Transcriptome analysis of colored calla lily (Zantedeschia rehmannii Engl.) by Illumina sequencing: de novo assembly, annotation and EST-SSR marker development

Colored calla lily is the short name for the species or hybrids in section Aestivae of genus Zantedeschia. It is currently one of the most popular flower plants in the world due to its beautiful flower spathe and long postharvest life. However, little genomic information and few molecular markers are available for its genetic improvement. Here, de novo transcriptome sequencing was performed to produce large transcript sequences for Z. rehmannii cv. ‘Rehmannii’ using an Illumina HiSeq 2000 instrument. More than 59.9 million cDNA sequence reads were obtained and assembled into 39,298 unigenes with an average length of 1,038 bp. Among these, 21,077 unigenes showed significant similarity to protein sequences in the non-redundant protein database (Nr) and in the Swiss-Prot, Gene Ontology (GO), Cluster of Orthologous Group (COG) and Kyoto Encyclopedia of Genes and Genomes (KEGG) databases. Moreover, a total of 117 unique transcripts were then defined that might regulate the flower spathe development of colored calla lily. Additionally, 9,933 simple sequence repeats (SSRs) and 7,162 single nucleotide polymorphisms (SNPs) were identified as putative molecular markers. High-quality primers for 200 SSR loci were designed and selected, of which 58 amplified reproducible amplicons were polymorphic among 21 accessions of colored calla lily. The sequence information and molecular markers in the present study will provide valuable resources for genetic diversity analysis, germplasm characterization and marker-assisted selection in the genus Zantedeschia.


INTRODUCTION
the favored markers for the detection of allelic diversity. Among these markers, SSR markers have gained attention due to their multi-allelic nature, abundance throughout the genome, high reproducibility and polymorphism, adaptability to automation and high-throughput genotyping (Morgante, Hanafey & Powell, 2002). Nevertheless, progress in developing SSR markers in section Aestivae and the genus Zantedeschia is limited compared to that in other crop plants. Only one previous effort has produced 43 polymorphic SSRs from 4,394 EST sequences of Z. aethiopica, representing the first dataset of polymorphic EST-SSR markers for the genus Zantedeschia (Wei et al., 2012). This conventional method of EST-SSR development has a high development cost and a low throughput, which may restrict the use of SSR markers in further genetic breeding (Morgante, Hanafey & Powell, 2002;Varshney, Graner & Sorrells, 2005).
Recent advances in RNA sequencing (RNA-seq) technology and de novo assembly provide an excellent strategy for the efficient and cost-effective development of highthroughput EST-SSRs. This strategy can enrich the large amounts of expressed sequence data for non-model organisms for which the reference genome or transcriptome is not available (Martin & Wang, 2011). Several non-model organisms in the Araceae family, including Z. aethiopica (Cândido et al., 2014) Arum concinnatum (Onda et al., 2015), Anthurium andraeanum (Tian et al., 2013;Yang et al., 2015;Li et al., 2015), Pinellia ternate (Wang et al., 2014), Amorphophallus konjac and A. bulbifer (Zheng et al., 2013;Diao et al., 2014), have been recently studied by transcriptome sequencing, which has provided a better understanding of these crops. EST-SSR markers have been successfully characterized and developed in P. ternate (Wang et al., 2014), A. konjac and A. bulbifer (Zheng et al., 2013). A total of 14,468 and 19,596 EST-SSRs were identified in 12,000 and 16,027 nonredundant transcriptome unigenes, respectively, of P. ternate and two Amorphophallus species (Wang et al., 2014;Zheng et al., 2013). Furthermore, 320 primers were synthesized and used to validate the amplification and assessment of polymorphisms in 25 individual Amorphophallus spp. plants (Zheng et al., 2013), of which 275 primer pairs yielded PCR amplification products and 205 were polymorphic. This strongly demonstrated that de novo assembly based on RNA-seq can offer a simple, direct and reliable approach for the identification and development of massive unigene-based microsatellite markers with diverse motifs. In addition to the development of molecular markers, transcriptome sequencing has also been used for the discovery, profiling and quantification of RNA transcripts and novel genes. For example, three lipid transfer proteins (LTP) that are potentially involved in defense against pathogens or predators were identified by the in silico screening of the 83,578 transcriptome contigs of Z. aethiopica (Cândido et al., 2014). This represents the first transcriptome dataset for the genus Zantedeschia.
Here, we presented the generation of a large expressed sequence dataset based on Illumina HiSeq TM 2000 sequencing data from mixed tissues of colored calla lily, Z. rehmannii cv. 'Rehmannii'. The objectives were as follows: (1) to characterize and annotate the transcriptome information of colored calla lily; (2) to identify all of the candidate genes encoding enzymes or putative transcription factors that are involved in its flower spathe development; (3) to analyze the frequency and distribution of SSRs and SNPs in its transcribed regions; (4) to develop a large set of and (5) to evaluate the

RNA extraction, cDNA library construction and Illumina sequencing
To achieve a comprehensive collection of expressed transcripts of colored calla lily, ten 'Rehmannii' tissue types, including root, tuber, stem, leaf, bud, spathe, pistillate inflorescence and staminate inflorescence, immature seed and mature seed, were separately harvested at the anthesis and wilting stages. All of the tissue samples were immediately frozen in liquid nitrogen and stored at −80 • C until RNA extraction. Total RNA Trizol Reagent (Invitrogen, USA) was used to extract RNA from all of the samples, following the manufacturer's instructions. The RNA purity and concentration were determined using a NanoDrop spectrophotometer (Thermo, USA). Equal quantities of total RNA from each sample were then pooled together and used for transcriptome sequencing.
cDNA library construction and Illumina-based sequencing were performed by the Shanghai Majorbio Bio-pharm Biotechnology Co. Ltd. (Shanghai, China) according to the manufacturer's instructions. The paired-end cDNA sequencing libraries with approximate average insert lengths of 200 base pairs were prepared from the total RNA, as per the protocol of the Genomic Sample Prep kit (Illumina, San Diego, CA). This process started with mRNA fragmentation, followed by reverse transcription, first-and second-strand synthesis, paired-end adapter ligation and PCR amplification. Library quantification and quality assessment were performed on an Agilent 2100 Bioanalyzer and an ABI StepOnePlus Real-Time PCR System. Finally, the cDNA library was sequenced as 101-mer×2 on an Illumina HiSeq 2000 using paired-end sequencing chemistry.

De novo transcriptome assembly and functional annotation
The raw sequencing image data were transformed into raw reads and stored in FASTQ format. These raw data were then filtered and deposited in the National Center for Biotechnology (NCBI) Sequence Read Archive (SRA) under accession number SRR3310941. All of the reads with adaptor contamination, empty reads, non-coding RNA (such as rRNA, tRNA and miRNA), ambiguous nucleotides comprising more than 5% or low quality value (QV), i.e., an average QV of less than 20 (QV < 20), were discarded or filtered. Then, the clean and high-quality transcriptome sequence data were de novo assembled using the short reads assembling program Trinity (Haas et al., 2013) with default settings. An assembled transcripts database (File S1) was finally achieved and arranged according to the gene family clustering analysis. These final assembled transcripts were used for further bioinformatics analysis.
The assembled sequences were functionally annotated by BLASTx against a series of databases, including the Nr, Swiss-Prot, COG, and KEGG databases, with a common significance threshold cut-off of E-value 1e-10 −5 . Based on Nr annotation, Blast2GO (https://www.blast2go.com/) and WEGO (http://wego.genomics.org.cn/cgi-bin/wegol) software were used to retrieve and classify GO annotation categories defined by molecular function, cellular component and biological process.

Marker locus detection and SSR primer pair design
MIcroSAtellite identification tool (MISA, http://pgrc.ipk-gatersleben.de/misa/misa.html) was used to identify microsatellites in the assembled transcripts. The minimum number of repeats used to select the SSRs was ten for mononucleotide repeats, six for dinucleotide repeats, and five for tri-, tetra-, penta-, and hexanucleotide repeats. Primer Premier 5.0 (PREMIER Biosoft International, Palo Alto, CA) was then used to manually design 200 pairs of PCR primers to randomly select sequences with SSR loci. The design criteria of the primers were as follows: primer length 18-24 bp; GC content 40-65%; melting temperature 50-65 • C; and expected product size 100-300 bp with no secondary structures. All 200 primer pairs were synthesized by Sangon Biological Engineering Technology (Shanghai, China).

DNA extraction and EST-SSR marker amplification
The young leaves were collected from 21 accessions (Table S1) grown at BPFG. The total genomic DNA was extracted using the DNeasy Plant Mini Kit (Zexing Biotech, China) following the manufacturer's protocol. The quantity and quality of DNA were evaluated using a Nanodrop ND 1,000 spectrophotometer (Thermo Scientific, USA). The DNA was adjusted to a concentration of 20 ng/µl and stored at −20 • C until use.
The SSR amplification reactions were conducted as described previously by Wei et al. (2012) with little modification. The PCR reaction mixtures (10 µl) contained 1× PCR buffer, 20-30 ng of template DNA, 0.8 mM MgCl2, 15 µM dNTPs, 0.25 µM each primer, and 0.2 U of Taq DNA polymerase (Zexing Biotech, China). PCR amplifications were performed in the GeneAmp PCR System 9700 (Applied Biosystems). The thermal profile included an initial denaturation at 95 • C for 5 min, followed by 20 cycles of 95 • C for 30 s, 50-60 • C for 45 s, and 72 • C for 60 s and a final extension at 72 • C for 10 min. The PCR products were separated on 8.0% polyacrylamide non-denaturing gels and visualized by silver staining (Wei et al., 2012). The product sizes were determined by comparison with a 34-to 501-bp pUC19/MspI DNA marker (Zexing Biotech, China).

Illumina sequencing and de novo assembly
To provide a comprehensive transcriptome database for calla lily, a mixed cDNA library of cultivar 'Rehmannii' from ten sampled tissues, i.e., root, tuber, stem, leaf, bud, spathe, pistillate inflorescence and staminate inflorescence, immature seed and mature seed, was constructed and sequenced using Illumina paired-end technology. A summary of the sequencing output statistics is shown in Table 1. The sequencing yielded approximately 59.9 million raw reads with a total of more than 6.1 billion nucleotides. After removing ambiguous, low-quality reads and reads with adaptors, 46.3 clean reads were obtained for further assembly. The de novo assembly yielded 62,382 transcripts and 39,298 unigenes. The sequences of all the assembled unigenes were provided in FASTA format in File S1. These sequences contained 40.8 Mb of sequence with an average size of 1,038 bp and an N50 of 1,476 bp. The lengths of the unigenes ranged from 351 to 15,521 bp. Of these unigenes, 67.3% (26,455) were shorter than 1,000 bp, 20.3% (7,963) ranged from 1,000 to 2,000 bp, and the remaining 12.4% (4,880) were longer than 2,000 bp (Table 1). These unigenes formed a potential pool for identification of genes and functional molecular markers in colored calla lily.

Functional classification of the unigenes
All of the assembled unigenes were subjected to a search against the COG, GO and KEGG databases for further functional prediction and classification (Table S2). The COG database can provide phylogenetic classification of proteins encoded by several complete genomes of bacteria, archaea and eukaryotes (Roman et al., 2000). The COG function classification of the 'Rehmannii' sequences is shown in Fig. 2A. In total, 6,731 of the 39,298 unigenes showing Nr hits were functionally annotated and classified into 24 COG categories, including cellular structure, biochemistry metabolism, molecular processing, and signal transduction, among others. Given that 9,379 COG-annotated putative proteins were obtained, some of these unigenes were assigned to multiple COG classifications. The cluster for general function prediction only (1,791, 26.6%) represented the largest group, followed by replication, recombination and repair (904, 13.4%) and transcription (836, 12.4%). Additionally, only a few unigenes were assigned to cell motility (9, 0.13%) and nuclear structure (1, 0.01%). The GO database, an internationally standardized gene functional classification system, offers dynamic and updated gene ontology that defines gene products in terms of their associated cellular component, molecular function, and biological process in any organism (Ashburner et al., 2000). Based on Nr annotation, 15,552 (39.6%), unigenes in the present study were assigned to GO classes with 15,964 functional terms. A summary of the 'Rehmannii' unigenes classified to each GO Slim term is shown in Fig. 2B. The annotated gene sequences that belong to the cellular component, molecular function, and biological process categories were divided into 47 functional groups. Under the cellular component category, cell (4,254, 27.4%) and cell part (4,253, 27.4%) were the most highly represented group, followed by organelle (3,331, 21.4%). For the molecular function category, the top two groups were binding (3,512, 22.6%) and catalytic activity (3,313, 21.3%). However, the majority of the groups, including antioxidant activity, receptor activity, protein binding transcription factor activity, nutrient reservoir activity, etc., contained only a few unigenes (655, 4.2%). For biological processes, the majority of the unigenes were involved in cellular processes (2,344, 15.1%) and metabolic processes (1,814, 11.7%), indicating that important metabolic and cellular activities occur in 'Rehmannii'. Genes involved in other important biological processes, such as single-organism processes (1,337, 8.6%), stimulus response (712, 4.6%) and biological regulation (416, 2.7%), also accounted for a large fraction of the annotated transcripts. Furthermore, we found that a portion of the unigenes were assigned to developmental processes (226, 1.5%), reproductive processes (63, 0.4%) and growth (25, 0.2%) and may be involved in flower-related biological processes, such as flower-type development and formation, in colored calla lily.
The KEGG database (Kanehisa et al., 2004) determines the biological pathways in which protein and small-molecule interactions occur. A total of 4,532 assembled sequences (11.5%) were consequently annotated and assigned to 117 predicted metabolic pathways. The number of sequences in each pathway ranged from 1 to 160. Fig. 2C shows the top 20 KEGG metabolic pathways represented by unique transcripts of calla lily. Ribosome represented the largest number of transcript sequences in our dataset (160), followed by RNA transport (154) and plant hormone signal transduction (152). Among the annotated sequences, more than one-third (1,661) were classified into 46 metabolic groups, such as purine metabolism (134), pyrimidine metabolism (107), starch and sucrose metabolism (75), and so on. This indicates that diverse metabolic processes are active and a variety of metabolites are synthesized in the tissues and organs of colored calla lily.
Overall, 21,077 unique sequence-based annotations using the selected Nr, Swiss-Prot, COG, GO and KEGG databases were assigned to assembled transcripts of the cultivar 'Rehmannii'. The functional analysis revealed that it is feasible to obtain transcriptome sequences through high-throughput technology, even for non-model plants with large genomes. Furthermore, all of these functional annotation assignments provide valuable information for colored calla lily to investigate specific biochemical and developmental processes and potential gene structures, functions and pathways.

Unigenes related to flower development in colored calla lily
Enzymes functioning in flower development have been well documented in many plants. Based on Nr annotation results of colored calla lily unigenes, we identified a total of 171 candidate transcripts encoding enzymes related to pigment biosynthesis (61), floral organ development (26), flowering regulation (63) and flower senescence (31). The EST sequences of all these unigenes are listed in Table S3. Flower development is a complex process controlled by an integrated network of multi-genetic pathways in higher plants. The putative gene homologs identified in the present study were involved in eight pathways, including the anthocyanin biosynthesis pathway (52), carotenoid biosynthesis pathway (9), specification of floral organ identity (26) , photoperiod pathway (22), vernalization pathway (11), gibberellic acid pathway (11), autonomous or other pathways (9) and ethylene biosynthesis pathway (31). The identification and analysis of these key genes will provide a foundation for understanding the potential molecular genetic mechanisms controlling different aspects of floral development of colored calla lily in the future.

SSR and SNP loci discovery
Transcriptome sequencing is important for identification and development of molecular markers, such as SSRs and SNPs. For the development of new markers for colored calla lily, all of the 39,298 unigenes generated in this study were used to mine potential microsatellites using MISA software. A total of 9,933 potential EST-SSRs were identified in 7,997 unigenes, of which 1,556 sequences contained more than one SSR, and 580 SSRs were present in compound form (Table 2 and Table S4). Considering that approximately 40,780 kb was analyzed, we detected a frequency of at least one SSR per 4.1 kb in the expressed fraction of the 'Rehmannii' genome.
In addition to EST-SSRs, a total of 7,162 potential high-quality SNPs were identified by mapping against 39,298 reference unigenes. The overall frequency of all types of SNPs was one SNP per 5.69 kb. The predicted SNPs included 4,450 transitions and 2,712 transversions ( Table 4). The most abundant SNPs detected were C/T (2,262, 31.6%), followed by A/G (2,188, 30.6%) and C/G (790, 11.0%). The numbers of the remaining three SNP types (A/T, A/C, and T/G) were similar, each accounting for less than 10%. Potential SNPs are shown in Table S5.

SSR primer design, polymorphism detection and phylogenetic analysis
A total of 200 EST-SSR loci (repeat motif >1) with appropriate flanking sequences were randomly selected for the design and synthesis of PCR primer pairs. Detailed information   non-specific products, of which 23 markers generated PCR products larger or smaller than expected and 37 generated more than one band (Table S4). Specific EST-SSR markers were used to assess the genetic diversity and relationships among the 21 accessions of colored calla lily from New Zealand, the Netherlands, the United States and China. Of these tested markers, 58 (or 75.3%) were polymorphic, while the others were monomorphic. The polymorphic EST-SSRs consisted of 22 di-, 18 tri-, 6 tetra-, 4 penta-motif-and 8 compound-motif-based markers. The raw data and characteristics of the 58 polymorphic EST-SSR markers in 21 accessions are listed in Table S7 and Table 5. The number of alleles per marker (Na) ranged from 2 to 6, with 174 alleles in total. The average effective number of alleles per locus (Ne) was 2.163, with a maximum of 4.762 and a minimum of 1.156. The observed heterozygosity (Ho) varied from 0 to 1.000, whereas the expected heterozygosity (He) varied from 0.135 to 0.790. The mean Ho and He values were 0.483 and 0.491, respectively. The polymorphic information content (PIC) values ranged from 0.130 to 0.757 with a mean value of 0.420. Nr annotation results (Table S6) showed that most of the polymorphic SSR-containing sequences shared significant homology to protein-encoding genes from V. vinifera (Vitaceae), T. cacao (Sterculiaceae), Z. mays and O. sativa Indica (Gramineae), among others. These positive-hit homologous genes, including nucleic acid binding protein, glycine-rich protein, abscisic acid, zinc-finger transcription factor, had hypothetical or putative functions in defense and stress, transporters and metabolic processes.
Based on the shared allele distance, we conducted a cluster analysis to assess the genetic relationships between 21 accessions of colored calla lily (Fig. 3). The dendrogram clearly grouped these accessions into three major clusters (I, II and III) at an approximate genetic distance of 0.18. Cluster I consisted of 13 accessions collected from the Netherlands, the United States, New Zealand and China. Twelve accessions in clusters I (Pink Diamond, Rose Gem, Super Gem, Greta, Parfait, Rehmannii, Odessa, Majestic Red, Xiangyuan Red, Captain Reno, Allure, and Captain Romance) had red, purple and pink spathes, whereas Captain Murano had an orange spathe. Clusters II and III consisted of eight yellow-flower accessions. The accessions in the former cluster (Jingcai Yangguang, Black Magic, ZH, Gold affair, Sunny Baby and Elmaro) originated from New Zealand and China, while those in the latter (Goldilocks and Butter Gold) were collected from New Zealand. The results may indicate a potential association between the genetic relationship of the tested accessions and the spathe color.

Illumina transcriptome sequencing and de novo assembly
Zantedeschia spp. are perennial bulb plants known worldwide for their ornamental uses. These plants are even commonly used in traditional medicine in some African regions (Letty, 1973;Singh, 1996). However, little genomic information is available for the plants of the genus Zantedeschia, even with significant advances in DNA sequencing technologies. The genomic researches of Zantedeschia spp. may be hindered by the large genome size of these plants. The nuclear DNA contents of colored calla lily Z. elliottiana and white calla lily Z. aethiopica (Ghimire et al., 2012) were estimated to be 1.17 ± 0.50 and 3.72 ± 0.10 pg DNA/2C, respectively, equivalent to genome sizes of 1.15 and 3.64 Gbp, approximately 2.5 and 7.8 times larger than the rice (O. sativa) genome of 0.47 Gb. Moreover, Zantedeschia spp. may be a genetically high-heterozygosity bulb-flower species. Since these species having been bred for more than one century, a larger number of outstanding cultivars and hybrids of white and colored calla lilies in a range of sizes and colors have been cross-bred, named and released by European and American breeders. Thus, it is obviously not feasible to consider whole-genome sequencing for this perennial plant because of the high cost and time consumption.
For organisms with large, heterozygous and complex genomes containing repetitive sequences, RNA-seq-based de novo transcriptome analysis is an attractive alternative to examine the properties of a transcriptome as a proxy for the whole genome (Martin & Wang, 2011;Onda et al., 2015). A transcriptome study of Z. aethiopica spathe (Cândido et al., 2014) also demonstrated that transcriptome sequencing is a useful molecular biology tool to rapidly build comprehensive sequence resources of expressed genes for in-depth analysis at high resolution. To date, de novo transcriptome analysis has successfully been applied to a wide range of plant species (Martin & Wang, 2011), including crops, fruits, vegetables, forest plants, flowers, and medicinal plants, for various purposes, such as gaining fundamental insight into biological processes, generating different dynamic views of interesting gene expression, discovering novel genes or transcripts, and developing and validating molecular markers. In the present study, Illumina paired-end RNA-seq technology was used to sequence the pooled RNA from ten tissues of 'Rehmannii'. More than 59.9 million 101-bp paired-end reads were yielded, encompassing 6.1 G nucleotides. Trinity software (Haas et al., 2013) was then used to de novo assemble these short reads to generate a total of 39,298 unigenes with an average length of 1,038 bp. The results indicated that de novo transcriptome assembly is a feasible strategy to provide genome information for non-model plant species. In sequence assembly, the N50 value is an important statistical measure that is used to evaluate transcriptome assemblies in which a high number corresponds to high quality. The N50 length of the 'Rehmannii' unigenes was 1,476 bp, which is comparable to a recent report of the assembled transcriptome sequences of Z. aethiopica (N50 = 1,600 bp) using a similar method (Cândido et al., 2014). However, it should be noted that Z. aethiopica has approximately 1.5 times (91.2 million) as many short reads as does 'Rehmannii'. These values are also larger than those reported for related species in the Araceae family, such as A. andraeanum (N50 = 1,172 bp), A. konjac and A. bulbifer (N50 = 381 and 534; N50 = 372 and 524, respectively) (Zheng et al., 2013;Diao et al., 2014). Previous reports have demonstrated that an accurate and effective assembly tends to have a longer mean length and a larger N50 value (Chen et al., 2015), which suggests the high quality of our colored calla lily transcriptome sequences.

Functional annotation and classification of unigenes
To predict the biological functions of as many assembled transcripts as possible, various protein databases, including Nr, Swiss-Prot, COG, GO and KEGG, were employed. In total, 21,077 unigenes (Table S2) showed significant hits in the above five databases. The annotation rate of 'Rehmannii' unigenes was 53.6%, which was higher than that of Z. aethiopica (39.7%) using same method. The higher percentage in this study may be attributed to the higher frequency of unigenes longer than 500 bp in the assembled transcripts. In general, the longer unigenes were more likely to have BLAST matches in the protein databases (Parchman et al., 2010). It was estimated that the percent of unigenes >500 bp in 'Rehmannii' and Z. aethiopica was 65. 9% and 42.1%, respectively. Novaes et al. (2008) reported that a high number of next-generation sequencing (NGS) short reads often cannot be matched to known genes because the significance of sequence similarity partially depends on the length of the query sequence. Therefore, the non-significantly annotated unigenes that are likely novel transcripts unique to Z. rehmannii 'Rehmannii' may be too short to allow for statistically meaningful matches.
A species-based distribution of the best hits from the BlastX search against the Nr database showed that 32.0% of the annotated sequences of 'Rehmannii' had similarity with dicotyledonous grape (V. vinifera). This finding was confirmed by other transcriptome reports on Z. aethiopica (Cândido et al., 2014), A. andraeanum (Yang et al., 2015Li et al., 2015), A. konjac and A. bulbifer (Zheng et al., 2013). In fact, calla lily and grape are a monocotyledonous herb and a dicotyledonous woody vine, respectively. So they are distant from each other genetically and evolutionarily. One possible reason for this perceived similarity is the absence of whole-genome sequences in public databases for any species of Araceae. These assembled transcriptome sequences may provide an important data resource for future studies on taxa-specific phenomena in the family Araceae. Consistent with previous studies (Cândido et al., 2014;Yang et al., 2015;Li et al., 2015;Zheng et al., 2013), the annotated unigenes were classified into 24 COG and 47 GO sub-terms or sub-categories, indicating that our transcriptome data represented a broad diversity of transcripts in colored calla lily. In addition, a total of 4,532 unigenes were annotated and mapped to 117 KEGG pathways. Approximately 80% of top 20 hit pathways were involved in genetic information processing and metabolism, while the others were related to pathways involved in plant hormone signal transduction, plant-pathogen interaction, phagosome, and endocytosis (Fig. 2C). The most highly represented pathway was related to genetic information processing and metabolism, reflecting the fact that calla lily devotes an enormous investment to gene transcription control and capacity, cell maintenance and defense capacity (Cândido et al., 2014). Unigene annotation, together with the predicted pathways, also facilitated the discovery of some key genes involved in flower development and function in colored calla lily. In total, we identified 117 homologous sequences involved in eight pathways, including the anthocyanin biosynthesis pathway, carotenoid biosynthesis pathway, photoperiod pathway, vernalization pathway, gibberellic acid pathway, autonomous or others. These captured unigenes again demonstrated that a relatively accurate and high-coverage genomic database can be produced by RNA-seq-based de novo transcriptome analysis for non-model plant species.

Identification of EST-SSR marker frequency and type
Markers based on expressed sequences are useful and attractive for the detection of functional variation and gene-based analysis. However, the available EST-SSR markers are insufficient for the genus Zantedeschia at the present time. Wei et al. (2012) identified 209 EST-SSRs from 2,175 non-redundant ESTs derived from cDNA libraries of developing spathe in Z. aethiopica. Among these, a total of 166 primer pairs flanking the EST-SSRs could be designed. Note here that these EST-SSRs were identified based on Sanger sequencing. Very limited expressed sequence data can be generally produced via the Sanger sequencing of cDNA libraries compared to RNA-based transcriptome sequencing. In our study, 9,933 potential EST-SSRs were identified in the de novo transcriptome sequences obtained by Illumina sequencing (Table 2). These data confirm that transcriptome sequences are excellent resources for the development of numerous SSR markers. In the present study, it was estimated that approximately 20.3% (7,997) of the assembled unigenes possess SSR loci, and the abundance of SSRs was one SSR locus per 4.1 kb. The density of SSR-containing sequences in 'Rehmannii' was higher than that in the transcriptome reports for other species in Araceae, such as Amorphophallus spp. (11.8%) and P. ternate (16.24%) (Wang et al., 2014;Zheng et al., 2013). The frequency of SSRs was higher than that in P. ternate (4.3 kb) (Zheng et al., 2013) but lower than that in Amorphophallus spp. (3.6 kb) (Wang et al., 2014). The difference in the abundance estimation and frequency of SSRs among various species could partially be due to the SSR search criteria, the size of the unigene assembly dataset, the database-mining tools and the sequence redundancy, in addition to actual differences between species (Wang et al., 2014;Zheng et al., 2013;Chen et al., 2015).
The SSRs identified in this study were not uniformly distributed in the 'Rehmannii' transcriptome database (Table 3). When mononucleotide repeats were excluded, di-(59.8%) and trinucleotide repeats (38.8%) were the most abundant, whereas hexa-(1.1%), penta-(0.2%) and tetranucleotide repeats (0.1%) were rare. This result is identical to the previous findings of di-and trinucleotide motifs as the most frequent SSR motif types in the transcriptome sequences of many other plants, including Amorphophallus spp. and P. ternate (Wang et al., 2014;Zheng et al., 2013). As shown in Table 3, the AG/CT motif was the most abundant dinucleotide repeat (27.9%), followed by AT/TA (3.6%) and AC/GT (3.3%). The predominant AG/CT motif repeats were also observed in Amorphophallus spp. and P. ternate (Wang et al., 2014;Zheng et al., 2013). The most abundant trinucleotide repeat motif in calla lily was AGG/CCT, closely followed by AAG/CTT, similar to reports in Amorphophallus spp. but different from those in P. ternate (Wang et al., 2014;Zheng et al., 2013). This difference possibly arises from the SSR search parameters and search algorithms. CCG/CGG was the most frequent trinucleotide motif in P. ternate but the third and fourth most common repeat type in Amorphophallus spp. and colored calla lily 'Rehmannii', respectively. Anyhow, these abundance results corroborated with the suggestion that the trinucleotide motif CCG/CGG is common in monocots. In addition, we also noticed that GC-rich trinucleotide motifs (ACC/GGT, ACG/CGT, AGC/CTG, AGG/CCT and CCG/CGG comprised > 65%) were more abundant than AT-rich trinucleotide motifs (AAG/CTT, AAT/ATT, ACT/AGT, and ATC/ATG comprised < 35%) in these three Araceae species. These results strongly support the fact that the high GC content and consequent codon usage bias are specific features of monocot genomes (Morgante, Hanafey & Powell, 2002).

Evaluation of genetic diversity and relationships among colored calla lily accessions
Cultivar 'Rehmannii' in the present study is a hybrid of Z. rehmannii. It has the desired attributes for potted flower purposes, such as pink spathes, perfect trumpetshaped inflorescence, lanceolate and semi-erect leaves, and high productivity. Similar characteristics were also observed in another four cultivars, including Super Gem, Rose Gem, Xiangyuan Red and Allure. Hence, all five varieties here were initially used to evaluate EST-SSR marker usefulness in monitoring polymorphisms. A total of 200 primer pairs were synthesized and tested, of which 137 (68.5%) successfully yielded amplicons in these five cultivars. Of the abovementioned working primer pairs, 77 (56.2%) produced PCR products of the expected fragment size used to screen for polymorphisms among 21 individual accessions. Finally, 58 polymorphic EST-SSR markers were obtained with a polymorphic proportion of 75.3%. EST-SSR markers are advantageous to SSRs in nontranscribed regions due to their higher amplification rates and cross-species transferability (Varshney, Graner & Sorrells, 2005). The results here indicate that these transcript-based SSRs are conserved in the germplasms of colored calla lily, suggesting that they will have a broad utilization in taxonomic and cultivar identification, as well as comparative mapping. The amplification and polymorphic rate of the EST-SSRs developed in our study (59.3%) is much higher than that obtained in Z. aethiopica using Sanger sequencing (Wei et al., 2012), suggesting that the de novo transcriptome sequence based on Illumina RNA-seq was accurate and of high quality. In Z. aethiopica (Wei et al., 2012), 68 (40.9%) of the EST-SSR primer pairs yielded PCR amplification products in 24 accessions, of which 43 (63.2%) exhibited polymorphisms. However, the result is lower than that reported in Amorphophallus spp. (Zheng et al., 2013). In A. konjac and A. bulbifer, 270 (84.4%) primer pairs produced amplicons in two wild accessions, and 205 (89.1%) EST-SSR markers were polymorphic between 25 wild and cultivated accessions. The low amplification and polymorphism rate in colored calla lily may be attributed to various factors (Varshney et al., 2007;Varshney, Graner & Sorrells, 2005), including the presence of introns in the corresponding cDNA, SNPs or InDels (insertion-deletions) in the primers, assembly errors in the de novo transcriptome sequences, and the high heterozygosity of the calla lily genome. Nonetheless, the failed amplification may be remedied if the PCR amplification conditions are re-optimized, such as by applying a lower annealing temperature and/or using gradient PCR.
Most calla lily cultivars have been bred following the intra/inter-specific hybridization of the species Z. rehmannii, Z. albomaculata, Z. elliottiana and Z. pentlandii within the section Aestivae, producing plants and flowers with a broad range of shapes and colors. The species Z. rehmannii, Z. albomaculata, Z. elliottiana and Z. pentlandii are closely related to each other based on cytogenetic karyotypes (Yao, Cohen & Rowl, 1994). However, the genetic relationship among commercial hybrids or varieties is still not well defined. Characters based on which species or even varieties have been previously separated, for example, the degree of spotting on the leaf, the presence of bristles on the petioles and peduncles and, to some extent, the leaf shape, have been unreliable when a wide range of hybrids is examined (Letty, 1973;Singh, 1996). Molecular markers, such as RAPD and ISSR (Hamada & Hagimori, 1996;Zhang et al., 2009;Lu et al., 2012;Chen et al., 2013;Lu et al., 2014), therefore, were used for cultivar identification and to evaluate the genetic diversity and relationship of germplasm resources of colored calla lily. However, very limited and uncertain information has been obtained until now. EST-SSR markers facilitate better cross-genome comparisons and genetic diversity and relationship evaluation because their target coding domains are more likely to be conserved between relatives (Morgante, Hanafey & Powell, 2002;Varshney, Graner & Sorrells, 2005). In the present study, 174 alleles were identified in 21 accessions using 58 polymorphic EST-SSR markers with an average of 3.0 alleles per locus. The discriminating power, as determined by the PIC value, ranged from 0.130 to 0.757 with a mean value of 0.420. Wei et al. (2012) reported 43 EST-SSR markers with an average gene diversity of 0.446 (PIC) in 24 Z. aethiopica individual plants. The polymorphism level here was comparable to that of the EST-SSR based study in Z. aethiopica.
A cluster analysis based on 58 polymorphic EST-SSR markers was then used to assess the genetic relationships among 21 accessions. Three major groups were identified at a cut-off genetic distance index of 0.18. Cluster I consisted of thirteen accessions with a complex set of colored spathes, including red, purple, and pink, among others. The cultivated species Z. rehmannii cv. Rehmannii and another four cultivars (Super Gem, Rose Gem, Xiangyuan Red and Allure) used in the initial validation of EST-SSR markers were grouped into several sub-clusters. These sub-clusters may reflect that Z. rehmannii is easily crossed with other species (Letty, 1973;Funnell, 1993;Singh, 1996), giving rise to hybrids with lobed leaves and spathes that vary in color and shape. Surprisingly, eight accessions with yellow flowers were separated into Clusters II and III. The cultivars Black Magic in Cluster II and Best Gold in Cluster III are intra-species hybrids of Z. elliottiana and Z. pentlandii, respectively, representing the close relationship between Clusters II and III, as well as Z. elliottiana and Z. pentlandii. The species Z. elliottiana and Z. pentlandii, also known as Yellow Arum (Letty, 1973;Singh, 1996), have leaves that are hastate to cordate at the base and differ in their consistency of yellow spathes. These species are distant from the other species Z. albomaculata and Z. rehmannii. Snijder, Santiago & Tuyl (2007) determined the plastome composition of Z. aethiopica, Z. rehmannii, Z. albomaculata subsp. albomaculata, Z. albomaculata subsp. macrocarpa, Z. elliottiana and Z. pentlandii with species-specific CAPS markers developed from the plastidial intergenic region of trnD and trnC (DC). They revealed that Z. elliottiana and Z. pentlandii showed a DC-AluI and a DC-HaeIII restriction pattern that differed from that of Z. rehmannii and Z. albomaculata. The results presented here also support the above findings and suggest that EST-SSR makers developed from de novo transcriptome analysis are potential tools for taxonomy and cultivar identification in the genus Zantedeschia. Interestingly, the genetic relationships among colored calla lily accessions seem to be related to the spathe color. However, we should note that this conclusion was drawn from limited numbers of Zantedeschia accessions in section Aestivae. A higher number of accessions and more individual cultivated plants will be essential for verifying the abovementioned relationship in future studies.

CONCLUSIONS
This study was an attempt to present the transcriptome of colored calla lily using Illumina next-generation sequencing and de novo assembly. A total of 39,298 unigenes with an average length of 1,038 bp were generated, of which 53.6% (21,077) were annotated using the Nr, Swiss-Prot, COG, GO and KEGG databases. Based on the transcriptome dataset, we identified a total of 117 unique transcripts related to flower development, including pigment biosynthesis, floral organ development, flowering regulation and flower senescence. Moreover, a large number of SSRs and SNPs were mined and identified, and high-quality primers of 200 SSR loci were designed and demonstrated for their amplification and cross-species transferability in germplasm resources of colored calla lily. Finally, a relatively distinct genetic relationship among 21 accessions in section Aestivae was elucidated via 58 EST-SSR markers. The enrichment results highlight the potential of a de novo transcriptome dataset for functional genomics studies and molecular marker development. The EST-SSR markers generated in this study will enhance the current repository for the genus Zantedeschia and will be useful for taxonomic study and crop improvement programs.