population of 125 recombinant inbred lines (RILs), named LR5, which is derived from a cross between Jemalong A17 and a cold-sensitive line, F83005.5. At a supra-optimal temperature (20°C), we used a population of 178 recombinant inbred lines, named LR4, which is derived from a cross between Jemalong 6 and DZA315.16, the latter genotype germinating more slowly at 20°C and above. Both populations were F8 derived lines. In this paper Jemalong-6 and Jemalong-A17 are both denoted Jemalong as no genotypic or phenotypic differences were identified between them (Thoquet et al. 2002). Numerous genetic and genomic tools have been developed using Jemalong and the three parental lines belong to the M. tr. nested core collections (Ronfort et al. 2006).

The seeds of LR5 were produced in greenhouses by INP ENSAT-CNRS (Toulouse, France) and seeds of LR4 by the M.tr. Biological Resource Centre (Montpellier, France, http://www.montpellier.inra.fr/BRC-MTR/). For all the experiments, seeds were scarified with sand paper to break seed coat dormancy. Seeds were considered as germinated when the radicle protruded from the seed coat (Bewley and Black 1994). Hard seeds that escaped rubbing and remained unimbibed were excluded from the results. Seed mass (SM) was measured in order to test the possible correlations between this seed characteristic and genotypic differences in germination or seedling growth.

For the germination experiments, three replicates of 30 seeds were placed in plastic boxes (5.5 × 12 × 18 cm) on filter paper imbibed with 14 ml of deionised water and incubated in a climatic chamber at 5°C without light at the National Seed Testing Station (SNES, Angers France). Germination was recorded once a day for 3 weeks. For the seedling growth study, only final hypocotyl length (HL) and width (HW) were recorded as previous results had shown that there were no differences in radicle growth between the parental accessions (Brunel et al. 2009). Seedlings were grown at 10°C under dark conditions to mimic heterotrophic growth, and sampled 24 and 31 days after sowing to ensure that the seedling lengths had reached their plateau value. Seeds were sown in pots (height 10 cm, diameter 7.5 cm) filled with 500 g sand and watered with a nutrient solution (Saglio and Pradet 1980) to ensure that seedling growth conditions were not limiting. Pot water content was maintained constant at 0.20 g g⁻¹. The experimental design was a randomized complete-block design with three blocks. All lines were present in one block, with one pot sown with seven seeds for each line for each sampling date (24 and 31 days). Measurements were thus made on three replicates of 14 seedlings for each genotype.

Two experiments were carried out at 20°C. For the first experiment performed at SNES in Angers (denoted SNES), three replicates of 25 seeds were used for each line. For image acquisition and analysis we used an automated set-up consisting of a Jacobsen table with three cameras positioned above the table (Ducournau et al. 2004, 2005). Seeds of 16 lines were positioned on filter paper under each camera, i.e. 48 lines for each sowing date. The time to reach full germination in a given line varied from 3 to 6 days. The experiment was designed as an incomplete block trial. Each block represented a given sowing date. Each of the three replicates of each line was sown at a different date. Thus, measurements of the three replicates of the 178 lines took 9 weeks. Each replicate was sown under a different camera, i.e. at a different place along the table, to ensure that differences in germination between lines could not be due to putative differences in thermal or water availability along the Jacobsen table. Temperature was recorded hourly with sensors placed on the table under each camera. Pictures were taken automatically every 2 h. An automatic process using Image J software (Rasband 1997) measured individual information on each seed (Demilly et al. 2007). We recorded the initial seed surface area (ISSA, mm²). The imbibition rate (IR, mm² h⁻¹) was estimated from the increase in seed surface area from 0 to 8 h after sowing, which had previously been shown to be correlated with seed water uptake (Demilly et al. 2007). Germination of each seed was determined from changes in the coordinates of its inertia centre. The germination times (Ti, number of hours to reach i% of germination, i < 90) were calculated from fittings to the germination time course on each replicate of 25 seeds. Final germination percentages (FGP) were recorded manually at the end of the experiment. The early embryo axis elongation rate (ELR, mm² h⁻¹) was determined by the difference in the embryo axis lengths measured at 4 and 10 h after germination, i.e. lengths of less than 5 mm.

Germination was also recorded during a second experiment (denoted LEG) carried out at INRA Dijon (Research Unit for Genetic and Ecophysiology of Grain Legumes). Here we took advantage of an experiment dedicated to phenotyping LR4 during vegetative growth. We measured seed mass (SM), Ti and FGP before the transfer of germinated seeds into pots for analysis of further growth. The seedlot used in the LEG experiment was different from the one used in the SNES experiment. Imbibed seeds, 50–90 per line, were vernalized for 3 days at 5°C before being transferred at 20°C. FGP was measured 10 days after sowing.

For each replicate, germination records were fitted to a Gompertz function (Richards 1959):\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ {\text{G}}\left( t \right) = a \times { \exp }\left[ {\left( { - b/c} \right) \times { \exp }\left( { - ct} \right)} \right] $$\end{document}where G(t) is the cumulative germination rate at time t after sowing, a the maximum cumulative germination rate, and b and c are shape parameters. The fittings were used to calculate the time to reach 20, 50, 70 and 90% germination (T20, T50, T70, and T90).

All statistical analyses were performed with SAS 8.1 (SAS 2000). Data were subjected to analysis of variance using the GLM procedure. Regarding seedling elongation at low temperature (LR5), the absence of differences in HL measurements at 24 and 31 days was first checked and then all the data were combined. As a significant block effect was observed, we generated adjusted means by regression on the block effect by using the least squares method (LSmeans) in subsequent analyses. For the SNES experiment carried out at 20°C on LR4, heterogeneities could originate from the sowing position along the table (table position effect) and from variations over the 9 weeks (block effect). As a significant table position effect was observed, adjusted means were generated by regression on the table position effect by using the least squares method (LSmeans) for subsequent analyses. Simple Pearson correlation coefficients (r) were calculated between all the variables measured either on LR5 or on LR4.

The genetic maps of LR5 and LR4 used for QTL mapping were based on 89 and 110 microsatellite markers, respectively (T. Huguet et al. unpublished). The markers are well distributed over the chromosomes and the framework maps span 665 and 608 cM, respectively, with an average distance between markers of 7.5 and 5.5 cM, respectively. These maps globally cover the entire genome of the species, the integrated genetic map of M. tr. covering 596 cM (http://www.medicago.org/). Before QTL analysis, the normality of distributions of all traits was tested by computing skewness and kurtosis indices. A trait was declared normally distributed when skewness and kurtosis values ranged between −2 and +2. Means or adjusted means (LSmeans) were used depending on the trait. QTL analysis was carried out with the PLABQTL software package (Utz and Melchinger 1996) using composite interval mapping (CIM). For each variable, a critical LOD score was determined using 1,000 permutations (Churchill and Doerge 1994) with a genome-wise error rate of α < 0.05. For CIM, PLABQTL uses the regression approach proposed by Haley and Knott (1992), using selected markers as cofactors. These cofactors, representing potential QTLs, were selected by forward stepwise regression with the statement cov SEL. The proportion of phenotypic variation accounted by each QTL (R²) was calculated by multiple regression in PLABQTL. When several QTLs were detected for one trait, an adjusted R² (R²adj), representing the proportion of the phenotypic variation explained by all QTLs, was calculated as described by Hospital et al. (1997). For each QTL, a support interval was automatically generated by the LOD fall of 1.0 method (Lander and Botstein 1989). We then searched for epistasis with the Epistat program (Chase et al. 1997). We used an automated search for both all-pairwise and conditional interactions with a log likelihood ratio threshold >6 as suggested by the authors. We subsequently used the Montecarlo simulation program associated with Epistat to analyze 10⁶ resampled subgroups, as a way of estimating P values. To avoid false positive interactions, epistasis was only considered to be significant when P < 0.0005. The BioMercator software (Arcade et al. 2004) was used to generate a consensus genetic map of the two RIL populations. This was done by iterative projection of loci from the LR5 map onto the LR4 map. QTLs were projected onto this map. They were designated as trait abbreviations followed by the chromosome number (1–8) and the name of the genetic background (LR4 or LR5) from which they were identified. The genetic map and the QTLs were then drawn using MapChart, version 2.1 (Voorrips 2002).

First, all the genes present in the QTL support intervals were listed in silico. Then, some were selected in the light of specific a priori hypotheses we made at cellular and ecophysiological scales. The length of an organ (e.g. an internode, or a leaf, Tardieu and Granier 2000) at a given time L(t) can be described as:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ L(t) = \sum\limits_{i = 1}^{N(t)} {Li(t)} $$\end{document}where N(t) is the number of cells at time t, and Li(t) the length of each cell i at this time. Based on previous studies (Gendreau et al. 1997; Barroco et al. 2005; Gimeno-Gilles et al. 2009), we hypothesized that germination and hypocotyl heterotrophic elongation were mainly related to cell elongation and not to an increase in the number of cells. Cell length which is related to changes in cell wall properties or in vacuole size can be altered not only by several environmental conditions, in particular temperature, but also by water availability and mechanical obstacles during heterotrophic growth. Such changes are strongly influenced by phytohormones (Hopkins and Huner 1999), especially abscissic acid (ABA) and gibberellins (GA), and depend on molecular cascades, including signal sensing and transduction, transcription control and stress responsive mechanisms that could be regulated differently in different genotypes. The list of Bacterial Artificial Chromosomes (BACs) included within the support interval for all QTLs was established from the positions of the upper and lower markers flanking the QTLs, by using the integrated M. tr. genetic map from the website http://www.medicago.org/genome/map.php. The chromosomal positions of the BACs on which given genes were annotated were confirmed by referring to the Oklahoma University website devoted to M. tr. (http://www.medicago.org/genome/index.php). We also used the website developed by the Kazusa DNA Research Institute for red clover (http://clovergarden.jp) to confirm certain BAC positions, because red clover is phylogenetically close to M. tr. (Choi et al. 2004b) and because the two species have a close syntenic relationship (Sato et al. 2005). The M. tr. gene Tentative Consensus (TCs) sequences were searched by keyword queries in the DCFI Medicago gene index (http://compbio.dfci.harvard.edu/tgi/cgi-bin/tgi/Blast/index.cgi). Additionally, the M. tr. EST sequences obtained in the TIGR Medicago genome (IMGAG loci) browser were blasted between Arabidopsis (http://www.arabidopsis.org/Blast/index.jsp-BlastX) and the red clover EST bank (http://est.kazusa.or.jp/en/plant/red_clover/EST/).

Table 1 gives mean values for seed mass, germination and seedling growth characteristics measured on LR5 at low temperatures, on LR4 at 20°C, and on their parental lines. It also includes skewness and kurtosis values that gave information on trait distribution shapes. Analysis of variance indicated highly significant differences between the RILs for seed mass and all the variables measured for germination and heterotrophic growth in LR5 and LR4 (SNES experiment only, as no replicate was made for the LEG experiment).Table 1

Seed characteristics, germination and heterotrophic growth (a) at sub-optimal temperature in the LR5 population (Jemalong × F83005.5) and (b) at supra-optimal temperature in the LR4 population (Jemalong × DZA315.16; two experiments SNES and LEG)

In both experiments, all germination times (T20 to T90) were positively correlated (r > 0.5***). At 5°C (LR5), the final percentage of germination (FPG) varied and was also negatively correlated with germination times (−0.74*** < r < −0.6***). This indicated that a RIL that began germinating early also germinated rapidly and reached a high final percentage. Germination times and FPG were thus all indicators of the same trend in the germination curve of a RIL. Seed mass was not correlated with germination time courses, FPG or HL. It was only positively correlated with HW (r = 0.59***), whereas HL and HW were not correlated. In the LR4 SNES experiment, no variables were correlated with SM except ISSA (r = 0.49***). The negative correlation observed between ISSA and IR (r = −0.65***) indicated a higher imbibition rate for small seeds. Weak but significant negative correlations (r < −0.25**) appeared between germination times and ELR, even though measurements were made for each seed so as to carefully separate these two variables.

Regarding germination, one QTL was found for FPG on chromosome 3 and one for T90 on chromosome 8. QTLs for other Ti were not significant but co-localized with the QTL for T90 (Table 2a; Fig. 1). The latter QTL explained 15% of phenotypic variation and showed a high additive effect, the F83005.5 alleles decreasing the trait value (Table 2a).Table 2

QTLs identified for seed mass, germination and heterotrophic growth, (a) at sub-optimal temperatures in the LR5 population (Jemalong × F83005.5), (b) at 20°C, a supra-optimal temperature in the LR4 population (Jemalong × DZA315.16; two experiments SNES and LEG)

Regarding heterotrophic growth, QTLs were identified for both HL and HW. The three HL QTLs each contributed to a high percentage of variation (>17%). They were located on chromosomes 1, 2 and 8 (Table 2a; Fig. 1). For all QTLs associated with variations in HL, the F83005.5 alleles decreased the trait value. The HW QTL was localized on chromosome 2, the F83005.5 alleles increasing HW but with a low additive effect. For this trait, a significant epistatic interaction (P < 0.0002) was found between MTE16 (chromosome 2) and MTE136 (chromosome 3). This interaction implicated additive QTL-linked markers associated with hypocotyl elongation at low temperature (QTL HW2-LR5 nearby MTE16 and QTL HL8-LR5 nearby MTE136). When combined, F83005.5 alleles increased HW. QTLs identified for the different variables were not co-localized. In particular the QTLs for germination and heterotrophic growth were distinct.

In the SNES experiment, QTLs were found for SM, IR, Ti (i < 70%) and for ELR and one to four QTLs were identified for each trait (Table 2b; Fig. 1). QTLs for the different traits all mapped into distinct genomic positions. Considering the germination times, two QTLs were identified on chromosome 5, one at position 8 (T20 and T50.5-LR4) and the other at position 48 (T70.5-LR4). The LOD value of the last QTL was lower. The DZA315.16 alleles increased the Ti values for both QTLs, i.e. they decreased the germination rate. Concerning heterotrophic growth, one QTL was mapped for early ELR on chromosome 3 in a distinct chromosomal region from the QTLs for SM and for traits characterizing germination. In the LEG experiment, two QTLs involved in the variation in T90 were detected on chromosomes 5 and 7. These QTLs were distinct from those identified with SNES data.

SM QTLs identified in LR5 were located on chromosomes 5 and 8 (Table 2a; Fig. 1) each explaining more than 10% of SM variation. In LR4 (SNES), four QTLs were identified on chromosomes 1, 5, 7 and 8 (Table 2b; Fig. 1). The QTL on chromosome 7 had the highest LOD value (7.6) and the highest explained variance (18.5%). The genetic effect at this QTL was negative, i.e. the DZA315.16 alleles decreased SM values. SM QTLs identified with the LEG seed lot on chromosomes 1 and 7 were a subset of those found with the SNES seed lot and could therefore be considered as stable QTLs in the LR4 genetic background. The support intervals of the QTLs identified on chromosome 8 for both populations overlapped but their peaks were not in the same marker intervals, suggesting they may be different.

Whatever the RIL population used, SM QTLs were not co-localized with any QTL of other variables in spite of the correlations observed between SM and other traits (HW, ISSA). The QTLs detected for SM and Ti (i < 50%) on chromosome 5 in the LR4 SNES experiment overlapped, suggesting a pleiotropic effect or tight linkage between them. If so, alleles that resulted in a high SM at these QTLs would result in a longer germination time (lower germination rate). However, these QTLs were probably distinct as the two variables were not correlated and the peaks of both QTLs were not in the same marker intervals.

Depending on the size of the QTL support interval (ranging from 4 to 16 cM), 10 to 34 BACs and 20 to 40 BACs were analyzed per QTL in LR5 and LR4, respectively. A total of 204 BACs were analyzed in LR5 and 151 in LR4. The number of genes analysed in support intervals was 700 in average. Up to now, about 50% of M. tr. genes have unknown function (http://www.medicago.org/). Table 3 lists the candidate genes selected for SM and for the other traits, from imbibition to elongation, based on a priori hypotheses we had made at cellular and ecophysiological scales.Table 3

Genes identified in the support intervals of QTLs, by data mining and BLAST results for similarity with A. thaliana genes (a) for QTLs identified at sub-optimal temperatures in the LR5 population and (b) for QTLs identified at a supra-optimal temperature in the LR4 population