A tandem array of CBF/DREB1 genes is located in a major freezing tolerance QTL region on Medicago truncatula chromosome 6

Background Freezing provokes severe yield losses to different fall-sown annual legumes. Understanding the molecular bases of freezing tolerance is of great interest for breeding programs. Medicago truncatula Gaertn. is an annual temperate forage legume that has been chosen as a model species for agronomically and economically important legume crops. The present study aimed to identify positional candidate genes for a major freezing tolerance quantitative trait locus that was previously mapped to M. truncatula chromosome 6 (Mt-FTQTL6) using the LR3 population derived from a cross between the freezing-tolerant accession F83005-5 and the freezing-sensitive accession DZA045-5. Results The confidence interval of Mt-FTQTL6 was narrowed down to the region comprised between markers MTIC153 and NT6054 using recombinant F7 and F8 lines. A bacterial-artificial chromosome (BAC) clone contig map was constructed in an attempt to close the residual assembly gap existing therein. Twenty positional candidate genes including twelve C-repeat binding factor (CBF)/dehydration-responsive element binding factor 1 (DREB1) genes were identified from BAC-derived sequences and whole-genome shotgun sequences (WGS). CBF/DREB1 genes are organized in a tandem array within an approximately 296-Kb region. Eleven CBF/DREB1 genes were isolated and sequenced from F83005-5 and DZA045-5 which revealed high polymorphism among these accessions. Unique features characterizing CBF/DREB1 genes from M. truncatula, such as alternative splicing and large tandem duplication, are elucidated for the first time. Conclusions Overall, twenty genes were identified as potential candidates to explain Mt-FTQTL6 effect. Their future functional characterization will uncover the gene(s) involved in freezing tolerance difference observed between F83005-5 and DZA045-5. Knowledge transfer for breeding improvement of crop legumes is expected. Furthermore, CBF/DREB1 related data will certainly have a large impact on research studies targeting this group of transcriptional activators in M. truncatula and other legume species. Electronic supplementary material The online version of this article (doi:10.1186/1471-2164-14-814) contains supplementary material, which is available to authorized users.


Background
Freezing is an adverse abiotic factor with severe negative impacts on plant health and productivity [1]. Most temperate plants can increase their freezing tolerance after exposure to low temperatures, a process called cold acclimation [1][2][3]. CBF/DREB1 genes play a key role in the regulation of the transcriptome during cold acclimation [4]. They belong to the AP2/EREBP family of transcription factors and were first isolated from Arabidopsis thaliana (L.) Heynh. [5][6][7]. Up to now, CBF/DREB1 genes have been identified in numerous herbaceous and woody plant species [8][9][10][11] and different studies have reported their significant role in freezing tolerance [12][13][14][15][16].
Grain legumes are an important source of food and feed worldwide. With their seeds containing 20% to 30% protein, they largely contribute to the protein requirements of humans and animals [17]. Forage legumes are a valuable source of protein, fiber and energy for livestock as well [17]. Besides their nutritional importance, legumes are an exceptional component to sustainable agriculture. In fact, most legume species are able to establish nitrogen-fixing symbioses with rhizobial bacteria [17,18] which reduces the need for fertilizer application in crop rotation systems. Nowadays, the susceptibility to low temperatures and freezing conditions still represents a major constraint to the cultivation of legumes in many agro-ecological zones. The identification of the genes responsible for the natural freezing tolerance variation can enhance the breeding progress and subsequently the release of novel freezing-tolerant cultivars.
Medicago truncatula is an annual legume of Mediterranean origin [19]. It is not only a valuable forage crop as in Southern Australia [20] but has been chosen as a model molecular-genetic system for legume biology [21]. A large array of genomic tools has been developed for the reference accession A17 [22] and a draft sequence of the euchromatic portion of its genome comprising approximately 94% of all genes has been released [23]. These resources can accelerate the identification of the molecular determinants of various traits in M. truncatula including freezing tolerance. Knowledge can be then transferred to different agronomically and economically important crops considering their phylogenetic closeness to M. truncatula and the remarkable synteny between their genomes and that of M. truncatula [24][25][26][27][28].
Studies on cold acclimation and freezing tolerance in M. truncatula are scarce [29][30][31][32][33][34]. Few biochemical changes have been reported in response to low non-freezing temperatures for this species [30,31,34]. At the molecular level, transcripts corresponding to M. truncatula cold acclimation-specific gene 15, a candidate CBF/DREB1 target, have been found to accumulate 6 hours after exposure to low temperature [32]. Four CBF/DREB1 genes, i.e. MtCBF1-4, have been reported to be rapidly induced in response to cold stress [32,34,35]. Furthermore, the overexpression of MtCBF3 (also named MtDREB1C) has been shown to improve the freezing tolerance of transgenic M. truncatula accession Jemalong [14]. MtCBF1 and MtCBF4 are located on M. truncatula chromosomes 5 and 1, respectively, but no information on the physical localization of MtCBF2-3 has yet been obtained. Apart from MtCBF1-4, no other CBF/DREB1 genes have been described and characterized from M. truncatula even though expressed sequence tag (EST) and complementary DNA (cDNA) potentially representing transcripts from other putative CBF/DREB1 genes are available [36,37].
Recently, Avia et al. [38] have uncovered natural variation for freezing tolerance among 15 M. truncatula accessions after a cold acclimation period. The genetic bases for freezing tolerance variation between the contrasted accessions F83005-5 and DZA045-5 have been identified using a quantitative trait locus (QTL) mapping approach [38]. A major freezing tolerance QTL (Mt-FTQTL6) accounting for 40% of the phenotypic variation has been mapped to a region of M. truncatula chromosome 6 [38] coinciding with an assembly gap in the A17 euchromatic sequence [39]. The present study aimed to fine map Mt-FTQTL6 and to identify genes located in the corresponding genomic region.

Results
Freezing tolerance evaluation of recombinant lines validates Mt-FTQTL6 position and narrows down its confidence interval Eighteen F 7 or F 8 lines were evaluated for freezing tolerance after a cold acclimation period ( Figure 1). Based on markers used in this study (Additional file 1), screened lines correspond to 9 different haplotypes; 7 of which carry recombination events within or next to Mt-FTQTL6 confidence interval ( Figure 1). Phenotypic data from F83005-5 and DZA045-5, included as check plants, were significantly different as expected. Comparisons of the phenotypes and marker genotypes validated Mt-FTQTL6 confidence interval previously located between markers NT6001 and NT6019 [39]. Freezing damage scores from the self-pollinated progenies of lines 76-08-04 and 76-06-187-02 showed that F83005-5 inserts upstream NT6001 have no phenotypic effect. The same conclusion was reached concerning F83005-5 inserts downstream NT6019 based on the progenies of line 76-05-06. Furthermore, Mt-FTQTL6 interval could be delineated to a smaller region of 0.4 cM between markers MTIC153 and NT6054 (Figure 1 and 2; Additional file 2). A critical recombination event in line 76-02-86, as seen in lines 76-02-86-07 and 76-02-86-12, suggested MTIC153 as the left marker for Mt-FTQTL6 confidence interval. The right marker, NT6054, was identified based on information from the progenies of lines 76-09-04 and 76-11-108-04.

Twenty non-transposon genes reside in Mt-FTQTL6 confidence interval
Twenty non-transposon genes with significant matches to GenBank's non-redundant nucleotide (nr/nt), expressed sequence tag and protein databases, including those corresponding to markers MTIC153 and NT6054, could be identified in the candidate sequence for Mt-FTQTL6 ( Figure 2; Table 1). Spacing between adjacent genes varies from 1,586 bp (between MtCBF5 and MtPERLD) up to 318,549 bp (between MTR_054s0019 and MTR_6g089580).
Coding sequences have an overall G + C content of 42.15%. Sixteen genes are supported by cognate EST and additionally 7 of them by cDNA sequences (Table 1). Comparisons between genomic and EST/cDNA sequences indicate that four genes (i.e. MtCBF8, MtCBF6, MtPERLD and MTR_054s0019) have 2 transcript isoforms each. All positional candidate genes for Mt-FTQTL6 were found to have highly similar counterparts across a set of dicotyledonous species for which genome sequence information is available (Additional file 7). All, except for homologous genes to MTR_054s0001, are contained in syntenic blocks exhibiting large gene order conservation with Mt-FTQTL6 (Additional file 7). Three chromosomal segments in A. thaliana and in Solanum lycopersicum L., two segments in Populus trichocarpa (Torr. & Gray) and a single segment in Vitis vinifera L. share colinearity with Mt-FTQTL6. Among legumes, two additional M. truncatula, six Glycine max (L.) Merr. and three Lotus japonicus L. regions are colinear with Mt-FTQTL6. Few rearrangements could be noted among conserved regions including gene loss, local duplication or also translocation events (Additional file 7).
Twelve candidate genes organized in a tandem array belong to the CBF/DREB1 group of the AP2/EREBP transcription factor family CBF/DREB1 genes represent 60% of all candidates for Mt-FTQTL6 (Table 1). They are organized in a tandem cluster spanning a region of approximately 296 Kb ( Figure 2). No intervening spacer genes, except those for retrotransposons, exist in between. MtCBF12 and MtCBF13 are pseudogenes. A deletion of a guanine at position 314 in MtCBF12 coding sequence causes a frameshift and a premature stop codon at the 148 th amino acid. MtCBF13 coding region shows an in-frame 848-bp insertion and two point mutations leading each to a premature stop codon. MtCBF2-3;5-11;14 are likely to represent functional protein-coding genes. The corresponding proteins share 61 to 92% overall amino acid identity (Additional file 8) and all show typical features of the CBF/DREB1 group ( Figure 3). They have each a 58-amino acid long AP2/ERF nuclear signaling and DNA binding domain [42][43][44] and an acidic C-terminal region [isoelectric point (pI) ranging between 3.94 and 5.04] that may function in transactivation [45]. MtCBF11 contains in addition to the 58-amino acid AP2/ERF domain, a 17-amino acid long sequence most likely corresponding to the C-terminal end of a second AP2/ERF domain (Figures 3 and 4). CBF signature sequences comprising PKKP/RAGRxKFxETRHP and DSAWR motifs bracketing the AP2/ERF domain [8] are largely conserved among MtCBF2-3;5-11;14. These latter also show A(A/V)xxA(A/V)xxF [46] and LWSY motifs [47], reported to be conserved in CBF/DREB1 proteins in different plant species. Variants at the CBF/DREB1 characteristic amino acid sequences are listed in Additional file 9. The (L/Y)(L/Y)x(N/S)(M/L)A(E/Q)G (M/L)(L/M)xxPP sequence, previously suggested as a CBF/DREB1 conserved motif [46] is not present in MtCBF2-3;5-14 except for the NMA motif. These proteins have rather a NM(A/V) LMSPTHS conserved sequence at the same position.
Distinct levels of polymorphism exist between MtCBF2-3;5-12;14 alleles from F83005-5 and DZA045-5 MtCBF2-3;5-12;14 specific primers yielded positive amplification from F83005-5 and DZA045-5 indicating that all 11 genes are present in these accessions. No amplification could be obtained with MtCBF13-specific primers from both accessions despite using different primer combinations and PCR conditions. For MtCBF2, two distinct PCR products were obtained from F83005-5 (named MtCBF2a and MtCBF2b) using distinct forward primers but not from DZA045-5, thus indicating that this gene has undergone duplication at least in F83005-5's background. Overall, 14,953 bp (F83005-5) and 15,038 bp (DZA045-5) of comparable sequences were generated (Additional file 10). They show a total of 402 single nucleotide polymorphism (SNP), 41 insertion-deletion (indel) and 9 simple sequence repeat (SSR) polymorphisms with 206 SNP, 8 indel and 3 SSR located in the coding sequences (Additional file 10). Contrary to A17 and F83005-5, DZA045-5 has an insertion of a 160-bp short interspersed nuclear element (SINE) in the coding   [48]. Solid squares indicate the conserved WLG motif in the middle of the AP2/ERF domain which is a common feature with almost all A. thaliana, G. max and Oryza sativa L. genes carrying a single AP2/ERF domain [48,49]. Solid and open triangles show conserved amino acid residues involved in the interaction of the AP2/ERF domain with target DNA sequences and the sugar phosphate backbone, respectively [50]. The A(A/V)xxA(A/V)xxF [46] and LWSY motifs [47] are underlined with crosses and the CBF/DREB1 signature sequences bracketing the AP2/ERF domain [8] with asterisks. Open circles indicate arginine and phenylalanine residues in the PKK/RPAGRxKFxETRHP signature sequence reported to be critical for DNA binding [44]. The C-terminal clusters of hydrophobic residues [45]  sequence of MtCBF14 and another SINE element in the 5′ upstream region of MtCBF8. F83005-5 has a specific insertion of a putative miniature inverted-repeat transposable element in the 5′ upstream region of MtCBF6. Figure 4 illustrates allelic variation between F83005-5 and DZA045-5 in the coding regions of MtCBF2-3; 5-12;14. As it could be noticed, nucleotide changes leading to differences at the amino acid level are not evenly distributed amongst genes. Amino acid substitution difference between F83005-5 and DZA045-5 ranges between 1 (MtCBF3) and 34 (MtCBF5). MtCBF14 and MtCBF8 are likely to encode truncated proteins in freezing-sensitive accession DZA045-5 but not in freezing-tolerant accessions F83005-5 and A17. In contrast, MtCBF12 is likely to encode a full-length protein in DZA045-5 which is not the case for F83005-5 and A17.

Discussion
M. truncatula is a valuable forage crop and also a prominent model for legume genomics. Freezing tolerance QTL have been previously identified in M. truncatula [38]. In the current study, Mt-FTQTL6 has been finely mapped by genetically and physiologically characterizing recombinant haplotypes carrying recombination events within the QTL confidence interval. Genomic data have been developed to fill the genome assembly gap in Mt-FTQTL6 region and have permitted to reveal twenty positional candidate genes. Interestingly, twelve genes belong to the CBF/DREB1 group of the AP2/EREBP transcription factor family and occur in tandem array.
Current knowledge on the functions of the twenty positional candidate genes for Mt-FTQTL6 The fine mapping step has significantly reduced Mt-FTQTL6′s confidence interval and has subsequently allowed the identification of 20 candidate genes. Functional data regarding these candidate genes are still limited at present (Additional file 11  [32,34]. Furthermore, transgenic M. truncatula lines overexpressing MtCBF3 were shown to exhibit an improved freezing tolerance compared to the wild-type [14]. However, even if the CBF/DREB1 genes can be considered as the most likely candidates for the Mt-FTQTL6 effect, none of the non-functionally characterized genes can be excluded at present. Further knowledge needs to be gained regarding the different candidate genes and the QTL of interest itself.
Co-location between a freezing tolerance QTL and CBF/DREB1 genes in a legume species CBF/DREB1 genes were found to co-locate with QTL for freezing tolerance in several plant species including A. thaliana [59,60], Triticum monococcum L. [61], Triticum aestivum L. [62] and Hordeum vulgare L. subsp. vulgare [63]. A large deletion in the promoter region of A. thaliana CBF2 [59] and co-locations of freezing tolerance QTL in T. monococcum [61] and H. vulgare subsp. vulgare [63] with QTL for expression of cold-induced CBF/DREB1 target gene COR14b and accumulation of COR14b protein suggested CBF/DREB1 genes being responsible for the QTL effects. The identification of CBF/DREB1 genes among positional candidates of Mt-FTQTL6 provides an additional example of a co-location between a freezing tolerance QTL and this group of genes but for the first time in a legume species. It will be of great interest to inspect such colocation in other legumes especially P. sativum having a freezing damage QTL syntenic to Mt-FTQTL6 [39].

Large size of the CBF/DREB1 group in M. truncatula
This study has permitted the physical positioning of twelve M. truncatula CBF/DREB1 genes through the assembly of Mt-FTQTL6 region. Apart from MtCBF2 and MtCBF3 that were identified by Pennycooke et al. [32] and DREB1A whose cDNA sequence was directly submitted to GenBank (GB acc DQ778006), the other CBF/ DREB1 genes associated with Mt-FTQTL6 were not previously annotated or described ( Table 1). As the name "MtCBF4" has been recently assigned to a CBF/DREB1 gene located on M. truncatula chromosome 1 [35], the novel genes including DREB1A were here given consecutive numbers from 5 to 14 starting from the most proximal to MtPERLD (Figure 2). MtCBF2-3;5-14 occur in tandem array and are in direct orientation ( Figure 2) which suggests that they are most likely derived from an unequal crossing over mechanism [64,65]. However, the contribution of transposable elements to this duplication event cannot be totally excluded [65], especially with several transposon and retrotransposon insertions being evident between CBF/DREB1 genes (data not shown).
According to Young et al. [23], M. truncatula has experienced high rates of local gene duplication compared to other plant genomes. It may thus not be surprising to have such a large tandem array organization for the MtCBF2-3;5-14 locus. Similar tandem duplications have been described for nucleotide-binding site-leucine-rich repeat genes [66] and lipoxygenase genes [67] on M. truncatula chromosomes 6 and 8, respectively. However, the main particularity of the tandem cluster in Mt-FTQTL6 region is that it concerns CBF/DREB1 genes. Large groups of CBF/DREB1 genes have always been thought to be limited to monocotyledons. This study reports for the first time that the CBF/DREB1 genes can also be highly duplicated in legume species. M. truncatula has at least 17 CBF/DREB1 genes (considering MtCBF2a and MtCBF2b from F83005-5) all located in homoeologous regions (Additional file 7). A recent report has similarly shown that Eucalyptus grandis genome contains 17 CBF/DREB1 genes [68]. It will be next important to determine if other legumes have as many genes that belong to the CBF/ DREB1 group as M. truncatula. First evidences denote that there may not be a general trend at this level. As shown in Additional file 7, G. max does not have any duplication of CBF/DREB1 genes in Mt-FTQTL6 syntenic regions. In contrast, L. japonicus has 7 tandemly-arrayed putative CBF/DREB1 genes on chromosome 4.
Complexity of the CBF/DREB1 locus on M. truncatula chromosome 6 As for other duplication modes, tandem duplication generates identical genes that evolve in different ways [65,[69][70][71][72][73][74][75]. In the case of the locus harboring MtCBF2-3;5-14, some evidences of pseudogenization are apparent: premature stop codons leading to truncated proteins exist within the corresponding coding sequences of MtCBF13 and MtCBF12 in A17, MtCBF12 in F83005-5 and MtCBF8 and MtCBF14 in DZA045-5 ( Figure 4). However, most CBF/DREB1 duplicates are likely to encode functional proteins. Up to date, information regarding the expression profiles of MtCBF2-3;5-14 is still greatly lacking. M. truncatula gene expression atlas [76] provides only a first glimpse on the expression of some genes among MtCBF2-3;5-14 under unstressed conditions (Additional file 12). It is thus crucial to conduct a complete survey on the expression patterns (including under cold stress) of all CBF/DREB1 genes on M. truncatula chromosome 6. This will determine whether tandemly duplicated CBF/ DREB1 genes have evolved or not different expression patterns and so possibly distinct functions as already reported for different CBF/DREB1 genes in other plant species [10,41,77,78]. The characterization of MtCBF11 deserves a particular attention as its predicted protein product contains a full-length and an additional 17-amino acid long partial AP2/ERF domain (Figure 3 and 4). MtCBF11 has most likely arisen from an inframe fusion of the 5′ portion of the coding region from a first CBF/DREB1 gene with the 3′ portion of the coding sequence from a second gene.
Alternative splicing events can diverge between duplicates, including organ-and stress-specific differences, and result in functional variation [79]. Based on genomic sequence alignment with EST and cDNA data, MtCBF6 and MtCBF8 have two transcript variants each, generated by intron retention/splicing events (Table 1). MtCBF5 and MtCBF7 contain each an intron at the same position as for MtCBF6 and MtCBF8 intronspliced isoforms but no evidence yet suggests that MtCBF5 and MtCBF7 may have intron-retaining transcript forms similar to MtCBF6 and MtCBF8. As EST databases can be incomplete with regard to alternative splicing transcript variants, reverse transcription PCR or transcription analysis using second generation sequencing would allow determining alternative splicing patterns with more precision. This is the first study to provide data on the presence of introns and alternative splicing forms in CBF/DREB1 genes. Each of the isoforms of MtCBF5-8 will need to be characterized apart in order to determine if these are functional and then compare them to other CBF/DREB1 genes and alternatively spliced forms if any. Any possible association of alternative splicing of CBF/ DREB1 genes with M. truncatula response to low temperatures should be inspected. Recent reports from grasses have shown the existence of alternatively spliced isoforms for AP2/EREBP family members [80][81][82] homologous to A. thaliana DREB2 [6] which is known to be activated by dehydration and high-salt stresses. The alternative splicing of these genes was suggested to be important for reducing the activity of their target genes under unstressed conditions and thus preventing negative effects on plant growth.

Conclusions
The present study has permitted to narrow down the confidence interval of a major freezing tolerance QTL on M. truncatula chromosome 6 and to identify genes located therein. No similar studies related to legume freezing tolerance have to date been reported. Candidate genes represent important elements toward positional cloning of Mt-FTQTL6. Information gained from M. truncatula being a model species should be transferred to agronomically and economically important crop legumes. The BAC clone contig map constructed for the assembly gap coinciding with Mt-FTQTL6 will be interesting in efforts to clone important loci mapped to the same region as Mt-FTQTL6 including a locus for resistance to Subterranean clover mottle virus [83] and a QTL for the number of secondary branches [84]. When sequenced from F83005-5 and DZA045-5, MtCBF2-3;5-12;14 coding sequences showed distinct levels of indels and non-synonymous mutations which suggest that they have evolved in a heterogenous manner. Allelic variation identified on coding and immediate flanking sequences of MtCBF2-3;5-12;14 will be useful for future studies to establish the adaptative value of the different CBF/DREB1 duplicates.

Development of plant material
M. truncatula LR3 population comprises 178 F 2:5 recombinant inbred lines (RIL) obtained by single seed descent from a cross between the freezing-tolerant accession F83005-5 and the freezing-sensitive accession DZA045-5 [85]. F83005-5 derives from a natural population collected in France and DZA045-5 from an Algerian population [86]. Genotypic data from the LR3 population [38] indicate that twelve RIL (RIL4, 13, 17, 22, 76, 90, 101, 151, 161, 165 and 241) have residual heterozygosity at Mt-FTQTL6 peak marker, MTIC153. Among them, RIL76 is particularly important because: (1) it is homozygous at the confidence intervals of both other freezing tolerance QTL from the LR3 population mapped on chromosomes 1 and 4 [38]; (2) it carries the sensitive parent's alleles at both these QTL; and (3) it shows a low overall genome heterozygosity. RIL76 sister lines were therefore chosen for the development of inbred lines showing recombination events at Mt-FTQTL6. Twelve F 5 lines were grown and submitted to a genotypic analysis using: SSR markers evenly spaced on the eight M. truncatula chromosomes (3 to 6 markers per chromosome; 33 markers in total), the nearest SSR marker to the peak of each of the freezing tolerance QTL on chromosomes 1 and 4, and SNP and SSR/ indel markers closely linked to Mt-FTQTL6 (6 markers in total). Information regarding Mt-FTQTL6-linked markers is available in Additional file 1. The complete list for the rest of markers is available upon request. Similarly to RIL76, sister lines did not show heterozygosity for any of the background markers and were all homozygous carrying alleles from the freezing-sensitive parent at the freezing tolerance QTL on chromosomes 1 and 4. Five heterozygous recombinants at Mt-FTQTL6 were identified, namely 76-02, 76-05, 76-07, 76-08 and 76-09. Taking advantage of their heterozygosity for all markers associated with Mt-FTQTL6, 76-06 and 76-11 were selected for the construction of large segregating populations. Line 76-02 was also used for this purpose because it is only homozygous for two markers at the upper border of Mt-FTQTL6, namely NT6005 and NT6001 (Figure 1). F 6 populations that were obtained through self-pollination of 76-02, 76-06 and 76-11 comprised 447, 241 and 232 plants, respectively. Progenies were genotyped using 7 SNP and 6 SSR/indel markers located on both sides of marker MTIC153 (Additional file 1). For recombinant individuals, genotyping was further completed with 13 additional SNP markers (Additional file 1). A high-resolution linkage map constructed from a subset of marker data scored on 76-06 and 76-11 progenies is described in Tayeh et al. [39] (Additional file 2). F 5 and F 6 heterozygous recombinant individuals were self-pollinated. Homozygous (F 6 or F 7 ) recombinants were selected from their progenies using corresponding markers. F 7 and F 8 plants were finally obtained through the self-pollination of homozygous recombinant plants and were submitted to freezing tolerance tests (Figure 1).

Plant growth
Seeds of M. truncatula were scarified using sand paper in order to rupture the seed coat. Scarified seeds were soaked in distilled water for 6 hours and spread on moistened filter papers in Petri dishes. After a 3-day dark storage at 6°C to break embryo dormancy and synchronize germination, Petri dishes were held at 20°C for three other days. Seedlings were then transplanted in 2 L pots containing a mix of peat and compost or in pre-moistened 41 × 42-mm jiffy-7 pellets (Jiffy France S.A.R.L., Trevoux, France), depending on the experimental need. In case of nongerminated seeds, especially when they were recently harvested, 3 to 5 parts per million of 2-chloroethylphosphonic acid were applied to Petri dishes to break embryo dormancy and seedlings were transplanted 24 hours later. Except for freezing tolerance tests, plants were finally left to grow in the greenhouse at an average temperature of 18/ 14°C (day/night) and a 16-hour photoperiod. Seeds were extracted from ripe pods using in most cases a homemade extraction system based on a rubber mat and a plaster hawk as described by Garcia et al. [87]. Large extractions were performed using a seed mill.

Marker development and genotyping
In addition to markers from Tayeh et al. [39], 8 new markers associated with Mt-FTQTL6 were developed in this study (Additional file 1). Primers were designed from BAC-end sequences using the Primer3 tool [88] with default parameters. Maximum product length was limited to 500 bp. Total genomic DNA was isolated from fresh leaflets harvested on one-month-old plants using Nucleospin 96 Plant II (Macherey Nagel, Germany) according to manufacturer's instructions. Genotyping of the plant material from this study with background and Mt-FTQTL6-associated markers was conducted as described in Tayeh et al. [39]. Capillary gel electrophoresis was employed for SSR and indel markers. The highresolution melting technique (in case of F 7 /F 8 lines) and the competitive allele specific PCR method (in case of 447-plant 76-02-derived population) were used for SNP markers.

Evaluation of freezing tolerance
Freezing tolerance tests were performed in a climatecontrolled walk-in 2.6 × 2.3 × 2 m growth chamber. Two experiments, consisting of two replications each, were carried out. Each replicate was composed of: (1) F 7 or F 8 plants carrying one (or two) recombination event(s) within Mt-FTQTL6-containing chromosomal region, (2) plants obtained from the same parental lines as the recombinant ones or from sister parental lines but that ideally do not have recombination events within Mt-FTQTL6 region and (3) parental accessions F83005-5 and DZA045-5 ( Figure 1). Ten individuals from each of the above-cited lines were evaluated per replicate. Seedlings, transplanted in Jiffy pellets (see plant growth section), were placed in 10 × 10 well-containing homemade Styrodur plates to ensure root insulation. Four different phases were applied during each freezing tolerance test with the following environmental conditions: (1) nursery phase-18 days: 20°C/ 14°C, 14-hour photoperiod and 250 μmol.m -2 .s -1 photosynthetically active photon flux density (PPFD) provided by fluorescent lights; (2) cold acclimation phase-20 days: 8°C/2°C, 14-hour photoperiod and 250 μmol.m -2 .s -1 PPFD; (3) freezing phase-8 days: 4°C/-6°C, 10-hour photoperiod and 150 μmol.m -2 .s -1 PPFD; and (4) recovery phase-16 days: 16°C/5°C, 14-hour photoperiod and 250 μmol.m -2 .s -1 PPFD. Continuous acquisition of chamber air temperature was carried out using a Campbell CR10x system equipped with a Humitter 50Y integrated humidity and temperature transmitter (Vaisala, Vantaa, Finland). Temperatures at leaf-and root-levels were also continuously surveyed using eight and four homemade T-type thermocouples connected to the acquisition system, respectively. Photosynthetically active radiation was checked twice per test phase by means of a LiCor 190SA quantum sensor (Licor Biosciences, Lincoln, Nebraska, USA). Irrigation was performed manually and stopped 2 days before and all along the freezing phase. Developmental stages were scored plant-by-plant six times during the nursery and cold acclimation phases. At the 16 th day of rewarming, plants were evaluated for freezing tolerance using a visual rating scale based on leaf injuries and ranging from 0 (no freezing injuries) to 5 (all leaflets are yellow and dry: dead plant). Plants were scored 1 when less than one quarter of the leaflets were visually damaged, 2 for one quarter of the leaflets being affected, 3 when freezing damage was observed on half of the leaflets and 4 when three-quarter of the leaflets were concerned. Pairwise comparisons of damage scores from recombinant, non-recombinant and parental lines were performed by one-way analysis of variance followed by a Tukey test at P ≤ 0.05 using R [89] ( Figure 1). For each of these pairwise comparisons, the recombinant line and its non-recombinant counterpart(s) were compared between each other and with the closest plants of F83005-5 and DZA045-5 within the growth chamber.
Full BAC sequencing BAC DNA was isolated using Nucleospin fast purification Kit (Macherey-Nagel, Düren, Germany). Full BAC sequencing was performed using 454 multiplexing technology on a GS-FLX Titanium sequencer (454 Life Sciences, Roche Diagnostics, Branford, Connecticut, USA). Raw reads were cleaned using PyroCleaner [90]. Reads sharing more than 95% homology with BAC vector or Escherichia coli genome sequence were excluded. de novo assembly of the cleaned reads was performed using Newbler (version 2.3; 454 Life Sciences, Roche Diagnostics). The resulting contigs were, when possible, ordered and oriented based on marker sequences located therein. End sequences from overlapping BAC clones (Additional file 5) were used for the same purpose. Complementarily, end sequences of the different contigs were submitted to blastn/blastx [91] search against M. truncatula nr/nt, high-throughput genomic sequence, EST, genome survey sequence and protein databases in GenBank [36] at the National Center for Biotechnology Information. End sequences sharing more than 99% identity with the same target sequence(s) were considered as indicative of adjacent contigs.
Sequencing of CBF/DREB1 genes from F83005-5 and DZA045-5 The isolation and sequencing of MtCBF2-3;5-14 from F83005-5 and DZA045-5 was undertaken in order to determine if all genes composing the CBF/DREB1 cluster in A17 are similarly present in these accessions and to identify the polymorphism in the coding and immediate flanking sequences of these genes. Two to 3 specific forward and also 2 to 3 specific reverse primers were designed per gene. PCR were carried out using a touchdown protocol with the following thermal cycling conditions: 4 minutes at 94°C; 5 cycles at 94°C for 30 seconds, annealing temperature (Ta) [-1°C/cycle] for 30 seconds, 72°C for 90 seconds; 35 cycles at 94°C for 30 seconds, [Ta-5°C] for 30 seconds, 72°C for 90 seconds; and 10 minutes at 72°C. Different primer combinations and PCR conditions (Ta and/or MgCl 2 concentration) were tested to optimize the amplification result. PCR products from the optimal conditions were purified from agarose gel using Nucleospin Extract II (Macherey-Nagel, Düren, Germany) following manufacturer's instructions and Sanger-sequenced using BigDye Terminator v3.1 chemistry (Applied Biosystems, Foster City, California, USA) on a 3130×l Genetic Analyzer (Hitachi/Applied Biosystems). Primers and amplification conditions that were used in the sequencing step are provided in Additional file 10. For 9 out of the 12 CBF/DREB1 genes, one or two internal primers were needed to ensure full sequence coverage and/or overcome sequencing difficulties caused by simple sequence repeats [(A) n or (T) n ] present in 5′-or 3′-non-coding flanking regions (Additional file 10). Base calling was performed with Sequencing Analysis Software (version 5.4; Applied Biosystems). Electropherograms were manually trimmed of poor sequence data.

Availability of supporting data
The data sets supporting the results of this article are included within the article and its additional files. All sequences were submitted to GenBank [36]. Insert sequences of BAC clones mth2-92O15, mte1-60A22 and mth2-221P20 were deposited in the "high throughput genomic sequences" division under accession numbers KF006382-84. BAC-end sequences were submitted to the "genome survey sequences" division and have been assigned the accession numbers JY974377-472. CBF/DREB1 gene sequences from A17, F83005-5 and DZA045-5 were deposited in the "plant, fungal, and algal sequences" division with accession numbers KC997199-225.