Network analysis of the metabolome and transcriptome reveals novel regulation of potato pigmentation

Highlight An integrated approach of metabolomics and transcriptomics was applied to understand regulatory networks associated with biosynthesis of anthocyanins that are differentially regulated in light-red- and dark-purple-colored potato cultivars.


Introduction
To date, more than 635 anthocyanins have been identified in numerous fruits, vegetables, and flowers (Wu and Prior, 2005;He and Giusti, 2010;. The anthocyanin derivatives of delphinidin, petunidin, and malvidin are sources of purple and dark colors, whereas the derivatives of cyanidin and pelargonidin are the main pigments in bright-red-colored fruits (Jaakola, 2013). Although several genes encoding proteins implicated in anthocyanin biosynthesis and regulation have been identified, insights into the regulation of each anthocyanin biosynthesis pathway have remained future objectives. Recently, the integration of large-scale datasets derived from high-throughput functional genomics techniques have been applied successfully to studies on the functions of genes regulating tissue development (Persson et al., 2005), environmental responses (Wang et al., 2006;Cho et al., 2008), and plant metabolism (Hirai et al., 2007;Gutiérrez et al., 2008). In particular, transcript and metabolite datasets have been combined through correlation and clustering analyses and further represented as connection networks between genes and metabolites in many plants (Urbanczyk-Wochniak et al., 2003;Hoefgen and Nikiforova, 2008;Saito et al., 2008), including Arabidopsis (Hirai et al., 2004(Hirai et al., , 2007, tomatoes (Alba et al., 2005;Mounet et al., 2009), and potatoes (Stushnoff et al., 2010). The acquisition and integration of an "omics" dataset (i.e. transcriptome, proteome, and metabolome) represents a useful approach for the establishment of a strategy to identify potential genes regulating the determination of pigmentation in potatoes.
Colored potatoes have attracted research interest due to their anthocyanin content with enhanced antioxidant capacities (Truong et al., 2009;Stushnoff et al., 2010;Chong et al., 2013). Solanum tuberosum cvs Hongyoung and Jayoung are pigmented potato cultivars originated from a cross made between the white-colored Atlantic and deep dark-purplecolored AG34314 cultivars through the potato breeding program of the National Institute of Highland Agriculture Research Center in 2003 (Park et al., 2009a, b). Hongyoung has a light-red skin and light-red flesh, while Jayoung has dark-purple skin and dark-purple flesh (see Supplementary  Fig. S1 available at JXB online).
Most anthocyanins have been identified by the combined methods of ultraviolet/visible (UV/Vis) spectrometry and mass spectrometry (MS). In positive ionization mode, the [M + ] ion and mass fragmentation patterns of anthocyanins are the same as the [M+H] + ion and fragmentation patterns of flavonol (Wu and Prior, 2005;Sun et al., 2012). Anthocyanins (480-540 nm) and non-anthocyanin phenolic compounds (<400 nm) have maximum absorbance at different UV/Vis wavelengths. Thus, a liquid chromatography (LC) mass spectrometer equipped with a photodiode array detector is usually used to distinguish anthocyanin and flavonol glycosides (Lin et al., 2011;Sun et al., 2012). A recent study showed that MS data acquired in the negative ionization mode using ultrahigh-performance LC with high-resolution MS provided a series of characteristic ions for anthocyanins (e.g. [M-2H] -, [M-2H+H 2 O] -, and formic acid adducts) (Sun et al., 2012), suggesting that the integrative analysis of mass ions and mass fragmentations acquired from both the positive and negative modes can distinguish and identify anthocyanin and flavonol glycosides.
In this study, we explored the regulatory networks of anthocyanin biosynthesis in colored potatoes at the level of the transcriptome and metabolome. We focused on the differential expression of anthocyanin metabolites and their regulatory genes in light-red Hongyoung and dark-purple Jayoung potatoes compared with those of a white Atlantic potato cultivar. Connection networks were mapped on the basis of correlation analyses between metabolites and transcripts to highlight the regulatory genes associated with anthocyanin metabolites. Our findings provide new insights into the molecular mechanisms associated with the biosynthesis and regulation of anthocyanin in the pigmentation of potatoes, and highlight the usefulness of an integrated approach for understanding this process.

Plant material
Medium-sized (80-150 g) potato tubers from three different potato cultivars, Hongyoung, Jayoung, and Atlantic, were stored at a low temperature (4 °C) for 4 months after harvesting in the Dae-Gwal-Lyeong area (800 m above sea level), Korea. After storage, sprouts of potato tubers were induced at room temperature for 1 month with scattered light conditions. Whole sprouts were collected, immediately frozen in liquid nitrogen, and then stored at -80 °C prior to metabolite extraction.

Metabolite profiling using ultraperformance LC quadrupole timeof-flight tandem MS (UPLC-Q-TOF-MS)
Metabolite profiling was conducted using a UPLC system (ACQUITY UPLC; Waters, Milford, MA, USA) and hybrid Q-TOF tandem mass spectrometry (Triple-TOF-MS) (Triple TOF 5600 system; AB SCIEX, Concord, ON, Canada). Chromatographic separation was performed on an ACQUITY UPLC BEH C18 column (2.1 mm×100 mm×1.7 μm; Waters) using mobile phase A (0.1% formic acid in deionized water) and mobile phase B (0.1% formic acid in acetonitrile). Mobile phase B was increased linearly from 3% at 0 min to 50% at 3 min to 70% at 4 min to 100% at 10 min, and then held at 100% until 10.5 min. Finally, solvent B was decreased to 3% at 11 min and held at 3% until 12 min. The flow rate was maintained at 0.4 ml min -1 . Mass data acquisition was performed in both positive [electrospray ionization-positive (ESI + )] and negative (ESI -) modes using the following parameters: ion spray voltage of 5.5 kV in ESI + and -4.5 kV in ESI -; nebulizer gas (gas 1) of 55 psi; heater gas (gas 2) of 65 psi; curtain gas of 30 psi; turbo spray temperature of 600 °C; and declustering potential of 100 V in ESI + and -100 V in ESI -. For TOF MS 2 data, information-dependent acquisition was used with the following conditions: survey scans of 250 ms; product ion scans of 70 ms; high-resolution mode; declustering potential of 90 V; collision energy of 35 V in ESI + (-35 V in ESI -); and collision energy spread of 15 V. The TOF-MS and information-dependent acquisition scan was operated with the mass range of m/z 50-1600. TOF-MS and product ion calibration was performed in both highsensitivity and high-resolution modes using a calibrant delivery system prior to analysis. LC-MS data files (Wiff format files) including MS and MS 2 spectra data were converted to mzXML files using MSConvert in the Proteowizard software (version 3.04999) (Patti et al., 2012). The converted raw data were further processed using MZmine software (version 2.10) and outputted as a retention time m/z dataset.

Multivariate statistical analysis
The intensities of mass peaks for each sample were sum-normalized and Pareto-scaled using the SIMCA-P + software package (version 12.0;Umetrics,Umeå,Sweden). Principal component analysis (PCA) and orthogonal partial least-squares discriminant analysis (OPLS-DA) with data from 18 samples (three cultivars×six biological replicates) were performed to observe differences in metabolic composition among the three potato cultivars. The reliability correlation [p(corr)] values of all metabolites from the S-plot of the OPLS-DA were extracted using the first component. We selected metabolites satisfying the following criteria as potential markers: (i) high confidence [|p(corr)|>0.6] in discriminations between Hongyoung and Atlantic, between Jayoung and Atlantic, and between Jayoung and Hongyoung; (ii) mean intensities in one potato cultivar significantly different from those of another cultivar (P<0.05); and (iii) a minimum of a 2-fold change. The P value was calculated using an independent two-sample t-test.

Targeted selection of anthocyanins and their intermediates
Targeted selection of anthocyanins and their intermediates was performed based on follow information: (i) molecular formula and exact mass information of the compounds on phenylprophanoid and flavonoid pathways were established throughout the KEGG (http://www.genome.jp/kegg) and PMN (plant metabolic network, http://www.plantcyc.org) databases and literary references (see Supplementary Fig. S2 and Table S2 available at JXB online) Stushnoff et al., 2010;Jaakola, 2013); (ii) specific daughter ions of anthocyanins through literary references on similar compounds (Sun et al., 2012); and (iii) MS 2 spectra of standard compounds and metabolome databases, including METLIN (http://metlin.scripps.edu/) and LIPD MAPS (http://www.lipidmaps.org/). Finally, fragmented or adducted mass features were analyzed for selected ions. The identification of anthocyaninrelated compounds is summarized in Supplementary Table S3 available at JXB online.
RNA sequencing (RNA-seq) data analysis RNA-seq paired-end libraries were prepared using the Illumina TruSeq RNA Sample Preparation Kit version 2 (Illumina, San Diego, CA, USA). Starting with total RNA isolated using PureLink ® Plant RNA Reagent (Life Technologies Korea LLC, Seoul, Korea), mRNA was purified using poly(A) selection or rRNA depletion; next, RNA was chemically fragmented and converted into single-stranded cDNA using random hexamer priming. Next, the second strand was generated to create double-stranded cDNA. Library construction began with the generation of bluntend cDNA fragments from double-stranded cDNA. Then, an A base was added to the blunt ends to make them ready for the ligation of sequencing adapters. After size selection of the ligation products, the ligated cDNA fragments that contained adapter sequences were enhanced via PCR using adapter-specific primers. The library was quantified with a KAPA library quantification kit (Kapa biosystems KK4854) following the manufacturer's instructions. Each library was loaded onto the Illumina Hiseq2000 platform and high-throughput sequencing was performed to ensure that each sample met the desired average sequencing depth. Image analysis and base calling were performed using the Illumina pipeline with default settings.
For mRNA sequencing, total RNA (10 μg) was isolated from sprouts of Atlantic, Hongyoung, and Jayoung using a PureLink ® RNA Mini kit (Life Technologies Korea LLC) and used to create normalized cDNA and PCR-amplified datasets according to the Illumina RNA-seq protocol; then, the RNA was sequenced by Illumina HiSeq2000 (242M 100 bp paired-end reads). Sequence data with base-pair qualities in the upper Q ≥20 were extracted by SolexaQA. Trimming resulted in reads with a mean length of 80.14 bp across all samples; a minimum length of 25 bp was applied during sequence trimming. The gene annotation used S. tuberosum Group Phureja DM1-3 516R44 (CIP801092) Genome Annotation version 3.4 mapped to the pseudomolecule sequence (PGSC_DM_v3_2.1.10_pseudomolecules.fa) downloaded from Solanaceae Genomics Resource at Michigan State University (http://solanaceae.plantbiology.msu.edu/pgsc_download.shtml) (Consortium, 2011) Transcript profiles and annotation mRNA libraries generated from each sample were sequenced using Illumina HiSeq2000 (100 bp paired ends). Reads for each sequence tag were mapped to the reference with the Bowtie software (Langmead et al., 2009). The number of mapped clean reads for each gene was counted and normalized using the DESeq package in R (Anders and Huber, 2010). Only the genes that mapped with read counts of 100 or above in all experimental samples were retained for further analysis. Fold change and binomial tests were used to identify differentially expressed genes between each sample. The false discovery rate calculated via DESeq was applied to identify the threshold of the P value in binomial tests and analyses.
Gene Ontology (GO) and KEGG pathway functional enrichment analyses were performed via the Gene Ontology Database and DAVID (http://david.abcc.ncifcrf.gov/tools.jsp), respectively (Huang et al., 2008). GO consists of terms that provide a more global representation of gene functions using a controlled vocabulary; DAVID comprises web-accessible programs that provide a comprehensive set of functional annotation tools that can be used by investigators to understand the biological meaning behind a large list of genes. The gene lists generated by annotated TAIR (The Arabidopsis Information Resource) ID of transcripts of up-and down-regulated differentially expressed genes were classified into MapMan BINs using the MapCave tool (http://mapman. gabipd.org/web/guest/mapcave), which is linked with three databases (Arabidopsis thaliana TAIR8, Arabidopsis thaliana TAIR9, and TAIR release 10) (see Supplementary Table S4 available at JXB online).

Integrative analysis of metabolome and transcriptome
Pearson correlation coefficients were calculated for metabolome and transcriptome data integration. For this, the mean of all biological replicates of each cultivar in the metabolome data and the mean value of expression of each transcript in the transcriptome data were calculated. The fold changes in each pigmented potato (Hongyoung and Jayoung) were then calculated in both the metabolome and transcriptome data and compared with the control cultivar (Atlantic). Finally, the coefficients were calculated from log 2 (fold change) of each metabolite and log 2 (fold change) of each transcript using the EXCEL program (see Supplementary Table S5 available at JXB online). Correlations corresponding to a coefficient with R 2 >0.9 were selected (see Supplementary Table S6 available at JXB online). Metabolome and transcriptome relationships were visualized using Cytoscape (version 2.8.2).

Metabolic differences among the three potato cultivars
To compare the metabolite composition involved in the pigmentation of the three different potato cultivars, datasets obtained from UPLC-Triple-TOF-MS in the ESI + (ESI -) mode were subjected to PCA. The results showed that the three potato cultivars were clearly separated in the PC1×PC2 score plots (Fig. 1A, B). Indeed, the first principal component (PC1) in ESI + mode (31.1% of the total variables) and PC1 and PC2 in ESI -(38.3 and 35.9%, respectively) were clearly separated between Hongyoung and Atlantic. The differences between Jayoung and Atlantic resulted from PC2 (26.3% variables) in ESI + mode and PC1 (38.3%) in ESImode. Furthermore, score plots and S-plots of OPLS-DA were used for modeling the differences between two potato cultivars (see Supplementary Fig. S3 available at JXB online). The selection of variables responsible for the differences was performed through statistical analysis as described in Materials and methods. A total of 556 (651), 441 (470), and 454 (466) mass ions were selected between Hongyoung and Atlantic, between Jayoung and Atlantic, and between Jayoung and Hongyoung in the ESI + (ESI -) mode, respectively. In total, 841 and 895 mass ions were selected in the ESI + and ESImodes, respectively ( Fig. 1C, D).

Differential accumulation of anthocyanin derivatives between Hongyoung and Jayoung
Anthocyanins are glycosides and acylglycosides of anthocyanidin aglycones that are biosynthesized through the flavonoid pathway via the phenylpropanoid pathway (Stushnoff et al., 2010;Jaakola, 2013). Cyanidin (Cy), delphinidin (Dp), pelargonidin (Pg), peonidin (Pn), petunidin (Pt), and malvidin (Mv) are six common anthocyanins that are grouped according to the hydroxyl pattern or methoxy substitutions of B ring ( Supplementary Fig. S2). Among them, the anthocyanin derivatives Dp, Pn, and Mv are sources of purple and dark colors, whereas the derivatives of Cy and Pg are the main pigments in bright-red-colored fruits (Jaakola, 2013).
Among the mass ion peaks detected in our metabolomics analysis, anthocyanins and their intermediates were selected using product-ion scanning and precursor-ion scanning based on their molecular formula in MS 1 and MS 2 spectral data, respectively (see Supplementary Fig. S4 available at JXB online). The selected peaks were identified by interpretation of their MS 2 fragment patterns and are summarized in Supplementary Table S3. The identified anthocyanins and their relevant compounds were rearranged to their corresponding positions in an anthocyanin biosynthesis pathway established based on KEGG, PMN, and literature references. Figure 2 shows that the composition of compounds on the anthocyanin biosynthesis pathways were specifically different depending on the potato cultivars (i.e. white Atlantic, lightred Hongyoung, and dark-purple Jayoung). In particular, the compositions of flavonoids downstream of the phenylpropanoid pathway were highly different between Hongyoung and Fig. 1. PCA score plot of colored potatoes and numbers of potential markers for each. PCA score plots were derived from metabolite ions acquired from ESI + (A) and ESI -(B) mode. Potential markers were selected by comparing quantitative differences of mass ions in ESI + (C) and ESI -(D) mode between Hongyoung and Atlantic, between Jayoung and Atlantic, and between Hongyoung and Jayoung. Jayoung, indicating that these metabolites might play a crucial role in determining the pigmentation in potato. Apigenin (Ap), kaempferol (Ka), dihydrokaempferol (Dk), and Pg derivatives were shown to be most abundant in Hongyoung, whereas Pn, Pt, Dp, and Mv derivatives were shown to be the more abundant in Jayoung.

Correlation analysis between transcripts and anthocyanin derivatives reveals the differential regulatory network of anthocyanin biosynthesis in Hongyoung and Jayoung
To understand the regulatory network of anthocyanins implicated in the differential distribution of anthocyanin derivatives between Hongyoung and Jayoung, we carried out correlation tests between quantitative changes of metabolites and transcripts in the three different colored potatoes. For this, derivatives of Ka, Dk, Dp, Pg, Pn, Pt, and Mv detected in this study and transcripts categorized into flavonoid metabolism, hormone metabolism, regulation of transcription, and cell signaling were selected from the 1044 genes ( Supplementary Fig. S5) differentially expressed in the three   Table S6). Based on the result, interaction networks between the 22 metabolites and 119 transcripts were organized in Hongyoung and Jayoung ( Fig. 3 and Supplementary Table S6). The networks showed that the 119 transcripts were grouped into five clusters (I-V) ( Table 2) and the 22 metabolites were divided into four groups (A-D) ( Table 3). Metabolites in group A containing derivatives of a Pg, two Kas, a Dk, and an Ap were more predominant in Hongyoung compared with Jayoung, and were highly correlated with the transcripts in clusters I and II. On the other hand, group D metabolites including three Pt derivatives and one Dp derivative were more predominant in Jayoung and were showed to be highly connected with the transcripts in clusters IV and V. Group B and C metabolites were highly increased in both Hongyoung and Jayoung compared with Atlantic, although the increase of     group C metabolites containing a Pg, three Pt, and an Mv derivative were a little more prominent in Jayoung than Hongyoung. The metabolites in groups B and C were shown to be strongly correlated with the transcripts in cluster III (Table 3 and Fig. 3). Validation of differential gene expression was performed for 14 genes using quantitative PCR (qPCR) with gene-specific primers (see Supplementary Table S1 available at JXB online). Real-time qPCR analysis with RNA isolated from sprouts of Hongyoung, Jayoung, and Atlantic showed that genes in clusters I and II (JAZ8, JAZ10, WRKY75, ERF1, and MYB9) were highly expressed in Hongyoung, whereas MYB3 and UFGT (cluster IV) were highly expressed in Jayoung. The expression levels of most genes in cluster III (CCOMT, two CHI, two CHS, and LODX) were similarly increased in both Hongyoung and Jayoung ( Supplementary Fig. S6 available at JXB online).

Anthocyanin biosynthesis
The distribution of anthocyanins in the colored potatoes showed that Pg-derivative anthocyanins were abundant in AGI, Arabidopsis Genome Initiative Number; TAIR, The Arabidopsis Information Resource. a Regulatory genes were clustered according to gene and metabolite correlation in Fig. 3, which was calculated using Pearson correlation coefficients (R 2 ). b BINs of genes generated according to MapMan classification using the MapCave tool (http://mapman.gabipd.org/web/guest/mapcave) light-red-colored Hongyoung, whereas Pn, Dp, Pt, and Mv derivatives of anthocyanins were enriched in dark-purple Jayoung. Transcripts related to anthocyanin biosynthesis were differentially regulated in each cultivar. Among the 10 genes related to anthocyanin biosynthesis in clusters I and II, homologs for leucoanthocyanidin dioxygenase (LDOX), acyltransferase, a UDP-glucose:flavonoid O-glycosyltransferase (UFGT: PGSC0003DMT400020466), and hydroxycinnamoyl-CoA shikimate/quinate hydroxycinnamoyl transferases (HCT: PGSC0003DMT400066505) were more strongly up-regulated in light-red Hongyoung compared with Jayoung. In contrast, two homologs from cluster V [phenylalanine ammonia-lyase (PAL2) and NAD(P)-linked oxidoreductase] and six homologs from cluster IV (two UFGTs, two chalcone synthases (CHSs), a chalcone isomerase (CHI) and an ascorbate oxidase) were more strongly up-or down-regulated in dark-purple Jayoung compared with Hongyoung.

Hormones
Of the 12 genes implicated in hormone response in cluster I and four in cluster II (Table 2 and Fig. 3), homologs of a SAUR-like auxin-responsive protein (SAUR71), an IAA-amido synthase (GH3.6), and more axillary branches 1 (MAX1) were significantly down-regulated in Hongyoung compared with Atlantic. However, nine homologs (five in cluster I and four in cluster II) were significantly up-regulated in Hongyoung (Table 2). Hormone-related transcripts of clusters IV and V were highly correlated with group C and D metabolites that were highly increased in Jayoung compared with Hongyoung. The significant down-regulation of four transcripts was observed in Jayoung, including homologs of GA requiring 3 (GA3), GH3.1, brassinosteroid-insensitive 2 (BI2), and an NAD(P)-linked oxidoreductase protein. The four transcripts including gibberellin-regulated family proteins (AT1G22690 and AT5G59845) were up-regulated.

Transcription factors
Transcription factors in clusters I and II were demonstrated to be strongly connected with group A metabolites abundant in Hongyoung. Of these, four transcripts including A20/ AN1-like zinc finger family protein were down-regulated, and 21 were up-regulated including NAC2, jasmonate zim domain protein 8 (JAZ8), JAZ10, WRKY75, MYB9, and MYB62 (Table 2 and Fig. 3). In cluster III, which was connected with both group B and C metabolites, a homolog of jumonji (jmj) family protein (PKDM7D) was significantly down-regulated and the genes encoding set domain 1 (SET1), TT8, and a transcription factor were up-regulated in both Hongyoung and Jayoung. Moreover, 13 transcripts encoding transcription factors (five in cluster IV and eight in cluster V) were highly correlated with group C and D metabolites abundant in Jayoung (Table 2 and Fig. 4). Of the five transcripts in cluster IV, transcripts for MOS4-associated complex subunit 5A (MAC5A) and a transcription factor were down-regulated and three transcripts for ovate family protein 8 (OFP8), an ERF/AP2 transcription factor, and MYB3 were significantly up-regulated in Jayoung compared with Atlantic. In cluster V, the up-regulation of MYB12 and the down-regulation of seven transcripts including WRKY65, two zinc finger transcription factors, response regulator 2 (RR2), and GATA transcription factor 26 (GATA26) were observed in Jayoung.

Signaling
The expression levels of 30 transcripts encoding proteins implicated in signaling pathways were correlated with flavonoid levels in Hongyoung and/or Jayoung (Table 2 and Fig. 3). Six homologs (four in cluster I and two in cluster II) of genes were significantly up-regulated and five were down regulated both in Hongyoung and Jayoung. The up-or downregulation of transcripts related to signaling in clusters I and II were more significant in Hongyoung than in Jayoung. Furthermore, the significant down-regulation of genes encoding a phototropic-responsive NPH3 family protein, calmodulin 5 (CAM5), wall-associated kinase 3 (WAK3), and a protease inhibitor/seed storage/LTP family protein in cluster IV were observed in Jayoung. Eight transcripts (three in cluster IV and five in cluster V) were significantly up-regulated, including homologous genes encoding early flowering 4 (ELF4), early light-inducible protein (ELIP), and a leucinerich repeat protein kinase in cluster IV and a calcium-binding EF-hand family protein (CBP) and exordium (EXO) in cluster V (

Discussion
In this study, as an effort to elucidate the differential regulation of anthocyanin biosynthesis involved in differential pigmentation of potatoes, a correlation test was performed in light-red-colored Hongyoung and dark-purple Jayoung with 22 anthocyanins and 167 genes categorized to flavonoid metabolism, hormones, transcription factors, and signaling. Of the 167 genes, 119 genes were strongly correlated with the 22 anthocyanins, and the correlation network showed that the genes and metabolites were divided into five clusters (I-V) and four subgroups (A-D) (Fig. 3, Tables 2 and 3). Many of the differentially expressed genes between white and colored potatoes coincided with a recent report performed with potato cultivars "Xin Dang" (white skin and white flesh) and "Hei Meiren" (purple skin and purple flesh) (Liu et al., 2015). In particular, a CHI gene (PGSC0003DMT400030430), two CHSs (PGSC0003DMT400049165 and PGSC0003 DMT400076178), an LDOX (PGSC0003DMT400058554), a flavonoid 3′,5′-hydroxylase (F3′5′H; PGSC0003DMT 400001124), and a basic helix-loop-helix (bHLH) DNAbinding superfamily protein, transparent testa 8 (TT8: PGSC0003DMT400033569) that are increased in the skin and flesh of the purple potato "Hei Meiren" (Liu et al., 2015) were also found to be increased in Hongyoung and Jayoung. However, the relative expression levels of these genes were differential between Hongyoung and Jayoung. Each gene cluster might be functionally connected to the anthocyanin subgroups, regulating the biosynthesis of anthocyanin derivatives that determined the colors of potatoes. Genes in cluster III were shown to be strongly connected with anthocyanins in groups B and C, which were highly accumulated in both lightred Hongyoung and dark-purple Jayoung. The genes in cluster III contained CHS, CHI, and F3′5′H. Expression of these genes was similar in Hongyoung and Jayoung; thus, these genes might be commonly involved in the biosynthesis of anthocyanins in both Hongyoung and Jayoung. Furthermore, genes encoding TT8 and WD40-repeat protein (WD40) in cluster III were strongly up-regulated in both Hongyoung and Jayoung. TT8, a bHLH-type regulation factor, forms a ternary complex with WD40-repeat protein and R2R3-MYB (WD40/bHLH/R2R3-MYB complex); these proteins are involved in the regulation of flavonoid pathways, and specifically in anthocyanin and pro-anthocyanin biosynthesis in Arabidopsis, purple cauliflower, and purple strawberry (Chiu and Li, 2012;Jaakola, 2013;Schaart et al., 2013;Xu et al., 2013). Studies on TT8 promoter activity using WD40 (ttg1), bHLH (tt8, gl3, and egl3), and R2R3-MYB (tt2, myb5, pap1, and pap2) in Arabidopsis mutants showed that the TT8 promoter activity is differentially regulated by various WD40/ bHLH/R2R3-MYB complexes (Chiu and Li, 2012;Xu et al., 2013). In our study, the up-regulation of homologous genes encoding TT8 and WD40 in both Hongyoung and Jayoung indicated that the accumulation of anthocyanin in red-lightand dark-purple-colored potato cultivars was commonly regulated by the TT8-mediated pathway.
Of the genes in clusters I and II, we observed that jasmonic acid (JA) signaling-related genes, including JIH, JAZ8, and JAZ10, were up-regulated in light-red Hongyoung compared with dark-purple Jayoung. JA has been known to increase anthocyanin production and to stimulate gene expression of CHS and UFGT. In Arabidopsis, JA activates the degradation of JAZs, a negative regulator of JA, in a SCF COl1 complex-dependent manner, to abolish the interaction between JAZs and the bHLH/R2R3-MYB complexes, and to stimulate activation of the WD40/bHLH (GL3, EGL3, and TT8)/R2R3-MYB (GL1 and MYB75) complexes, thereby activating the expression of anthocyanin biosynthesis-related genes (Qi et al., 2011(Qi et al., , 2013. Of JA and its oxylipin derivatives, JA-Ile (but not JA, methyl jasmonate, or 12-oxo phytodienoic acid) promotes degradation of JAZ by the formation of the SCF COl1 -JAZ complexes (Thines et al., 2007). JIH catalyzes the cleavage of the JA-Ile conjugate, generating 2-hydroxy-JA. NaJIHsuppressed transgenic tobacco plants showed a dramatic increase in JA-Ile levels during herbivore attacks, thereby enhancing their resistance compared with that of wildtype tobacco (Woldemariam et al., 2012). Thus, the upexpression of JIH most likely reduces JA-Ile levels. Taken together, the transcriptional up-regulation of homologous genes encoding JIH, JAZ8, and JAZ10 probably indicates that JAZ degradation-mediated anthocyanin biosynthesis might be inactivated in light-red Hongyoung.
In this study, most auxin-related genes were shown to be negatively correlated with anthocyanin levels. Genes encoding SAUR66, SAUR71, GH3.6, and MAX1 in clusters I and II were negatively correlated with group A or B anthocyanins that are abundant in light-red Hongyoung. Moreover, we observed a negative correlation between the levels of GH3.1 in cluster V and anthocyanins in group D that are enriched in dark-red Jayoung. Auxin-sensitive SAUR66 and SAUR71 genes are responsive to GH3.1 and GH3.6 genes that catalyze the conjugation of amino acids to auxin. MAX1, a member of the CYP711A cytochrome P450 family, has been known to down-regulate genes involved in the flavonoid pathway, including CHS, CHI, F3H, F3′H, FLS, DFR, ANS, and UFGT (Lazar and Goodman, 2006). Liu et al. (2014) showed that auxins regulated the expression levels of anthocyanin biosynthesis genes in red pap1-D Arabidopsis cells, including genes for six transcriptional factors (TTG1, EGL3, MYBL2, TT8, GL3, and PAP1) and four structural genes (PAL1, CHS, DFR, and ANS) (Liu et al., 2014). The results demonstrated the involvement of auxin in anthocyanin biosynthesis. In other words, the anthocyanin biosynthesis in light-red Hongyoung and dark-purple Jayoung might be regulated by the inactivation of negative regulators, including MAX1.
In contrast to the down-regulation of auxin-related genes, genes involved in ethylene were up-regulated in the two pigmented potatoes compared with the white Atlantic, including homologous genes encoding 1-aminocyclopropane-1-carboxylate oxidase 4 (ACO4) in cluster I, an ethylene response factor 1 (ERF1) in cluster II, and an ERF/AP2 transcription factor in cluster IV. ACO4 converts 1-aminocyclopropane-1-carboxylic acid to ethylene, and ERF1 promotes ethylene production via the ethylene signaling cascade. The up-regulation of ACO4 and ERF1 in both Hongyoung and Jayoung indicated that ethylene may be related to the pigmentation of the two pigmented potatoes. In a recent study, exogenous treatment with the ethylene-releasing compound 2-chloroehtylphosphonic acid was reported to result in the accumulation of anthocyanins in grape skins and to stimulate the long-term expression of CHS, F3H, ANS, and UFGT (El-Kereamy et al., 2003). These results indicated that ethylene was involved in anthocyanin biosynthesis. However, a gene in cluster I encoding NAC2 (also called ANAC092 and ORE1), a positive regulator of ethylene-mediated leaf senescence, was observed to be down-regulated in both Hongyoung and Jayoung compared with Atlantic. Transcriptional expression of NAC2 has been shown to be up-regulated by ethylene insensitive 2 (EIN2), which activates ethylene signaling and induces the expression of senescence-associated genes (Woo et al., 2013). Therefore, the result indicates that anthocyanin biosynthesis in the two pigmented potatoes may not be induced in a senescence-dependent manner activated by the EIN2-NAC2 pathway. Indeed, it has been reported that ethylene suppresses sugar-induced anthocyanin accumulation in Arabidopsis by suppressing the expression of positive regulators of the WD40/bHLH/R2R3-MYB complex and stimulating the expression of the negative R3-MYB regulator MYBL2 (Jeong et al., 2010). Thus, ethylene differentially regulates anthocyanin biosynthesis according to developmental and environmental stimuli. However, studies into their regulation mechanisms are lacking.
In addition to the above genes, Fig. 3 showed that a large number of genes were connected with diverse anthocyanins. Quantitative changes in the secondary metabolites in groups A and B, which were most abundant in the light-red Hongyoung, showed negative or positive correlations with transcriptional changes in genes in clusters I and II. Moreover, the anthocyanin contents in groups C and D, which were most abundant in the dark-purple Jayoung, were negatively or positively correlated with the expression levels of genes in clusters IV and V (Table 2 and Fig. 3). In contrast, the expression level of two F3′5′Hs was not significantly different between Hongyoung and Jayoung (Table 2), while the Dp and Pt derivatives of anthocyanin were highly increased in Jayoung (Table 3). As the biosynthesis of blue Dp-type anthocyanins are known to be driven by the activity of F3′5′H (Ishiguro et al., 2012), there might be other genes responsible for the Dp-based anthocyanin biosynthesis in Jayoung. The more significant up-regulation of LODX, UFGT, CHS, and CHI in cluster IV and PAL2 and NAD(P)linked oxidoreductase (AT1G59960) in cluster V might have role in the more significant accumulation of Dp and Pt derivatives. LODX converts the colorless leucoanthocyanidins into the colored anthocyanidins, which are inherently unstable under physiological conditions (Lo Piero, 2015). The addition of a glucose moiety in the 3-OH positions of anthocyanidins by UFGT increases the hydrophilicity and stability of anthocyanidins, conveying the flux of flavonoid intermediates towards the synthesis of anthocyanins (Lo Piero, 2015). Putting these together, with the significant increase of early-step genes including PAL2, CHS, and CHI in clusters IV and V, it can be postulated that that UFGT and LODX might have role in driving the flux and accumulation of Dp-type anthocyanins in Jayoung.
In conclusion, we explored the regulatory network connected to anthocyanin biosynthesis using integrated analysis of the metabolome and transcriptome in sprouts of three different colored potatoes: light-red Hongyoung, dark-purple Jayoung, and white Atlantic. Correlation analysis between metabolites and regulatory genes identified the regulatory genes associated with anthocyanin metabolites and provided new insight into the regulatory mechanism underlying the biosynthesis of anthocyanin accumulation in colored potatoes. Moreover, a connection network between changes in transcriptional expression and metabolite levels according to the pigmentation was obtained. The dataset could be harnessed by researchers to utilize genetic approaches to clarify the mechanism of anthocyanin regulation.

Supplementary Data
Supplementary data are available at JXB online. Table S1. Primer list used in qPCR analysis. Table S2. Exact mass of aglycones, sugars, and acylated groups found in anthocyanins and flavonoid glycosides. Table S3. Identification of anthocyanin biosynthesisrelated compounds with MS/MS spectra obtained in ESI + and ESI − modes. Table S4. List of 167 genes categorized to hormones, signaling, transcriptional regulation, and flavonoid metabolism. Table S5. The correlation matrix of metabolites (anthocyanins) and gene expression levels. Table S6. Interaction value between 22 metabolites and 119 genes that has a strong correlation coefficient (R 2 >0.9). Fig. S1. Tubers and sprouts of Solanum tuberosum cvs Atlantic, Hongyoung, and Jayoung. Fig. S2. Structure and molecular formulae of anthocyanidins. Fig. S3. Score plots and S-plots of orthogonal partial leastsquares discriminant analysis (OPLS-DA) in positive (A) and negative (B) modes. Fig. S4. Identification of anthocyanin derivatives using MS 2 fragmentation. Fig. S5. The number of differently expressed genes among Hongyoung, Atlantic, and Jayoung. Fig. S6. Quantitative real-time RT-PCR (qPCR) analysis of genes involved in anthocyanin biosynthetic pathway and putative transcriptional regulators according to different color potato cultivars.