The BANYULS gene encodes a DFR-like protein and is a marker of early seed coat development

Characterization of the T-DNA insertion site in ban plants

The banyuls mutant was originally isolated during a screen for the embryo-defective (emb) phenotype and has been described by Albert et al. (1997) . The mutation causes a precocious pigment accumulation in the seed coat resulting in a purple colour of the immature seeds. The initial transformant exhibited both the ban and the emb phenotype and had complex T-DNA insertions. In order to segregate the two mutations and the multiple copies of T-DNA, we identified plants presenting the ban phenotype and containing a single T-DNA insert using Southern blot analysis on plants from the progeny of the cross ban × tt2 ( Albert et al. 1997 ). PCR walking ( Devic et al. 1997 ) allowed us to amplify and characterize about 500 bp of plant genomic sequences flanking each border of the T-DNA of the transgenic plants. The corresponding genomic sequences of the wild-type were amplified in two rounds of PCR walking using the ban-5′ and wsban-5′ nested primers for the 5′ end, and the ban-3′ and wsban-3′ nested primers for the 3′end. A contig of 2023 bp corresponding to the complete gene was constructed and sequenced. T-DNA integration into the BAN gene resulted in a deletion of 16 bp of plant genomic DNA. Southern blot analysis demonstrated that BAN is present as a single copy in the Arabidopsis genome (data not shown).

Comparisons with various databases revealed similarities to known genes encoding the dihydroflavonol reductase (DFR) of Rosa×hybrida and many other plant species and with the vestitone reductase of Medicago sativa. This information helped us to predict the exon–intron boundaries and to design primers for the cloning of the corresponding cDNA by Marathon-based PCR (Clontech). cDNA fragments were amplified from a Marathon-like library made from RNA of immature siliques using the ban-5′ (for the 5′ portion) and ban-3′ (for the 3′ portion) primers in combination with the adaptor primers. The 5′ extremity was amplified using the ban-5′ and banATG nested primers on the same library and by sequencing up to 10 independent clones. The complete cDNA is 1213 bp long with 45 bp of 5′ untranslated leader sequence and a 3′ end of 142 bp. It encodes a protein of 342 amino acids with a molecular mass of 38 kDa. Its cellular localization is predicted to be the cytoplasm (PSORT program, Nakai & Kanehisa 1992). Alignment of the genomic and cDNA sequences showed that the BAN gene is composed of six exons and five introns. The T-DNA was found to have integrated into the third intron.

Regular submissions of large genomic sequences to the databases have prompted us to perform routine searches with the BAN cDNA sequence. Recently, a near identical sequence to BAN has been found in a BAC clone assigned to chromosome I. The rare differences in nucleotides may be due to the ecotype, WS for BAN and Col for the BAC clone (Federspiel et al. 1998, unpublished). These data validate the genetic positioning of BAN ( Albert et al. 1997 ). BAN is located on the BAC T13M11 (AC005882).

BANYULS is a member of the NADPH-dependent oxido-reductase superfamily

The amino acid sequence of the BAN protein is very similar to NADP(H) binding oxido-reductases and can be aligned over its entire length to DFR, an enzyme of the anthocyanin biosynthesis pathway, CCR (cinnamoyl CoA reductase) and CAD (cinnamoyl alcohol dehydrogenase), enzymes of the lignin pathway, and vestitone reductase, an enzyme of the isoflavone pathway ( Fig. 2). The highest scores for DFR were found with the DFR of Rosa×hybrida ( Tanaka et al. 1995 ; D85102, 41% identity) and Arabidopsis ( Shirley et al. 1992 ; P51102, 40%). In terms of other enzymes, BAN showed similarity with the vestitone reductase of Medicago sativa ( Guo & Paiva 1995; U28213, 39%), and with CCR (Pichon, unpublished data, X98083, 36%) and CAD (Goffner, Van Doorsselaere and Boudet, unpublished data, X88797, 35%) of Eucalyptus gunnii. The motif of 13 amino acid residues common to DFRs ( Fig. 2, DFR-Rh and DFR-At: residues 132–144) and thought to define their substrate specificity ( Beld et al. 1989 ) is not found in BAN and other members of the family. There is also an insertion of one amino acid residue (in BAN, G168) within the motif KNWYCYGK ( Fig. 2, CCR: residues 158–165), a sequence thought to be involved in the catalytic site of CCR ( Lacombe et al. 1997 ). The positions of the five introns are well conserved between BAN and DFR, although the introns have no sequence homology. Thus, the conserved gene structure of BAN, DFR and CCR ( Lacombe et al. 1997 ) suggests a common ancestor for these proteins.

To a lesser extent, and within a portion of the protein only, BAN also shares homologies with the NADPH-dependent reductase of barley which inactivates HC toxin ( Han et al. 1997 ; HVU7746, 32%), the UDP-galactose 4-epimerase ( Adams et al. 1988 ; P13226, 26%) of yeast, and the 3-β-hydroxysteroid dehydrogenase of hamster ( Rogerson et al. 1995 ; HAHMSD3B, 19%). The NADPH binding site in BAN (15–35) is well conserved among this family of oxido-reductase proteins ( Fig. 2). In addition, BAN exhibits a perfect leucine zipper consensus (192–213) that is not be found in other members of the family ( Fig. 2).

A phylogenetic tree was constructed using the DARWIN program ( Gonnet et al. 1992 ) and is presented in Fig. 3. The tree summarizes the theoretical evolutionary distances among the different NADPH-dependent oxido-reductase superfamily members and BAN. The DFR proteins from several plant species represent a separate cluster from which BAN is excluded. We can conclude that BAN belongs to the mammalian 3β-hydroxysteroid dehydrogenase/plant dihydroflavonol reductase superfamily as described by Baker et al. (1990) and Baker & Blasco (1992) and that BAN does not correspond to a second DFR in Arabidopsis.

Mutations are present in different ban alleles

Allelism tests were performed between ban plants and two mutant plants, F36 and F52, from the Kranz and Röbbelen Arabidopsis Information Service (AIS) collection exhibiting a similar phenotype. The three mutations were found to be allelic. The complete cDNA corresponding to the BAN gene was amplified by RT–PCR from RNA of immature siliques of F36 and F52 using the RT5′ and RT3′ primers and were sequenced. The wild-type BAN cDNA from Enkheim-1 (En-1) was also sequenced in order to distinguish ecotype sequence variation from mutation in the gene sequence. A striking ecotype difference was the absence of the two terminal amino acids in the En-1 ecotype compared to the WS sequence. In the ban cDNA of the F36 mutant, we observed a small deletion of 14 bp that removed five amino-acids (217-SFITG-221) and created a frameshift. In the ban cDNA of the F52 mutant, a base substitution of C to T introduced a stop codon at position Q307. These mutations were verified by sequencing several clones derived from each mutant line. In both F36 and F52, Q at position 32 is replaced by K. Although this substitution is located within the NADPH binding domain, the residue at position 32 is apparently not conserved ( Fig. 2), and this variation will probably not result in a loss of function of the protein. This substitution was not found in the wild-type En-1 sequence and may also be due to the mutagen, or, most probably, the En-1 ecotype that we have used is not the same wild-type ecotype in which F36 and F52 were generated. The mutagen used for the AIS collection is unknown. The conversion C→T is consistent with ethyl methyl sulfonate (EMS) mutagenesis since EMS has been shown to cause primarily base substitution of G:C to A:T. These results demonstrate that mutations in the BAN gene are responsible for the ban phenotype and that the mutation is tagged by the T-DNA insertion in the ban mutant plant. The mutation in the T-DNA tagged allele might be null, since there is no evidence of the presence of a transcript corresponding to the altered BAN gene (data not shown). In the case of F36 and F52, the mutations are probably null since they produce truncated proteins and no difference in strength of the phenotype of the three alleles was noticeable.

Expression of the BAN gene during plant development

The qualitative expression of the BAN gene was studied by RT–PCR using the RT5′ and RT3′ primers and was compared with the expression of chalcone synthase (CHS) which is involved in the biosynthesis of flavonoids. The expression of the histone H2A variant (H2A) was used as a constitutive control. CHS transcripts were amplified from leaf tissues, flower buds, flowers and young siliques samples ( Fig. 4). No CHS transcripts were detected during seed maturation. The faint band in last two lanes (torpedo and cotyledon) corresponds to amplification from residual genomic DNA since the amplimer is of a slightly higher molecular weight due to the presence of an intron. In contrast, the BAN transcripts were detected only in flowers and young siliques and were absent from the other samples ( Fig. 4). Increasing the number of PCR cycles to 45 did not result in the detection of BAN transcripts from leaf cDNA (data not shown).

The localization and transient expression of the BAN transcript were analysed in detail by in situ hybridization during wild-type seed development ( Fig. 5a–d). The BAN transcripts were detected specifically in the endothelium of the seed coat, the cell layer that precociously accumulates purple pigments in the ban mutant seeds ( Fig. 5a). During seed coat development, the presence of the BAN transcripts in the endothelium was transient and corresponded to pre-globular embryo stages ( Fig. 5a). From the early globular stage of embryo development until desiccation, the transcripts could no longer be detected ( Fig. 5b–d). Since no signal was observed in the ovules prior to fertilization, we infer that the expression of the BAN gene may be triggered by fertilization (data not shown). The profile of the expression of the BAN gene is consistent with the observed phenotype of the ban seeds.

The temporal expression of the flavonoid genes is similar during wild-type and ban seed coat development and is triggered by fertilization

Most of the flavonoid genes are present as a single-copy gene in the Arabidopsis genome, and, in contrast to BAN, they are not endothelium-specific but are expressed in many tissues. Since the study of their expression during seed development has not already been described in detail, we analysed their expression by in situ hybridization. Several key genes were used as probes: the CHS gene, encoding the first enzyme leading to the flavonoid pathway, the CHI gene, the second gene; the DFR gene, for the first enzyme of the anthocyanin-specific pathway; the LDOX gene, for catalysis of the leucoanthocyanidins to anthocyanidins thereby producing the purple–red pigments ( Fig. 1). The results of in situ hybridization with the LDOX probe are presented in Fig. 5(e–h). Similar results were obtained with the three other probes (data not shown). The transcripts for each of these four genes were detected in the endothelium soon after fertilization ( Fig. 5e) and persisted until the heart stage ( Fig. 5e–g). From the torpedo stage onwards, transcripts were no longer visible ( Fig. 5h). The expression of these four genes of the flavonoid pathway was induced at a similar time in comparison to the induction of the BAN gene; however, in contrast to BAN, their expression differed in that gene expression persisted until the torpedo stage and was not seed coat-specific.

Since the ban mutation induced a precocious accumulation of anthocyanins in the seed coat, we also studied the expression of the flavonoid genes in the ban seed coat. Only the results of the LDOX gene expression are presented in Fig. 5(i–l) since the four probes gave similar results. We have found that the temporal and spatial pattern of expression of CHS, CHI, DFR and LDOX genes in the ban seed coat did not differ from their normal pattern of expression in the wild-type seed coat. Using the in situ hybridization (ISH) technique, it is not possible to ascertain whether quantitative differences may exist between transcript levels in the wild type and ban seed coat.

Catechins are not produced in ban seeds

A cytological comparison of ban and wild-type seeds revealed cytological differences in the endothelium. The characteristic brown granules that start to accumulate at the pre-globular stage in the wild-type endothelium ( Fig. 5a–h) are absent in the ban seed coat ( Fig. 5I–L). These granules are not coloured in non-treated wild-type seeds. The coloration is due to treatment of the samples necessary for microscopic studies. To obtain a superior image of the structure of the seed coat, the seeds were embedded into resin used for making thin sections. The results of this study are presented in Fig. 6. When stained with toluidine blue, these granules are blue and can be observed only in the endothelium of wild-type ( Fig. 6a) but not the ban testa ( Fig. 6b). This difference is clearly visible at higher magnification in Fig. 6(c,d). In order to define the nature of these products, we studied the presence or absence of these granules in the developing seed coat of various mutants. Granules were absent from the testa of tt3, tt4 and tt5, as in ban ( Table 1). Therefore, the products are probably not isoflavones or flavanones. Most probably, the flavonoids that accumulate at early embryogenesis are catechins, precursors of tannins.

Using barley grains, Kristensen & Aastrup (1986) have developed a vanillin test for detection of catechins. Catechins are produced by the leucoanthocyanidin reductase (LCR) from leucoanthocyanidin precusors. The catechins are coloured red in the endothelium cell layer in this in vivo test. Staining with vanillin confirmed the nature of the granules in wild-type Arabidopsis seeds ( Fig. 6e) as catechins, precursors of tannins, and confirmed the absence of these products in ban seeds ( Fig. 6f). Similarly, no red staining appeared in immature seeds of the F36 and F52 alleles ( Table 1). There is a strict correlation between the presence of the granules in the endothelium and the vanillin coloration in all the mutants tested ( Table 1). Immature seeds of tt1, tt9 and tt10, as well as double mutants ban/tt were also tested with vanillin. In these double mutants, the ban phenotype is epistatic to the tt phenotype for anthocyanin accumulation. The results presented in Table 1 show that the ban/tt1 seeds synthesize both anthocyanins and catechins and can be distinguished from single-mutant ban seeds, while the other double mutants are similar to ban. In the tt1 seed coat, the cathechins are present mainly as traces at the base of the seed, surrounding the suspensor and the nucellus. A closer inspection of the wild-type seed coat ( Fig. 6a) shows the presence of fewer granules in the cell layer adjacent to the endothelium at the level of the suspensor (arrow). In tt1, only this second cell layer produces granules and vanillin staining. This result suggests an independent regulation of the synthesis of tannins in the different cell layers of the seed coat. In ban/tt1 seeds, the distribution of cathechins is more similar to that of the wild-type although at lower level.

Taken together, these observations strongly suggest that BAN encodes LCR.

Temporal and spatial expression of the BAN gene is consistent with the predictions of the genetic analysis

A previous study on the genetic and phenotypic characterization of the BAN gene led to the hypothesis that BAN is a negative regulator of anthocyanin accumulation during seed coat development ( Albert et al. 1997 ). Since no effect of the mutation was detected in other parts of the plant or during different developmental programmes of the plant life cycle, we predicted that BAN transcripts would be present exclusively in the endothelium of the seed coat during early development. The results obtained by RT–PCR and in situ hybridization are in agreement with our hypothesis. The BAN gene is expressed specifically in the endothelium at the onset of fertilization and persists only until the pre-globular stages. In contrast, the period of expression of the known flavonoid genes at the level the endothelium extends to the torpedo stage. The BAN gene can thus be considered as an early marker of fertilization and seed coat development.

BAN is an enzyme of the mammalian 3β-hydroxysteroid dehydrogenase/plant dihydroflavonol reductase superfamily

Protein sequence analysis demonstrated that BAN belongs to a superfamily of oxido-reductases as defined by Baker & Blasco (1992). Identification of several residues V15, G17, G20, L30 and S132 (position in the BAN sequence) that are strictly conserved in all the members of the family, indicates a requirement for a NADPH co-factor in the reaction catalysed by BAN. Since the best score of similarities with BAN was obtained with DFR, the first enzyme committed to the biosynthesis of anthocyanins and tannins, the most simple conclusion is that BAN is a second DFR gene, specifically expressed in the endothelium. However, this hypothesis can be easily ruled out for the following reasons. Firstly, the phylogenetic tree built by the DARWIN program based on 30 members of the superfamily positions BAN close to, but distinct from the group of DFR sequences isolated from various species. Secondly, DFR has been previously characterized as a single-copy gene in Arabidopsis and mutations in this gene resulted in the tt3 phenotype ( Shirley et al. 1992 ). It is unlikely that each of the three alleles at the BAN locus, which exhibit loss of function characteristics (recessiveness and the nature of mutations e.g. disruption of BAN ORF), lead to over-expression of a gene with a similar function as DFR. In addition, BAN possesses a perfect leucine zipper consensus that is absent in the other members of the family. The leucine zipper is responsible for homo- or heterodimer formation between protein subunits. We conclude that BAN is most probably a NADPH oxido-reductase, functioning as a homo- or heterodimeric protein in the cytoplasm.

BAN function is positioned at the branch point between anthocyanin and tannin production

Since BAN does not seem to act as a transcription factor or to be involved in the regulation of transcription factors that control flavonoid biosynthesis, we favour the hypothesis that BAN is a structural enzyme of the phenylpropanoid pathway. The main function of BAN would be to synthesize seed coat-specific products during early testa development. In the absence of BAN enzymatic activity, these components are not produced, but their absence may not lead directly to an obvious phenotype. The precocious accumulation of anthocyanins in ban seed coat could be considered as an indirect effect of the mutation. Several attempts to manipulate the phenylpropanoid pathway have identified the existence of metabolic channelling and highlighted the problem of making predictable changes since alternative biosynthetic routes exist ( Sewalt et al. 1997 ). A significant degree of similarity was evident between the sequence of BAN and the vestitone reductase of Medicago sativa on the isoflavones branch. We used the cytological observation that ban seed coat has a modified endothelium morphology to test whether BAN is involved in isoflavone biosynthesis. Light microscopy study of the seed coat showed that the brown granules were present in wild-type seed but absent in the ban seed coat, and BAN is probably involved in the production of these granules. Since the granules are absent in the endothelium of tt3, tt4 and tt5 mutant seed coats, this argues against the isoflavone nature of these granules since these products would have been present in the tt3 endothelium but absent from the tt4 and tt5 testa ( Fig. 1 and Table 1).

Positioning BAN as a structural enzyme of the flavonoid pathway leading to condensed tannins (LCR) would reconcile a negative regulator with an enzymatic function. Competition may exist between LDOX leading to anthocyanin production and LCR leading to tannins for their common substrate, the leucoanthocyanidins. The direct precursors of tannins are catechins that can be coloured by vanillin. The vanillin test revealed that the brown granules are composed of catechins and demonstrated that no tannins were synthesized in ban seed coat in accordance with the pale brown colour. Thus, the pink colour of the ban seed at early development is due only to anthocyanin accumulation and not to a combined production of anthocyanins and tannins. Since the enzyme in the branch leading to tannin production is a reductase, it is tempting to postulate that BAN is LCR. LCR activity has been studied in detail and is currently purified from barley grains ( Jende-Strid 1993) and legume plants ( Skadhauge et al. 1997 ). Furthermore, Tanner & Kristiansen (1993) have demonstrated that the enzymatic reduction of leucocyanidin to catechin is NADPH-dependent. The analysis of the ban/tt double mutants is in agreement with the BAN identity of LCR, except in the case ban/tt1. However, the results obtained with ban/tt1 are unequivocal. The ban phenotype is epistatic to the tt1 phenotype, thus the immature seeds are purple. When purple seeds are stained with vanillin, anthocyanins are released from the cells and the red precipitate of cathechin is formed. If these seeds were actually ban seeds, then no red coloration would have appeared. This observation suggests that there may be more than one LCR gene, one regulated positively by TT1 and one independent of TT1 and expressed in the second cell layer of the testa. Alternatively, BAN is not LCR but regulates its activity in part with TT1.

The negative regulatory effect of BAN on the anthocyanin biosynthesis pathway does not affect the temporal and spatial expression of flavonoid genes

Since in ban endothelium, there is a precocious accumulation of pigments compared to the wild-type testa, one could expect an early induction of the transcription of all or at least some of the flavonoid genes. Because DFR and LDOX lead specifically to the synthesis of anthocyanins and tannins, while CHS and CHI synthezise common intermediates involved not only in pigments but also in the production of flavones, isoflavones and flavonols ( Fig. 1), it may be predicted that regulation acting at the transcriptional level would act at the latest on the LDOX gene. This enzyme catalyses the transition from colourless anthocyanidins (leuco-) to red–purple anthocyanidins. However, our results revealed no evidence of such a change in the induction time of transcription of the CHS and CHI genes, nor of the DFR and LDOX genes during wild-type and ban seed coat development. Furthermore, our results demonstrate that in the wild-type seed, there is a long time gap between the disappearance of flavonoid transcripts (heart–torpedo stage) and the appearance of the visible coloration of the seed coat (maturation–desiccation stages), suggesting a possible regulation at a post-transcriptional level. In the endothelium of the ban mutant, there were no qualitative differences in the induction time and duration of transcription of the genes encoding enzymes of the biosynthesis of flavonoids compared to wild-type; however, due to the small size of Arabidopsis seeds, it is not possible to obtain quantitative data. In contrast, during the development of the hypocotyl, a distinction could be made between early and late genes regulated predominantly at the level of transcription ( Kubasek et al. 1992 ; Pelletier & Shirley 1996; Pelletier et al. 1997 ; Shirley et al. 1995 ).

In conclusion, our results strongly support the hypothesis that BAN is functioning at the branch point between anthocyanin and tannin production and is involved directly or indirectly in metabolic channelling. BAN most probably encodes LCR or regulates LCR activity. In the absence of LCR activity, the pathway will be directed by default to the synthesis of anthocyanins rather than cathechins and tannins. A regulator of this enzymatic step should have the same mutant phenotype as a mutant in LCR activity. The final proof of the identity of BAN will be obtained when the enzymatic activity of the recombinant BAN protein is tested in vitro.

Plant material and growth conditions

Arabidopsis thaliana (L.) Heynh., ecotype Wassilevskija (WS), and banyuls (ban, ABW1) mutant seeds, WS ecotype, were obtained from INRA, Versailles, France. F36 (N323) and F52 (N334) ban alleles seeds, originated from the Kranz and Röbbelen AIS collection, and the Enkheim (En-1) wild-type seeds (N1136) were provided by the Nottingham stock centre (UK). T-DNA-transformed ban seeds were selected by germination on Murashige & Skoog (1962) medium containing kanamycin (Sigma) at 100 mg l^–1. Non-transformed seeds were sown directly into soil. After a cold treatment of 2–4 days at 4°C, pots were transferred to a growth chamber set at 22°C with a photoperiod of 16 h light/8 h dark.

Identification of the T-DNA integration site

Genomic DNA from homozygous ban plants was extracted by a CTAB method ( Doyle & Doyle 1989) and used to construct PCR walking libraries as described by Devic et al. (1997) . Two fragments of 800 and 600 bp were amplified, respectively, from the ScaI and PvuII libraries using a pair of nested primers specific to the T-DNA left border, LB-BAR1 and LB-BAR2. At the right border, fragments of 450 bp and 900 bp were amplified from EcoRV and ScaI libraries with the nested primers RB-GUS1 and RB-GUS2 and sequenced. Two primers, designed from plant genomic DNA, ban-3′ (5′-AACCAATCTCGAAGGTGCTAGC-3′), and ban-5′ (5′-GTGCCGGAATCACGGTTACGAG-3′), were synthesized and used to characterize the T-DNA target site in wild-type genomic DNA by PCR-walking with a single gene-specific primer. The complete BAN gene sequence was re-constructed by PCR walking in two steps. In the first step, using ban-3′ as a single gene-specific primer, a fragment of 500 bp was obtained from the EcoRV library, and using ban-5′, a fragments of 450 and 200 bp respectively were obtained from the DraI and PvuII libraries. The primers wsban-5′ and wsban-3′ were designed from the sequence of these fragments and used for a second round of PCR walking. Using the nested primers, ban-5′ and wsban-5′ (5′-CAAGAGAGACCTCGGGATCTTCGGA-3′), fragments of 2.5 and 1.1 kb were amplified from the Ssp1 and DraI libraries, respectively, each of which contained the initiation methionine and the promoter region. For the cloning of the 3′ end of the gene, the nested primers ban3′ and wsban-3′ (5′-CACAAGTGTTCCAGA- GATTGCGGA-3′) were used to specifically amplify fragments of 600 and 500 bp from the DraI and ScaI libraries. The amplified fragments were cloned into the pMosblue vector (Amersham) and several clones corresponding to each amplimer were sequenced in order to eliminate errors that might have been introduced by the DNA polymerase. A contig of 2023 bp was constructed using the SEQUENCHER program (Gene Codes).

Characterization of the BAN cDNA

The cDNA corresponding to the BAN gene was cloned by a technique based on the Marathon technique, a technique commercialized by Clontech. One microgram of poly(A)⁺ RNA from immature siliques was used for the construction of an oligo(dT)-primed double-stranded cDNA. After polishing of the cDNA ends, the same adaptor duplex as used for the production of genomic walk libraries was ligated to the cDNA extremities to constitute the cDNA library stock. The 3′ region of the cDNA was amplified using the ban-3′ and AP1 primers in the first PCR reaction and ban-3′ and AP2, in the second reaction. Several fragments of approximately of 900 bp were cloned and sequenced. For the 5′ region, the primers ban-5′, AP1 and AP2 were used in a similar manner to amplify fragments of 350 bp. The sequence revealed the presence of a putative initiation methionine. In order to increase the size of the 5′ untranslated region, PCR on the cDNA walking library was performed with nested primers ban-5′ and AP1 for the first PCR reaction and banATG (5′-TCCCGTGCCACCAATGACACAAGCC-3′) and AP2 for the second PCR. The longest clones were selected and sequenced. A cDNA contig of 1080 bp was constructed using the SEQUENCHER program. The accession number for the cDNA sequence in Genbank is AF092912. For rapid cloning of the BAN cDNA from the different alleles and ecotypes and the RT–PCR experiments, the primers RT5′ (5′-AACAACTAAATCTCTATCTCTGTA-3′) and RT3′ (5′-GAATGAGACCAAAGACTCATATAC-3′) were used.

Characterization of ban alleles and RT–PCR experiments

RNAs were extracted according to Kay et al. (1987) . The same RT–PCR conditions were used for the cloning of ban alleles and for the study of the BAN gene expression during silique development. Five micrograms of total RNA were used for the synthesis of the first-strand cDNA under the conditions of the RT–PCR Stratagene kit. An aliquot of 1.25 μl of first-strand cDNA was added to a 20 μl PCR reaction. The reaction was submitted to 30 or 45 cycles. Using the primers RT5′ and RT3′, a specific band of 1 kb corresponding to the BAN transcript was amplified. The primers CHSintron (5′-TCCTGACTACTCTTCCGCATCAC-3′) and CHSwalk (5′-AGCTATCCAGAAGAGGGAGTTCCA-3′) specific to the chalcone synthase cDNA and gene were used as control. A fragment of 800 bp was amplified for the CHS cDNA as predicted. The primers for the amplification of the histone H2A variant cDNA were H2 A5′ (5′-GGGGAAACCCAGCGGTACG-3′) and H2 A3′ (5′-GATAGATAAATCGACTGGCG-3′).

In situ hybridization and histological studies

The protocol for in situ hybridization was based on Cox et al. (1984) except for the labelling of the probes and the detection of the signal. Probes were synthezised and labelled using the Boehringer digoxigenin system. Detection was performed using the BM purple AP substrate (Boehringer). The CHS, CHI and DFR probes (a gift from Dr Brenda Shirley) were synthesized from full-length cDNA clones in the pBluescript vector. The BAN probe was synthesized from the product of an amplification with ban-3′ and RT3′ primers cloned into the pCRscript vector (Stratagene). The LDOX clone and probe were obtained in a similar manner using the gene-specific primers LDOX3 (5′-CAGGATACGCAAGATGAAAG-3′) and LDOX5 (5′-GGTTGCGGTTGAAAGAGTTG-3′).

Histological sections were prepared as follows: immature siliques of wild-type and ban plants were harvested and fixed in sodium phosphate buffer, 200 m m pH 7.2, containing 2.5% glutaraldehyde and 2.5% formaldehyde. After embedding in Technovit 7100 (Kulzer) resin, 2 μm sections were stained with toluidine blue and observed under the light microscope. The vanillin test ( Kristensen & Aastrup 1986) was performed directly by incubation of fresh immature seeds in 1% vanillin in 5 N HCl for 30 min on a microscopy slide. During this incubation period, the anthocyanins present in the ban and double-mutants seed coat are completely removed. Under these conditions, the red colour due to the vanillin test can be unambiguously distinguished from the pink colour due to anthocyanin accumulation in the ban and double mutants.

Sequence analysis

The sequence analyses were performed on double-stranded plasmid DNA with an ABI 373A automatic sequencer using fluorescent dye terminators. Genomic and cDNA sequences were analysed by the computer programs SEQUENCHER (Gene Codes), DNA STRIDER ( Marck 1988) and BLAST ( Altschul et al. 1990 ).