The Arabidopsis TDS4 gene encodes leucoanthocyanidin dioxygenase (LDOX) and is essential for proanthocyanidin synthesis and vacuole development


For correspondence (fax +61 02 62465000; e-mail

These authors contributed equally to this paper.


The anthocyanin and proanthocyanidin (PA) biosynthetic pathways share common intermediates until leucocyanidin, which may be used by leucoanthocyanidin dioxygenase (LDOX) to produce anthocyanin, or the enzyme leucoanthocyanidin reductase (LAR) to produce catechin, a precursor of PA. The Arabidopsis mutant tannin deficient seed 4 (tds4-1) has a reduced PA level and altered pattern PA accumulation. We identified the TDS4 gene as LDOX by complementation of the tds4-1 mutation either with a cosmid encoding LDOX or a 35S:LDOX construct. Independent Arabidopsis lines with a T-DNA insertion in the LDOX gene had a similar phenotype, and one was allelic to tds4-1. The seed phenotype of ban tds4 double mutants showed that LDOX precedes BANYULS (BAN) in the PA pathway, confirming recent biochemical characterisation of BAN as an anthocyanidin reductase. Double mutant analysis was also used to order the other TDS genes. Analysis of the PA intermediates in tds4-1 revealed three dimethylaminocinnamaldehyde (DMACA) reacting compounds that accumulated in extracts from developing seeds. Analysis of Arabidopsis PA and its precursors indicates that Arabidopsis, unlike many other plants, exclusively uses the epicatechin and not the catechin pathway to PA. Transmission electron microscopy (TEM) showed that the pattern observed when seeds of tds4 were stained with DMACA was a result of the accumulation of PA intermediates in the cytoplasm of endothelial cells. Fluorescent marker dyes were used to show that tds4 endothelial cells had multiple small vacuoles, instead of a large central vacuole as observed in the wild types (WT). These results show that in addition to its established role in the formation of anthocyanin, LDOX is also part of the PA biosynthesis pathway.


Flavonoids are plant secondary metabolites whose roles include attracting pollinators, pathogen defence, plant-microbe signalling processes and UV protection. Products of the flavonoid pathway include the monomeric anthocyanins and the polymeric proanthocyanidins (PA; Figure 1). In Arabidopsis, anthocyanin accumulates in tissues such as hypocotyls of emerging seedlings, leaves, stems, siliques and seed coat; however, PA accumulation is confined to the endothelial cell layer of the seed coat. PA consists of a variety of flavonoid monomers that are polymerised together mainly through a C4-C8 link but occasionally through a C4-C6 link (Figure 1). Monomers consist of variations of two different series of isomers, namely isomers containing one, two or three hydroxyls on the B-ring, and epimers that vary in the configuration of the 3-hydroxyl of the C-ring, typified by 2,3-trans catechin and 2,3-cis epicatechin (Figure 1; Stafford, 1990). The monomers are further distinguished into the flavon-3-ol monomers such as catechin and epicatechin that initiate the polymerisation process by reacting with the flavon-3,4-diols (e.g. leucocyanidin) to form a PA dimer (Stafford, 1990). The polymerisation is believed to continue by progressive addition of flavon-3,4-diols (Stafford, 1990). The C-ring 3-hydroxyl whose orientation differs between catechin and epicatechin is introduced into the flavonoid nucleus by the 2-oxoglutarate-dependent oxygenase flavanone 3β-hydroxylase (F3H) in the 2,3-trans configuration (Britsch and Grisebach, 1986). The PA initiator catechin is synthesised from 2,3-trans leucocyanidin by the enzyme leucoanthocyanidin reductase (LAR; Tanner and Kristiansen, 1993), recently found to be a member of the isoflavone reductase (IFR) group of enzymes (Tanner et al., 2003). The discovery that the Arabidopsis BANYULS (BAN) protein (Devic et al., 1999) is an anthocyanidin reductase that reduces anthocyanidin to epicatechin (Xie et al., 2003) explains the origin of the PA initiator epicatechin. An outstanding problem in the biosynthesis of PA has been the origin of the 2,3-cis 3,4 diol extention units, which constitute the bulk of PA subunits in most plants (Stafford, 1990). It is also not known if the extension units of Arabidopsis PA are composed of either 2,3-trans or 2,3-cis forms or both.

Figure 1.

Anthocyanin and PA synthesis pathway in Arabidopsis.

The anthocyanin and PA pathway from chalcone synthase is shown. The enzymes LDOX and BAN act on the substrates leucocyanidin and cyanidin, respectively, to produce epicatechin. The origin of the extension unit shown in the polymer is not known. There does not appear to be an LAR homologue in the Arabidopsis genome, nor does Arabidopsis appear to make catechin, so this is indicated by a dashed line. The difference between 2,3-cis and 2,3-trans isomers is shown using catechin and epicatechin as examples (inset). Abbreviations: CHS, chalcone synthase; CHI, chalcone isomerase; F3H, flavanone 3β-hydroxylase; F3′H, flavonoid 3′ hydroxylase; FLS, flavonol synthase; DFR, dihydroflavonol 4-reductase; LDOX, leucoanthocyanidin dioxygenase; BAN, BANYULS (anthocynidin reductase); UFGT, UDP glucose-flavonoid 3-O-glucosyl transferase.

Although the early steps in the synthesis of anthocyanin and PA occur in the cytoplasm, the end products accumulate within the vacuole. Enzymes of the pathway such as dihydroflavonol reductase (DFR), LAR and BAN are located in the cytoplasm, together with the other substrate NADPH, so monomer formation takes place in the cytoplasm. As polymerised PA interacts more strongly with proteins than the monomers (Hagerman and Butler, 1981), it is likely that monomers are transported into the vacuole and polymerised there. The TT12 gene has been identified as encoding a multi drug and toxic compound extrusion (MATE) transporter specifically involved in the PA synthesis (Debeaujon et al., 2001), suggesting that this protein transports intermediates into the vacuole. We do not know if this protein transports both initiating and extension units or has a more specific function, and requires the participation of other transporters to allow vacuolar assembly of PA. The process of vacuole formation in plants is not well understood, although it appears that provacuoles are formed by budding from the endoplasmic reticulum (Bethke and Jones, 2000; Marty, 1999). These provacuoles fuse with each other or with the tonoplast by a process that involves specific membrane-targeting and fusion events (Lin et al., 2001; Sanderfoot and Raikhel, 1999; Sanderfoot et al., 2001). The molecular events necessary for vacuole biogenesis are poorly characterised, and it is not known what signals are involved in the initiation of vacuole formation.

Genetic analysis of PA-deficient mutants in Arabidopsis is proving a powerful way to dissect and define the PA pathway. The Arabidopsis ban mutant described by Devic et al. (1999) was characterised by its ability to produce enhanced levels of anthocyanin at the expense of PA. Because of the similarity of BAN to DFR, as well as the ban mutant phenotype, BAN was considered as a candidate LAR in Arabidopsis (Devic et al., 1999), acting at the branch-point between anthocyanin and PA synthesis pathways. However, BAN has recently been characterised as an anthocyanidin reductase, utilising the substrate cyanidin to form epicatechin (Xie et al., 2003), an intermediate in PA synthesis. As the leucoanthocyanidin dioxygenase (LDOX) reaction produces cyanidin (Saito et al., 1999), the substrate for BAN, this places LDOX before BAN in the pathway (Figure 1). In addition to this, an LAR cDNA has been recently cloned from Desmodium uncinatum (Tanner et al., 2002), and although being a member of the reductase-epimerase-dehydrogenase family like DFR and BAN, it is most closely related to the IFR group of this family. Although species such as Medicago sativa (Koupai-Abyazani et al., 1993) and Lotus corniculatus (Foo et al., 1996) use both catechin and epicatechin as initiators in PA synthesis, Arabidopsis appears to use only epicatechin (Abrahams et al., 2002). Furthermore, there does not appear to be an LAR orthologue in the Arabidopsis genome (Tanner et al., 2002). These findings suggested that LDOX, previously thought to be involved in only anthocyanin synthesis, might also be involved in PA synthesis, when the intermediate epicatechin is utilised. An LDOX cDNA has been cloned (Pelletier et al., 1997), and the position of the LDOX gene is known from mapping (Pelletier et al., 1997) and from Arabidopsis genome sequence annotations. There are also reports in the literature mentioning transparent testa18 (Xie et al., 2003) and transparent testa19 (Winkel-Shirley, 2001), both being ldox mutants.

We have previously identified six tannin deficient seed (tds) mutants involved in the formation of PA in the Arabidopsis seed coat (Abrahams et al., 2002). This paper describes the characterisation of the TDS4 gene from a mutant identified by its distinctive pattern of PA accumulation when stained with dimethylaminocinnamaldehyde (DMACA; Abrahams et al., 2002), its identification as the LDOX gene, and its role in PA biosynthesis. Seed coat endothelial cells of the tds4 mutant accumulate small PA-containing globules, suggesting that transport processes are uncoupled as a consequence of the mutation. We also report the ordering of the TDS genes in the pathway by double mutant analysis, showing genetically that LDOX functions before BAN. Finally, chemical analysis of Arabidopsis PA shows that both PA-initiating and extension units contain only 2,3-cis isomers. The presence of a single 2,3-cis PA pathway in Arabidopsis rather than the apparently redundant 2,3-cis and 2,3-trans pathways in other plants probably contributes to the success of Arabidopsis as a model to study PA biosynthesis.


TDS4 maps to chromosome IV

Although tds4-1 was isolated from the T-DNA tagged lines from Institut National de la Recherche Agronomique (INRA), the seed did not grow on selective media, nor was there any detectable hybridisation on Southern blots with T-DNA probes, indicating that the TDS4 gene was not tagged. To map the tds4-1 mutation to a particular chromosome, five markers were selected that mapped to the middle of each chromosome. Results of scoring these markers using 21 mutant recombinant lines showed that TDS4 clearly mapped to chromosome IV, with a score of 15 lines identical to tds4-1 parent, six heterozygous and none resembling the Col-7 parent. A total of 139 F2 recombinant lines with the tds4-1 DMACA-stained seed phenotype were analysed with a selection of markers (Table S1). The relative positions of markers between C57K190 and C58K163 that defined the region of the tds4-1 mutation are shown in Figure 2. No other polymorphisms were found near these markers to reduce this region.

Figure 2.

Mapping of tds4-1.

Chromosome IV is represented on the left. An expanded view shows markers between contig C45 and C62 used in mapping the region. A 170-kb region defined by recombinants mapped with markers C57K190 and C58K163 is illustrated. The cosmids (T) that were used for transformation are also shown.

Molecular complementation of tds4-1

As LDOX was known to be involved in anthocyanin/PA biosynthesis, and the mapping work indicated that the TDS4 gene is mapped close to LDOX (Pelletier et al., 1997), we decided to sequence the LDOX gene in the tds4-1 mutant. When the coding region of the LDOX gene was PCR amplified from Ws-4 and tds4-1 DNA and sequenced, we found that a GGT codon was mutated to GAT, changing the amino acid 228 in the predicted protein from a glycine to an aspartate. Evidence that TDS4 is the LDOX gene came from molecular complementation of the tds4-1 allele. Four cosmids containing 25-kb genomic inserts (Figure 2) were used to transform tds4-1. The left panel of Figure 3 shows mature dry Arabidopsis seed, and the right panel shows DMACA-stained aliquots of the same seed. The seed phenotype of Ws-4 (Figure 3a,b) and tds4-1 (Figure 3c,d) is shown. When the T46 cosmid containing the LDOX gene was introduced into tds4-1, several transformants had a phenotype similar to the wild types (WT), typified by gLDOX-T46 shown in Figure 3(e,f). The adjacent cosmids T190, T17 and T97 were unable to complement the tds4-1 mutant (data not shown). The ends of the T46 cosmid were sequenced, and the cosmid was found to include a disrupted At4g22850 gene, an unclassified expressed protein (At4g22860), a truncated LDOX gene (At4g22870; see the section under Discussion), the full-length LDOX gene (At4g22880) and a disrupted At4g22890 gene. A 35S:LDOX construct was also able to complement the mutant (Figure 3g,h). Several of the transformants from the 35S:LDOX complementation experiment had phenotypes intermediate between those of tds4-1 and the WT (results not shown).

Figure 3.

Molecular complementation of tds4-1.

The left series of figures show unstained mature seed of (a) Ws-4, (c) tds4-1, and (e) tds4-1 complemented with the genomic gLDOX cosmid and (g) the 35S: LDOX construct. The right series of figures (b, d, f, h) show aliquots of the same seed stained with DMACA. The bar represents 0.05 mm (a–h).

tds4-1 is an allele of the LDOX gene

While the mapping and complementation of tds4-1 was proceeding, we had obtained seeds of T-DNA insertion lines in the LDOX gene from Arabidopsis seed stock centres. The anthocyanin-accumulating phenotype of the ban mutant (Devic et al., 1999) suggested that anthocyanin synthesis was competing with PA synthesis for leucocyanidin intermediates, so we examined the seed of LDOX T-DNA insertion lines for evidence of increased PA production. Instead of increased PA production, we found the ldox mutant seed had much reduced PA and looked very similar to our tds4-1 mutant. Figure 4(a) shows the LDOX gene and the relative positions of T-DNA insertion sites or G228D mutation in the alleles of tds4. Figure 4(b) shows in detail the region of the LDOX protein affected by the G228D mutation, and its proximity to the active site motif H232, X, D234 (HXD). The DMACA phenotype of the tds4-1, tds4-3 and tds4-4 alleles is similar (Figure 4d,f; tds4-4 not shown), whereas that of tds4-2 was marginally darker and the pattern of staining less distinct (Figure 4e). The subtle difference observed between tds4-2 and other alleles might be because of the T-DNA insertion site in tds4-2 being 3′ in the gene relative to that of tds4-3 and tds-4, resulting in a less extreme phenotype. The DMACA reagent reacts with PA-related intermediates such as leucocyanidin, epicatechin and PA polymers, giving a characteristic blue/green colour when the concentration is within the linear range of an epicatechin standard curve (0.05–10 µg). Other flavonoid compounds in the pathway can also react with DMACA to give a pale blue colour, but approximately 100–1000-fold more of the compound is required to give a positive reaction. When mature dry seeds are stained with DMACA, a positive reaction is dark purple/brown, because of other pigments in the seed and the quantity of PA being detected. The similarity observed between alleles of tds4 suggested that the DMACA phenotype observed in tds4-1 was caused by a mutation in the LDOX gene and not by another mutation elsewhere in the genome. Because of the similar appearance of DMACA-stained seed of tds4-1 and LDOX T-DNA insertion lines, we crossed tds4-1 with tds4-2 to perform a complementation test. As expected for a gene expressed in the seed coat, the resulting F1 seed had the same phenotype as that of the maternal parent. The DMACA phenotype of the F2 seed was the same as that of tds4-1, confirming that TDS4 and LDOX were the same gene.

Figure 4.

The insertion sites or point mutations creating alleles of tds4.

Sketch of TDS4 gene (a) showing the two exons and one intron arrangement of the gene and insertion sites for the T-DNA lines and point mutation.

(b) A section of the TDS4 protein showing conserved amino acids underlined. The position of the mutated Gly 228 amino acid is indicated by an arrow, and the location of His 232 and Asp 234 of the HXD motif indicated by the triangles.

(c) Mature WT seed stained with DMACA is uniformly stained under the conditions used. Seeds of (d) tds4-1, (e) tds4-2 and (f) tds4-3 show a characteristic pattern when stained with DMACA. The bars represent 0.4 mm.

tds4 mutant cotyledons lack anthocyanin

The expression of anthocyanin is a variable character, dependent on growth conditions, including temperature and light levels. Previously, when tds4-1 plants were grown in soil in a growth cabinet under high light conditions to induce anthocyanin synthesis, anthocyanin was produced at 30% of the amount in WT leaves (Abrahams et al., 2002). We were therefore interested to observe the anthocyanin phenotype in cotyledons of tds4 alleles. Seedlings were grown for 4 days on plates containing 5% sucrose. The tds4 alleles were in Ws-4, Col-2 and Landsberg erecta (Ler) ecotypes, which all had dark purple hypocotyls and anthocyanin in the cotyledons (Figure 5a,d,g). tds4-1 and tds4-3 do not make anthocyanin in the cotyledons under these conditions (Figure 5b,h), and are similar to tt3-1 (Figure 5i), which lacks DFR and is therefore unable to make anthocyanin. However, tds4-2 appears to make intermediate levels of anthocyanin, which is consistent with the intermediate levels of PA observed in this allele when seed is stained with DMACA (Figure 3e). The 35S:LDOX-5 transformant (Figure 5c) has levels of anthocyanin similar to that of the WT, indicating that, in addition to the seed PA phenotype (Figure 3g,h), the leaf anthocyanin phenotype has also been complemented. As expected, the ban mutant looked similar to the WT (Figure 5f).

Figure 5.

Anthocyanin phenotype of 4-day-old seedlings.

Photographs of seedlings show upper hypocotyl and cotyledons. Ws-4, tds4-1, 35S:LDOX-5, Col-2, tds4-2, ban F36, Ler, tds4-3 and tt3-1 (a–i, respectively).

Double mutant analyses shows LDOX precedes BAN in the PA pathway

The phenotype of a double mutant will be the same as that of the mutant gene encoding a protein upstream in a linear metabolic pathway, and the phenotype of the upstream gene is regarded as epistatic to that of the downstream gene. Thus, double mutant analysis can order genes in a pathway provided they have distinct phenotypes. Crosses of tds4-1 with ban, tds1, tds2, tds3, tds5 and tds6 were made to determine the order in which these genes acted in the PA pathway (Table 1). Significantly, when tds4-1 was crossed with ban, we found that the double mutant DMACA-stained seed phenotype was that of tds4-1, suggesting that TDS4 occurred before BAN in the pathway. Devic et al. (1999) had previously reported that tt3 (dfr) was epistatic to ban. This placed TDS4 before BAN and probably after DFR in the PA synthesis pathway, and suggested that DFR, the producer of leucocyanidin, and BAN were not immediately sequential in the pathway.

Table 1.  Genotype and DMACA-stained seed phenotype of double mutants
Double mutant genotypePhenotype
tds1 tds4tds4
tds2 tds4tds4
tds3 tds4tds4
tds5 tds4tds4
tds6 tds4tds4
ban tds4tds4
tds1 banban
tds2 banban
tds3 banban
tds5 banban
tds6 banban
tds1 tds3tds3
tds2 tds3tds3
tds5 tds3tds3
tds6 tds3tds3

When tds4-1 was crossed with the remaining tds mutants, we found that the double mutants tds1 tds4, tds2 tds4, tds3 tds4, tds5 tds4 and tds6 tds4 all showed the tds4 DMACA-stained seed phenotype (Table 1), suggesting that the products of these genes act after TDS4 (LDOX). Similarly, the double mutants tds1 ban, tds2 ban, tds3 ban, tds5 ban and tds6 ban all showed the ban DMACA phenotype (Table 1), indicating that they act after BAN in the PA pathway.

The tds3 mutant seed can be distinguished from that of the other tds mutants, because tds3 does not stain with DMACA at the chalazal end of the seed, whereas the other tds mutants do. When tds3 was crossed with the other tds mutants, we found that the DMACA phenotype of the double mutants tds1 tds3, tds2 tds3, tds5 tds3 and tds6 tds3 was that of tds3. When taken together with the above double mutant seed phenotypes, this suggests that the order of these genes in the PA pathway is TDS4, BAN, TDS3 and then the remaining TDS genes (TDS1, TDS2, TDS5 and TDS6), whose order cannot be easily defined as we are unable to distinguish between mutants in these genes on the basis of their DMACA phenotype.

tds4 alleles produce cytoplasmic globules of PA intermediates and multiple small vacuoles

Three alleles of tds4 gave similar phenotypes when seeds were stained with DMACA (Figure 5). Because the DMACA staining pattern suggested that PA intermediates were accumulating at sites other than the vacuole, we investigated the ultra-structural detail of the PA deposits in tds4-1 using transmission electron microscopy (TEM). Samples of WT- and tds4-1-developing seeds were treated with osmium tetroxide, sectioned and then examined by TEM (Figure 6). The endothelial cell layer was easily identified by the characteristically strong reaction of PA with osmium tetroxide in these cells (Figure 6a). In WT sections, the entire cell is occupied with a PA-containing vacuole, although, within the vacuole, there are regions that do not contain osmium-reacting material. In tds4-1, there are numerous small vacuoles and large dark globules of osmium-reacting PA intermediates (Figure 6b). The size range of the globules was usually between 0.2 and 2 µm, with occasional globules up to 5 µm in size. The globules appeared not to be surrounded by a membrane, and to be distinct from the vacuoles. Higher magnification of the globules (Figure 6c,d) revealed that they were made of many smaller aggregated vesicles. These vesicles were approximately 15 nm in diameter. Towards the chalazal end of the seed, small vacuoles with an osmium-reacting tonoplast were observed (Figure 6c), which is typical of small quantities of PA within and at the periphery of the vacuole (Amelunxen and Heinze, 1984; Hilling and Amelunxen, 1985), and is observed in similar sections of WT seeds at very early stages of PA accumulation (not shown).

Figure 6.

TEM of osmium-treated developing seeds.

Wild-type cells of the endothelial layer of the seed coat (a) produce a single large vacuole containing osmium-reacting PA. The endothelial cells in tds4-1, at the chalazal end of the seed (b), contain globules of osmium-reacting material. The cells appear to form multiple small vacuoles that do not fuse to form a large vacuole. The bars represent 5 µm in (a) and (b). Higher magnification of the osmium-reacting material located in the cytoplasm is shown in (c, d). The enclosed area in (c) is enlarged in (d) to show the fine structure of the globules. The bar in (c) represents 1 µm and that in (d) 0.1 µm. v, vacuole; CW, cell wall; PA, proanthocyanidin.

The TEM of tds4-1-developing seeds had shown that there were multiple small vacuoles in the endothelial cell layer instead of a large central vacuole found in WT. We used vacuolar marker dyes to confirm this observation in the developing seed, and to determine whether vacuole morphology in other organs was also affected by mutations in TDS4 (Figure 7). The fluorescent dye 5-carboxy-2′,7′-dichlorofluorescein diacetate (DCFDA) is often used as a marker for the vacuolar lumen. Developing WT seed endothelial cells treated with carboxy-DCFDA for 2 h (Figure 7a) appeared to contain a large circular fluorescent body, shown in green pseudocolour, occupying almost the entire contents of the cell, characteristic of the large central vacuole observed in these cells. Occasional, small, more intensely labelled structures could also be seen. Starch granules appeared as very small red structures, mostly around the periphery of the cell. Dissecting open the developing seeds and removing the embryo made it easy to distinguish between inner and outer seed coat cell layers, and ensured that we were observing the endothelial cells. In contrast to the WT, the endothelial cells in tds4-1 and tds4-3 showed numerous small fluorescent bodies when treated with carboxy-DCFDA (Figure 7b,c). We have interpreted these to be small vacuoles, also observed in the TEM sections of tds4-1 (Figure 6b). Endothelial cells were also treated with the vacuolar membrane marker MDY-64. These samples also showed a single large vacuole in the WT and numerous small vacuoles in tds4-1 (results not shown).

Figure 7.

Vacuole marker dyes in developing seed and leaf tissue.

DCFDA loaded vacuoles in (a) WT, (b) tds4-1 and (c) tds4-3 endothelial cells. WT cells contain a single large vacuole, whereas tds4-1 and tds4-3 cells contain multiple small vacuoles. MDY targets the vacuolar membrane of leaf epidermal cells, which appear to be the same in (d, e) WT and (f) tds4-1 leaf tissue. The bars represent 10 µm.

As LDOX is also expressed in leaf tissue, we were interested to examine for any changes in vacuole structure in tds4-1 leaves. We used the fluorescent vacuole membrane marker MDY-64 to analyse WT and tds4-1 leaves (Figure 7d–f), and showed that there is no obvious difference in vacuole size or morphology between WT and tds4-1 leaves. Root tissue also did not show any changes to vacuole morphology (results not shown), as might be expected, because root tissue does not accumulate anthocyanin or PA.

Analysis of the intermediates accumulating in tds4-1

Apart from PA, the intermediate epicatechin is normally the only DMACA-reacting compound, isolated from 70% acetone extracts of developing or mature WT seeds of Arabidopsis (Abrahams et al., 2002). However, similar extracts of tds4-1-developing seed contain three DMACA-reacting intermediates when analysed by thin layer chromatography and sprayed with DMACA (Figure 8a). We have previously shown that none of these intermediates are epicatechin (Abrahams et al., 2002). These intermediates do not appear to stain with DMACA with the same sensitivity as compounds such as catechin appear (Figure 8a). The catechin standard reacts immediately with DMACA on thin layer chromatography (TLC) plates, giving a strong distinctive colour, whereas the intermediates in tds4-1 extracts take 5–10 min to develop a colour, and are much paler than the catechin standard. A similar reaction occurs with compounds such as leucocyanidin, naringenin, eriodictyol and dihydroquercetin, which are structurally similar to the catechins, but do not react with the same sensitivity as the catechins react with DMACA (results not shown). Presumably, the DMACA-reacting intermediates we see on TLC plates are the osmium-reacting intermediates that accumulate in the globules observed in the TEM sections.

Figure 8.

Biochemical analyses of tds4-1 and Arabidopsis PA.

TLC of DMACA-reacting intermediates in tds4-1 and catechin (cat) standard (a). The Rf values for the intermediates were 0.75, 0.53 and 0.46; Rf for catechin and epicatechin were 0.75 and 0.65, respectively. The HPLC trace of phloroglucinol hydrolysis of WT PA including the structure of epicatechin (ecat) and epicatechin-phloroglucinol (ecat-phloroglucinol) PA derivatives (b).

Composition of Arabidopsis PA

As the composition of Arabidopsis PA has not yet been analysed, we used acid hydrolysis in the presence of phloroglucinol to determine the initiating and extension units of the PA polymer (Kennedy and Jones, 2001). To distinguish between catechin-(4α-6)-phloroglucinol and epicatechin-(4β-6)-phloroglucinol, both potential products of hydrolysis of Arabidopsis PA, the standards procyanidin B3 and B2 were used (Porter, 1989). Procyanidin B3 is a dimer of catechin, and can yield only catechin-(4α-6)-phloroglucinol, whereas procyanidin B2 is a dimer of epicatechin, and can yield only epicatechin-(4β-6)-phloroglucinol. Hydrolysis in phloroglucinol of Arabidopsis PA yields two new dominant peaks on HPLC chromatograms, which together account for over 90% of the total peak area (excluding ascorbate, phloroglucinol and unhydrolysed PA) (Figure 8b). The major peak resulting from Arabidopsis PA hydrolysis, representing the extension units, co-chromatographs with epicatechin-(4β-6)-phloroglucinol produced by hydrolysis of procyanidin B2. This peak is resolved from catechin-(4α-6)-phloroglucinol, produced by hydrolysis of procyanidin B3 (results not shown). The minor peak co-chromatographs with authentic epicatechin (10.5% of peak area), and represents the terminal or initiator units of the polymer. The mean degree of polymerisation (Dp) for Arabidopsis PA was calculated to be 8.


Prior to this work, LDOX had been mapped to chromosome IV, less than 1 cM from AG and 10.7 cM from M600 (Pelletier et al., 1997). However, no mutant has been characterised that corresponds to an alteration in this gene. The data presented here demonstrate that TDS4 is the LDOX gene (At4g22880). The original tds4-1 mutant has a G228D mutation that is close to the enzyme active site. Three lines with T-DNA insertions in the At4g22880 gene showed a tds4 phenotype, and one of these insertion mutants was shown to be allelic to tds4-1. Finally, complementation of the tds4-1 mutant by transformation with either a genomic fragment containing LDOX or a 35S:LDOX construct indicate that a mutation in the At4g22880 gene is responsible for the tds4 mutant phenotype.

The annotated At4g22870 gene preceding LDOX on chromosome IV is very similar to LDOX. At4g22870 encodes a putative protein of 112 amino acids that is 97% identical and collinear with At4g22880 from residue 245 to its C-terminal amino acid 356. The nucleotide similarity to the LDOX gene and flanking regions starts six bases before the predicted initiating Met and ends about 20 bases past the stop codon. This region of 367 nucleotides, 2540 nucleotides downstream of the LDOX stop codon, is 96% identical to the LDOX gene. At4g22870 is probably a recent partial duplication of the LDOX gene, rather than partial truncation of an ancient duplication, because the sequences are very similar.

Recently, Xie et al. (2003) demonstrated that BAN is capable of using cyanidin as a substrate to produce epicatechin, suggesting that the enzyme producing cyanidin (i.e. LDOX) precedes BAN in the PA pathway. The double mutant analyses reported here confirms the order of enzymes in the PA pathway to be LDOX and then BAN. Thus, combined biochemical (Xie et al., 2003) and genetic evidence suggests that LDOX plays a role in PA synthesis. Furthermore, by analysing the seed DMACA phenotype of double mutants, we have also shown that TDS3 acts after BAN in the PA pathway, and that TDS1, TDS2, TDS5 and TDS6 act after TDS3 in the pathway. At present, we are unable to further distinguish between tds1, tds2, tds5 and tds6 mutant seed phenotypes, and thus, we cannot identify double mutants of these genes.

On the basis of our allelic and molecular complementation data, the phenotype of the double mutants, and biochemical evidence, we propose the PA synthesis pathway in Arabidopsis as is shown in Figure 1. Significantly, the enzyme LAR is not included in this scheme. Our group has recently purified and cloned LAR from D. uncinatum (Tanner et al., 2003). The LAR sequence is most closely related to eight IFR-like homologues in the Arabidopsis genome, but is less than 50% identical to any of them. In a phylogenetic analysis, the D. uncinatum LAR is distinct from the Arabidopsis IFR genes that cluster with defined IFR-related genes from other plant species such as chickpea and Pinus taeda. Given the lack of any obvious LAR orthologue in the Arabidopsis genome, and that Arabidopsis appears to make only epicatechin and not catechin, we conclude that the enzyme LAR is absent from Arabidopsis. In Arabidopsis, the location of the branch-point between anthocyanin and synthesis of the initiating PA monomer is at cyanidin, and not at leucocyanidin as previously thought. In other species that produce PA containing both catechin and epicatechin (and contain both LAR and BAN), there are two branch-points, the first at leucocyanidin, utilised by LAR to make catechin, and the second at cyanidin, utilised by BAN to make epicatechin. It is fortunate that a simplified version of the pathway occurs in Arabidopsis, which made it possible to readily identify the role of LDOX in PA biosynthesis. If Arabidopsis also possessed LAR, the mutant phenotype might be masked by this alternate route to PA.

The finding that LDOX is part of the PA biosynthesis pathway in Arabidopsis clarifies some previously intriguing and confusing observations about the pathway. Nesi et al. (2001) found that TT2 regulated not only the expression of DFR, BAN and TT12 but also LDOX mRNA. Devic et al. (1999) found that LDOX mRNA was present in the endothelial layer of the seed coat. These results suggested that TT2 positively regulated both the anthocyanin and the PA branches of the flavonoid pathway. Leucocyanidin has at least three fates in flavonoid metabolism. It can be converted to catechin by LAR (but not in Arabidopsis), to anthocyanidin by LDOX, and is also believed to be the extension unit that reacts with the PA-initiating units, catechin and epicatechin to start polymerisation of PA. Extension units must necessarily be more abundant than initiating units when synthesising polymers, and as the extension units are precursors of the initiating units, the pathway at this step must be carefully regulated to maintain a sufficient pool of leucocyanidin. The reported TT2-dependent synthesis of LDOX that would deplete the leucocyanidin pool was therefore surprising. We had previously analysed Arabidopsis seeds for anthocyanin and PA content to characterise the tds mutant phenotypes (Abrahams et al., 2002), and found that there was very little anthocyanin being produced in WT seeds. Devic et al. (1999) had also noted that there was a long time gap between the disappearance of transcripts for DFR and LDOX (heart to torpedo stage of embryo development) and the appearance of visible coloration in the seed coat. This result can now be interpreted as DFR and LDOX gene expression are associated with PA synthesis, rather than with anthocyanin synthesis. PA-containing cells in leaves of legumes such as D. uncinatum and Onobrychis are colourless, again suggesting that anthocyanin and PA synthesis do not occur in the same cell. In developing Vitis vinifera grapes, PA is made during early ripening, while anthocyanins are synthesised later, after veraison; yet LDOX mRNA is expressed at both stages (Boss et al., 1996). A more reliable indicator of anthocyanin synthesis than the expression of LDOX was found to be the expression of UDP-glucose flavonoid 3-O-glucosyl transferase, a later step in the modification of the primary anthocyanin structure (Bloor and Abrahams, 2002). These observations are consistent with LDOX participating in PA biosynthesis during early ripening and in anthocyanin synthesis during veraison.

In contrast to TT2, TT8, a basic helix-loop-helix transcription factor required for PA accumulation in the endothelial layer (Nesi et al., 2000), has been reported to be not essential for LDOX mRNA expression in developing seeds (Nesi et al., 2000; Pelletier et al., 1997). However, we find that two tt8 insertion alleles isolated in our search for tds mutants (Abrahams et al., 2002) had negligible amounts of LDOX transcript in developing seeds (Figure S1). This disparity could be attributed to the severity of the different tt8 alleles used. The revised position of the branch-point between anthocyanin and PA being at cyanidin clarifies the regulation of the DFR, LDOX and BAN genes. In Arabidopsis, all three genes are positively regulated by TT2 and TT8 because their protein products act sequentially in the epicatechin-based PA pathway, and not at a branch-point.

The tds4-1 mutant is the result of a G228D mutation in Gly 228, a very highly conserved residue in plant 2-oxoglutarate-dependent dioxygenases, and close to His 232 and Asp 234, which are co-ordinated to the iron in the active site of LDOX (Wilmouth et al., 2002). The crystal structure of LDOX indicates the involvement of His 232, Asp 234 and His 288 residues and 2-oxoglutarate in the binding of iron in the active site. Hence, conservation of the so-called HXD motif is fundamental to the catalysis by LDOX. The tds4-2 and tds4-3 alleles both are caused by the insertion of a T-DNA near the carboxy-terminal region of the protein. The C-terminal region of LDOX forms a loop-strand-helix structure that covers the active site, maintaining the flavonol substrate in the correct position during catalysis (Wilmouth et al., 2002). In the mutant alleles, inactive mutated or truncated proteins may be produced. We have found three DMACA-reacting intermediates in tds4-1 siliques that accumulate to much higher levels than in WT siliques. The substrate of LDOX in vivo is the reactive intermediate leucocyanidin, so we might expect its levels to build up in the absence of LDOX activity. Feedback inhibition of DFR, the enzyme that produces leucocyanidin, and the accumulation of dihydroquercetin isomers, or their glucoside derivatives, may account for some of the novel intermediates. Because leucocyanidin is reactive and unstable, it may transform into the other compounds, and thus, the chromatographically distinct DMACA-reacting species that we find in tds4-1 siliques may be in vivo artefacts. Alternatively, as recombinant native LDOX produces a wide array of products in vitro (Turnbull et al., 2000), the G228D LDOX may convert leucocyanidin to products other than cyanidin.

The phloroglucinol hydrolysis of Arabidopsis PA indicates that the terminal unit for PA is epicatechin, which is consistent with the identification of epicatechin in PA extracts of WT seeds (Abrahams et al., 2002). The extension units have the same 2,3-cis stereochemistry as epicatechin. The origin of the intermediate, which gives rise to the 2,3-cis extension units, is not known; however, the identification of Arabidopsis as a species that appears to use only one isomer for polymerisation should assist in the eventual identification of the source of this intermediate in PA synthesis. The Arabidopsis polymer is on average 8 units in length, which is longer than PA from L. corniculatus leaves, with a Dp of 6–7 (Foo et al., 1996), and Trifolium repens flowers, with a Dp of about 6 (Foo et al., 2000), and Medicago sativa seed coat, with a Dp of 5 (Koupai-Abyazani et al., 1993). Analysis of V. vinifera seed coat PA indicated a Dp of 2.6 and that of 4.4 for Chardonnay and Shiraz cultivars, shorter than Arabidopsis seed coat PA, whereas grape berry PA Dp was 15.7–12.1 for Chardonnay and Shiraz cultivars, respectively (Kennedy and Jones, 2001). The difference in polymer length between tissues might reflect a true difference in the length of the polymer between seed coat and grape berry, or a problem of seed coat PA that has oxidised during seed maturation, becoming resistant to extraction and/or hydrolysis (Downey et al., 2003).

The TEM study suggests that the unusual intermediates produced in tds4-1 accumulate in discrete regions within the cytoplasm, presumably because they are unable to be transported into the PA-forming vacuole. If this is the case, the mechanisms for the transport of PA intermediates into the vacuole are specific. Notably, the vanillin-stained phenotype of the tt12 mutant seed (Debeaujon et al., 2001), which is unable to transport intermediates into the vacuole, is superficially similar to that of tds4 stained with DMACA.

At present, we cannot explain why a mutation in the LDOX gene should be responsible for the lack of normal vacuoles in endothelial cells. The change in vacuole morphology in tds4-1 seems to be restricted to the endothelial cells that accumulate PA, as the vacuoles in the leaves are unaffected by the tds4-1 mutation. Perhaps the formation of PA intermediates acts as a signal to the cell to produce a vacuole to store the eventual end products of the pathway. The unusual vacuole morphology found in the tds4 seed coat cannot be simply attributed to the lack of PA to fill the vacuole, because the tds1, tds2, tds5 and tds6 mutants have a normal vacuole and produce the WT intermediate epicatechin, in the absence of PA polymer (Abrahams et al., unpublished). Our early observations suggest that the vacuoles in the endothelial cells of tds3 have characteristics that are intermediate between tds4 and the remaining tds mutants, which is consistent with the assigned position of TDS3 in the pathway as a result of the double mutant analysis. Together with observations from earlier studies that indicate a Golgi origin for PA provacuoles (Hilling and Amelunxen, 1985), our results suggest that the process of PA vacuole formation involves a combination of specific vacuolar docking and fusion events. Furthermore, it is possible that vacuole formation and PA biosynthesis may be coupled processes in the cell. The branch-point in the anthocyanin and PA pathways at which the cell becomes dedicated to the synthesis of PA, rather than anthocyanin, would be the most likely point at which to signal the requirement for a PA vacuole. The analysis of leaf vacuole morphology in a tissue that accumulates flavonols and anthocyanin, but not PA, indicated that the leaves of tds4-1 created a normal vacuole. This type of coupling might explain the formation of PA-specific idioblasts in leaves of plants that accumulate PA such as L. corniculatus (Robbins et al., 1992) and D. uncinatum. Our finding that LDOX is involved in PA synthesis as well as anthocyanin synthesis, together with the implication that both BAN and flavonoid 3-glucosyl transferase can use cyanidin as a substrate, is an indication that there is still a lot to learn about the PA biosynthesis pathway and its regulation.

Experimental procedures

Plant material and growth conditions

Arabidopsis insertion lines defined by the insert flanking sequence were obtained from the Arabidopsis Biological Resource Center (ABRC; Ohio State University, Columbus, USA) or Cold Spring Harbor Laboratory (CSHL; Sundaresan et al., 1995). Homozygous mutant plants were selected by their pale seed phenotype and confirmed by staining progeny seed with DMACA. The mutant lines of LDOX (At4g22880) were Salk_028793 (tds4-2), CSHL GT9767 (tds4-3) and Salk_073183 (tds4-4). Insertion mutant information was obtained from the SIGnAL website at The tds4-2 and tds4-3 lines contained a functional nptII gene and were resistant to kanamycin. tds4-1 was identified from a pool of seed from Institut National de la Recherche Agronomique (Abrahams et al., 2002). The ban mutant F36 was used for performing the crosses for the double mutant analyses.

Seed was sterilised using 0.1% (w/v) mercuric chloride for 15 min or 70% (v/v) ethanol for 1 min, followed by 15% (v/v) household bleach for 10 min, washed with H2O, germinated on MS media containing 0.8% agar under constant light (100 µmol photons m−2 sec−1), and then transferred to soil after 2 weeks. To promote anthocyanin accumulation in emerging seedlings, the seeds were placed on half strength MS containing 5% (w/v) sucrose and 1.2% (w/v) phytagel (Sigma-Aldrich, Australia). Plates were placed vertically in constant light (150 µmol photons m−2 sec−1).

Plants were grown in 5- or 8-cm pots containing an equal mix of peat, sand and perlite with the addition of Osmocote (Short Crop) in a growth cabinet or room, with 16 h, 24°C light (150 µmol photons m−2 sec−1) and 8 h, 18°C dark cycles, maintaining a continuous supply of water. Harvested material was either frozen in liquid nitrogen or fixed immediately in glutaraldehyde. Siliques for analysis were measured in length to estimate the stage of maturity, and then checked by microscopic observation, either in sections or by dissection.

Positional cloning of TDS4

F2 mutant individuals from a cross between tds4 plants in a Ws-4 background and Col-7 were selected on the basis of seed colour and DMACA phenotype. DNA was extracted from mutant plants and used to map the TDS4 locus using PCR and specific primers. Some polymorphisms based on PCR product size could be scored directly on an agarose gel, while others were CAPS markers. Markers for each chromosome were initially selected from information available at When markers were not available, new markers were found by amplifying and directly sequencing the PCR products from non-coding regions of the genome in the appropriate area to identify polymorphisms between Col-7 and Ws-4. Markers used for chromosome IV are shown in the Supplementary Material.

Complementation of the tds4 mutation

Each chapter of the binary cosmid library (Schulz et al., 1994; was plated to a density of 2000–3000 cfu per plate. DNA was extracted from the combined colonies (Sambrook and Russell, 2001) and pooled in a cross-gridded fashion. To identify positive pools, PCR was performed using primers for markers in the region of TDS4 and pooled DNA samples. Colonies from positive pools were plated at a density of 500–1000 cfu per plate, lifted and probed with labelled PCR products. The 35S:LDOX construct was made by cloning a genomic fragment, which included 150 nucleotides of 5′ UTR, the coding region of the LDOX gene and 130 nucleotides of 3′ UTR into the pART7/pART27 vectors (Gleave, 1992). The LDOX gene was PCR amplified using Pfu polymerase (Stratagene, USA), Col-7 genomic DNA template and the primers 5′-TTCATTTACTTGCAACCAATTAC and 5′-AGTTTGCTTGTTAAGCTCCGC.

Agrobacterium strain LBA4404 was transformed with gLDOX cosmids or 35S:LDOX and used to transform tds4-1 mutant plants (Bechtold et al., 1993). Transformed seedlings were selected on agar plates with half strength MS containing 35 mg l−1 kanamycin. DNA was extracted from one leaf of each plant, and checked for the mutant tds4-1 sequence (to verify the parental origin of the seed) and the presence of the construct, using primers specific for the vector and the insert. Seeds from these were examined for colour and stained with DMACA.


Double-stranded DNA sequencing was performed using Big Dye version 2 or 3 and an ABI 3700 automatic sequencer at the Institute of Medical and Veterinary Science, Adelaide, or CSIRO Plant Industry, Canberra. Sequence analysis was performed using Sequencher, blast (Altschul et al., 1990) and GCG analysis programs at

Double mutant analysis

Double mutants of tds4 tds1, tds4 tds2, tds4 tds3, tds4 tds5, tds4 tds6, ban tds1, ban tds2, ban tds3, ban tds4, ban tds5 and ban tds6 were identified by crossing the homozygous recessive mutant lines and using DMACA to determine phenotypes of progeny seed through three generations. As the seed coat is maternal tissue, all F2 seeds appeared WT and were grown to collect F3 seeds. For a homozygous recessive mutation, the segregation ratio of 9 : 3 : 3 : 1 (WT:m1:m2:m1m2) is observed in F3 seed populations. F3 seeds with either an m1 or m2 phenotype were grown to collect F4 seeds, which were DMACA stained to determine whether they segregated for the seed phenotype. The double mutant phenotype was revealed when the progeny of F3m1 seed segregated for both m1 and m2, but not the progeny of the F3m2 seed from the same cross.

Determination of Arabidopsis PA subunit composition

Extraction of PA and TLC analysis of PA monomers was carried out as previously described by Abrahams et al. (2002), except that cellulose high performance thin layer chromatography (HPTLC) plates were used (Merck), and the solvent to develop chromatograms was water-saturated butanol. PA was extracted from 15 g of mature WT Col-7 seed, purified on Sephadex LH20 (Tanner et al., 1994), and subjected to acid hydrolysis in the presence of phloroglucinol (adapted from Kennedy and Jones, 2001). An aliquot containing between 2 and 250 µg of purified PA in methanol was dried under nitrogen and hydrolysed for 20 min at 50°C in 10 µl of solution containing 1.0% (w/v) sodium ascorbate and 5.0% (w/v) phloroglucinol in 0.1N methanolic HCl. The products of hydrolysis were neutralised with an equal volume of 200 mm acetate buffer (Na+, pH 7.2), diluted with 80 µl of water and injected onto an HPLC column (Hamilton PRP-1, 150 mm × 4.1 mm, 5 µm, 75 A, polystyrene-divinylbenzene), and eluted at 1 ml min−1 with a linear gradient from solvent A (2% acetic acid) to 35% solvent B (methanol) for 40 min at 35°C. After each run, the column was flushed with 100% solvent B and re-equilibrated with solvent A. Flavon-4-ols were identified by comparing with authentic standards. Phloroglucinol adducts were identified by comparing with retention times of known hydrolysis products from procyanidin B2, B3, or C2, or PA purified from M. sativa (Koupai-Abyazani et al., 1993) and Hordeum vulgare (Kristiansen, 1984). The peak area was calibrated against catechin, and molar ratios were calculated from published values of molar response relative to a weighed catechin standard (Kennedy and Jones, 2001).


For light microscopy, seeds were stained with DMACA as previously described by Abrahams et al. (2002). For TEM, developing siliques were harvested and fixed in 100 mm phosphate buffer (Na+, pH 7.0) containing 3% glutaraldehyde and treated with 2% osmium tetroxide in phosphate buffer (Nielson and Griffith, 1978). Specimens were embedded in LR white resin. Ultra-thin sections were prepared for TEM and analysed using A JEOL 100CX TEM at 80 kV.

The vacuole lumen marker carboxy-DCFDA (5-(and-6)-carboxy-2′,7′-dichlorofluorescein diacetate) and yeast vacuole membrane marker MDY-64 were obtained from a Yeast Vacuole Marker Sampler Kit (Molecular Probes, USA). These dyes were prepared as described by the manufacturer. For carboxy-DCFDA loading, developing seeds were dissected from siliques submerged in 50 mm citrate buffer (Na+, pH 5), containing 2% w/v glucose, using 25G needles. Seeds were further dissected to remove the embryo and to expose the endothelial cells to dye. For MDY-64, samples were treated in 10 mm HEPES buffer, pH 7.4, containing 5% w/v glucose. Seeds and leaves were incubated in 10 µm carboxy-DCFDA for 2 h and in 10 µm MDY-64 for 30 min, respectively, at 25°C, washed in fresh buffer and mounted on slides for observation using a Leica SP2 confocal microscope. For visualisation of fluorescence in MDY-64 treated leaves, the sample was excited at 458 nm and the emitted fluorescence from 465 to 525 nm was collected. For visualisation of fluorescence in carboxy-DCFDA-treated seeds, the sample was excited at 488 nm and the emitted fluorescence from 500 to 550 nm was collected. Autofluorescence was monitored by collecting emissions from 600 to 720 nm. Under these conditions fluorescence from chloroplasts in leaves was minimal.


We thank the Salk Institute Genomic Analysis Laboratory for providing the sequence-indexed Arabidopsis T-DNA insertion mutants, and the CSHL and ABRC for providing seeds and the cosmid library used in this research. Thanks to members of the respective labs for advice and helpful discussions, Veronique Cheynier for the authentic procyanidin B2, Klaus Kristiansen for procyanidin B3 and C2, Justine Chambers for help with photography, and Celia Miller and Rosemary White for their help with TEM and CLSM. This work was supported by Pioneer Hi-Bred International (Dupont) and Meat and Livestock Australia. Funding for the SIGnAL indexed insertion mutant collection was provided by the National Science Foundation.

Supplementary Material

The following material is available from

Figure S1. Reverse transcriptase (RT)-PCR analyses of the expression of CHS, DFR, LDOX and BAN in tt8 alleles.

Siliques and leaves were harvested, placed in liquid N2, and stored at −80°C until use. RNA was extracted using an SV. Total RNA Isolation System (Promega Corporation, Australia) and 10 µg RNA reverse-transcribed using moloney murine leukemia virus (MMLV) reverse transcriptase (Promega). One-tenth of the cDNA was used as a template for PCR, using KlenTaq polymerase (Clontech Laboratories, Inc., USA), and 30 rounds of amplification. PCR primers were designed to amplify across an intron to distinguish between products amplified from cDNA and residual genomic DNA. The optimum annealing temperature for each primer pair was determined using a gradient PCR block (Hybaid Limited, UK). The WT for tt8-4 and tt8-5 was Col-7, whereas tt8-1 was Landsberg erecta (Ler).

Table S1 Table shows markers, contig numbers, primer sequences and relative size of PCR products in each used to map TDS4

The tds4-1 DNA sequence has the accession number AJ564262.