Gene coexpression network for trans-1,4-polyisoprene biosynthesis involving mevalonate and methylerythritol phosphate pathways in Eucommia ulmoides Oliver

Abstract

Eucommia ulmoides, a deciduous dioecious plant species, accumulates trans-1,4-polyisoprene (TPI) in its tissues such as pericarp and leaf. Probable TPI synthase (trans-isoprenyl diphosphate synthase (TIDS)) genes were identified by expressed sequence tags of this species; however, the metabolic pathway of TPI biosynthesis, including the role of TIDSs, is unknown. To understand the mechanism of TPI biosynthesis at the transcriptional level, comprehensive gene expression data from various organs were generated and TPI biosynthesis related genes were extracted by principal component analysis (PCA). The metabolic pathway was assessed by comparing the coexpression network of TPI genes with the isoprenoid gene coexpression network of model plants. By PCA, we dissected 27 genes assumed to be involved in polyisoprene biosynthesis, including TIDS genes, genes encoding enzymes of the mevalonate (MVA) pathway and the 2-C-methyl-D-erythritol 4-phosphate (MEP) pathway, and genes related to rubber synthesis. The coexpression network revealed that 22 of the 27 TPI biosynthesis genes are coordinately expressed. The network was clustered into two modules, and this was also observed in model plants. The first module was mainly comprised of MEP pathway genes and TIDS1 gene, and the second module, of MVA pathway genes and TIDS5 gene. These results indicate that TPI is likely biosynthesized by both the MEP and MVA pathways and that TIDS gene expression is differentially controlled by these pathways.

Eucommia ulmoides Oliver (Eucommiaceae) is a species producing trans-1,4-polyisoprene (TPI) throughout its tissues. Because this plant species has been artificially planted widely in East Asia (China, Korea, and Japan) and has ideal characteristics for efficient production of TPI, such as fast growth, high TPI content, and high molecular weight of TPI, it has been proposed as a candidate source for commercial TPI production (Nakazawa et al. 2009). TPI is a non-petroleum-based material that possesses more elasticity with resistance to biological degradation than cis-polyisoprene (natural rubber), and based on these characteristics, TPI has been used for cables, golf balls and other materials (Nakazawa et al. 2009; Tangpakdee et al. 1997; Tsujimoto et al. 2014).

TPI is an isoprenoid metabolite thought to be synthesized using isopentenyl diphosphate (IPP) as substrate (Bamba et al. 2010). IPP is synthesized by two pathways: the mevalonate (MVA) pathway in the cytoplasm (mitochondria and endoplasmic reticulum) and the 2-C-methyl-D-erythritol 4-phosphate (MEP) pathway in the chloroplast (Vranová et al. 2012). Based on analysis of expressed sequence tags (ESTs) of E. ulmoides, both isoprenoid synthesizing gene candidates and putative trans-isoprenyl diphosphate synthase (TIDS) genes were identified (Suzuki et al. 2012). Among the TIDSs, TIDS2 and 4 were confirmed as having farnesyl diphosphate synthase (FDPS) activity (Kajiura et al. 2017; Suzuki et al. 2012). Other putative genes, TIDS1, 3, and 5, which have sequences similar to FDPS, do not complement FDPS genes; these genes might be long-chain trans-polyprenyl diphosphate synthases in E. ulmoides (Suzuki et al. 2012). However, these genes have not been characterized and the mechanism of TPI biosynthesis through the MVA and MEP pathways is unclear.

In contrast to model plants, progress in identifying the functions of genes in trees is slow because of the limited research approaches and the small number of researchers. Under such circumstances, microarray analysis has been used to analyze the global gene expression pattern in various plant organs, and functionally similar genes and organ-specific genes have been grouped by multivariate analysis, such as principal component analysis (PCA) (Chow et al. 2007; Fasoli et al. 2012). Based on PCA, unique functional genes in specific organs have been identified (Fasoli et al. 2012). Furthermore, because genes in the same metabolic pathway are often coexpressed, genes of unknown function are sometimes identified as putatively related to the pathway based on coexpression network modeling (Hirai et al. 2007; Wille et al. 2004). As several studies of Arabidopsis thaliana have revealed a specific coexpression pattern in both the MVA and MEP pathways (Wille et al. 2004), using multivariate analysis and comparing the isoprenoid coexpression network of model plants to the non-model plant E. ulmoides might reveal the functions of uncharacterized genes including TIDS1, 3, and 5 and address a major aspect of TPI biosynthesis.

Our goals of this study were (1) to classify the gene expression pattern into several functional gene groups by PCA by searching for putative trans-1,4-polyisoprene biosynthesis related genes in transcriptional data of various samples, and (2) to assess the gene coexpression network of the two isoprenoid pathways of E. ulmoides by comparing it to the isoprenoid related gene networks in two model plants: A. thaliana and Oryza sativa.

To obtain expression data, 102 samples from two individual E. ulmoides plants including both male and female were obtained in 2008 and 2009. The individuals were planted in soil at an elevation 1.65 m above sea level at a Hitachi Zosen Corporation factory (34°37′56″N, 135°27′27″E) in Osaka, Japan. The tree heights were approximately 10 m. Among the samples, 56 were comprised of six types of tissue: six inner stem tissues, six outer stem tissues, 20 female reproductive organs (flower or immature fruit) sampled during different seasons, three male flowers, 12 male leaves, and nine female leaves. The other 46 samples were comprised of tissues following various hormone, temperature, and light treatments: 25 fruit were given one of five treatments including three hormones (indole-3-acetic acid (IAA), naphthylphthalamic acid (NPA), combination of IAA and NPA, abscisic acid (ABA), and no-hormone control) for five treatment times (0, 3, 6, 12, and 24 h); 15 leaves were treated by combinations of two temperatures (27 and 37°C) and two lighting conditions (light and dark); six leaves were treated at one of two temperatures (15 and 42°C) under field lighting conditions. For a detailed description of samples, refer to Supplementary Table 1.

For dissecting gene expression characteristics, total RNA was prepared from each frozen sample. Each sample was ground into powder using a mortar and pestle, and RNA was extracted from about 200 mg using an RNeasy Plant Mini Kit (Qiagen, Hilden, Germany). The extracted RNA was treated with DNase I using an RNase-Free DNase Set (Qiagen), and RNA quantity and quality were checked by a NanoDrop 1000 spectrophotometer (Thermo Fisher Scientific, Waltham, MA, USA) and an Agilent Bioanalyzer 2100 electrophoresis system (Agilent Technologies, Inc., Santa Clara, CA, USA), respectively.

The DNA microarray was designed using the eArray program (Agilent Technologies) using EST information for this species (accessions FY896671-FY925126). The custom array (4×44 K platform) contains 10456 60-mer probes, and the probe sequences on the microarray were automatically designed by the algorithms of the eArray program based on 10,456 non-redundant tentative transcribed sequences (contigs and singlets) of this species (Suzuki et al. 2012). RNA was labeled by Cy3 dye using a Low RNA Input Linear Amplification Kit (Agilent Technologies) and the resultant cRNA probe quality was analyzed using an RNA 6000 Nano Assay Kit, and hybridization and washing were conducted using Gene Expression hybridization kit and wash buffer kit, respectively (Agilent Technologies). Hybridized slides were scanned using an Agilent G2505B scanner at a resolution of 5 µM at a wavelength of 532 nm, corresponding to the Cy3 emission wavelength. The microarray images were imported into Agilent Feature Extraction software v.9.5.3, and the signal intensity of each probe was obtained. The data discussed in this publication have been deposited in NCBI’s Gene Expression Omnibus (Edgar et al. 2002) and are accessible through GEO Series accession number GSE97899. The obtained data were normalized using median and first and third quantiles for further analysis.

To separate the 10,456 genes of the 102 samples into similar expression groups, principal component analysis (PCA) was conducted by R ver. 3.2.0 software (R Core Team 2015). Annotations of the contributing genes for the first to third principal components (PCs) were assigned by the BLASTX program against proteins with an E-value of 1.0E⁻¹⁰ or better.

The first PC (PC1) represents 75.2% (standard deviation, 8.80) of the variability and this score was higher by far than the other PCs: the second and third PCs (PC2 and PC3) explained 5.7% (standard deviation, 2.40) and 3.7% (standard deviation, 1.97) of the variability, respectively. Genes related to constitutive functions, such as cytochrome, ubiquitin, glutamine synthetase, photosynthesis (chlorophyll A–B binding protein gene), were over-represented in the list of genes contributing to the PC1 (Table 1). Hormone induced genes and stress response genes were represented in the list of genes contributing to the PC2 and other genes, such as lignin biosynthesis pathway genes and dehydration-related genes (i.e. dehydrin, a dehydration-responsive protein), had a high rank for PC2 (data not shown) (Table 1). Genes involved in secondary metabolism were represented in the list of genes contributing to the PC3 (Table 1). The rank of the most of polyisoprene synthesis genes in PC3 was higher than in other PCs. All TIDS genes positively contributed to PC3, however, some of these genes contributed negatively to the other PCs (Supplementary Table 2). Thus, the polyisoprene synthesis genes might contribute more to PC3 than to the other PCs.

Table 1. Top 10 unigenes, showing scores for three principal component and putative function assigned from BLASTX search results.

Rank in each PC	Scores on each PC	Unigene	Putative function of unigene	Accession number	E-value*
PC1
1	220.7	RB013_E23	Hypothetical protein	BAD46202	4.0.E-29
2	164.4	RB013_C03	Probable cytochrome P450 monooxygenase	T02955	3.0.E-60
3	141.2	Contig347	RNA-binding gricine-rich protein-1 (RGP-1c)	BAA03743	2.0.E-65
4	140.1	Contig2324	Polyubiquitin UBQ10	AAM98141	0.0.E+00
5	138.7	RB011_P05	Hypothetical protein PY01929	EAA21331	2.0.E-14
6	125.5	Contig529	Surface protein SdrI	AAM90673	1.0.E-22
7	122.8	Contig1709	Hypothetical protein AN5245.2	XP_662849	2.0.E-04
8	122.0	Contig1928	Glutamine synthetase	AAB61597	0.0.E+00
9	116.5	Contig343	PM28B protein	CAB56217	1.0.E-152
10	114.2	Contig1209	Light harvesting chlorophyll a/b-binding protein	BAA25396	1.0.E-102
PC2
1	72.7	Contig529	Surface protein SdrI	AAM90673	1.0.E-22
2	41.3	Contig2875	Allergenic isoflavone reductase-like protein Bet v 6.0102	AAG22740	1.0.E-143
3	35.1	Contig105	Putative transcription factor	AAK69513	7.0.E-49
4	32.6	Contig343	PM28B protein	CAB56217	1.0.E-152
5	28.9	Contig505	Auxin-repressed protein-like protein ARP1	AAX84677	7.0.E-40
6	28.7	Contig3064	Auxin-repressed protein-like protein ARP1	AAX84677	7.0.E-30
7	28.6	Contig1628_1	Metallothionein-1 like protein	AAB70560	8.0.E-32
8	28.6	Contig117	DRM1 (Dormancy-associated protein 1)	NP_849720	1.0.E-30
9	28.6	Contig1726	Auxin-repressed protein-like protein ARP1	AAX84677	7.0.E-45
10	26.7	RT022_C16	Transposon protein, putative, CACTA, En/Spm sub-class	ABA99001	6.5.E-02
PC3
1	35.8	RB032_I18	Metallothionein-like protein	CAA69624	7.0.E-21
2	30.9	Contig1454	Metallothionein-like protein	CAA69624	1.0.E-23
3	27.1	Contig1928	Glutamine synthetase	AAB61597	0.0.E+00
4	24.4	Contig1912	Major allergen Pru ar 1	O50001	4.0.E-49
5	24.2	Contig655	Inositol-3-phosphate synthase (Myo-inositol-1-phosphate synthase)	Q9LW96	0.0.E+00
6	23.6	RB013_H06	Hypothetical protein TTHERM_00411540	EAS00603	2.3.E+00
7	23.4	Contig459	RD22-like protein	AAV36561	1.0.E-120
8	21.2	Contig1474	Lipid transfer protein	AAQ96338	1.0.E-38
9	19.9	Contig2834	Cinnamic acid 4-hydroxylase	BAB71716	0.0.E+00
10	19.6	Contig2875	Allergenic isoflavone reductase-like protein Bet v 6.0102	AAG22740	1.0.E-143

* E-values of the best hit in BLASTX are shown.

Because the first three PCs explain 84.6% of the gene expression variability, global gene expression of the samples was plotted using the three PCs: fruit, hormone-treated fruit and stressed leaves showed convergence, while stem samples of both outer and inner tissues, female leaves, ovules and stamens were distributed widely (Figure 1). PC1 had a much larger variance (75.2%) than other PCs; this indicated that most of this data could be explained by PC1. We used the six types of tissue with different seasons for this analysis (Supplementary Table 1). Female reproductive organ and leaves had some convergence on higher scores of PC1 axis, however, inner and outer stem tissues, male flower and some female leaf distributed widely on lower scores (Figure 1). It might be suggested that the our analyzed samples with different tissues and sampling season resulted in the large differences of gene expression related to constitutive function, and the gene expression characteristics of each tissue resulted in large variances of PC1. On the other hand, PC2 represents hormone related genes and lignin synthesis genes, and hormone-treated reproductive organs and both outer and inner stem tissues had positive scores for PC2. For PC3, hormone-treated fruit and female reproductive organs had positive scores. In a previous study, each plant tissue also showed convergence in the PCA plot and each PC represented organ-specific gene expression (Fasoli et al. 2012). In E. ulmoides, accumulation of TPI is particularly observed in fruit (Nakazawa et al. 2009), so PC3 might reflect fruit-specific genes including those related to TPI biosynthesis.

Figure 1. Score plot for PCA. Each colored point represents an individual tissue sample.

The global gene expression pattern provides fundamental insight into how specific genes are involved in a biological event (Fasoli et al. 2012; Kang et al. 2011), and gene grouping is also useful for isolating genes involved in the same biological process such as a metabolic pathway (Basso et al. 2005; Hamada et al. 2011; Srinivasasainagendra et al. 2008). The multivariate analysis approach of PCA classifies multivariate gene expression data into mutually uncorrelated axes, and each of them has a different aspects of the samples (Ringnér 2008); thus, this method has been utilized for the grouping of the genes at the first step of classification of gene expression data without much computational cost (Ma and Dai 2011; Ringnér 2008). However, another study has shown that gene grouping is not conducted well for PCs that account for many variations in the data (Yeung and Ruzzo 2001). In this study, PCA was conducted to extract gene groups from over 10,000 genes into three PCs, and PC3 had a much smaller variance (3.8%) than PC1 (75.2%), and successfully identified secondary metabolism genes including TPI biosynthesis related genes.

To detect the gene coexpression network of isoprenoid biosynthesis related genes, genes within PC3 were used for further network analysis. Referring to BLAST results, 27 genes related to isoprenoid biosynthesis were extracted from the genes within PC3; they included seven MVA pathway genes (acetyl-CoA C-acetyltransferase (ACAT), 3-methylglutaryl-CoA synthase (HMGCS), 3-hydroxyl-3-methylglutaryl-CoA reductase (HMGCR), mevalonate kinase (MVK), and mevalonate diphosphate (MVD)), one MEP pathway gene, 1-deoxy-D-xylulose 5-phosphate synthase (DXPS), the genes involved in branches of both pathways (isopentenyl pyrophosphate isomerase (IDI), geranylgeranyl diphosphate synthase (GPPS), FDPS and geranylgeranyl polyphosphate synthase (GGPPS)), and three natural rubber biosynthesis related genes (small rubber particle protein (SRPP), rubber elongation factor (REF), and major latex protein (MLP)); these genes have already been identified as isoprenoid biosynthetic genes by Suzuki et al. (2012). Three other polyisoprene synthesis genes were identified in this study: one from the MEP pathway, 4-hydroxy-3-methyl-2-butenyl diphosphate reductase (HDR) (Unigene: RT019_M04.b, Accession No: ABB55395, E-value: 3.0E⁻⁴³), another encoding major latex-like protein (MLP-like) (Unigene: RT046_E14.b, Accession No: CAC83581, E-value: 8.0E⁻³⁹), and the other encoding rubber elongation factor-family protein (REF3) (Unigene: Contig4051, Accession No: PF05755, E-value: 2.0E⁻³²) (Table 2). Using these 27 genes, Pearson correlation coefficients (PCC) among all pairs of genes were calculated (Table 2). To search the specific coexpression network among isoprenoid biosynthesis related genes of E. ulmoides, the network structure was confirmed in 0.05 increments between PCC values of 0.50 (weak connection) and 0.70 (strong connection).

Table 2. List of genes obtained from coexpression network analysis, showing scores for third principal component (PC3).

Rank in PC3	Scores on PC3	Unigene ID	Putative product	Gene name	Module
45	9.28	RT019_M04	4-Hydroxy-3-methyl-2-butenyl diphosphate reductase	HDR*	1
50	8.57	Contig861	Isopentenyl diphosphate isomerase	IDI	1
52	8.55	Contig4112	Rubber elongation factor 2	REF2	1
93	5.81	Contig988	trans-Isoprenyl diphosphate synthase	TIDS1	1
104	5.47	Contig1438	trans-Isoprenyl diphosphate synthase	TIDS5	2
172	3.87	RE047_D18	Acetyl-CoA acetyltransferase 1	ACAT1	2
193	3.35	Contig1491	3-Hydroxy-3-methylglutaryl CoA synthase 1	HMGCS1	1
197	3.30	Contig1073	Rubber elongation factor 1	REF1	2
281	2.40	Contig3771	3-Hydroxy-3-methylglutaryl CoA synthase 2	HMGCS2	2
305	2.22	RT020_C19	Geranylgeranyl pyrophosphate synthase 2	GGPPS2	1
314	2.16	RT050_P19	Acetyl-CoA acetyltransferase 2	ACAT2	2
561	1.02	RB058_N12	1-Deoxy-D-xylulose 5-phosphate synthase 3	DXPS3	2
573	0.98	RT032_A10	3-Hydroxy-3-methylglutaryl CoA reductase 2	HMGCR2	2
1018	0.44	Contig3459	Diphosphomevalonate decarboxylase	MVD	2
1095	0.38	RT016_K07	Geranylgeranyl pyrophosphate synthase 1	GGPPS1	—
1115	0.38	Contig298	Mevalonate kinase	MVK	2
1281	0.30	RT023_E11	Farnesyl diphosphate synthase	TIDS4	2
1532	0.21	Contig3911	Farnesyl diphosphate synthase	TIDS2	—
1632	0.19	Contig2465	3-Hydroxy-3-methylglutaryl CoA reductase 1	HMGCR1	—
1862	0.14	RB033_C15	trans-Isoprenyl diphosphate synthase	TIDS3	—
2125	0.10	Contig3476	1-Deoxy-D-xylulose 5-phosphate synthase 1	DXPS1	—
2161	0.09	Contig2314	Geranylgeranyl diphospahte synthase	GPPS	—
2710	0.04	RB015F22	1-Deoxy-D-xylulose 5-phosphate synthase 2	DXPS2	2
2956	0.03	RT016_L11	1-Deoxy-D-xylulose 5-phosphate synthase 4	DXPS4	—
2973	0.03	Contig274	Major latex protein	MLP	—
3128	0.02	RT046_E14	Major latex protein-like	MLP-like*	—
3292	0.02	Contig4051	Rubber elongation factor-family protein	REF3*	—

* Newly identified genes in this study.

To compare the coexpression network for E. ulmoides polyisoprenoid biosynthesis to that of model plant species, we examined the gene coexpression network among isoprenoid synthesis genes of A. thaliana and O. sativa. Data were downloaded from the Gene Expression Omnibus for fitting to samples of E. ulmoides (Edgar et al. 2002). The data obtained from A. thaliana covered samples of 54 stems or shoots, 22 reproductive organs, 24 leaves, and 38 seedlings at different developmental stages or treated with heat or hormones; in total, there were 138 samples (Supplementary Table 3). Data from O. sativa covered samples of 14 stems or shoots, 45 reproductive organs, 43 leaves, and 10 seedlings at different developmental stages or treated with hormones; in total, there were 112 samples (Supplementary Table 4). From the data, 31 A. thaliana genes and 17 O. sativa genes related to isoprenoid biosynthesis were extracted (Supplementary Table 5) and the PCC was calculated among all pairs of these genes; the network structures were confirmed in 0.05 increments between PCC values of 0.50 and 0.70.

In E. ulmoides, 78% (21 of 27 genes) of the genes formed two coexpression network modules at a threshold PCC value of 0.65. The first module was comprised of HDR, IDI, TIDS1, REF2, HMGCS1, and GGPPS2 genes and the second was comprised of ACAT, HMGCS2, HMGCR2, MVK, MVD, DXPS, REF1, TIDS4, and TIDS5 genes (Figure 2). In A. thaliana, two network modules were also formed independently. The first module was comprised of seven genes of the MEP pathway (DXPS, DXR, ISPD, CDPMEK, ISPF, HDS, and HDR) and GGPPS8. The second module was mainly comprised of six genes of the MVA pathway (ACAT3, HMGCS, HMGCR, MVK, and MVD1 and 2), IDI, and FDPS2 (Supplementary Figure 1). O. sativa also formed two independent network modules. The first module was comprised of five genes of the MEP pathway (DXR, ISPD, CDPMEK, HDS, and HDR) and GPPS. Four genes in the MVA pathway (ACAT, HMGCS, HMGCR, and MVK) and FDPS2 constructed the second module (Supplementary Figure 2). In E. ulmoides, two modules were connected at a PCC value of 0.60; however, the modules of model plants did not connect at PCC scores from 0.50 to 0.70 (data not shown) suggesting that the connections of two modules of model plant species were weaker than that of E. ulmoides.

Figure 2. Coexpression network of the first modules found for isoprenoid pathway genes and potential rubber genes with the threshold PCC value of 0.65 or higher. Identified and unidentified unigenes are enclosed by solid and dotted lines, respectively. Metabolic flow is shown by arrows and coexpressed genes are connected by colored solid lines; green represents different and purple the same connections found in the networks of model plants. Genes with no connections to any genes have been omitted from the figure. For details of the genes used for network analysis, see Table 2.

The network structure of all three species was divided into two modules. The first and second modules of the two model plants corresponded with the gene groups of the MEP and MVA pathway, respectively (Supplementary Figures 1 and 2). However, a few MEP pathway genes were identified in E. ulmoides and the structure of the first module of E. ulmoides did not correspond well to the MEP pathway genes. On the other hand, GPPS or GGPPS constituted the first module of all three plant species, and previous research also shows the coexpression network of isoprenoid pathway genes of A. thaliana divided into two modules corresponding to the MEP and MVA pathways (Wille et al. 2004). The coexpression network of isoprenoid biosynthesis tends to be divided into modules of MEP and MVA pathway genes, and thus, the first module of E. ulmoides may also represent the MEP pathway gene groups.

Connections were observed between the MEP and MVA pathway genes of E. ulmoides: HMGCS1 is coexpressed with an MEP pathway gene via REF2 in the first module, and DXPS is coexpressed with ACAT (Figure 3). Although genes from the two pathways are distributed in different organelles (MEP pathway genes in the chloroplast; MVA pathway genes in the cytosolic, endoplasmic reticulum and peroxisome compartments (Vranová et al. 2012)), cross-talk via IPP is confirmed in several species (Hemmerlin et al. 2003; Laule et al. 2003) and the previous study of the coexpression network in A. thaliana also found a connection between genes in the MEP and MVA pathways: between the gene for DXPS and one of the MVA pathway genes, MVD, and between genes encoding IDI and HDS of the MEP pathway (Wille et al. 2004). E. ulmoides also shares IPP between the MEP and MVA pathways, indicating connections between these pathways (Bamba et al. 2010); the results of our network analysis supported these connections between the MEP and MVA pathways in E. ulmoides. Furthermore, these connections were not observed in the model plants of our study. In contrast to E. ulmoides, these model plants do not biosynthesize TPI or other polyisoprenes, indicating that the interaction between the MEP and MVA pathways or the cross-talk of IPP may be more active in E. ulmoides than in these model plants.

Figure 3. Coexpression network of the second module of isoprenoid pathway genes and potential rubber genes above the threshold PCC value of 0.65. For explanations of the lines and gene names, see the legends of Figure 2 and Table 2, respectively.

Different REF and TIDS genes constituted each module: the first module has REF2 and TIDS1; the second module, REF1 and TIDS5 (Figures 2 and 3). The functions of TIDSs are still unknown; however, rubber formation related genes, REFs, contribute to forming cis-polyisoprene in over one micrometer aggregates (Berthelot et al. 2014). Some products downstream of the MEP and MVA pathways in model plants differ (MEP pathway, monoterpenoids, carotenoids; MVA pathway, squalene (Dubey et al. 2003; Vranová et al. 2012; Wille et al. 2004)); however, several products such as sesquiterpenes and polyprenols are synthesized via both pathways (Vranová et al. 2012). A previous study found that ¹³C labeled TPI of E. ulmoides is obtained from both pathways, indicating that TPI biosynthesis is conducted through two pathways (Bamba et al. 2010). The module connections between each gene in the pathways and the set of each TIDS and REF gene indicate that a combination of TIDS1 and REF2 is presumably involved in TPI production using IPP mainly from the MEP pathway, and a combination of TIDS5 and REF1 using IPP mainly from the MVA pathway. On the other hand, it was not suggested which is the primary pathway of TPI biosynthesis. Previous EST analysis shows that the High TIDS1 expression is observed in mature leaves and stems and TIDS5 expression is observed in young leaves and stems (Suzuki et al. 2012). These results suggest that the primary pathway of TPI biosynthesis could be changed among the tissues with different growing stages.

Coexpression network analysis has demonstrated the power to identify genes involved in the same biological process (Hirai et al. 2007; Liu et al. 2009; Sasaki-Sekimoto et al. 2005; van Waveren and Moraes 2008; Walhout et al. 2002). Since genes involved in the same biological process are often co-regulated in a similar spatiotemporal manner, coexpression analysis can identify critical genetic information associated with the process. In particular, genes involved in secondary metabolite pathways are more likely to be detected than in primary metabolite pathways because primary metabolites constitutively active, whereas secondary metabolism tends to be activated under particular circumstances (Hirai et al. 2007). Hirai et al. (2007) suggest that secondary metabolites are controlled by a small number of regulatory elements, and this might allow the detection of genes related to secondary metabolites. Indeed, TPI is likely to be accumulated in plant tissue when fertilizers with Mg²⁺ and Ca²⁺ are supplied, and these divalent cations and thiamine pyrophosphate are essential for activation of genes in the MEP and MVA pathways (Dubey et al. 2003; Eisenreich et al. 2004). Thus, these factors affecting expression of genes in isoprene biosynthesis may allow identification of extraction of corresponding gene groups by PCA and detection of genes of unknown function such as the TIDSs and REFs involved in both the MEP and MVA pathways.

These results might confirm the suitability of PCA and subsequent coexpression network analysis based on PCC scores in non-model plants to estimate the function of unknown genes before the molecular characterization of each gene. However, only two genes in the MEP pathway of E. ulmoides have been identified: DXPS, in Suzuki et al. (2012), and HDR, in this study, so an overview of the MEP pathway gene coexpression network is still incomplete. Further studies of the coexpression of the entire genome, including the MEP pathway genes, and characterization of TIDSs and REFs might reveal the details of TPI biosynthesis.

Abbreviations

ABA

abscisic acid

ACAT

Acetyl-CoA C-acetyltransferase

DMAPP

dimethylallyl diphosphate

DXPS

1-deoxy-D-xylulose 5-phosphate synthase

FDPS

farnesyl diphosphate synthase

GAPDH

glyceraldehyde 3-phosphate dehydrogenase

GPPS

geranylgeranyl diphosphate synthase

GGPPS

geranylgeranyl pyrophosphate synthase

HDR

4-hydroxy-3-methyl-2-butenyl-diphosphate reductase

HMGCR

3-hydroxyl-3-mehylglutaryl-CoA reductase

HMGCS

3-hydroxy-3-methylglutaryl-CoA synthase

IAA

indole-3-acetic acid

IDI

isopentenyl diphosphate isomerase

MEP

the 2-C-methyl-D-erythritol 4-phosphate

MVK

mevalonate kinase

MLP

major latex protein

MVA

mevalonate

MVD

Diphosphomevalonate decarboxylase

NPA

naphthylphthalamic acid

PCA

principal component analysis

REF

rubber elongation factor

SRPP

small rubber particle protein

TIDS

trans-isoprenyl diphosphate synthase

TPI

trans-1,4-polyisopren