Copy number variation at the HvCBF4–HvCBF2 genomic segment is a major component of frost resistance in barley

A family of CBF transcription factors plays a major role in reconfiguring the plant transcriptome in response to low-freezing temperature in temperate cereals. In barley, more than 13 HvCBF genes map coincident with the major QTL FR-H2 suggesting them as candidates to explain the function of the locus. Variation in copy number (CNV) of specific HvCBFs was assayed in a panel of 41 barley genotypes using RT-qPCR. Taking advantage of an accurate phenotyping that combined Fv/Fm and field survival, resistance-associated variants within FR-H2 were identified. Genotypes with an increased copy number of HvCBF4 and HvCBF2 (at least ten and eight copies, respectively) showed greater frost resistance. A CAPS marker able to distinguish the CBF2A, CBF2B and CBF2A/B forms was developed and showed that all the higher-ranking genotypes in term of resistance harbour only CBF2A, while other resistant winter genotypes harbour also CBF2B, although at a lower CNV. In addition to the major involvement of the HvCBF4-HvCBF2 genomic segment in the proximal cluster of CBF elements, a negative role of HvCBF3 in the distal cluster was identified. Multiple linear regression models taking into account allelic variation at FR-H1/VRN-H1 explained 0.434 and 0.550 (both at p < 0.001) of the phenotypic variation for Fv/Fm and field survival respectively, while no interaction effect between CNV at the HvCBFs and FR-H1/VRN-H1 was found. Altogether our data suggest a major involvement of the CBF genes located in the proximal cluster, with no apparent involvement of the central cluster contrary to what was reported for wheat.

transcriptome in response to low-freezing temperature in temperate cereals. In barley, more than 48 13 HvCBF genes map coincident with the major QTL FR-H2 suggesting them as candidates to 49 explain the function of the locus. Variation in copy number (CNV) of specific HvCBFs was 50 assayed in a panel of 41 barley genotypes using RT-qPCR. Taking advantage of an accurate 51 phenotyping that combined Fv/Fm and field survival, resistance-associated variants within FR-52 H2 were identified. Genotypes with an increased copy number of HvCBF4 and HvCBF2 (at 53 least 10 and 8 copies, respectively) showed greater frost resistance. A CAPS marker able to 54 distinguish the CBF2A, CBF2B and CBF2A/B forms was developed and showed that all the 55 higher-ranking genotypes in term of resistance harbour only CBF2A, while other resistant 56 winter genotypes harbour also CBF2B, although at a lower CNV. In addition to the major 57 involvement of the HvCBF4-HvCBF2 genomic segment in the proximal cluster of CBF 58 elements, a negative role of HvCBF3 in the distal cluster was identified. Multiple linear 59 regression models taking into account allelic variation at FR-H1/VRN-H1 explained 0.434 and 60 Introduction 73 74 Low freezing temperature is one of the primary limiting factors that affect yield potential of fall-75 sown crops such as barley (Hordeum vulgare L.) and wheat (Triticum aestivum L.). In 76 temperate cereals serious damages to plant cells and tissues are caused by intracellular or 77 extracellular ice formation during the freeze events (Levitt 1980), and cold winters with large 78 temperature fluctuations may result in severe winterkill. In view of possibly higher and faster 79 temperature variations, expectedly due to climate change, improving fast resilience to frost is 80 fundamental for minimizing the so-called yield gap (the gap between potential and actual yield), 81 increasing yield stability, and guaranteeing crop sustainability (Rizza et al. 2011;Visioni et al. 82 2013). During the last 10 years, barley has been extensively used as a model system for genetic 83 dissection of low temperature tolerance in fall-sown cereals (Pecchioni et al. 2012). There is 84 indeed abundant variation for this trait within the primary gene pool and an ever expanding set 85 of genomics tools is available, from high-resolution maps (Poland et al. 2012) to the structured 86 whole-genome context describing the barley gene-space (International Barley Genome 87 Sequencing Consortium 2012). As in other members of the Triticeae, barley shows also genetic 88 variation for growth habit, which is broadly classified as 'winter' or 'spring'. In general, winter 89 genotypes are low-temperature tolerant, vernalization requiring, and photoperiod sensitive; 90 whereas, spring ones have minimal low temperature tolerance capacity, do not require 91 vernalization, and are typically insensitive to short day photoperiod. These combinations of 92 characters have a recognized role in winterhardiness, i.e. the capacity of a plant to survive the 93 winter, and the genetic basis of each component trait has been extensively reviewed (Kosová et 94 al. 2008;Galiba et al. 2013). According to the epistatic interaction between the two major genes 95 that determine vernalization requirement in Triticeae, i.e. VRN-1 and VRN-2, a third growth 96 habit classification, 'facultative', has been proposed (Tranquilli and Dubcovsky 2000;von 97 Zitzewitz et al. 2005). Facultative barleys represent thus a subclass of the winter genotypes that 98 are low-temperature tolerant, have winter (recessive) alleles at VRN-H1, but lack the repressor 99 encoded by VRN-H2. 100 Phenotypic evaluations of plant survival after exposure to natural or artificial freezing events 101 display a wide range of ability to acclimate and cope with frost in small grain cereals (Limin 102 and Fowler 2006;Akar et al. 2009;Rizza et al. 2011). Freezing resistance in fact depends on a 103 period of cold acclimation that involves a highly integrated physiological response and in which 104 the expression of structural, regulatory, and developmental genes is finely modulated (for a 105 review, see Galiba et al. 2013). In the Triticeaeas in Arabidopsissuch transcriptome 106 reconfiguration acts through major "regulatory hubs" and relies on the action of CBF/DREB 107 (hereafter CBF) proteins, fundamental components of the frost resistance regulon (Thomashow 108 2010). The CBF genes are intron-less and encode AP2/ERF-type transcription factors that 109 recognize the CRT/DRE regulatory element in the promoters of many CBF-targeted COld Regulated (COR) and Dehydrin (DHN) genes. The corresponding COR and DHN proteins 111 operate as effector components of the stress response, and are involved in protecting the cell 112 against the harmful effects of frost (Thomashow 2010). Overexpression of specific CBFs from 113 wheat and barley in the susceptible barley cultivar 'Golden Promise' accelerates COR transcript 114 accumulation during cold acclimation and results in an improved frost resistance (Morran et al. 115 2011;Jeknić et al. 2014). The CBF gene family in temperate cereals is particularly large and 116 complex, with up to 25 different members classified into ten groups that share common 117 phylogenetic origin and similar structural features Badawi et al. 2007). 118 Another common characteristics of the genome of grasses is the organization of CBFs in 119 syntenic clusters of tandemly duplicated paralogs that belong to the phylogenetic subgroups 120 CBF3/CBFIII and CBF4/CBFIV (Badawi et al. 2007;Tondelli et al. 2011). In barley, 121 quantitative genetic studies have demonstrated that a large part of the observed phenotypic 122 variation in frost resistance is explained by two major loci (QTL), FR-H1 and FR-H2, located 123 approximately 25 cM apart on chromosome 5H (Francia et al. 2004;von Zitzewitz et al. 2011;124 Tondelli et al. 2014). FR-H1 overlaps the VRN-H1 vernalization response locus; FR-H2 125 segregates with a cluster of at least 13 CBFs, co-localizes with QTL influencing COR protein 126 accumulation, and correlates with mRNA levels of the CBF genes within the locus (Francia et 127 al. 2004(Francia et 127 al. , 2007Stockinger et al. 2007). Recent investigation of the signal transduction pathway 128 leading to cold acclimation in the frost tolerant cultivar 'Nure' indicates a diversity in regulation 129 between different CBF factors (Marozsán-Tóth et al. 2015). Such induction pattern specificity is 130 in accordance with phylogenetic relationships ) and suggests functional and 131 regulatory specialization of the CBF elements in addition to their structural separation at FR-H2. 132 High-resolution mapping in barley and diploid wheat was used by Francia et al. (2007) and 133 Miller et al. (2006) to identify informative recombinant events within FR-H2 and FR-A m 2 loci, 134 respectively. In diploid wheat (Triticum monococcum L.), Knox et al. (2008) resolved the 135 cluster of TmCBF genes into three clusters at FR-A m 2, referred to as proximal (containing 136 TmCBF2,4,9,17),central (TmCBF14,15,12) and distal (TmCBF16,13,3,10). Recently, 137 Pasquariello et al. (2014) physically mapped, sequenced and assembled ca. 1.5 Mb of sequence 138 that harbour the entire HvCBF cluster in the barley reference cultivar 'Morex'. Representing 139 more than 90% of the target genomic region, and with a general survey of Repetitive Elements, 140 this sequence can serve as a useful genomic tool for structural and evolutionary comparisons of 141 FR-H2 in germplasm accessions. In fact, even though mutations within individual CBF-coding 142 sequences were firstly proposed as causative of frost resistance variation in diploid wheat (Knox 143 et al. 2008), a differential expansion of the gene cluster together with copy number variation 144 (CNV), has been put forward as a structural explanation for the functional role played by the 145 CBFs at FR-2 in different genotypes. Evidences in favour of this hypothesis start to accumulate 146 in barley and hexaploid wheat Pearce et al. 2013;Jeknić et al. 2014;Zhu et 147 al. 2014).
Copy number variants (CNVs) are unbalanced changes in the genome structure and represent a 149 large category of genomic structural variants, which include by definition insertions, deletions 150 and tandem or interspersed duplications (Alkan et al. 2011). In the wake of the human genetics 151 research, plant structural variation has increasingly been recognized not only as a common 152 feature and evolutionary force of genomes but also as a genetic mechanism regulating relevant 153 adaptive traits (Zmieńko et al. 2013;Francia et al. 2015). Evidences of structural variation at the 154 FR-2 locus in Triticeae entail CNV of some CBF genes, however apparently involving different CBF-B14 and CBF-B15, is associated with reduced frost resistance in both tetraploid durum (T. 168 turgidum ssp. durum) and hexaploid (T. aestivum) bread wheat (Pearce et al. 2013). Increased 169 copy number of the central cluster has also been observed, and this was frequently associated 170 with haplotype variation at VRN-A1 and with higher TaCBF-A14 transcript levels (Zhu et al. 171 2014). Taken together, the expression and genetic data suggest that the central cluster plays a 172 critical role in frost resistance in wheats. 173 The recent availability of a complete FR-H2 genomic sequence in the barley reference genotype 174 'Morex' (Pasquariello et al. 2014) enables a thorough CNV survey at this locus. Several well-175 surrounding genomic landscape. Accordingly, our objectives were to (i) determine variation in 179 copy number of HvCBF genes at the FR-H2 locus using RT-qPCR, (ii) identify resistance-180 associated CNVs at specific clusters of HvCBF genes, (iii) unravel whether the effect of FR-H2 181 is linked to a single gene variation or to a sort of "CBF-number game", and its dependency from  Table 1). The majority of genotypes originated from 11 192 European countries while ten were from Australia (1), Syria (2), Turkey (5) and USA (2). 193 Cultivars identified in different field and laboratory experiments as highly tolerant to freezing 194 were included in the panel, moreover some of them were already characterized for adaptation to 195 Mediterranean environments (Rizza et al. 2011;Visioni et al. 2013). In addition, four 'Nure' 196 (resistant) x 'Tremois' (susceptible) recombinants were used. They represent a subset of F5:6 197 progeny derived from F2 recombinants within the HvCBF cluster used by Francia et al. (2007) 198 and Pasquariello et al. (2014) for fine mapping the FR-H2 genomic region. Two additional 199 Doubled Haploid lines -NT-42 and NT-64-from a 'Nure' x 'Tremois' DH population (Francia 200 et al. 2004), harbouring reciprocal alleles at FR-H2 and FR-H1/VRN-H1 QTLs were tested as 201 control genotypes. 202

DNA extraction 204
Leaves from two-week-old plants of each genotype were selected and manually ground in liquid 205 Nitrogen. 300-mg of grinded material was mixed with 860 µl of extraction buffer (10 mM Tris-206 HCl, pH 8.0, 150 mM NaCl, 2 mM EDTA, 1% SDS) and 100 µl of 5 M guanidinium chloride. 207 After incubation at 60°C for 3 h, samples were centrifuged for 10 min at 14,000 rpm. RNase (5 208 µl at 500 μg μl -1 ) was added to 500 µl of the supernatant, and the preparation was incubated at 209 37°C for 10 min to digest contaminant RNA. The extracted DNA was purified according to the 210 Wizard® Genomic DNA Purification Kit (Promega) and eluted with 55 µl of buffer (10 mM 211 Tris-HCl, pH 9.0). DNA concentration was measured at 260 and 280 nm using a Nanodrop ND-212 1000 spectrophotometer (Thermo Scientific).  Table 2). Eleven out of 13 primer pairs targeted HvCBF genes residing at FR-219 H2 (GenBank KF686739; Pasquariello et al. 2014). Primers were designed using the sequences 220 available in GenBank for the cultivars 'Morex', 'Nure' and 'Tremois' to enable amplification of 221 either A and B forms for CBF2, 4, 10 and 15. Primers for both CBF2A and CBF2B were not 222 able to amplify CBF2C form identified in 'Morex' (Pasquariello et al. 2014) due to significant 223 sequence differences. Two gene-specific amplicons were designed on HvNUD (GenBank AP009567) and HvSAMDC (GenBank AK252992) and were assumed to amplify control 225 (reference) segments typically known as single copy genes per haploid genome. The absence of 226 secondary structures in the region in which the primers anneal were verified via MFOLD 227 (http://frontend.bioinfo.rpi.edu/applications/mfold/cgi-bin/dnaform1.cgi) using the theoretical 228 melting temperature of primers and the appropriate Na + and Mg 2+ concentrations as in the used 229 SYBR Green kit. Each primer pair and the corresponding amplicon sequence were checked for 230 homology also against 'Morex' KF686739 to verify their specificity and uniqueness. In 231 addition, an empirical validation of the assays was performed using four-fold dilution series of 232 'Nure' genomic DNA. Using the direct method of calibration dilution curve and slope 233 calculation, real-time qPCR efficiencies were determined and data are summarized in 234 Supplementary Table 2. 235 RT-qPCR was carried out in a total of 25 μl consisting of 12.5 μl of SYBR Green PCR Master 236 Mix (QuantiFast SYBR Green PCR kit, Qiagen), 2 μl of both the forward and the reverse primer 237 (10 μM) (Sigma-Aldrich), 5 μl of template DNA solution, and 3.5 μl of water. PCR was 238 performed on a 7300 Real Time PCR System (Applied Biosystems) using the following cycling 239 protocol: 50°C for 2 min; 95°C for 10 min; and 40 cycles of 95°C for 15 s and 60°C for 1 min. 240 A dissociation curve was included to confirm amplification of single gene products. 241 For copy number calculation, quantification cycle (Cq) values were imported into qbasePLUS 242 v2.4 (http://www.qbaseplus.com). The 2 -ΔΔCq method was used to determine CNV at the target 243 gene assays by normalizing to the single-copy reference genes HvNUD and HvSAMDC. The

Statistical analysis 300
For investigating either simple association or cause-and-effect relationships between CNV at 301 barley CBFs and frost resistance traits, correlation and linear regression statistical tests were 302 applied. All data were analyzed through the software GenStat 17 th Edition (Payne 2014), while 303 charts and graphs were obtained with SygmaPlot v10 (Systat). Initially, correlations were 304 computed for any pair of quantitative variables (i.e. CNRQ values, Fv/Fm and field survival); 305 then, the "All-subsets regression" procedure was used to efficiently fit all possible regression 306 models explaining Fv/Fm and field survival data. Selection of the more effective CBF variable(s) 307 was based on the convergence of three different statistics: 1) the adjusted R-squared, 2) the 308  1,051,128-1,051,228). The assay for HvCBF13 hit several sequences other than the intended 336 target (data not shown) and was excluded from further analysis. According to the relative physical and genetic position of the HvCBF genes (Pasquariello et al. 2014), the designed 338 assays spanned an approximate distance of 1,089 kb. The inter-probe space between two 339 consecutive amplicons ranged in size from 14 to 438 kb (for the HvCBF4-HvCBF2 and 340

Mallows
HvCBF2-HvCBF14 intervals, respectively); while the average probe density was 97.9 kb. For 341 CNV calculation the copy number was fixed as 2 for 'Morex' (reference genotype) for all genes 342 apart from HvCBF6 were the copy number in 'Morex' was fixed as 4 as this primer pair 343 targeted both the gene and its pseudogene. Copy number relative quantity (CNRQ) data 344 revealed that HvCBF4 and HvCBF2 in the proximal cluster of Fr-H2 displayed extensive CNV, 345 with an average of 4.38 and 4.00 copies per diploid genome, respectively (Table 1). Such 346 variability was evident from the much higher coefficient of variation of the two genes (64% and 347 78%, respectively) compared to the other target genes (ranging from 15.2% for CBF15 to 28.0% 348 for HvCBF10). Conversely, the two reference genes HvNUD and HvSAMDC showed a 349 difference <0.5 in CNRQ mean (1.82 and 2.22, respectively) and the smallest coefficient of 350 variation (9.8% and 9.3%, respectively). If we consider winter/facultative and spring genotypes 351 separately, the copy number of both HvCBF4 and HvCBF2 was double in the former group, 352 namely 6.20 vs. 2.81 and 5.60 vs. 2.61 copies per diploid genome, respectively (Table 1). 353 Covariance between HvCBF4 and HvCBF2 is also graphically illustrated by Fig. 1, where a  354 positive relationship between the numbers of copies of the two genes was observed along with a 355 clear separation of the spring types from the winter and facultative ones. 356 Relationships among CNRQ values at the ten assayed HvCBFs were explored by pairwise 357 correlation analysis, and the strongest association (r = 0.89***) was once again observed 358 between HvCBF4 and HvCBF2 in the proximal cluster of FR-H2 locus. Other positive 359 association were evident, although at lower values of correlation coefficient. In particular, 360 HvCBF9 CNRQ correlated with the genes belonging to the central cluster (HvCBF15 and 361 HvCBF12) and to the distal cluster (HvCBF16, HvCBF3, HvCBF10 and HvCBF6), as well as 362 HvCBF12 copy number correlated with CNRQ of HvCBF15 and of all the genes in the distal 363 cluster (Supplementary Table 3). 364

CNV at specific HvCBF genes is associated with frost resistance 366
To identify resistance-associated CNVs at specific HvCBFs, relationships among phenotypic 367 data were firstly explored by correlation. Phenotypic records obtained from three consecutive 368 experiments of freezing survival in controlled conditions (Fv/Fm) were more interrelated than 369 field-laboratory assessments of winter survival (%) from three consecutive trials with mean r 370 coefficient 0.77 and 0.48, respectively. Frost resistance data were then averaged separately for 371 Fv/Fm and for field survival score, and used for investigating relationship of phenotypic values 372 to CNV effects of the single CBFs (Table 1). Both traits showed high positive correlation with 373 CNRQ variation at HvCBF4 and HvCBF2 in the proximal cluster (r ranging from 0.56*** to 374 0.42*). A lower yet significant negative correlation was observed between HvCBF3 CNV and frost resistance traits. No other statistical association between CNV at the assayed genes and 376 frost resistance was found (Table 1). The association between frost resistance and copy number 377 variation at HvCBF4 and HvCBF2 is well illustrated by the scatter plots in Fig. 2 and Fig. 3  378 respectively. In both cases, when linear regression fits were plotted separately for the 379 winter/facultative and spring accessions, genotypes with increased copy number of HvCBF4 380 and HvCBF2 similarly showed greater frost resistance in the two breeding groups.

Combination of HvCBF genes showing CNV at FR-H2 better explain frost resistance 403
With the final aim of unravelling whether the effect of FR-H2 is linked to CNV for a single 404 gene within the genomic region under study or to the involvement of multiple HvCBFs in the 405 cluster, single and multiple linear regression analysis models were applied. As a first step, 406 prediction of Fv/Fm and winter survival phenotypes through single linear regression confirmed 407 the major involvement of HvCBF4 and HvCBF2 and, to a lesser extent, of HvCBF3. The fitting 408 of the model also allowed to estimate the proportion of variance accounted for by either single 409 genes or by different combinations of copy number (i.e. sum and/or difference) of multiple 410 genes together. The effect of combined gene copy change at HvCBF4, -2, and -3 was generally 411 associated to greater values of adjusted R-squared rather than taking CNV at the single genes 412 separately (Supplementary Table 5). A cumulative effect of the proximal cluster was confirmed 413 in barley, while again no significant effect of the central and distal clusters was observed. 414 Owing to Stockinger et al. (2007), Akar et al. (2009), andZhu et al. (2014), a crosstalk between 415 vernalization and cold acclimation pathways through the interaction of FR-1/VRN-1 and FR-416 2/CBF loci should not be overlooked. Therefore, the allelic state at VRN-H1 was included as 417 covariate in the regression models by classifying the accessions in winter/facultative or spring 418 types and setting the 'Tremois' (spring, susceptible) allele as reference. As expected, a greater 419 explanatory power of the model was observed in all cases, and the effect of combined gene copy 420 change at HvCBF4, -2, and -3 was generally associated to larger values of adjusted R-squared 421 (0.403-0.414 and 0.371-0.445 for Fv/Fm and winter survival, respectively) rather than taking 422 CNV at the genes separately (Supplementary Table 5). 423 Generalized Linear Model regression analysis was further exploited using the "All subsets 424 regression" method, and this allowed identifying the best combinations of predictors explaining 425 the two frost resistance traits. The allelic variation at FR-H1/VRN-H1 was also included in the 426 final multiple linear regression model with groups (gMLR), but no significant interaction effect 427 between CNV at the HvCBFs and FR-H1/VRN-H1 were found (data not shown). For Fv/Fm, the 428 combined copy number of HvCBF4 and HvCBF2 (i.e. SUM4+2) together with variation at 429 HvCBF3 in winter/facultative or spring genotypes could explain up to 0.434*** of the variance 430 accounted for by the fitted model (Table 2). Likewise, winter survival was better explained by 431 fitting copy number variation at HvCBF4 and HvCBF3, and this resulted in an adjusted R-432 squared of 0.550*** (Table 3). Overall, this data showed that genotypes more tolerant to frost 433 carried a higher copy number of the HvCBF4-HvCBF2 segment in combination with a lower 434 CNV at HvCBF3, and that roughly 50% of the trait could be explained after the VRN-H1 allelic 435 effect is fixed in the model. 436 During recent years, high-resolution mapping in barley was used by Francia et al. (2007) and 437 Pasquariello et al. (2014) to identify informative recombinants among HvCBF gene elements at 438 FR-H2 in a 'Nure' x 'Tremois' (NxT) experimental population. Accordingly, in this work, 439 selected F5:6 lines were tested to obtain an independent proof of the CNV involvement in frost 440 resistance observed in the barley collection. The recombinants were phenotyped both for the 441 photochemical capacity of photosystem II (Fv/Fm parameter) after freezing and for winter 442 survival in the field, and compared to the parents and to two NxT doubled haploid lines 443 harbouring alternative alleles at FR-H1/VRN-H1 and FR-H2/CBF loci (Supplementary Fig. 1). 444 Results of restricted maximum likelihood (REML) ANOVA agreed with multiple linear 445 regressions. In particular, a recombinant line (Rec 15-12) carrying HvCBF4 and HvCBF2 from 446 'Nure' in combination with HvCBF3 from 'Tremois' was more resistant in terms of Fv/Fm than 447 the three recombinants harbouring reciprocal alleles at the same loci. An equivalent trend was 448 observed in a one-year field trial experiment. Overall, these results confirmed the major role 449 played by the genes of the proximal cluster combined with a minor effect of a single element of 450 the distal cluster, while no effect was observed for the HvCBFs of the central cluster. investigations. In crops a causal relationship between copy number and phenotypic variation 468 linked to single genomic regions has been firstly demonstrated for the barley Bot1 boron 469 transporter gene (Sutton et al. 2007). Later on, the study of the soybean Rhg1 cyst nematode 470 resistance locus (Cook et al. 2012), of the maize efflux transporter MATE1 (Maron et al. 2013), 471 and the barley flowering time determinant FT1 (Nitcher et al. 2013) showed that different 472 phenotypes in evolutionary distant plant species could be ruled by CNV gene dosage effects. A 473 similar phenomenon was also hypothesized for the FR-H2 locus by ). This 474 QTL in the Triticeae defines a region on the homeologous chromosome group 5 that harbour 475 several duplicated CBF genes and in barley reference cultivar 'Morex' the region entails a 476 genomic segment of ca. 1.5 Mb Pearce et al. 2013;Pasquariello et al. 2014). 477 The physical structure of the locus suggests that this gene cluster is subject to dynamic 478 expansions and contractions potentially entailing for a more flexible response to cold stress 479 (Zhu et al. 2014). Accordingly, in the present study this was investigated through a CNV survey 480 in 41 barley varieties representing a broad genetic basis (Supplementary Table 1). 481 Although nowadays the methods of choice for genome-wide identification of CNVs are those 482 based on high-throughput platforms (Li and Olivier 2013), for targeted locus studies a RT-483 qPCR could be preferred as amenable for analysis of a large number of samples and no need of 484 post-amplification labor processing. RT-qPCR presents also some constraints that should be 485 taken into consideration, such as a relative analysis method requiring a control with 2 copies of 486 measuring 1.3) making interpretation more difficult (Ma and Chung 2014). Here RT-qPCR was 488 adopted to determine how HvCBF gene copy number varies at FR-H2 and, as expected, allowed 489 a thorough investigation of the copy number relative quantity of the different CBF genes (Table  490 1). No CNV was observed for the majority of the investigated genes, whereas HvCBF4 and 491 HvCBF2 in the proximal cluster of FR-H2 displayed extensive CNV (Fig. 2 and Fig. 3). The 492 values obtained for CBF10 should putatively attributed to the presence of CBF10A and CBF10B 493 forms in the tested collection as already suggested by Knox et al. (2010) who compared 494 'Morex' and 'Nure' lambda phage genomic sequences. On the other hand, in case of CBF6 it 495 was not possible to design a primer pair that would no amplify the pseudogene due to the 496 sequence identity. In fact the observed copy number was double respect to the reference genes 497 (CNRQ 4.01 in Table 1). 498 Knox et al. (2010) stated that the genotypes harbouring the vrn-H1 winter allele display two 499 distinct CBF2 paralogs (2A and 2B), while genotypes carrying the Vrn-H1 spring allele have 500 single copies of CBF2 and CBF4 paralogs. However, in the present work the results obtained 501 using a CAPS protocol for the whole collection revealed that not all the winter accessions do 502 harbour the CBF2B form identified in 'Nure' and 'Dicktoo' . Two winter 503 varieties, 'Cetin' and 'Bazant', have only the CBF2A form associated with a high degree of 504 CNV (7.6 in Supplementary Table 4), while CBF2B was observed together with CBF2A also 505 among the spring accessions ('Angela', 'Mellori', 'Ponente', 'Aldebaran', 'Tidone') with a 506 lower average CNV value of 5.6. This finding shows that, contrary to what stated by Knox et al. 507 (2010), the presence of CBF2B form is not able to discriminate winter from spring barleys. As a 508 general rule however it could be confirmed that the number of copies of both HvCBF4 and 509 HvCBF2 was double in winter/facultative vs. spring accessions, namely 6.20 vs. 2.81 and 5.60 510 vs. 2.61 copies per diploid genome, respectively (Table 1). 511 Several studies have revealed that the cereal CBF family is large and complex with at least 17 512 CBFs in barley Francia et al. 2007) and 13 in diploid einkorn wheat (Knox 513 et al. 2008). An even more complex organization of the family is observed in the hexaploid 514 wheat genome, which contains at least 65 CBF genes, 27 of which are paralogs with 1-3 515 homeologous copies for the A, B and D genomes (Mohseni et al. 2012). In addition, in the 516 Triticeae inter-and intra-specific variability has been observed in terms of presence/absence of 517 specific CBF paralogs that reside at FR-2. While some genes are conserved, some others seem 518 to be species-specific. For example, CBF17 has never been isolated in barley and is not present 519 in the 'Morex' reference genome as well as orthologs of barley and T. monococcum CBF13, 520 CBF16 have yet to be identified in rye and bread wheat (Skinner et al. 2006;Campoli et al. 521 2009;Pearce et al. 2013;Pasquariello et al. 2014). This large diversity in Triticeae seems to 522 represent an evolutionary adaptation to temperate climate habitats with a wide range of 523 temperature changes. The present study suggests that also as far as the CNV within the FR-2 524 locus is regarded, different clusters or subgroups of CBF paralogs are involved in barley and wheat. In wheat, the copy number of CBF-A14, CBF-A15 and CBF-A12 was found to be higher 526 in winter than in spring wheats revealing a crucial role of the central cluster (Dhillon and 527 Stockinger 2013;Zhu et al. 2014). Here the availability of the 'Morex' reference genome 528 allowed to obtain a much more comprehensive picture of the locus than the one obtained in 529 wheat by Zhu et al. (2014) where only some TaCBFs were investigated and showed that none of 530 those CBFs was affected by CNV in barley (Table 1). On the other hand, an increase in copy 531 number of two genes belonging to the proximal cluster reported by Knox et al. (2010) was fully 532 confirmed highlighting a very high number of copies in winter/facultative over spring genotypes 533 (Fig. 1). 534 Accurate phenotyping is an essential step for studying the implications of gene/genomic CNV in 535 the genetic architecture of underlying complex traits. Frost resistance data showed high positive 536 correlation with CNRQ variation at HvCBF4 and HvCBF2. When linear regressions fits were 537 plotted separately for the winter/facultative and spring accessions, genotypes harbouring more 538 copies of HvCBF4 and HvCBF2 showed greater frost resistance ( Fig. 2 and Fig. 3). In 539 particular, two winter ('Bazant' and 'Cetin') and three facultative ('Lunet', 'Pamina' and 540 'U259') genotypes that were highly resistant, ranked in the five top positions at both genes for 541 CNRQ. All of them harboured at least 10 copies of HvCBF4 and at least 8 copies of HvCBF2 542 per diploid genome. Interestingly, these cultivars originated from different countries and did not 543 share any common ancestor (Supplementary Table 1 Table 4). 547 Noteworthy, all facultative (including the highly resistant genotypes 'Lunet', 'Pamina' and 548 'U259') and the best ranking winter genotypes ('Bazant' and 'Cetin') harboured only the 549 CBF2A form, while other resistant winter genotypes harbouring also the CBF2B form had lower 550 CNV (4−6). As reported by Rizza et al. (2011) the groups of winter and facultative genotypes 551 showed the highest frost resistance when evaluated in standard freezing stress (e.g. −12 or 552 −14°C) in controlled conditions. Moreover, the interesting association of copy number of 553 HvCBF4 and HvCBF2, physically tightly linked in a genomic region of only ca. 14 kb (Knox et 554 al. 2010;Pasquariello et al. 2014), and the presence of CBF2A form was highlighted in the 555 present work thus extending the molecular understanding of the phenomenon. 556 One of the working hypotheses of the present paper was that the frost resistance effect ruled by 557 FR-H2 could depend on a sort of "CBF-number game" rather than to a genetic variation at 558 single gene element. Having obtained a global view of the locus highlighted not only interesting 559 relationships between paralogs within the proximal cluster, but also the involvement of the 560 distal one through the action of HvCBF3. This combined effect of gene copy change involving 561 HvCBF4, -2, and -3 was generally associated to greater values of adjusted R-squared rather than 562 taking CNV at the single genes separately (Table 2 and Table 3). This result was confirmed using recombinants of a 'Nure' x 'Tremois' experimental population. The line carrying the 564 HvCBF4-HvCBF2 segment from 'Nure' in combination with HvCBF3 from 'Tremois' was 565 significantly more resistant than recombinants harbouring reciprocal alleles at the same loci 566 ( Supplementary Fig. 1). Noteworthy, self-regulated feedback loops for CBFs action, as already 567 reported in Arabidopsis (Novillo et al. 2011), could be involved also in Triticeae. In the present 568 study this is suggested by the negative influence of HvCBF3 on frost resistance. No 569 involvement of the central cluster was detected, contrary to what was observed by Zhu et al. 570 (2014) for wheat. 571 Stockinger et al. (2007) hypothesized that VRN-H1/FR-H1 is a regulatory locus affecting the 572 expression of multiple CBFs at the chromosomally linked FR-H2 locus. Modulation of CBF 573 transcription at the promoter level exercised by the MADS box transcription factor HvBM5 574 (VRN-H1) has been recently hypothesized by Deng et al. (2015), in accordance with reduced 575 expression of CBF2, -4 and -9 during low-temperature treatment in lines with elevated VRN-1 576 activity ). These differences had been proposed to putatively depend either 577 on a different modulation at promoter level between 'Nure' and 'Tremois' CBF2 and CBF4 578 alleles, or a greater stability of the CBF2 and CBF4 mRNAs from the 'Nure' allele (Stockinger 579 et al. 2007). All the above findings suggested a qualitative link between frost resistance and 580 allelic state variation at VRN-H1/FR-H1 and a quantitative connection to the expression of CBF 581 genes. On the other hand, in the present study the relationship between CNV at FR-H2 and the 582 allelic state at VRN-H1/FR-H1 was investigated for the first time in barley. When the allelic 583 state at VRN-H1 was included as covariate in the regression models by classifying the 584 accessions in winter/facultative or spring types and setting the 'Tremois' (spring, susceptible) 585 allele as reference, the effect of combined gene copy change at HvCBF4, -2, and -3 was 586 generally higher than taking CNV at the genes separately (Table 2 and Table 3). Overall, our data show that genotypes more tolerant to frost carry higher copies of the 591 HvCBF4-HvCBF2 segment in combination with lower copies of HvCBF3, and that increased 592 copy number of CBF2A was observed in highly resistant facultative and winter genotypes. 593 Roughly 50% of the trait could be explained after the FR-H1/VRN-H1 allelic state is fixed in the 594 model. However when the allelic variation at FR-H1/VRN-H1 was also included in multiple 595 linear regression with groups, no significant interaction effect between CNV at the HvCBFs and 596 FR-H1/VRN-H1 was found, contrary to what was reported in both winter and spring wheat 597 (Zhu et al. 2014). The results obtained in the present study suggest the additive effect of those 598 two loci, as firstly observed in barley by Francia et al. (2004), where no interaction effects were 599 found between FR-H1 and FR-H2. Taken together the data show a major involvement of the 600 genes located in the proximal cluster, with no apparent involvement of the central cluster, confirming that the combinatorial game of HvCBFs at FR-H2 allows a more flexible response to 602 cold stress and suggest this gene cluster has shared ancestry generated by multiple duplication 603 events and subjected to dynamic expansions/contractions. It might be hypothesized that 604 duplication of different CBFs and putatively subsequent association of different clusters with 605 frost resistance took place in barley and wheat during their evolution preserving however the 606 functional role of the FR-2 locus as a whole. The present work could represent a useful example 607 of targeted identification of specific genetic determinants of stress resistance that might serve as 608 potential candidates for precise introgression of superior alleles into breeding programmes. 609 a : CNRQ is the copy number relative quantity; Range is maximum -minimum values; Lower Q (quartile) is the value l such that 25% of a sample are less than l, Upper Q (quartile) is the value u such that 25% of a sample are greater than u; CV% is coefficient of variation (Std dev/Mean x 100). b : For each gene Student's t-test (p<0.05) was used to compare mean value of two sets of data; only significant differences are indicated. c : r coefficient is calculated on 38 and 26 observations for Fv/Fm and winter survival data, respectively; *: p < 0.05, **: p < 0.01, ***: p < 0.001.