The Tarenaya hassleriana Genome Provides Insight into Reproductive Trait and Genome Evolution of Crucifers^,

A comparative analysis of the genome of the spider flower (Tarenaya hassleriana) from the Brassicaceae sister family, the Cleomaceae, with Arabidopsis and Brassica crops shows that genome evolution following ancient polyploidy and gene duplication events affect reproductively important traits, including floral development and self-incompatibility systems.

Abstract

The Brassicaceae, including Arabidopsis thaliana and Brassica crops, is unmatched among plants in its wealth of genomic and functional molecular data and has long served as a model for understanding gene, genome, and trait evolution. However, genome information from a phylogenetic outgroup that is essential for inferring directionality of evolutionary change has been lacking. We therefore sequenced the genome of the spider flower (Tarenaya hassleriana) from the Brassicaceae sister family, the Cleomaceae. By comparative analysis of the two lineages, we show that genome evolution following ancient polyploidy and gene duplication events affect reproductively important traits. We found an ancient genome triplication in Tarenaya (Th-α) that is independent of the Brassicaceae-specific duplication (At-α) and nested Brassica (Br-α) triplication. To showcase the potential of sister lineage genome analysis, we investigated the state of floral developmental genes and show Brassica retains twice as many floral MADS (for MINICHROMOSOME MAINTENANCE1, AGAMOUS, DEFICIENS and SERUM RESPONSE FACTOR) genes as Tarenaya that likely contribute to morphological diversity in Brassica. We also performed synteny analysis of gene families that confer self-incompatibility in Brassicaceae and found that the critical SERINE RECEPTOR KINASE receptor gene is derived from a lineage-specific tandem duplication. The T. hassleriana genome will facilitate future research toward elucidating the evolutionary history of Brassicaceae genomes.

INTRODUCTION

Studies of the model plant Arabidopsis thaliana and its close relatives in the Brassicaceae family have provided fundamental insight into the processes and patterns of plant evolution and function (Koornneef and Meinke, 2010; Hu et al., 2011; Wang et al., 2011). Comparative analyses between Brassicaceae and crop species have had profound influences on plant improvement and production. For example, knowledge about the control and evolution of plant reproductive traits, such as floral and fruit development and self-incompatibility (SI) systems, can be directly related to plant fitness and yield (Tanksley, 2004; Shen et al., 2005). The Brassicaceae have also been a model for understanding the dynamics and impacts of ancient polyploidy (genome doubling), considering that the entire family has undergone a whole-genome duplication (named At-α) and the Brassica crops have had an additional genome triplication (Br-α) (Blanc et al., 2003; Thomas et al., 2006; Wang et al., 2011). Genes retained in multiple copies due to these ancient polyploidy events, in addition to more recent tandem duplications, have played important roles in the evolution and regulation of key traits (Edger and Pires, 2009; Flagel and Wendel, 2009). However, the polyploid history of the Brassicaceae also complicates synteny and evolutionary inferences to distantly related crop species.

To fully exploit the fundamental trait and genome insights garnered from Brassicaceae systems and improve synteny analyses to more distant crops, we report the genome sequencing and analysis of Tarenaya hassleriana from the Brassicaceae sister family Cleomaceae. Currently, papaya (Carica papaya), a member of the order Brassicales, is the closest relative with a complete genome sequence; however, these two lineages diverged more than 70 to 110 million years ago (Ming et al., 2008; Beilstein et al., 2010). The Cleomaceae is the phylogenetic sister family to the Brassicaceae, with the two lineages having diverged only ∼38 million years ago (Schranz and Mitchell-Olds, 2006; Couvreur et al., 2010). Brassicaceae and Cleomaceae share many traits (Hall et al., 2002; Iltis et al., 2011), such as a preponderance of herbaceous species, the same general floral ground plan (four sepals, four petals, six stamens, and two fused carpels), and a replum in the mostly dehiscent fruits, referred to as capsules. There are also a number of key differences. Most of the 300 Cleomaceae species are restricted to the semitropics and arid desert regions and lack a genetic pollen-pistil SI system, whereas most of the 3700 Brassicaceae species largely radiated into cold temperate regions and possess a genetically regulated SI system (Guo et al., 2011). Another striking distinction is in floral symmetry: Cleomaceae have mostly monosymmetric flowers and Brassicaceae have mostly disymmetric flowers (Endress, 1999; Patchell et al., 2011). Cleomaceae also exhibit greater variation in the basic floral plan with increases in stamen number, petal dimorphisms, and stalks to the ovary, whereas Brassicaceae exhibit greater diversity in fruit morphology and dehiscence capabilities (Franzke et al., 2011). Comparative analyses can be used to elucidate the genomic basis of these differences. The Cleomaceae species we sequenced is T. hassleriana, often referred to as the spider flower, which is widely grown as an ornamental species and used as an educational model (Marquard and Steinback, 2009). This species was formerly named Cleome hassleriana (often erroneously labeled as Cleome spinosa), but the genus Cleome has undergone recent taxonomic revisions (Iltis and Cochrane, 2007).

Brassicaceae and Cleomaceae have undergone independent ancient polyploidy events. At least five ancient polyploidy events have occurred in the evolutionary history of Arabidopsis (Bowers et al., 2003; Van de Peer et al., 2009), four of which are shared with Cleomaceae: ζ near the origin of seed plants (Jiao et al., 2011), ε near the origin of angiosperms (Jiao et al., 2011), the ancient hexaploidy At-γ shared by nearly all eudicots (Jaillon et al., 2007; Vekemans et al., 2012), and At-β restricted to part of the order Brassicales as it is lacking from the papaya genome (Ming et al., 2008). The most extensively studied ancient polyploidy event is the more recent At-α genome duplication (Arabidopsis Genome Initiative, 2000; Bowers et al., 2003; Schnable et al., 2012) and is shared by all Brassicaceae species (Schranz et al., 2012). The crop genus Brassica has all the ancient polyploidy events in common with Arabidopsis but also has undergone an additional and more recent whole-genome triplication (hexaploidy) event (Br-α) after its split with Arabidopsis around ∼17 million years ago (Wang et al., 2011). Limited BAC and transcriptome sequencing revealed that Tarenaya lacked the At-α event and that it underwent an independent ancient genome triplication (Th-α) (Schranz and Mitchell-Olds, 2006; Barker et al., 2009). Thus, Tareneya provides a unique opportunity to contrast genome evolution from a common ancestor comparing three genomic equivalents in Tarenaya, two in Arabidopsis, and six in Brassica, and furthermore to contrast two independent ancient genome triplications (Th-α versus Br-α). We not only compare these polyploidy events and more recent tandem duplication events, but also show how they contributed to the genes regulating key reproductive traits (Van de Peer et al., 2009).

RESULTS

Genome Sequencing and Integration with Physical Map

The T. hassleriana genome is relatively small (∼290 Mb: 2n = 20) and within the range of sequenced Brassicaceae species: Schrenkiella parvula (formerly Thellungiella parvula), 140 Mb (Dassanayake et al., 2011); A. thaliana, 157 Mb (Bennett et al., 2003); Arabidopsis lyrata, 207 Mb (Hu et al., 2011); Capsella rubella, 210 Mb (Slotte et al., 2013); Aethionema arabicum, 240 Mb (Haudry et al., 2013); Sisimbrium irio, 262 Mb (Haudry et al., 2013); Eutrema salsugineum (formerly Thelungiella salsuginea), 314 Mb (Wu et al., 2012; Yang et al., 2013); Leavenworthia alabamica, 316 Mb (Haudry et al., 2013); and Brassica rapa 485 Mb (Wang et al., 2011). To generate a high-quality draft genome assembly, we used both sequenced paired-end libraries and constructed a Bacterial Artificial Chromosome (BAC) based whole-genome profiling (WGP) physical map. We used the Illumina next-generation sequencing platform to generate ∼70.2 Gb (245X genome-depth) raw data of paired-end reads ranging from 90 to 100 bp (see Supplemental Table 1 online) from seven libraries with various insert sizes (350 to 20 kb). Sequence data were filtered, yielding ∼40 Gb of high-quality sequence (∼139X coverage) (see Supplemental Table 2 online). Sequences were assembled using SOAPdenovo (Li et al., 2010) (version 2.21) (see Supplemental Table 3 online). We also sequenced and mapped over 4 Gb of transcriptome data to the assembly showing that >94% of the genic regions were covered (see Supplemental Table 4 online).

The physical map was made using the Keygene WGP fingerprinting technique (van Oeveren et al., 2011). We generated 192,000 BAC-based Illumina sequence tags from 19,200 BACs from two libraries (EcoRI and MseI) with an average insert size of ∼125 kb (giving a total of 32X genome equivalents) (see Supplemental Table 5 online). We identified 87,617 high-quality and unique WGP tag sequences that allowed us to uniquely identify 15,567 BACs (with an average of 40.3 tags per BAC). These were used to build a high-stringency map assembly using modified Finger Printed Contigs (FPC) software (Engler et al., 2003) to generate 786 contigs using data from 9396 BACs (see Supplemental Table 6 online). We integrated the WGP physical map scaffolds with the Short Oligonucleotide Analysis Package (SOAP) sequence scaffolds to produce 77 superscaffolds from the integration of 349 sequence scaffolds (integrating between two and 21 scaffolds per superscaffold with an average of 4.5 scaffolds per superscaffold) through Basic Local Alignment Search Tool (BLAST) mapping of the WGP anchors (tags). We evaluated the quality of the integration between physical map and superscaffolds by manually checking the ordering and orientation of connected scaffolds by analyzing collinearity relative to A. lyrata (example shown in Supplemental Figure 1 online), confirming that all Tarenaya superscaffolds have extensive and extended synteny. We further validated our assembly by comparing it with four previously published Sanger-sequenced BACs (Schranz and Mitchell-Olds, 2006; Navarro-Quezada, 2007), revealing nearly identical assemblies (see Supplemental Figure 2 online). The final assembly statistics of the integrated dataset are summarized (Table 1; see Supplemental Table 7 online). With this integration, the N50 was increased by more than 2.6-fold (N50 = 1.26 Mb) due to the merger of most of the de novo assembled scaffolds into superscaffolds.

Table 1. Summary of the genome sequencing, assembly, and annotation.

Assembly
	N50 (size/number)	N90 (size/number)	Total sizes
Contigs	21.58 kb/2761	2.7 kb/13591	222 Mb
Scaffolds	551.9 kb/98	64.8 kb/622	256.5 Mb
Superscaffolds	1.26Mb/40	7.4kb/1014	273Mb
Annotation
	Glean	RNA-Seq supported	Homologous with Arabidopsis
# Genes	28917	20337	24245
	LTR	DNA transposons	Total size
TE sizes, Mb (%)	97.28 (38.19)	11.8 (4.62)	110 (43.3)

Gene Annotation

Gene annotation was conducted using a pipeline that integrates de novo gene prediction, homology-based alignment, and RNA-seq data. In total, we conducted >4 Gb of RNA-seq, of which 77.4% of reads could be confidently mapped onto the genome (see Supplemental Table 8 online). To analyze the overall gene expression patterns and to provide basic gene expression information, transcriptomes were generated for several T. hassleriana tissues (see Supplemental Table 9 online). A principal component analysis showed that stamen, root, and seed profiles separate most, a result similar to the pattern detected in A. thaliana where these three tissue types separate most over the first two components (see Supplemental Figure 3 online) (Schmid et al., 2005).

The prediction of gene models by various techniques was summarized (see Supplemental Table 10 online), with a large range in the number of predicted genes. To be conservative, we used the models predicted by GLEAN to have a high posterior probability out of the various gene prediction techniques, which resulted in the identification of 28,917 highly supported gene models with an average transcript length of 2216 bp, coding sequence size of 1169 bp, and 5.27 exons per gene, both similar to that observed in A. thaliana and B. rapa (see Supplemental Figure 4 online). A total of 92.9% of gene models have a homolog match or conserved motif in at least one of the public protein databases, including Swissprot (McMillan and Martin, 2008), 71.1%; the Translated European Molecular Biology Laboratory database (Boeckmann et al., 2003), 92.5%; InterPro (Zdobnov and Apweiler, 2001), 74.9%; the Kyoto Encyclopedia of Genes and Genomes (Kanehisa and Goto, 2000), 55.2%; and Gene Ontology (GO) (Ashburner et al., 2000), 55.7% (see Supplemental Table 11 online), and 97.1% are represented among the public Expressed Sequence Tag (EST) collections or de novo Illumina mRNA-Seq data. In addition to protein-coding genes, we also identified 220 microRNA, 862 tRNA, and 685 small nuclear RNA genes in the T. hassleriana genome (see Supplemental Table 12 online). Orthologous clustering of proteomes predicted for T. hassleriana and three Brassicaceae species (A. thaliana, A. lyrata, and B. rapa) revealed 15,112 genes in 12,689 families in common (Figure 1). We found that 20,926 T. hassleriana genes clustered with at least one of the three genomes. Furthermore, 1556 Tarenaya-specific genes in 748 families were identified, most of which were enriched for genes of unknown function and for which 34% have EST supported annotation.

Figure 1.

Venn Diagram Illustrating the Shared and Unique Gene Families from T. hassleriana (Cleomaceae), A. thaliana, A. lyrata, and B. rapa (Brassicaceae).

In total, we predicted 28,917 well-supported gene models for T. hassleriana, of which 22,482 could be placed into one of the 14,505 gene families. A total of 87.5% of gene families found in Tarenaya were present in all three Brassicaceae species and 7.4% in one or two Brassicaceae species, and only 5% of gene families were unique to Tarenaya with many of these associated genome-specific retrotransposons. Thus, comparative functional and evolutionary analysis of well-characterized Arabidopsis and Brassicaceae genes is feasible using Tarenaya as an outgroup.

To evaluate the status of our genomic assembly, we compared it to the nine published crucifer genomes (Haudry et al., 2013) (see Supplemental Table 13 online). Our T. hassleriana assembly has the highest assembled sequence versus expected genome size with an estimated 93% genome size completeness. For comparison, the B. rapa assembly only has 51.6% genome coverage. Our assembly also has the largest scaffold N50 (1.26 Mb) of the Illumina-only sequenced genomes (T. hassleriana, Ae. arabicum, S. irio, Leavenworthia alabamica, and B. rapa). The number of predicted genes is on par with that in the other crucifers and the number of A. thaliana orthologs is slightly lower than average, as expected from an organism outside the Brassicaceae. Eighty-eight percent of ultraconserved core eukaryotic genes (Parra et al., 2009) were found, suggesting a slightly lower covered gene space. This might be due to the fact that the percentage of transposable elements (TEs) is relatively high, which can cause difficulties assembling regions where TEs and genes are intermixed.

Because of our high genome sequencing coverage, we were able to identify more than 43% of the 293-Mb T. hassleriana genome as being composed of transposons. For comparison, only 31% of the 529 Mb B. rapa genome has been identified as transposons (see Supplemental Table 13 online). The discrepancy is because of the great difference in the coverage of the assemblies (93% versus 56%). Thus, the percentage of transposons in Brassica is largely underestimated. To examine the contiguity of the genome assembly (causes for gaps between contigs in the scaffolds), we analyzed the distribution of contigs versus transposon long-terminal repeats across the largest 491 scaffolds (see Supplemental Figure 5 online). By doing so, we can demonstrate that regions with a high density of long terminal repeats (LTRs) correspond to smaller contigs in scaffolds and thus generate regions of lower contiguity. Both de novo repeat identification and homology-based methods were applied to predict transposable elements (TEs) (see Supplemental Table 14 online). The majority of repetitive sequences were Class I long-terminal repeat retrotransposons, constituting 36.6% of the genome compared with 27.1% in B. rapa. The overall lower percentages of annotated transposons in Brassica are likely due to its lower genome sequence coverage (see Supplemental Table 13 online) because the B. rapa genome is nearly 200 Mb larger than Tarenaya. Most of the repeats were located in the intergenic regions.

Comparative Analysis of Ancient Polyploidy Events

The Brassicaceae-Cleomaceae system allowed us to compare genome evolution after several rounds of independent ancient polyploidy (Figure 2). We confirmed that the Cleomaceae polyploidy event (Th-α) occurred independently of, and more recently than, the Brassicaceae-specific duplication event detected in Arabidopsis and Brassica (At-α). We also detected the nested B. rapa ancient hexaploidy (triplication of the genome) event (Br-α) and show that it is of approximately the same age as Th-α. For all three taxa, we detected the diffuse signal of the older and shared events (At-β, At-γ, ε, and ζ) (Figure 2). The Th-α and Br-α events are of approximately the same age and, as discussed below, represent independent ancient hexaploidy events. Using whole genome intragenomic dot plots of Tarenaya, we showed many triplicated blocks (see Supplemental Figure 6 online), and analysis of syntenic depth with QuotaAlign (Tang et al., 2011) shows that 49.4% of genes are found at 3× coverage (see Supplemental Table 15 online). To illustrate the homologous relationships and the evolutionary history of triplicated/duplicated segments in Cleomaceae and Brassicaceae, we integrated intra- and intergenomic analyses (Figure 3). We analyzed synteny relationships both within and between genomes (see Supplemental Figures 6 to 9 online). For T. hassleriana, 86, 83, and 85% of the protein-coding genes were homologous to the genes in the A. thaliana, A. lyrata, and B. rapa genomes, respectively (see Supplemental Table 16 online). By making these comparisons of Tarenaya versus A. thaliana, A. lyrata, and B. rapa genomes (Figure 3), we found significant, 3:2, 3:2, 3:4, 3:5, and 3:6 homologous patterns, respectively, which is consistent with the polyploid history of the species. To illustrate this pattern, we highlighted two ancestral blocks (A1 and A2) (Figure 3, two small insets). Note that the three Tarenaya blocks show almost perfect collinearity, whereas one of two Arabidopsis regions is broken across two chromosomes, suggesting a Brassicaceae-specific rearrangement(s) after At-α. Since synteny analysis has been extensively performed within Brassicaceae, we also show our results with the collinear blocks color-coded according to the current Brassicaceae conventions (Schranz et al., 2006) (see Supplemental Figure 10 online).

Figure 2.

Relative Timing of the Polyploidy Events and Lineage Splitting Based on Divergence of Fourfold Degenerate Sites (4DTv) for Duplicated Genes within A. thaliana, B. rapa, and T. hassleriana and Orthologous Genes between A. thaliana and T. hassleriana.

All plots detect broad overlapping peaks between 0.5 and 1.0, representing shared older polyploidy events (At-β, At-γ, ε, and ζ). The divergence of the Brassicaceae-Cleomaceae lineages is seen by the differentiation of Arabidopsis and Tarenaya homologs at the peak centered at ∼0.35 (highlighted by red star). The divergence between paralogs from the At-α duplication event occurred slightly after the lineage splitting and is detected by the peaks centered at ∼0.3 for both Arabidopsis and Brassica. The At-α peak is lacking from Tarenaya, proving At-α is Brassicaceae specific. Nearly overlapping distributions between 0.15 and 0.25 were detected for Brassica and Tarenaya, representing the independent Br-α and Th-α ancient hexaploidy events, respectively.

Figure 3.

Homologous Genome Blocks within and between Genomes for Cleomaceae and Brassicaceae.

The largest 20 (plus additional two smaller scaffolds) color-coded superscaffolds of T. hassleriana are taken as the reference, such that any region homologous to the 22 scaffolds is colored accordingly. A, Self-alignment of Tarenaya superscaffolds, with the inner circle showing links of syntenic blocks. Over 47% of the genome is found in three copies, supporting the conclusion that it experienced an ancient hexaploidy event (Th-α = triplication). Rings within a genome (inner gray bars) and blocks homologous to Tarenaya (outer color-coded bars) for completed Brassicaceae genomes: B, A. thaliana; C, A. lyrata; and D, B. rapa. The inner gray bars show a clear pattern related to the ancient polyploidy events of the Brassicaceae (At-α = duplication) and nested Brassica-specific lineage (Br-α = triplication). The color-coded outer rings of homology relative to Tarenaya show a complex pattern due to the independent polyploidy events between families. The two small insets illustrate examples of the three Tarenaya to two Arabidopsis to six Brassica genome equivalents due to ancient polyploidy events.

We inferred the putative “A ancestor” (pre-At-α) shared by A. thaliana and A. lyrata, the “B ancestor” of B. rapa (pre-Br-α but post-At-α ancestral genome state), and the “T ancestor” of T. hassleriana (pre-Th-α ancestral genome state) (see Methods). We compared our identified homologous replicated blocks within and between genomes in Brassicaceae species and T. hassleriana with the results of an earlier analysis of conserved At-α blocks (Blanc et al., 2003; Thomas et al., 2006). First, we reconstructed our version of the pre-At-α ancestor (A ancestor) of A. thaliana (version TAIR9), which resulted in 64 ancestral regions involving 19,976 protein-coding genes (see Supplemental Table 15 online). The corresponding relationships to the blocks identified by Wolfe and colleagues (Blanc et al., 2003) (that we refer to as “Wolfe blocks”) are illustrated in Supplemental Figure 11 online. We used the same method on the minimized genomes of A. lyrata, B. rapa, and T. hassleriana and generated 61, 71, and 87 conserved ancestral blocks covering 24,373; 25,646; and 20,680 protein-coding genes, respectively, to represent the postulated A, B, and T ancestors (see Supplemental Figures 12 to 14 online). A table listing all genes included in each block for each genome is provided in the supplemental materials (see Supplemental Data Set 1 online). A comparison of the reconstructed T and A ancestor genomes revealed a 1:1 relationship, supporting our conclusion that the ancestral genome of the Brassicaceae and Cleomaceae was conserved before the independent duplication (At-α) and triplication (Th- α) events, respectively. Since B. rapa underwent a nested and specific triplication following the At-α, we see a 1:2 pattern when we compare the inferred T to B ancestors. A comparison of conserved ancestral genomic blocks across species is shown in Supplemental Figure 15 online. We then traced the extent of gene retention and fractionation in homologous blocks after polyploidy events. We partitioned the two subgenomes of Arabidopsis, three subgenomes of Brassica, and three subgenomes of T. hassleriana, respectively, by comparing them with the reference A ancestor of A. lyrata (see Supplemental Figures 16 to 19 online). These improvements to understanding genome evolution after independent ancient polyploidy events of Brassicaceae species will facilitate synteny analyses in distant crop species.

Comparative Analysis of Type II MADS Box Genes

The development of the four floral organ types and later the fruits is regulated by Type II MINICHROMOSOME MAINTENANCE1, AGAMOUS, DEFICIENS and SERUM RESPONSE FACTOR (MADS) domain proteins (Smaczniak et al., 2012) as described by the ABCDE model (Theissen, 2001). The types of MADS box genes that regulate development are remarkably well conserved across eudicots, with the molecular mechanisms of their action extensively studied in Arabidopsis. We found that the Tarenaya genome contains representatives of all the major Type II MADS box genes described in Arabidopsis and Brassica (see Supplemental Figure 20 online). We concentrated on the retention of the MADS box genes derived from At-α, Br-α, and Th-α and compared this with the polyploid origins of additional duplicates (At-β, At-γ, ε, T [tomato] [Tomato Genome Consortium, 2012], and Pt-α [poplar] [Tuskan et al., 2006]) (Figure 4). Theoretically, the At-α, Br-α, and Th-α events should have given rise to two Arabidopsis, six Brassica, and three Tarenaya gene copies (syntelogs) from a single ancestral gene. Of the 11 MADS box gene clades involved in floral, fruit, and inflorescence development that were likely present in the most recent common ancestor of Brassicaceae and Cleomaceae shown in Figure 4, we found that only three duplicate pairs are in fact maintained in Arabidopsis due to At-α: APETALA1 (AP1)/CAULIFLOWER (CAL) (A-function), SHATTERPROOF1 (SHP1)/SHP2 (D-function), and SEPALLATA1 (SEP1)/SEP2 (E-function). Thus, Arabidopsis has only three of 11 possible replicates (27.3% syntelog retention). This implies that during early Brassicaceae evolution (before the split of Arabidopsis-Brassica), there were 14 gene lineages. From these 14 lineages then there would be 42 possible syntelogs in Brassica due to Br-α triplication, of which 28 copies are considered as additional syntelogs. We find a remarkable 19 of these additional gene copies (67.8% syntelog retention). This includes all three possible copies maintained for the following seven Brassica gene families: SEP4, SEP3, AGAMOUS-LIKE79 (AGL79), FRUITFULL, AP1, PISTILLATA (PI), and SHP1. The only two genes to return to single copy in Brassica after Br-α are CAL and SHP2. When we also consider the At-α gene retention, then a maximum of four of six gene copies are found in the SEP1/2 clade, AP1/CAL clade, and the SHP1/2 clade. From the 11 ancestral loci, 33 syntelogs would be expected in Tarenaya due to the Th-α triplication, with 22 possible additional syntelogs. However, we only recovered six (27.3% syntelog retention) with no cases where all three possible copies are maintained (we do not count the additional tandem duplicate of Th-AP3 here, which is discussed below). Thus, we find more than double the syntelog retention in Brassica than in Tarenaya, despite the fact that both are ancient hexaploids of approximately the same age. The greatest differential in gene copy retention between Brassica and Tarenaya is for the AGL79 clade (3 versus 1) and the SHP1/2 clade (4 versus 1). The SHATTERPROOF genes in Brassicaceae regulate various traits during carpel and fruit formation (Colombo et al., 2010). The single-copy nature of the SHP homolog in Tarenaya is thus notable, since this is the only gene that is duplicated in Arabidopsis due to At-α but has returned to single copy in Tarenaya (see Supplemental Figure 21 online). In Cleomaceae, fruit morphology is less diverse than in Brassicaceae, and we hypothesize that the retention of SHP genes plays an important role in the morphological variability of Brassicaceae.

Figure 4.

Phylogenetic Tree of Type-II MADS Box Transcription Factor Genes Involved in Floral Organ Specification.

The alignment used to create this tree is available in the online materials (see Supplemental Data Set 2 online). The major floral MADS box genes cluster into five groups corresponding to the five main functional types (AP1-like genes, shown in yellow; AP3/PI-like genes, B-type shown in blue; AG-like genes shown in gray; STK-like genes shown in red; and SEP-like genes in green) according to the ABC(DE) model of floral development. Species included are A. thaliana (At), grape (Vitis vinifera, Vv), tomato (Solanum lycopersicum, Sl), poplar (Populus trichocarpa, Pt), B. rapa (Br), and T. hassleriana (Th). Tarenaya genes are indicated in red. The colored squares (duplication events) and circles (triplication events) placed on nodes represent gene lineage expansion(s) that can be associated with particular ancient polyploid events: Th-α, At-α, Br-α, At-β, At-γ, ε, T (identified by tomato genome sequencing), and Pt-α (identified by poplar genome sequencing). Type-II MADS box genes are often retained after ancient polyploidy events. From the tree above, we calculated a 27.3% syntelog (homolog generated by a polyploidy event) retention after At-α, 27.3% syntelog retention after Th-α, and a much higher (67.8%) syntelog retention after Br-α, despite the fact that Th-α and Br-α triplications are of approximately the same age. The Tarenaya B-class genes show unusual patterns in that the AP3 homologs (Ch02920 and Th02921) represent a recent tandem duplication, which is rare for floral MADS box genes, and there are two copies of PI homologs that are likely due to At-β with one lineage being lost in Brassicaceae. Shown is a maximum likelihood tree with 1000 replicate bootstrap values, of which the branches with a bootstrap value of >80 are presented, visualized topology only.

Comparative Collinearity Analysis of Floral Developmental Regulators

To assess the contributions of ancient polyploidy and tandem gene duplications to floral regulatory gene diversification, we conducted a more detailed analysis of gene synteny and expression patterns. Almost all floral MADS box genes show conserved synteny between Brassicaceae and Cleomaceae. For example, the A-class genes show stable duplicate retention; the loci containing the AP1 and CAL homologs in Tarenaya, Arabidopsis, and Brassica are syntenic to one another with little evidence of local rearrangements (see Supplemental Figure 22 online).

Compared with other MADS box genes, the B-class (PI and AP3) genomic regions show a more dynamic pattern (Figure 5). The split between PI and AP3 is an old duplication due to the angiosperm ε polyploidy event (Figure 4), with almost no detectable collinearity between these regions (Figure 5). Comparison of B-class Tarenaya genomic regions with Brassicaceae allowed us to detect two B-class duplication events: A recent Tarenaya tandem duplication of AP3 (Th02920 and Th02921) and an older PI duplication, likely due to At-β, which has been lost from Brassicaceae but is still retained in Tarenaya (Th17298) (Figure 5; see Supplemental Figures 23 and 24 online).

Figure 5.

Collinearity Analysis of B-Class Type II MADS Box Gene (AP3 and PI) Homologs Reveals Unusual Patterns of Gene Loss, Lineage-Specific Transpositions, and Local Tandem Duplications.

The placement of ancient polyploid events giving rise to gene duplicates is shown on appropriate nodes (At-β, At-α, Br-α, and Th-α). For the AP3 group genes in Brassicaceae (shown by red bars), there is only a single locus retained in A. thaliana and A. lyrata and two retained Brassica syntelogs derived from Br-α. Collinear homoeologous regions derived from At-α are detectable in Brassicaceae genomes; however, the AP3 syntelogs were lost (regions highlighted in red boxes). T. hassleriana has an unusual tandem duplication of AP3 genes in one of two homoeologous regions derived from Th-α. The AP3 genes and the neighboring Forkhead gene (EMB1967) are the only genes syntenic to the AP3 Brassicaceae region (see Supplemental Figure 25 online). The Cleomaceae AP3 region is syntenic with AP3 regions of all other eudicot genomes analyzed (see Supplemental Figure 23 online). Thus, we conclude that there was a lineage-specific transposition of AP3 and the neighboring Forkhead locus in the Brassicaceae. There is only a single copy of PI genes (red bars) in A. thaliana and A. lyrata and all three Br-α derived syntelogs in Brassicaceae. There is no detectable homoeologous region in Brassicaceae derived from At-α. In Tarenaya, we detected one syntenic PI gene and region to the Brassicaceae, but also a second region that is syntenic to other eudicots (see Supplemental Figure 24 online). We conclude that these two Tarenaya PI genes were generated due to the At-β ancient duplication event with the subsequent transposition of one locus into the region collinear between Brassicaceae and Cleomaceae and loss of the nontransposed locus from only the Brassicaceae lineage. The differences in genomic context and gene expression (see Supplemental Figure 27 online) may contribute to shifts in floral morphology and symmetry between families.

Strikingly, we also detected two gene transposition events: a Brassicaceae-specific AP3 transposition and a shared transposition event of one PI gene before the split of Brassicaceae and Cleomaceae (Figure 5). The Brassicaceae-specific AP3 transposition event also involved the flanking EMBRYO DEFECTIVE1967 (EMB1967) gene containing two conserved domains: the N-terminal region of microspherule protein (MCRS_N) and Forkhead-associated (FHA). Tarenaya AP3 is similarly flanked by a Forkhead-associated protein (Th02919) that has its highest match to EMB1967; however, the orientation of AP3 and the Forkhead genes is inverted between species (see Supplemental Figure 25 online) and is also detected in a distantly related Cleome species. In general, B-class genes are functionally highly conserved across angiosperms, whereas A-class gene function appears to be less conserved (Litt and Kramer, 2010). However, we have shown that it is in fact the B-class genes that have undergone transposition events.

Members of the TEOSINTE BRANCHED1, CYCLOIDEA, and PCF (TCP) gene family play an important role in the transition from polysymmetric to monosymmetric flowers (reviewed in Busch and Zachgo, 2009; Jabbour et al., 2009; Rosin and Kramer, 2009), including monosymmetric Brassicaceae (Busch and Zachgo, 2007; Busch et al., 2012). Cleomaceae floral morphology, especially in petal and stamen position, numbers, and asymmetry, is quite variable. However, the role of Tarenaya TCP homologs in monosymmetry has not yet been fully characterized. We find a pattern of conservation of genomic collinearity around the TCP1 locus between species (see Supplemental Figure 26 online). Arabidopsis contains only a single TCP1 locus, as does A. lyrata. Due to Br-α, Brassica has three syntenic copies of TCP1. We also can detect the syntenic regions in Brassicaceae species due to At-α (see Supplemental Figure 26 online) but find no At-α derived homologs of TCP genes, suggesting the loss of a TCP1 syntelog occurred early in Brassicaceae evolution. In Tarenaya, we find three genomic regions derived from Th-α, with two copies of TCP1 intact (Th21666 and Th24587) (see Supplemental Figure 26 online). The correlation between multiple copies of TCP members and monosymmetry has been noted across angiosperms (Rosin and Kramer, 2009).

Expression of A- and B-Class Homolog Genes

The expression of T. hassleriana homologs of A. thaliana major floral regulators was also analyzed with quantitative RT-PCR (qRT-PCR). The two putative T. hassleriana homologs of CAL/AP1 (Th-CAL/AP1-1 and Th-CAL/AP1-2) show similar expression in all bud stages, but Th-CAL/AP1-1 with only half of the transcript abundance of Th-CAL/AP1-2. Th-CAL/AP1-1 shows highest expression in sepals and ∼10 times lower expression in petals. Expression of neither homolog was detectable in stamens, petals, gynoecia, capsules, roots, or leaves. Th-PI-1, the homolog of PI in A. thaliana, is collinear with the Brassicaceae gene order and is expressed mainly in petals and stamens, with less expression in younger and higher expression in older stages (see Supplemental Figure 27 online). The second PI homolog of T. hassleriana, Th-PI-2, is expressed at a much lower rate than Th-PI-1, ranging from around 50% of the Th-PI-1 expression in stamens to only 10% of the Th-PI-1 expression in late bud states.

The second two putative floral homeotic B-function genes, Th-AP3-1 and Th-AP3-2, are highly similar in coding, 3′, and 5′ untranslated region sequence, and both are homologous to the AP3 gene in A. thaliana. Th-AP3-1 is expressed throughout the observed stages of bud development. In petals and stamens at anthesis, it is the most highly expressed gene of the MADS box genes analyzed (see Supplemental Figure 27 online). In petals, it is expressed at∼200% and in stamens around 400% higher level than Th-CAL, Th-PI-1, and Th-AP3-2. Expression of Th-AP3-2 in buds is around one-third lower than that of Th-AP3-1, and differential expression between both genes is detected in petals and stamens. While Th-AP3-1 shows higher expression in stamens than in petals, Th-AP3-2 has stronger expression in petals than in stamens (see Supplemental Figure 27 online). The divergence in expression of the B-class genes, along with the aforementioned gene transpositions, is indicative of the likely role in B-class gene functional differentiation and the regulation of the different floral morphologies between families. We acknowledge that these comparative analyses of Tarenaya to Brassicaceae are based on the analysis of the draft genome sequence of a single domesticated accession of one species; thus, future comparative analyses to other Cleomaceae species is needed for validation.

Evolution of the Brassicaceae SI Locus

Many Brassicaceae species possess a pollen-pistil recognition system that confers SI through the rejection of self-pollen (Boyes et al., 1997). This system is based on the interaction of the stigmatically expressed S-receptor kinase (encoded by SRK) with a small polymorphic peptide that is coded by the S-locus Cys-rich protein (SCR) gene, both located on the so-called S-locus of the genome. Many SCR alleles are likely derived from (partial) duplication and/or gene conversion from SRK alleles (Koornneef and Meinke, 2010; Guo et al., 2011). The cysteine-rich protein kinase (CRK) genes and ARK genes that belong to the S-locus are part of a larger family of receptor-like protein-kinases (RLK) genes, which have been shown to be mostly involved in oxidative stress and pathogen response (Chen et al., 2004; Wrzaczek et al., 2010). It should be noted that most ecotypes of A. thaliana are self-compatible due to pseudogenization of the SRK and/or SCR genes, a pattern seen in other self-compatible crucifers, but still an exception among Brassicaceae (Nasrallah et al., 2004). T. hassleriana does not possess a SI system, but it still contains an S-like locus that contains functional genes, and it is likely that this part of the genome is close to the ancestral state of this locus for Brassicaceae.

SRK and ARK on the S-locus are characterized by specific variations on the following protein domain compositions: B_lectin (B), S_locus_glyco (S), PAN-2 (Pa), Pkinase_Tyr (Pk), and Duf3403 (D1) and/or DUF3660 (D2) (Zhang et al., 2011). Using the Pfam database (Punta et al., 2012), we found that the S locus region in C. rubella, B. rapa, A. lyrata, and A. thaliana as well as the homologous region in T. hassleriana mostly contains SRK genes with a B-S-Pa-D1-Pk-D2 protein domain structure, followed by the B-S-Pa-Pk-D1/2 protein domain structure, which is more common across all SRK families (Figure 6; see Supplemental Figure 28 online). Three syntenic regions containing most of the genes of the S-locus were found in Tarenaya. One of these contained a gene with the exact B-S-Pa-Pk-D1/2 protein domain structure: Th11131. Our analysis of domain structure further found that another gene, Th22785, had a B-S-Pa-Pk protein domain structure, an architecture shared with many angiosperm genes but not with SI-specific SRK alleles. We have also found two homologs of the SI-modifier gene, Pub8, in two of the three Tarenaya syntenic regions. One of the homologs, Th22784, is adjacent to the Th22785 locus. The other homolog, Th25331, is contained in the syntenic region that completely lacks any S_locus_glyco Pfam-containing proteins. Interestingly, we do not find a Pub8 homolog in close proximity to Th11131. However, based on the alignment of all three regions, we can assume that the single-copy ancestral region contained homologs to Pub8, ARK3, and B120 (Figure 6). To ensure that syntenic genes that were not found were not due to gaps in our assembly coverage, we manually confirmed that all relevant regions had no large gaps. In Supplemental Table 17 online, a summary of the coverage in these regions can be found. No large gaps were present except for one 2.5-kb gap in the region around Th22784. Based on the syntenic order of S-locus genes in other species, it is extremely unlikely that a gene is present in this gap, let alone any important candidate gene for this region.

Figure 6.

Synteny and Protein Domain Analysis of the Brassicaceae SI-Like Regions with Cleomaceae and Inference of the Ancestral Genomic Region.

All regions presented here are drawn in more detail (including genetic coordinates) in Supplemental Figure 28 online. Genes are marked by block arrows and color coded according to their protein domain composition as listed in the legend. In the case of pseudogenes, the block arrows have a dashed border. To find protein domain composition in pseudogenes, the longest open reading frame was translated in silico (see Methods). Gene orientation is shown by block arrows pointing left for gene orientation toward the 5′ end and pointing right for gene orientation toward the 3′end. The At-α duplication and the Br-α and Th-α triplications have been marked on the tree with an orange box (At-α) and yellow and purple circles (Br-α and Th-α, respectively). Each branch corresponds to a subgenome resulting from such a polyploid event. Theoretically, B. rapa should have six subgenomes, but only the regions showing synteny are listed here for clarity. The bottom branch represents a hypothetical layout of this genome region in the common ancestor of these species before the At-α, Br-α, and Th-α polyploid events. From our results, we conclude that an ARK3-like gene underwent a Brassicaceae-specific tandem gene duplication generating the key SI receptor SRK.

We conclude that the syntenic Tarenaya gene Th11131, which is most closely related to ARK3 with which it shares protein architecture, is similar to the ancestral version of SI locus genes in Brassicaceae. We hypothesize that the origin of the S-locus is due to a local rearrangement/duplication of an ARK3-like gene and subsequent expansion and diversification of the S-locus. Since ARK3 in Arabidopsis is not expressed in the stigma, the regulatory domains needed for tissue-specific expression may potentially be derived from the concurrent duplication of elements from neighboring genes.

DISCUSSION

The completed genome of A. thaliana was a major milestone in plant biology and has provided a key tool for understanding plant gene function and genome organization (Arabidopsis Genome Initiative, 2000). Subsequently, there has been a great effort and interest to sequence other crucifer species to leverage the knowledge gained from Arabidopsis to other species in a comparative context. To date, there are more completed crucifer genomes (including the crop B. rapa; Wang et al., 2011) than any other eudicot plant family, and ambitious plans exist to sequence many more (such as the Brassica Map Alignment Project; Pires et al., 2013). We sequenced T. hassleriana, a phylogenetic outgroup of the Brassicaceae sister family, the Cleomaceae. We have shown that the vast majority of genes in Tarenaya have clear homologs within Brassicaceae. We provide several examples of how this sister-group genome can be used to elucidate patterns of gene, genome, and trait evolution within the Brassicaceae. Specifically, we focused on independent ancient polyploidy events, floral MADS box, and SI gene evolution. The genome of T. hassleriana will facilitate future research into the evolutionary and functional history of Arabidopsis genes and pathways.

While it has long been known that there are numerous recent polyploid plants (Jiao et al., 2011), with the arrival of the genomics era, it has become clear that there also is extensive evidence for ancient polyploidy across the tree of life (Kasahara, 2007; Jiao et al., 2011; Murat et al., 2012). Most ancient plant polyploid events that have been identified are ancient tetraploidy events (duplications such as At-α), but there are at least four published genome analyses of ancient plant hexaploidy (triplication events): at the base of the eudicots (At-γ) (Jaillon et al., 2007), in tomato (T) (2012), in Brassica (Br-α) (Wang et al., 2011), and this report of the ancient genome triplication in the Tarenaya (Tr-α). The Br-α and Tr-α events are of approximately the same age, allowing us to contrast independent ancient hexaploidy events from closely related lineages. The analysis of the retention of replicated genes (syntelogs) of the Type II MADS box genes provides a compelling example of what can be deduced by the comparison of these independent ancient triplications. We found that Brassica retains nearly twice as many Type-II MADS box genes as does Tarenaya. Genes retained after polyploidy often are dosage-sensitive gene complexes, whereby interacting partners must be maintained in the proper ratios (Edger and Pires, 2009). Considering the wealth of phenotypic diversity seen in the Brassica crops, we hypothesize that this great enrichment of morphological regulators in Brassica derived from Br-α may play a significant role. Also, the recently released genome of L. alabamica of the Brassicaceae also revealed an ancient hexaploidy (La-α) (Haudry et al. 2013). Future comparisons of the closely related La-α, Br-α, and Th-α events should be fruitful.

Comparative analyses can also be used to identify important gene transposition and deletion events. Type-II MADS box genes are remarkably resistant to gene transpositions and thus their collinearity is highly conserved across all angiosperms (Type-I MADS box genes are highly prone to transposition, but their functions are less known). When comparing Tarenaya to Brassicaceae, we found almost all Type-II MADS box genes are collinear, except for the B-class homologs of AP3 and PI. Specifically, we established that there was a Brassicaceae-specific transposition of AP3 and a rare tandem duplication of a floral MADS box gene of AP3 homologs in Tarenaya. The transposition of AP3 also involved a neighboring Forkhead gene, which may be coregulated and important for AP3 function. We further demonstrate that Tarenaya has two homologs of PI: one that is syntenic with other eudicots and for which the locus is lost from Brassicaceae and one that is syntenic with Brassicaceae PI homologs. Both PI and AP3 in Tarenaya have diverged in expression levels. MADS box B-class gene homologs in the AP3 lineage as well as TCP members have been implicated in the establishment of monosymmetric flowers in monocots, including Orchidaceae species (Tsai et al., 2004, 2008; Mondragón-Palomino and Theissen, 2009; Bartlett and Specht, 2010; Preston and Hileman, 2012), PI genes have contributed to floral diversification in Asterids (Viaene et al., 2009), and B-class genes have contributed to Aquilegia floral diversification (Kramer et al., 2007). Thus, it is possible that both B-class and TCP genes have an impact on floral monosymmetry in Cleomaceae. Furthermore, B-class gene diversification has also been implicated in regulating floral gender shifts (Ackerman et al., 2008) and could similarly have diversified in Cleomaceae.

By comparing Brassicaceae genomes with the Tarenaya genome, we established that SRK in the S-locus occurred via a local rearrangement/duplication of an ARK3-like gene and subsequent expansion and diversification in Brassicaceae. In Tarenaya, we can clearly identify syntenic regions that contain homologous functional genes, including an ARK3 and CRK homologs. The exact function of ARK3 is not known, but based on gene expression analysis, this gene is thought to function during development of the sporophyte, perhaps in processes related to organ maturation, the establishment of growth pattern transitions (Dwyer et al., 1994), and/or involvement in pathogen responses (Pastuglia et al., 2002). However, it is not expressed in the stigma. Further research on Brassicaceae and Cleomaceae ARK3 homologs is needed to establish the function of these genes. Interestingly, while Cleomaceae species do not have a SI system, many species, including T. hassleriana, are polygamous (trimonoecious) and can have flowers on the same inflorescence with different genders: male sterile, female sterile, or complete (Stout, 1923; Cruden and Lloyd, 1995; Machado et al., 2006), providing an alternative mechanism to reduce inbreeding.

Our results demonstrate the utility of the Tarenaya genome in complementing Brassicaceae genetic research in efforts to understand the function and evolution of genes and traits. The Tarenaya genome sequence will also pave the way for further studies of Cleomaceae traits not found in Brassicaceae, such as the evolution of C₄ photosynthesis (Brown et al., 2005; Marshall et al., 2007).

METHODS

Sample Preparation, Library Construction, Genome Sequencing, and Assembly

The purple-flowered Tarenaya hassleriana (Purple Queen) line selected for sequencing (ES1100) was first inbred by hand-pollination and floral bagging for four generations. Earlier generations of this line have previously been used for both BAC library construction and limited BAC sequencing (Schranz and Mitchell-Olds, 2006) and transcriptome sequencing and analysis (Barker et al., 2009); however, the material was referred to as being from Cleome spinosa. The two species are morphologically very similar, with only slight differences in stem-spine morphology, pubescence of sepals, ovary and capsules, and flower color; C. spinosa has only white flowers and the sepal and ovary are glandular-pubescent, whereas T. hassleriana can be white, pink, or purple and the sepals and ovary are glabrous (have no pubescence). Thus, many commercial seed providers erroneously label their T. hassleriana material as C. spinosa.

We extracted DNA from leaves of T. hassleriana and constructed seven pair-end libraries with insert sizes of 350 bp, 500 bp, 800 bp, 2 kb, 5 kb, 10 kb, and 20 kb. Illumina Hisequation 2000 was then applied to sequence those DNA libraries, and 70.22-Gb raw data were generated. Low-quality reads, reads with adaptor sequences, and duplicated reads were filtered, and the remaining high quality data were used in the assembly. SOAPdenovo2.21 was applied to assemble the genome in the procedure of contig construction, scaffold construction, and gap closure. After gap closure, the assembly was broken down into contigs again according to the position of Ns in the assembly.

WGP and Physical Map Construction

The BAC library was prepared using leaf material of T. hassleriana. Two BAC libraries were subsequently generated: the first using HindIII (CLEH library) and the second using EcoRI (CLEE library). Average insert sizes were 145 kb for the CLEH library and 130 kb for the CLEE library. The vector used for library construction was pCC1BAC (Epicentre). For each library, 9600 clones were picked and arrayed into 384 well plates. Together, the two libraries equal ∼8.8 genome equivalents (4.6 GE CLEH library and 4.2 GE CLEE library) at an estimated haploid genome size of 300 Mb. WGP was performed according to the methods detailed by van Oeveren et al. (2011). The resulting FPC map was used in further analysis.

FPC Map Assembly and Integration with de Novo Assembled Scaffolds

Sequence-based physical BAC maps were assembled using an improved version of FPC software (Keygene), capable of processing sequence-based BAC fingerprint (WGP) data instead of fragment mobility information as used in the original FPC software (Soderlund et al., 1997). The scaffolds from the SOAPdenovo assembly were then mapped to the physical contigs using nucleotide BLAST (Altschul et al., 1997). Hits were used only when they had a 100% identity match to the anchors. Subsequent filtering was performed to eliminate anchors with multiple hits and to establish superscaffold strand direction. The scaffolds were then ordered according to the mapped anchors and reassembled into superscaffolds.

RNA-Seq

A mixed sample from five tissues (buds, leaves, petioles, stems, and flowers) from Tarenaya flowering plants was used to isolate RNA. Total RNA was extracted using Trizol (Invitrogen). The isolated RNA was then treated with RNase-Free DNase and then subsequently with an Illumina mRNA-Seq Prep Kit, following the manufacturer’s instructions. The insert size of the RNA libraries was ∼200 bp, and the sequencing was done using Illumina GA II. Raw reads were filtered if there were adaptor contaminations and low quality (>10% bases with unknown quality). After filtering, all RNA reads were mapped back to the reference genome using Tophat Version 1.3.3 (Trapnell et al., 2009), implemented with bowtie66 version 0.12.7 (Langmead, 2010), and the transcripts were assembled according to the genome using Cufflinks (version 1.1.0) (Trapnell et al., 2012). Single libraries from eight different tissues were isolated with the Qiagen RNeasy plant mini kit and treated with RNase-free DNase. Illumina TruSeq Libraries were constructed according to the manufacturer’s suggestions and sequenced. Raw reads were filtered to remove adaptors and low-quality bases and mapped to the predicted coding sequences using Cufflinks.

Plant Genome Sources

In all analyses where plant genomes are used, source database and version information can be found in Supplemental Table 18 online.

Gene Annotation

Gene models were predicted following several steps: (1) De novo gene prediction. De novo predictions were performed on repeat masked genome assembly. AUGUSTUS (version 2.03) (Stanke and Morgenstern, 2005), GlimmerHMM (version 3.02) (Majoros et al., 2004), and SNAP (version 2.0) were used to perform the de novo annotation. (2) Homology gene prediction. The protein sequences from Arabidopsis thaliana, Brassica rapa, Carica papaya, Glycine max, Theobroma cacao, and Vitis vinifera were mapped to the Tarenaya genome using tBLASTn, by an E-value cutoff of 10⁻⁵, and then Genewise (version 2.2.0) (Birney et al., 2004) was used for gene annotation. (3) RNA aided annotation. All the RNA reads were mapped back to the reference genome by Tophat (version 1.0.14) (Trapnell et al., 2009), implemented with bowtie version 0.12.5, and the transcripts were assembled according to the genome using Cufflinks (version 0.8.2) (Trapnell et al., 2012). All the predictions were combined using GLEAN to produce the consensus gene sets.

The tRNA genes were identified by tRNAscan-SE (Lowe and Eddy, 1997). For rRNA identification, the Arabidopsis rRNA sequences were first downloaded from the National Center for Biotechnology Information (http://www.ncbi.nlm.nih.gov/nuccore). Then, rRNAs in the database were aligned against the Tarenaya genome using BLASTn to identify possible rRNAs. Other noncoding RNAs, including microRNA and small nuclear RNA, were identified using INFERNAL (Nawrocki et al., 2009) by searching against the Rfam database.

Gene Family Clustering

OrthoMCL (version 1.4) (Li et al., 2003) was used with default parameters followed by an all-versus-all BLASTP (E-value ≤ 1e-5) process using the protein sequence data sets from six plant species. Splice variants were removed from the data set (the longest protein sequence prediction was usually maintained), and the internal stop codons and incompatible reading frames were filtered. All gene families were classified according to the presence or absence of genes for specific species, and it was determined which gene families were species specific or genus specific.

A total of 184204 sequences from A. thaliana, Arabidopsis lyrata, B. rapa, C. papaya, Vitis vinifera, and T. hassleriana were clustered into 24,591 gene families. Of these, 9395 contained sequences from all six genomes, 2492 from Brassicaceae (A. thaliana, A. lyrata, and B. rapa), 1176 from plants as outgroups only bearing the At-γ event (C. papaya and V. vinifera), and 748 clusters were specific to Tarenaya. Of the 28,917 protein-coding genes predicted for Tarenaya, 22,482 were clustered in a total of 14,505 groups. The 748 Tarenaya-specific clusters contained 1556 genes, of which 529 have at least one INTERPRO domain. Singletons make up a total of 6435 genes, of which 2926 have at least one INTERPRO domain. Interestingly, many gene families that have decreased in number of genes in Tarenaya show bigger genes than others, containing more exons and more TE insertions. However, more Gene Ontology annotations are enriched for these contraction gene families. These results indicated that these contraction gene families may be functionally constrained.

Repeat Annotation

Repeats of the Tarenaya genome were identified by a combination of homology-based and de novo approaches. In the homology-based method, databases of known repetitive sequences were used to search against the genome assembly. In this way, RepeatMasker-3.2.9 and RepeatProteinMask software (Chen, 2004) were used to build the homology database and search the repeat sequence.

Furthermore, in the de novo approach, three de novo software packages (Piler-DF-1.0 [Edgar and Myers, 2005], RepeatScout-1.0.5, and LTR-FINDER-1.0.5 [Xu and Wang, 2007]) were used to build a de novo repeat database of the Tarenaya genome. RepeatMasker was then used to identify repeats using both the repeat database that we built in-house and Repbase. Finally, the de novo prediction and the homolog prediction of TEs according to the position in the genome were combined.

Phylogenetic Analysis and Species Divergence Time Estimation

The maximum likelihood phylogenetic tree of T. hassleriana and other plant genomes was constructed using whole-genome fourfold degenerate sites among species. Oryza sativa and Sorghum bicolor were taken as the monocot outgroups. The following steps were taken: First, all of the single-copy gene families were extracted from the OrthoMCL clustering results. Second, multiple sequence alignments were run for each single copy gene family using the protein-coding sequences. Third, for each aligned gene family, the CDS back-translation of the protein multiple alignments was performed from the original DNA sequences using in-house Perl scripts, and then the fourfold degenerate sites of orthologous genes in all single-copy gene families (concatenated into one supergene for each species) were extracted. The branch length represents the neutral divergence rate. The substitution model (GTR+gamma+I) and Mrbayes (Ronquist et al., 2012) were used to reconstruct the phylogenetic tree.

Synteny and Collinearity Analysis

First, a homology search within and between species was performed using BLASTP (E-value threshold 1e-7, top 20 hits). Tandem gene families and weak matches were removed using in-house Perl scripts for further analysis. Tandem gene families were defined as clusters of genes within 10 intervening genes from one another, and the longest model was maintained to represent each family. For the weak matches, only the top BLAST hits were retained by applying a C-score threshold of 0.8 [C-score(A, B) = score(A, B)/max(best score of A, best score of B)] (Putnam et al., 2007).

Then, based on these filtered BLAST results, the whole genome-wide sequence alignments within and between genomes using genes as anchors, which was to search syntenic blocks, were conducted by an in-house pipeline implementing Dynamic Programming (parameters: score_of_match, 50; penalty of mismatch, −5; penalty of indel, −5; penalty_of_extension_indel, −2; block_size, ≥5 gene pairs; gap_between_neighbor_blocks, 30 genes). The running time for each whole-genome sequence alignment using genes as anchors by Dynamic programming was ∼5 to ∼6 h. At the same time, i-Adhore 3.0 (Proost et al., 2012) (http://bioinformatics.psb.ugent.be/software) was used to identify syntenic and collinearity blocks (gap_size = 30, cluster_gap = 35, q_value = 0.75, prob_cutoff = 0.01, anchor_points = 5, alignment_method = gg4, level_2_only = false, table_type = family) within and between genomes, and all the syntenic blocks identified by i-adhore were contained in our results identified by the Dynamic programming method; however, the latter method is more sensitive and accurate. All the dot plot figures were plotted using a Support Vector Graphics package implemented perl scripts (see Supplemental Figures 1, 2, 6 to 9, and 11 to 14 online).

Ancestral Genome Reconstruction

Based on the paralogous duplicates within each genome, four minimized genomes were independently created for A. thaliana, A. lyrata, B. rapa, and T. hassleriana, by condensing local duplications to one gene, removing transposons, and including only genes within blocks defined by retained pairs. Each of the minimized genomes represents the ancestral state and predate the recent polyploidy event. At the same time, these four ancestral state genomes were compared with the Ken Wolfe’s 45 ancestral blocks of A. thaliana (see Supplemental Figures 11 to 14 online).

Partition of the T. hassleriana Genome into Three Subgenomes Following the Recent Polyploidy Event

To avoid the potentially confounding results with the independent ancient polyploidy events, the ancestor genome of A. lyrata (A ancestor) was used as the reference, and collinear blocks were identified using i-adhore 3.0 by aligning the four proteomes (A. thaliana, A. lyrata, B. rapa, and T. hassleriana) against the A ancestor genome. From the last common ancestor (the A ancestor represents this genome state) of these four species, both A. thaliana and A. lyrata experienced a whole-genome duplication event (At-α), B. rapa experienced one whole-genome duplication event (At-α) and an additional whole-genome triplication event (Br-α), and T. hassleriana experienced an independently whole-genome triplication event (Th-α). Therefore, quote ratios of 2:1, 2:1, 3/4/5/6:1, and 3:1 were observed for A. thaliana, A. lyrata, B. rapa, and T. hassleriana genomes. For T. hassleriana, the triplicated blocks identified were chained into three subgenomes compared with the A ancestor using dynamic programming. The main criteria are that the chained triplicated blocks are (1) nonoverlapping in the T. hassleriana genome; (2) have no more than 10% overlap between their orthologous A ancestor regions (results were similar using 0% overlap in A ancestor); and (3) maximize coverage of the T. hassleriana genome (annotated gene space). A similar strategy was taken for other genomes based on their polyploidy level from the last recent common ancestor, as shown in Supplemental Figures 16 to 19 online. For T. hassleriana, a total of 16,770 (63.2%) Tarenaya genes are in the triplicated blocks compared with the A ancestor, 688 (2.6%) genes are in the duplicated blocks, which indicate that another copy was possibly lost after the triplication event, and only 135 (0.5%) have one syntenic ortholog, which means that a very small subset of triplicated genes lost two copies. However, 8913 (33.6%) Tarenaya genes are not in any replicated blocks, which indicates species-specific deletion in A. lyrata or species-specific gains in Tarenaya after their divergence along evolutionary time.

Interproscan and Gene Functional Annotation

Interproscan version 4.5 was used to scan protein sequences against the protein signatures from InterPro (Hunter et al., 2009) (version 22.0) to infer functions for the protein-coding genes. This was done for the entire target proteomes involved in our main text analysis, including A. thaliana, A. lyrata, T. hassleriana, B. rapa, Solanum lycopersicum, V. vinifera, Prunus persica, Populus trichocarpa, and C. papaya. InterPro integrates protein families, domains, and functional sites from different databases: Pfam, PROSITE, PRINTS, ProDom, SMART, TIGRFAMs, PIRSF, SUPERFAMILY, Gene3D, and PANTHER. Interproscan integrates the searching algorithms of all these databases. In total, Interproscan identified 93,038 protein domains of 4733 distinct domain types. Seventy-five percent of the genes (21,829 out of 28,917 genes in total) have been assigned with at least one domain.

Genomic Analysis for Reproductive Traits

For the syntenic and protein domain analysis of SI genes, the genes annotated in A. thaliana, C. rubella, and A. lyrata were used as published in an extensive study into S-locus variation in Arabidopsis species (Guo et al., 2011). Homologs were then sought using top BLAST hits of these genes against T. hassleriana and B. rapa. All homology candidates were confirmed by manual inspection of the alignment in dot plots generated in MAFFT (Katoh et al., 2002) to confirm synteny of candidate regions. After compiling a definitive list of syntenic regions, the genes from these regions were analyzed through the PFAM online protein domain analysis program (Punta et al., 2012). Figure 6 was manually compiled from the results of this program.

qRT-PCR

For the qRT-PCR, total RNA was isolated from roots, leaves, three bud stages (1 to 5 mm, 5 to 10 mm, and 10 to 25 mm length), sepals, petals, stamens, carpels at anthesis, and three stages of siliques (10 mm, 10 to 30 mm, and 30 to 50 mm length) with the GeneJET plant RNA purification mini kit (Fermentas). First-strand cDNA was synthesized using 500 ng total RNA with the RevertAid H Minus First Strand cDNA synthesis kit (Fermentas) using random hexamer primers.

The qRT-PCR experiments were performed according to the MIQE guidelines (Bustin et al., 2009). Exon spanning primers were generated using PerlPrimer 1.1.21. (Marshall, 2004). A primer efficiency test was performed and the primers were tested with genomic DNA to ensure cDNA specificity. Standard dose response curves were constructed for all genes using serial dilutions (1:50 to 1:50,000) of 10- to 25-mm-long bud cDNA template to calculate amplification efficiency. The qRT-PCR assay was performed with the LightCycler 480 II (Roche) and the data analyzed with the LCS480 1.5.0.39 software. Each reaction was composed of 10 μL of 2× DyNAmo Flash SYBR Green Mastermix (Biozym Scientific), 2 μL each of 10 μM forward and reverse primers, 1 μL water, and 5 μL of 1:100 diluted template cDNA. Each reaction was performed in biological duplicate and technical triplicate along with water and RNA controls for each primer pair. The GLYCERALDEHYDE-3-PHOSPHATE DEHYDROGENASE C SUBUNIT and ELONGATION FACTOR 1-ALPHA genes served as internal controls. The following PCR program was used: 7 min at 95°C; 45 cycles of 10 s at 95°C, 15 s at 60°C, and 15 s at 72°C, followed by a melting curve of 5 s at 95°C, 1 min at 65°C, and 30 s at 97°C. The quantification cycles (C_q) were calculated according to the second derivative maximum algorithm. The raw C_q data were analyzed according to the Comparitive C_q method (ΔΔC_q) (Schmittgen and Livak, 2008). Gene expression was first normalized relative to the expression of the two reference genes in the respective tissues. The expression was further normalized with the expression of the reference genes in 10- to 25-mm long buds, which acted as an interassay calibrator. The relative expression was then calculated with reference to the expression of T. hassleriana CAL in stage 3 buds.

Accession Numbers

Sequence data from this article can be found in the Arabidopsis Genome Initiative or GenBank/EMBL databases under the following accession numbers: Th-CAL/AP1-1, Th01189; Th-CAL/AP1-2, Th13754; Th-PI-1, Th05675; Th-PI-2, Th17298; Th-AP3-1, Th02920; Th-AP3-2, Th02921; EMB196, AT3G54350; and TCP1, AT1G67260. The genomic reads of T. hassleriana, as well as RNA sequencing data, have been deposited into the NCBI Short Read Archive under accession numbers SRA058749 and GSM1008474. The information for the raw reads data can be found in Supplemental Table 1 online. The genome sequence and annotation data set have been deposited into NCBI (project ID PRJNA175230 [superscaffolds]; the accession number is AOUI00000000).

Supplemental Data

The following materials are available in the online version of this article.

• Supplemental Figure 1. Evaluation of Superscaffold13 Integrated by the Combination of SOAPdenovo Assembled Scaffolds and the Physical Map as an Example of Our Verification Method.

• Supplemental Figure 2. Comparison of the New T. hassleriana Assembly with Previously Sequenced BACs.

• Supplemental Figure 3. Principal Component Analysis of Gene Expression.

• Supplemental Figure 4. Comparison of the Distribution of Gene Features (Including mRNA Length, CDS Length, Exon Length, Intron Length) among Several Selected Angiosperms.

• Supplemental Figure 5. Graph Representing Contig Length versus LTR (from Retrotransposons) Density in the T. hassleriana Assembly.

• Supplemental Figure 6. Dot Plot Figure to Show Syntenic Duplicates within the T. hassleriana Genome.

• Supplemental Figure 7. Dot Plot Figure to Show the Collinear Blocks between A. thaliana and T. hassleriana Genomes.

• Supplemental Figure 8. Dot Plot Figure to Show the Collinear Blocks between A. lyrata and T. hassleriana Genomes.

• Supplemental Figure 9. Dot Plot Figure to Show the Collinear Blocks between B. rapa and T. hassleriana Genomes.

• Supplemental Figure 10. Homologous Blocks within and between Genomes Based on the Chromosome Layout of A. lyrata.

• Supplemental Figure 11. Dot Plot Figure to Show the Corresponding Homologous Relationship between “Wolfe” Blocks and the 64 Ancestor Genome Blocks of A. thaliana Identified in This Article and Named “A Ancestor.”

• Supplemental Figure 12. Dot Plot Figure to Show the Corresponding Homologous Relationship between the “Wolfe” Blocks and the 61 Ancestor Genome Blocks of A. lyrata Identified in This Article, Forming the Alternative “A” Ancestor Used in Subgenome Analyses.

• Supplemental Figure 13. Dot Plot Figure to Show the Corresponding Homologous Relationship between the “Wolfe” Blocks and the 71 Ancestor Genome Blocks of B. rapa Identified in This Article Named “B Ancestor.”

• Supplemental Figure 14. Dot Plot Figure to Show the Corresponding Homologous Relationship between “Wolfe” Blocks and the 87 Ancestor Genome Blocks of Tarenaya hassleriana Identified in This Article Named “C Ancestor.”

• Supplemental Figure 15. Multiple Homologous Relationships between the Ancestor Genome Blocks (“Wolfe” Ancestral Blocks, A Ancestral Blocks of A. thaliana, A Ancestral Blocks of A. lyrata, B Ancestral Blocks of B. rapa, C Ancestral Blocks of T. hassleriana) and the Reference Genome A. thaliana.

• Supplemental Figure 16. The Genome of A. lyrata Is Partitioned into Two Subgenomes (Based on the At-α Event) by Comparing against the Reference Ancestor Genome of A. lyrata.

• Supplemental Figure 17. The Genome of A. thaliana Is Partitioned into Two Subgenomes (Based on the At-α Event) by Comparing against the Reference Ancestor Genome of A. lyrata.

• Supplemental Figure 18. The Genome of B. rapa Is Partitioned into Three Subgenomes (Based on the Br-α Event) by Comparing against the Reference Ancestor Genome of A. lyrata.

• Supplemental Figure 19. The Genome of T. hassleriana Is Partitioned into Three Subgenomes (Based on the Ch-α Event) by Comparing against the Reference Ancestor Genome of A. lyrata.

• Supplemental Figure 20. Complete ML Tree of MADS Type II Homologs in All Sequenced Angiosperms.

• Supplemental Figure 21. SHP1, Two Gene Copies, and Synteny.

• Supplemental Figure 22. AP1/CAL Gene Copies and Synteny.

• Supplemental Figure 23. Synteny Plot of the Tarenaya Region Containing Th02920/Th02921 (AP3 Genes).

• Supplemental Figure 24. Synteny Plot of the Tarenaya Region Containing Th17298 (PI Gene).

• Supplemental Figure 25. Simplified Synteny Plots of AP3 and PI Genes in Tarenaya.

• Supplemental Figure 26. TCP1 Gene Copies and Synteny.

• Supplemental Figure 27. Histogram of Gene Expression in Various Tarenaya Tissues.

• Supplemental Figure 28. SI Locus Domain–Related Genes in Syntenic Block Context in Brassiaceae and Tarenaya.

• Supplemental Table 1. Library Construction and Sequencing (Raw data).

• Supplemental Table 2. Statistics of the Generation of the Clean Data.

• Supplemental Table 3. Summary Statistics of the Preliminary Genome Assembly.

• Supplemental Table 4. Evaluation of Gene Region Coverage by RNA Mapping.

• Supplemental Table 5. Summary of Results after Deconvolution and Filtering of the Whole-Genome Profiling Tags.

• Supplemental Table 6. FPC Assembly Results for the High, Reduced, and Low-Stringency WGP Assemblies.

• Supplemental Table 7. Final Contig and Scaffold Median Lengths after Superscaffold Construction through Integration with the Physical Map.

• Supplemental Table 8. Summary Statistics of RNA-Seq Sequencing Data and the Alignments Mapping to Genes and Genome.

• Supplemental Table 9. Mapping Statistics and Transcript Dynamics for Eight RNA Samples.

• Supplemental Table 10. Summary Statistics of the Results from Different Gene Annotation Strategies.

• Supplemental Table 11. Summary of Functional Annotations Derived from Interproscan.

• Supplemental Table 12. Summary of smRNA Annotation.

• Supplemental Table 13. Comparison of T. hassleriana Assembly with Nine Other Sequenced Crucifers.

• Supplemental Table 14. Classification of Transposable Elements within the T. hassleriana Genome.

• Supplemental Table 15. Statistics of the Ancestor Genome Construction within A. thaliana, A. lyrata, B. rapa, and T. hassleriana.

• Supplemental Table 16. Syntenic and Collinearity Analysis between Tarenaya and Other Species.

• Supplemental Table 17. Assembly Coverage of S-Locus-Like Regions in T. hassleriana.

• Supplemental Table 18. Plant Genome Source Databases.

• Supplemental Data Set 1. List of Syntenic Order of Tarenaya Genes in the Ancestor Block Configuration of the Genome.

• Supplemental Data Set 2. MADS Box Gene Alignment as Used for the ML Phylogenetic Tree in Supplemental Figure 20.

Glossary

SI

self-incompatibility

WGP

whole-genome profiling

TE

transposable element

qRT-PCR

quantitative RT-PCR

NCBI

National Center for Biotechnology Information