A genome-wide analysis of MADS-box genes in peach [Prunus persica (L.) Batsch]

MADS-box genes encode a family of eukaryotic transcription factors distinguished by the presence of a highly-conserved ~58 amino acid DNA-binding and dimerization domain (the MADS-box). The central role played by MADS-box genes in peach endodormancy regulation led us to examine this large gene family in more detail. We identified the locations and sequences of 79 MADS-box genes in peach, separated them into established subfamilies, and broadly surveyed their tissue-specific and dormancy-induced expression patterns using next-generation sequencing. We then focused on the dormancy-related SVP/AGL24 and FLC subfamilies, comparing their numbers and phylogenetic relationships with those of other sequenced woody perennial genomes. We identified 79 MADS-box genes distributed across all eight peach chromosomes and frequently located in clusters of two or more genes. They encode proteins with a mean length of 248 ± 72 amino acids and include representatives from most of the thirteen Type II (MIKC) subfamilies, as well as members of the Type I Mα, Mβ, and Mγ subfamilies. Most Type I genes were present in species-specific monophyletic lineages, and their expression in the peach sporophyte was low or absent. Most Type II genes had Arabidopsis orthologs and were expressed at much higher levels throughout vegetative and fruit tissues. During short-day-induced growth cessation, seven Type II genes from the SVP/AGL24, AGL17, and SEP subfamilies showed significant changes in expression. Phylogenetic analyses indicated that multiple, independent expansions have taken place within the SVP/AGL24 and FLC lineages in woody perennial species. Most Type I genes appear to have arisen through tandem duplications after the divergence of the Arabidopsis and peach lineages, whereas Type II genes appear to have increased following whole genome duplication events. An exception to the latter rule occurs in the FLC and SVP/AGL24 Type II subfamilies, in which species-specific tandem duplicates have been retained in a number of perennial species. These subfamilies comprise part of a genetic toolkit that regulates endodormancy transitions, but phylogenetic and expression data suggest that individual orthologs may not function identically across all species.


Background
Seasonal dormancy is an endogenous repression of meristematic growth exhibited by many perennial plants during the cold winter months. Endodormancy entrance and release are triggered by day length and/or temperature cues using a regulatory network that shares key features with the vernalization and photoperiodic flowering time pathways of Arabidopsis [1]. Nonetheless, precise mechanisms of endodormancy regulation in woody plants have not been characterized.
The peach evergrowing (evg) mutant has lost six tandem-duplicated dormancy-associated MADS-box (DAM) genes and does not form terminal buds or enter endodormancy under short day conditions [2]. The DAM genes are most closely related to Arabidopsis SVP and AGL24, both of which are involved in vernalization and flowering time regulation [1]. In peach, DAM gene expression tracks seasonal light and temperature cycles, and we have hypothesized that DAM genes integrate environmental cues to regulate the transition into and out of endodormancy [3].
The central role played by MADS-box genes in peach dormancy regulation has led us to examine this large gene family in more detail. MADS-box genes encode a family of eukaryotic transcription factors distinguished by the presence of a highly-conserved~58 amino acid DNA-binding and dimerization domain at the Nterminal (the MADS-box) [11]. In plants, MADS-box genes are best known as master regulators of flowering time and floral organ development, although they also function in the development of leaves, roots, fruit, seeds and gametophytes [12,13]. Members of the MADS-box gene family are found throughout higher eukaryotes and are divided into two classes, Type I and Type II, which arose from a single gene duplication before the divergence of plants and animals [14]. Type I genes are characterized by the presence of the MADS-box and by a simple intron-exon structure, while Type II genes possess additional conserved domains and a more complex gene structure [15,16].
In plants, Type II genes are termed MIKC (MADS Intervening Keratin-like C-terminal) genes in reference to the four recognized domains of their protein products. In addition to the MADS-box, MIKC proteins possess an intervening I domain (~30 aa) that contributes to dimerization specificity, a highlyconserved keratin-like K domain (~70 aa) that facilitates dimerization, and a variable C-terminal domain that plays a role in transcriptional activation and the formation of multimeric complexes [16]. MIKC genes are further divided into MIKC c and MIKC* classes, with the latter exhibiting an ancestral duplication within the K domain [17].
MIKC c genes are the best-studied plant MADS-box genes and have been divided into at least 13 subfamilies based on sequence similarity [18]. Several subfamilies form the basis for the ABCDE model of floral organogenesis, in which specific combinations of genes from the AP1, AP3/PI, AG, FUL and SEP subfamilies give rise to sepals, petals, stamens, carpels and ovules in Arabidopsis thaliana [19]. A subset of MIKC c genes from the FLC, SOC1 and SVP/AGL24 subfamilies control vernalization and flowering time in response to seasonal light and temperature cues in annual plants [20,21]. Genes from the FLC and SVP/AGL24 subfamilies also appear to regulate endodormancy transitions in perennial plants, using pathways that share significant features with those of vernalization [1,4,22]. In contrast to MIKC c genes, the functions of Type I and MIKC* genes are poorly understood. Recent work suggests that Type I genes are chiefly expressed in the female gametophyte and the developing seed of Arabidopsis [23]. Expression levels are often quite low, and there is evidence for considerable functional redundancy. MIKC* genes appear to function primarily in the Arabidopsis male gametophyte, where they control the expression of genes required for pollen maturity [24].
Here we present a genome-wide analysis of Type I and II MADS-box genes in peach, made possible by the availability of the peach genome sequence (Peach v1.0; [25]). We report the locations and sequences of Type I and II MADS-box genes in peach, separate them into established subfamilies, and broadly survey their tissue expression patterns. We then focus on the SVP/AGL24 and FLC subfamilies, comparing their numbers and phylogenetic relationships with those of other perennial species and quantifying their expression during the transition to endodormancy in peach. In particular, we test the hypotheses that (1) a similar expansion within the SVP/AGL24 subfamily has occurred in multiple perennial plant species and (2) genes from the SVP/AGL24 and FLC subfamilies are differentially expressed during the short-day dormancy transition in peach.

Sequence collection
Peach genome scaffolds, predicted peptides and ESTs were obtained from the Genome Database for Rosaceae (http://www.rosaceae.org/species/prunus_persica/genome_ v1.0, [25] [28], and Arora et al. 2007 [29], respectively. An exception occurred with the FLC genes from P. trichocarpa, which were incompletely annotated in the Populus v3.0 genome build. These sequences were curated manually and named according to the transcript ID containing their MADS box. Our revised Populus FLC protein sequences are given in Additional file 1.

Identification and annotation of peach MADS-box genes
The HMMER-3.0 software package [30] was used to build profile hidden Markov models from full Pfam alignment files for the MADS-box (SRF-TF PF00319) and K-box domains (K-box PF01486). Resulting models were used to search the database of predicted peach peptides and identify potential MADS-box proteins (E-value threshold 1 × e −10 , with manual inspection of sequences close to the threshold). The full peach genomic scaffolds were also queried with nucleic acid sequences from representative Arabidopsis and Vitis MADS-box genes using NCBI BLAST tools [31] to identify putative MADS-box genes not present in the predicted protein set.
A 15 kb region around each peach MADS-box was extracted, and the full gene structure was predicted using the FgenesH (Softberry, Inc., Mount Kisco, NY), Augustus [32] and SNAP [33] gene prediction programs within the DNA Subway annotation pipeline (http://dnasubway. iplantcollaborative.org/). Predicted models were refined by manual inspection and comparison with homologous Arabidopsis sequences and peach ESTs. Positions of MADS-box genes on peach genome scaffolds were visualized with MapChart software [34] and are provided as a gff3 file in Additional file 2.

Phylogenetic analyses
An initial phylogenetic analysis was performed to separate the peach MADS-box genes into Type I and Type II lineages. Fifty-eight amino acids from the MADS-box domain of each Arabidopsis and peach gene were aligned with Clustal W [35] and used to create a maximum likelihood phylogenetic tree in PhyML 3.0 [36]. Positions of MADS-box genes on the resulting tree classified them unambiguously as Type I or II, and these assignments were verified by confirming the presence of a K-box in the MIKC genes only.
Protein sequences of MIKC genes from peach and Arabidopsis were aligned with MAFFT v7 [37], and a phylogenetic analysis was performed with MrBayes v3.2 using the Jones amino acid substitution model [38]. Two independent runs with four Markov Chain Monte Carlo chains were run for 10 million generations and sampled every 1000 generations to achieve convergence (standard deviation of split frequencies < 0.02). After dropping the first 25% of the sampled trees as burn-in, results were visualized as a consensus tree with posterior probabilities indicated at each node. Trees were constructed in the same manner to partition Type I genes among Mα, Mβ, and Mγ clades and to analyze the relationships among genes from the FLC and SVP/AGL24 subfamilies across multiple species.
Tissue-specific expression analyses 75 base-pair paired-end Illumina RNAseq reads (llumina Inc., San Diego, CA) from root, expanded leaf, young apical leaf, fruit, pollen and cotyledon + embryo tissues were obtained as described in Verde et al. 2013 [25] and are available for download from the NCBI Sequence Read Archive (SRA053230). Reads were quality-trimmed using the default settings of ConDeTri [39] prior to read mapping and transcript quantification with the Cufflinks pipeline (Bowtie 1.0.0, TopHat 2.0.9, Cufflinks 2.1.0) and the peach v1.0 reference genome [25,40]. Estimated depth of transcriptome coverage was high but differed among the read sets. After filtering and trimming, the root, expanded leaf, young leaf, fruit, pollen and cotyledon + embryo read sets provided approximately 108X, 100X, 171X, 102X, 135X, and 67X coverage of the peach transcriptome, respectively. Reads from each tissue were mapped and quantified separately, using a gff3 file of peach MADS-box gene models as a reference and without assembly of additional transcripts (−G option in Cufflinks).
Resulting expression values (FPKM, i.e. fragments per kilobase of exon model per million mapped fragments) were log-transformed and used in an average linkage clustering analysis with Cluster 2.11 and TreeView 1.6 in order to visualize tissue-specific gene expression patterns [41]. All expression data are provided in Additional file 3.

Short-day expression analyses
Rooted peach cuttings were grown in a greenhouse for two months at 25°C under long days (LD, 16 h light/8 h dark). Cuttings were derived from wild type individuals in the F 2 population described in Jimenez et al. 2010 [9]. Plants were transferred to a growth room for two weeks of acclimation under LD, then shifted to SD conditions (8 h light/16 h dark) for two weeks. In the growth room, 250-300 μmol m −2 s −1 of light was provided at canopy height by AgroSun® Gold 1000 W sodium/halide lamps (Agrosun Inc, New York, NY, USA). Temperatures averaged 22.5°C (light) to 18.7°C (dark), and relative humidity ranged between 48% and 55%. Plants were watered every two days as needed.
At 0, 1, and 2 weeks after the transfer to SD, apical tips (youngest leaves and shoot apical meristems) from eight replicate plants per week were harvested and pooled for RNA extraction [42]. Following quantification and quality assessment on the Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara CA), 10 μg of ethanol-precipitated total RNA from each pooled sample was shipped to the Iowa State University DNA Facility for library preparation and 75 bp single-end sequencing on the Illumina Genome Analyzer II platform. Resulting sequence data were quality-filtered and trimmed as above prior to transcript assembly and quantification with the Cufflinks pipeline and average linkage clustering with Cluster and TreeView. Genes whose expression levels changed significantly through time were identified using the Audic and Claverie statistic implemented in IDEG6 with P <0.05 and a Bonferroni correction for multiple comparisons [43,44]. All expression data are provided in Additional file 3, and raw reads are available at the NCBI Sequence Read Archive (SRP046357).

MADS-box genes in peach
We used profile hidden Markov models to identify the positions and sequences of 79 MADS-box genes in the peach genome: 40 Type I and 39 Type II. Thirteen of these genes have been described previously, and two additional genes match peach ESTs available at NCBI (Additional file 4). They encode predicted proteins with a mean length of 248 ± 72 amino acids and include representatives from most Type II (MIKC) subfamilies, as well as members of the Type I Mα, Mβ, and Mγ subfamilies. Also identified were four probable pseudogenes with premature stop codons within the first two exons. These genes (PpeMADS02, PpeMADS05, PpeMADS68, and PpeMADS72) were dropped from further analysis. The majority of Type I genes had a single exon, while Type II genes had between 7 and 9 exons.

Chromosome positions
MADS-box genes are distributed across all eight chromosomes of peach ( Figure 1). Sixty percent of the peach MADS-box genes are clustered, i.e. present in groups of two or more genes separated by fewer than 200 kb [45]. The extent of clustering is particularly high in the Type I Mβ and Mγ subfamilies, 87.5% and 84.6% of whose genes are clustered. Clusters generally consist of close paralogs, but this is not always the case. PpeMADS66 (Mβ) and PpeMADS08 (MIKC c FLC-like) are located within 59 kb of one another on chromosome 3, while PpeMADS16 (Mα) is located within 86 kb of two tandem duplicated Mγs (PpeMADS73 and 74) on chromosome 7.
Several closely-adjacent pairs of distantly-related MADSbox genes are found multiple times in syntenic regions of the peach genome. There are three occurrences of a SEPlike gene located within 4 to 11 kb of a AP1/FUL-like gene within syntenic regions: PpeMADS18 and PpeMADS19 on chromosome 1, PpeMADS09 and PpeMADS10 on chromosome 3, and PpeMADS37 and PpeMADS38 on chromosome 5. Likewise, a SOC1 and an AGL6 homolog (PpeMADS22 and PpeMADS23, PpeMADS60 and PpeMADS61) are closely adjacent to one another on opposite strands at two positions on duplicated portions of chromosome 2. Such patterns have been reported previously [46] and suggest an ancient tandem duplication, followed by retention of the resulting paralogs and later duplication of the gene pair by polyploidization.

MADS-box protein phylogenies
Unrooted phylogenetic trees were constructed from full length protein sequences of Type I and Type II MADsbox genes of Arabidopsis and peach (Figures 2 and 3). Type I genes from both species grouped into the previously-identified Mα, Mβ and Mγ subfamilies with moderate support. While most Type I genes were present in species-specific monophyletic lineages, a small number of Arabidopsis Type I genes did have close peach orthologs. For example, the central cellexpressed Mα AGL61 (DIA) has two peach orthologs (PpeMADS29 and PpeMADS43), while its Mγ interaction partner AGL80 has five peach orthologs (Ppe-MADS06, PpeMADS07, PpeMADS42, PpeMADS48 and PpeMADS76).
Type II genes grouped into MIKC* and MIKC c clades, the latter containing members from 12 established subfamilies ( Figure 3; [18]). The majority of Type II subfamilies contained similar numbers of genes in Arabidopsis and peach. Exceptions occurred in the two subfamilies that play a pivotal role in Arabidopsis vernalization and flowering time: SVP/AGL24 and FLC. In Arabidopsis, the SVP/AGL24 subfamily contains only the two eponymous genes. In peach, the family is expanded to eight genes: the six DAM genes (AGL24 orthologs), PpeMADS57 (an SVP ortholog), and PpeMADS58, which has no Arabidopsis ortholog. Conversely, the FLC subfamily contains six members in Arabidopsis (FLC and MAF1-5) but only a single member in peach (PpeMADS08). To further investigate gene numbers and relationships within the SVP/AGL24 and FLC subfamilies, we created phylogenetic trees of SVP/AGL24 and FLC proteins from seven plant species with sequenced genomes and fullycatalogued MIKC c genes: Arabidopsis [26], peach, poplar [27], grapevine [18], maize [28], sorghum [28] and rice [29]. It is clear that multiple independent expansions have occurred within the SVP/AGL24 subfamily over the course of eudicot evolution (Figure 4). While the peach DAM gene expansion (PpeMADS49-54) occurred within the AGL24 lineage, expansions in poplar and grapevine have taken place in a separate lineage that contains Ppe-MADS58 and no Arabidopsis members. Monocot SVP/ AGL24 homologs form a completely separate lineage with 2-3 members per species.
The FLC subfamily is expanded in Arabidopsis by the presence of the 5 MAF genes, which have no orthologs in any other species examined ( Figure 5). The FLC subfamily contains two to three members in monocots, one in peach, two in grapevine and six in poplar. The single peach FLC-like gene (PpeMADS08) belongs to a lineage separate from that of Arabidopsis FLC and the MAFs, while five FLC-like genes from poplar form a species-specific clade. Expansions of the FLC gene family in Arabidopsis and poplar are clearly the result of separate evolutionary events.
Peach contains a single member (PpeMADS35) of the TM8 subfamily, a group of floral development genes present in many eudicots but lost in Arabidopsis [47]. Like many other eudicots, peach also has third member of the AP3/PI subfamily. Peach does not appear to contain members of the Bsister subfamily, represented by GOA and TT16 in Arabidopsis.

Tissue-specific gene expression
RNA-seq data were used to quantify the expression MADS-box genes in six peach tissues ( Figure 6). Expression of Type I genes was generally low or absent. Among 40 Type I genes, 14 showed no expression and only six were expressed at levels higher than 2 FPKM in any tissue. A notable exception to this pattern was Ppe-MADS27, an Mα gene detected at moderate levels in all tissues (  fruits (3.6 FPKM). Several other genes showed low-level expression across multiple tissues (e.g. PpeMADS06, PpeMADS31 and PpeMADS78). It should be noted that we did not specifically sample female gametophyte tissue, the location of most Type I gene expression in Arabidopsis.
In contrast to the extremely low expression of Type I MADS-box genes (0.4 FPKM averaged over all genes and tissues), expression of Type II genes was markedly higher (8.9 FPKM averaged over all genes and tissues). Only PpeMADS01 (MIKC*), PpeMADS04 (AGL17) and PpeMADS62 (AGL6) showed no expression in any tissue examined. The greatest number of Type II MADS-box genes was observed in roots (32 genes), followed by young leaves (30), fruit (27), expanded leaves (26), pollen (23), and cotyledon/embryo tissue (17).
We used average linkage clustering to group Type II genes by their tissue-specific expression patterns. A group of genes containing SEP and AG subfamily members was expressed almost exclusively in fruits, while a group of four SVP/AGL24-like genes constituted the most highly-expressed genes in cotyledon + embryo tissue. FLC, SOC1, SVP/AGL24 and AP1/FUL family members were highly expressed in leaves and roots. Genes with root-only expression included the AGL17 subfamily members PpeMADS59 and PpeMADS47, as well as the AGL12 subfamily member PpeMADS46. As expected, expression of the MIKC* genes was restricted mainly to pollen, as was expression of AGL15 and PI orthologs. Floral tissue was not represented in our RNA-seq read sets, precluding analysis of ABCDE-type floral homeotic gene expression in peach flowers.
Nonetheless, genes from each of the ABCDE gene categories were expressed in multiple peach tissues.

Gene expression during the short-day transition
In a second RNA-seq experiment, we quantified MADSbox gene expression in shoot apices before and after the transition to short day dormancy-inducing conditions (Figure 7). Seven Type II genes exhibited significant expression changes in the two weeks following the shortday transition, indicating that these genes may regulate the earliest stages of growth cessation, terminal bud set and endodormancy establishment.
The SVP ortholog PpeMADS57 was strongly downregulated, as was the SEP family member PpeMADS09. PpeMADS49 (DAM5) was down-regulated at week one and returned to its baseline by week two. Three additional DAM genes (PpeMADS51 [DAM3], PpeMADS52 [DAM6] and PpeMADS53 [DAM2]) were significantly up-regulated, and a similar trend was observed for Ppe-MADS50 (DAM4) and PpeMADS54 (DAM1). Among the DAM genes, the greatest magnitude of response was observed in PpeMADS51 (DAM3), whose expression increased 45-fold over the two-week interval. Expression of PpeMADS04 from the AGL17 subfamily also increased significantly from 0 to 137.15 FPKM during this time. The FLC subfamily member PpeMADS08 was expressed at low levels throughout the experiment and showed no significant change in the two weeks following the short day transition.

Type I and MIKC genes
We identified 40 Type I MADS-box genes and 39 MIKC MADS-box genes (4 MIKC* and 35 MIKC c ) in peach. The phylogenetic relationships, chromosomal distribution and expression patterns of these two gene families were quite different. Most Type I genes appeared to have arisen through tandem duplications after the divergence of the Arabidopsis and peach lineages. They generally formed species-specific clades and clustered in tandem-  duplicated groups on individual chromosomes [48,49]. In contrast, most MIKC subfamilies contained members from both species and appear to have been present in the most recent common ancestor of Arabidopsis and peach.
Differing patterns of Type I and MIKC gene evolution are not unique to peach and Arabidopsis but have recently been documented in MADS-box genes from 24 sequenced plant genomes [49]. Evidence suggests that MIKC genes mainly increase in number following periodic whole genome duplication events [50], whereas Type I genes experience faster rates of birth and death related to tandem duplication and loss [48].
Despite their possession of a similar~58 amino acid DNA-binding MADS domain, Type I and MIKC MADS-box genes actually share few common features. Type I genes have a very simple gene structure, generally consisting of only a single exon. Yeast two-hybrid screens in Arabidopsis suggest that many Type I proteins do not interact with other MADS-box proteins [51]. MIKC genes have a far more complex structure, containing up to 10 exons and three additional domains. Their protein products interact to form multimeric complexes, including the double dimers that specify floral organ identity in Arabidopsis [52][53][54].
The dosage imbalance that results from duplication of only one gene in a multi-protein complex is thought to incur a fitness cost [55]. As a consequence, one member of a gene pair that results from tandem duplication is often removed by purifying selection if its protein product functions as part of a higher level complex [56]. Genes that are less connected are not subject to the same dosage constraints and tend to undergo retention and subfunctionalization following tandem duplication. These trends are borne out in the patterns of evolution exhibited by Type I genes (relatively unconnected) and MIKC genes (highly connected). Exceptions occur, particularly within the SVP/AGL24 and FLC families (see below).
Connectedness may not be the only feature that drives differences in Type I and MIKC phylogenies. Given their short, simple structure, Type I genes may be more likely to be copied intact and in frame during tandem or segmental duplications. It has also been suggested that they exhibit particularly high transposition frequencies, although little direct evidence of transposition exists [49,57]. Their involvement in reproduction, female gametophyte development, and interspecific incompatibility may also promote retention and sub/neofunctionalization [23,49]. Whatever the underlying causes, the partitioning of Type I genes into species-specific clades limits the confidence with which we can functionally annotate peach Type I genes based on sequence similarities with Arabidopsis Type I genes.

Type I gene expression
Type I and MIKC genes generally differ in their tissuespecific expression patterns. In Arabidopsis, Type I gene expression is almost invariably low, detectable only with next generation sequencing or RT-PCR rather than blots or arrays [57,58]. Arabidopsis Type I genes are primarily expressed in the female gametophyte, developing embryo Figure 7 Expression profiles of Type I (left) and Type II (right) MADS-box genes from peach apical shoots at 0, 1 and 2 weeks after the transition to short days. FPKM expression values were log-transformed, and genes were grouped by average linkage clustering (see Methods). Asterisks denote genes whose expression level changed significantly over the course of the two-week experiment. and early endosperm, whereas MIKC c and MIKC* genes are primarily expressed in the sporophyte and male gametophyte, respectively [49].
We found very low levels of Type I gene expression in the sporophyte and male gametophyte tissues we sampled. In the tissue-specific expression dataset (Figure 6), fourteen Type I genes showed no expression, nineteen were expressed at levels lower than 1 FPKM and only seven were expressed at levels higher than 1 FPKM in any tissue. Similar results were seen in the short day transition dataset (Figure 7), where only nine Type I genes showed expression greater than 1 FPKM in shoot apices. The interpretation of such low FPKM values is problematic. On the one hand, 1 FPKM is a typical threshold used to separate expressed from nonexpressed genes in RNAseq experiments [59,60]. On the other hand, transcription factors such as the MADS-box proteins can exert their effects at very low expression levels. Low expression of Type I genes probably has little biological relevance, but given evidence that they may influence flowering time in Arabidopsis [61], significant roles for Type I genes in the peach sporophyte cannot be ruled out.
Five Type I genes were expressed at levels higher than 1 FPKM in both RNAseq datasets: PpeMADS27 and PpeMADS31 (Mα), PpeMADS06 and PpeMADS80 (Mγ) and PpeMADS78 (Mβ). Among these, PpeMADS27 is perhaps the most interesting, showing expression in all tissues examined and reaching expression levels as high as 19.3 FPKM in young leaves. PpeMADS27 is most similar to AGL102, whose expression has been reported in chalazal endosperm [58] and whose interaction partners include the Mβ genes AGL78, AGL82 and AGL103 [51]. Members of the AGL82 lineage (PpeMADS77 and PpeMADS78) were among the more highly expressed Type I peach genes, suggesting that a similar Mα/ Mβ interaction may also occur in peach.

MIKC* gene expression
In higher plants, MIKC* genes group into the S-and Pclades, members of which form interclade heterodimers and control pollen maturation in Arabidopsis [62]. Peach has two members of each clade, PpeMADS01 and PpeMADS55 in the S-clade and PpeMADS20 and PpeMADS36 in the P-clade. We saw no expression of PpeMADS01 in any tissue, perhaps indicating that it has become a pseudogene or that it is expressed during a stage of pollen development that we did not sample. The other S-clade gene, PpeMADS55, was expressed at high levels (57 FPKM) in pollen and very low levels elsewhere. In two independent RNAseq experiments, P-class genes were highly expressed in pollen (32-76 FPKM) but also expressed in the sporophyte (up to 11.8 FPKM). Expression of MIKC* genes outside the microgametophyte has been documented in non-seed plants [17], but does not appear to occur in Arabidopsis and rice [62]. As with persistent lowlevel expression of Type I genes throughout the plant, the functional significance of low MIKC* expression in the sporophyte remains unclear.
Floral homeotic genes MIKC c MADS-box genes from the A-, B-, C-, D-and Eclasses function as floral homeotic genes in angiosperms, specifying floral meristem and floral organ identity [53,63,64]. In general, the number and phylogenetic relationships among peach floral homeotic genes are similar to those reported for other eudicot species [18,26,27]. While we did not evaluate homeotic gene expression in flowers, we did measure the expression of A-, B-, C-, D-and E-class genes in other peach tissues.
The Arabidopsis A-class gene AP1, along with its partially redundant homolog CAL, helps to specify floral meristem identity and to direct the development of sepals and petals [65]. It is closely related to a third gene, AGL79, whose function is largely unknown [66]. Also in the AP1 clade is FUL, an Arabidopsis gene that acts redundantly with AP1 to determine floral meristem identity and plays additional roles in fruit and leaf development [65]. Peach has two FUL orthologs, PpeMADS37 and PpeMADS18. PpeMADS37 was expressed at high levels in fruit, roots, and leaves, while PpeMADS18 showed little expression in any tissue sampled. A third subfamily member, PpeMADS10, was highly expressed in roots and leaves. B-class MADS box genes include AP3 and PI, which form obligate heterodimers in Arabidopsis and specify the identity of petals and stamens [65,67]. The clade contains two members in Arabidopsis but three in many other eudicots, including peach: PpeMADS03, PpeMADS56 and PpeMADS44. Expression of Ppe-MADS56 was restricted almost entirely to pollen, likely indicating the presence of stamen tissue in the pollen sample rather than expression of this gene in the male gametophyte. PpeMADS03 was expressed throughout the plant, particularly in fruit, pollen and young leaves, and PpeMADS44 was found in all tissues except the embryo.
AG performs the C-class function of stamen and carpel specification in Arabidopsis and is a member of a subfamily that also includes the D-class ovule identity genes SHP1, SHP2 and STK [65]. The AG subfamily has three members in peach: PpeMADS24, PpeMADS11 and PpeMADS45, all of which were highly expressed in fruit.
The partially redundant E-class genes SEP1-4 assist in the formation of higher order complexes among other floral homeotic MADS-box proteins [65]. Their tomato and strawberry orthologs also function in fruit development and ripening [68,69]. Peach has four members of the SEP1-4 clade (PpeMADS19, PpeMADS38, PpeMADS21 and PpeMADS09), most of which were highly expressed in fruit and showed little expression elsewhere. It has been reported that genes from the AGL6 subfamily also exhibit Eclass activity. This family contains 2 members (AGL6 and AGL13) in Arabidopsis and 3 members (PpeMADS23, Ppe-MADS61 and PpeMADS62) in peach. Expression of these genes was negligible in all tissues we examined.

Other MIKC c gene families
The AGL17 clade, containing 4 genes in Arabidopsis (ANR1, AGL16, AGL17 and AGL21), has received significant attention for its role in controlling lateral root growth in response to nutrients [70,71]. These genes are largely root-expressed in Arabidopsis, although AGL16 is also expressed in leaves and stems, where it plays roles in stomatal development and flowering time regulation [13,72,73]. Recently, AGL17 has also been shown to function downstream of CONSTANS in the photoperiodic floral-induction pathway [74]. Peach has three members of this clade: PpeMADS59, PpeMADS47, and PpeMADS04, which was strongly induced under short days (see below). Both PpeMADS59 and PpeMADS47 were expressed almost exclusively in roots, while Ppe-MADS04 was expressed in apical shoots only following exposure to short days.
AGL12 constitutes its own subfamily in Arabidopsis and is highly expressed in roots, where it influences root meristem proliferation through its effects on auxin and cell cycle regulation [75]. It has been implicated in regulation of the floral transition, and its rice ortholog plays a role in stress response [75,76]. The single peach ortholog, PpeMADS46, was expressed almost exclusively in roots. It is interesting to note that roots expressed the largest number of different MADS-box genes, both Type I and MIKC, of all peach tissues examined. While MADS-box genes have received most attention for their role in floral development, they appear to have multiple, less appreciated functions belowground. For example, Moreno-Risueno et al. have recently demonstrated that oscillating expression of SOC1, SHP1 and SHP2 is involved in the establishment of Arabidopsis lateral root initiation sites [77].
AGL18 and AGL15 are expressed in the Arabidopsis endosperm and embryo, respectively, and also appear to function in the floral transition [13,65]. AGL18 is unique in being the only MIKC c gene expressed in an Arabidopsis gametophyte: it is found at high levels in pollen [16]. Peach members of this subfamily (PpeMADS17 and Ppe-MADS30) were also highly expressed in pollen.
Arabidopsis has no members of the ancient TM8 clade, but members are present in most other sequenced eudicots, including tomato, cucumber, poplar, grapevine and peach [64]. While their functions are poorly understood, expression data suggest a role for TM8 subfamily members in flower development [18,78]. The peach TM8 ortholog (PpeMADS35) showed minimal expression in all tissues examined and is perhaps chiefly expressed in floral tissues, as has been reported for grapevine.
SOC1 integrates information from multiple flowering time pathways in Arabidopsis and, together with AGL24, activates the flowering promoter FT [20,79]. It is also expressed elsewhere in the plant, particularly in the roots, where it may function in nutrient deficiency response [71]. In addition to SOC1 itself, the clade contains five more genes in Arabidopsis: AGL14, AGL19, AGL42, AGL71 and AGL72. AGL14 has been reported mainly in the roots [80], while AGL19 is induced by cold and promotes flowering in vernalized plants through a non-FLC pathway [81]. AGL42, AGL71 and AGL72 have also been shown to promote flowering through a gibberellin-dependent pathway [82]. Peach has single orthologs of SOC1 (PpeMADS22), AGL14/19 (Ppe-MADS64), and AGL42/71/72 (PpeMADS60), all which were expressed at high levels in roots and leaves. None of these genes showed significant changes in expression after two weeks under short day conditions.

FLC and SVP/AGL24 subfamilies
The two remaining MIKC c subfamilies, SVP/AGL24 and FLC, are best known for their roles in Arabidopsis flowering time regulation. It is within these last two families that we see the greatest differences in gene number and phylogeny among Arabidopsis, peach, and other woody perennials.
The SVP/AGL24 subfamily contains only two members in Arabidopsis: the flowering repressor SVP and the flowering promoter AGL24 [26]. Previous reports indicate that this subfamily is expanded in woody perennials [18,27], and our work suggests that multiple expansions have occurred within different branches of the SVP/ AGL24 subfamily over the course of plant evolution ( Figure 4). While peach contains the six tandemduplicated DAM genes (PpeMADS49 through Ppe-MADS54) that are most closely related to AGL24, poplar and grapevine exhibit expansions within a separate branch of the subfamily that has no Arabidopsis members and a single member in peach (PpeMADS58). The grouping of these genes in species-specific lineages suggests that the expansions have occurred independently.
Within the main poplar SVP/AGL24 expansion, three genes (PtMADS27, 28 and 29) are tandemly-arranged on chromosome VII, while two genes (PtMADS47 and 48) are closely adjacent to one another on a syntenic region of chromosome XVII [83]. This pattern suggests a complex history of both tandem and whole-genome duplications. Within the main grapevine SVP/AGL24 expansion (VvMADS3,4, and 5), two genes are located approximately 2 Mb apart on chromosome III, and one is located on chromosome 15 (Vitis genome data retrieved from http://www.phytozome.net on Nov. 29,2014).
It is interesting to note that poplar, grapevine, peach and Arabidopsis each contain only one true SVP ortholog: no expansions appear to have occurred within the SVP subclade itself. SVP/AGL24 homologs have been referred to as DAM genes in several perennial species and are implicated in endodormancy regulation [3,5,84,85]. Nonetheless, given their multiple, independent evolutionary origins, DAM genes from different species are unlikely to regulate dormancy in an strictly identical manner.
Expansions have also been reported within the FLC subfamily, although not in peach, which contains only a single subfamily member (PpeMADS08). The family is expanded in Arabidopsis by the presence of the five tandemly-duplicated MAF genes [86] and in poplar by a group of five FLC-like genes that form a separate subclade in our phylogeny ( Figure 5). Four of these genes are tandemly-arranged on poplar chromosome III, while the fifth is found in a syntenic region of chromosome I [83]. Again, independent FLC subfamily expansions have occurred in poplar and Arabidopsis, and there is no reason to conclude that FLC-like genes function identically across species. Indeed, while a poplar FLC homolog decreased in expression during the transition to endodormancy [87], our single peach FLC homolog showed no expression change after two weeks under dormancyinducing conditions. It is likely that different perennial species respond to dormancy-inducing conditions using a broadly similar genetic toolkit whose specific genes function in subtly different ways.
The questions remains, why have multiple tandem duplicates arisen and been retained within the SVP/AGL24 and FLC subfamilies in numerous plant lineages? The retention of tandem duplicates does not conform to the typical pattern of fractionation seen in highly-connected MIKC c genes. Do these genes perhaps function as homodimers, freeing them from gene dosage constraints? In a yeast two-hybrid study, AGL24 formed homodimers, while SVPfor which we found no evidence of tandem duplicate retention -did not [51]. In the same study, FLC exhibited no interactions with any other MIKC c proteins, and independent evidence suggests that FLC functions as part of a multi-protein complex containing at least two copies of the FLC protein itself [88]. Perhaps differences in connectivity and interaction among the SVP, AGL24 and FLC gene products have permitted retention and sub-/neo-functionalization of duplicates only within the latter two subfamilies.

Conclusions
Peach contains 79 MADS-box genes distributed across its eight chromosomes, often present in clusters of two or more genes. Most Type I genes appear to have arisen through relatively recent tandem duplications, whereas most Type II genes appear to have increased following whole genome duplication events. An exception to the latter rule occurs in the dormancy-related FLC and SVP/ AGL24 Type II subfamilies, in which species-specific tandem duplicates have been retained across a variety of perennial species. As new plant genomes are sequenced and additional expression data become available, we will undoubtedly learn more about the functions and relationships among these dormancy-related genes. Nonetheless, phylogenetic comparisons and expression data presented here suggest that we should proceed with caution when ascribing the specific functions of Arabidopsis SVP, AGL24 and FLC to related genes from other species.