Conditionally lethal mutant alleles of theFUSCA3(FUS3) gene ofArabidopsis thalianaare specifically defective in the gene expression program responsible for seed maturation.FUS3was isolated by map-based cloning and expression of theFUS3cDNA resulted in complementation of the Fus3– phenotype. In the predicted FUS3 gene product, a continuous stretch of more than 100 amino acids shows significant sequence similarity to the B3 domains of the polypeptides encoded byABI3(Arabidopsis) andVP1(maize).FUS3transcription was detected mainly in siliques and was found to be developmentally regulated during embryogenesis. Transcripts of abnormal sizes were observed infus3mutants due to aberrant splicing caused by point mutations at intron termini. Sequence analysis of mutant and wild-typeFUS3alleles, as well as sequencing offus3cDNAs, revealed small in-frame deletions at two different sites of the coding region. While a deletion between B3 and the C-terminus of the predicted polypeptide was found in conjunction with normal FUS3 function, another deletion located within the conserved B3 domain (as well as truncations therein) were associated with the Fus3– phenotype. It is apparent, therefore, that an intact B3 domain is essential for the regulation of seed maturation by FUS3.
The late stages of embryogenesis in higher plants are dominated by seed formation during which vegetative growth of the embryo is halted and storage reserves are accumulated. In addition, the embryo develops desiccation tolerance to prevent damage by the loss of water at the end of ripening ( Goldberg et al. 1994 ;Hughes & Galau 1989;Jürgens & Mayer 1994;West & Harada 1993). By means of seed formation, the plant embryo acquires a quiescent and protected state which allows it to outlast unfavourable environmental conditions such as drought. Under more suitable circumstances, germination can be initiated and further postembryonic vegetative development commences.
In Arabidopsis thaliana, recessive mutations in the FUSCA3 gene (FUS3) lead to a complex phenotype specifically affecting seed formation ( Bäumlein et al. 1994 ;Keith et al. 1994 ). Particularly, fus3 embryos are defective in producing 12S and 2S seed proteins as well as storage lipids but accumulate large amounts of anthocyanin. Mutant fus3 seeds are lethal when desiccated, although upon removal from immature siliques, homozygous fus3 seeds germinate precautiously on a humid substrate and give rise to normal plants without any obvious postembryonic defect.
In line with the conclusions of Hughes & Galau (1989), the developmental pleiotropic and stage-specific phenotype of fus3 mutants reveals a program of late embryonic and early postembryonic development consisting of subsequent, temporally arranged modular units. Accordingly, the FUS3 gene promotes the unit responsible for seed maturation following earlier steps of morphogenesis and tissue differentiation. Loss of FUS3 function can be interpreted as a tendency to shortcut the pathway. While the fus3 embryo is still contained within the silique, seed maturation is left out and the following unit, germination, becomes active prematurely. The presence of typical vegetative traits such as, for instance, an active shoot apical meristem between cotyledons carrying trichomes ( Meinke et al. 1994 ) supports this view by morphological criteria. On the molecular level, genes encoding MYB-family transcription factors which are typically expressed during germination and later, but not during the seed ripening phase, are actively transcribed in mutant fus3 seeds ( Kirik et al. 1998 ). Moreover, it has been shown that in the fus3 background seed-specific promoters of storage protein genes are inactive in embryonic tissues ( Bäumlein et al. 1994 ). Therefore, in analogy to Arabidopsis genes such as APETALA 1 or LEAFY which control the identity of the shoot apical meristem during postembryonic development ( Mandel & Yanofsky 1995;Weigel & Nilsson 1995), FUS3 activity is essential for the zygotic seed tissue to maintain its stage-specific identity of a maturing embryo at the expense of a default tissue identity corresponding to a vegetatively growing postembryonic plant.
The present paper reports the isolation of the FUS3 gene of Arabidopsis thaliana by positional cloning. The salient features of the deduced FUS3 amino acid sequence reveal a region of high sequence similarity with Vp1/ABI3-like proteins. Sequence analysis of mutant fus3 alleles predicts that even minor changes affecting this conserved region are associated with defective seed maturation.
Positional cloning of the FUS3 region
Summarizing the results, Fig. 1 depicts the genetic and physical map of the FUS3 region. The fus3 mutation was localized on the genetic map of molecular markers on the left arm of chromosome 3, within the interval given by GAP-A ( Konieczny & Ausubel 1993) on the telomeric side and g2440 ( Nam et al. 1989 ) on the centromeric side. The cosmid LFHL-1, which was obtained by library screening with the GAP-A amplification product, and g2440 were used as hybridization probes to screen a YAC library of genomic Arabidopsis DNA ( Grill & Somerville 1991). YAC clones EG17D10 (to the left of FUS3), EG14E3 and EG4D12 (to the right) were oriented on the genetic map by RFLP mapping of amplified YAC ends ( Gibson & Somerville 1992). The end probe from the right end of EG14E3 mapped closer to FUS3 on the centromeric (right) side, whereas on the telomeric side EG17D10 mapped equally distant to FUS3 as the GAP-A-derived cosmid LFHL-1. Since the end probe of EG17D10 failed to hybridize with LFHL-1 it was concluded that this YAC end was located closer to FUS3. The FUS3 region was found to reside within a stretch of approximately 80 kb on two YACs (EG8F10 and EG3D2) which bridged the flanking markers. Repeated screening of genomic phage and cosmid libraries with EG3D2 yielded a number of clones which were subsequently tested for hybridization with the inserts of the flanking YACs EG17D10 and EG14E3. Clones which failed to hybridize were used to produce a contig which spanned the fus3 locus as evidenced by recombination mapping. One phage clone and five different cosmid clones within this contig were found to contain a Sty I polymorphism which, in a mapping population comprising more than 1600 meiotic events on each side of the locus, showed absolute linkage. The Sty I cleavage site co-segregated with the Col-0 FUS3 allele whereas it was absent in the Di-G background of the mutant alleles fus3–1 and fus3–2.
cDNA, genomic sequence and expression of FUS3
The cosmid FHL-C4/2 was used to probe a silique-specific cDNA phage library ( Giraudat et al. 1992 ). One of the positive clones, HL1, hybridized to a transcript of low abundance which was mainly expressed in siliques and, to a much lower extent, in other tissues ( Fig. 2a). During wild-type embryogenesis ( Fig. 2b), the HL1-specific hybridization signal became detectable the second day after pollination (d.a.p.). Increasing signal strength was found during the first half of embryogenesis. Towards the end of embryogenesis the level of expression decreased; no hybridizing RNA was found in dry seeds (21 d.a.p., not shown). A comparison with the expression of ABI3 showed that both transcripts were induced at approximately the same developmental stage. However, the abundance of the HL1-specific RNA peaked shortly after mid-embryogenesis whereas, in line with an earlier study ( Parcy et al. 1994 ), the ABI3 transcript was expressed the strongest at the end of embryonic development.
The RNA band which hybridized to HL1 on Northern filters migrated below the 18S rRNA band (about 1.5 kb) corresponding to the size of the cDNA (1241 bp). On genomic Southern filters the hybridization pattern of HL1 was consistent with a single-copy gene (data not shown). A genomic EcoRI fragment spanning the entire cDNA, termed gHL1Ler, was subcloned from cosmid clone FHL-C4/2 (Ler), sequenced and aligned with HL1 (Col-0). The nucleotide sequence given in Fig. 3 represents the portion of gHL1Ler which corresponds to the cDNA plus the 5′ and 3′ adjacent stretches.
The reading frame of the encoded gene product starts at position 179 of the cDNA with two methionine codons. However, the neighbouring nucleotides hint that the second ATG at position 182 is the likely start of the coding sequence because its flankings resemble the consensus TAAACAATGGCT for translational start sites ( Joshi 1987a) more than those of the preceding ATG. Assuming that the second methionine is the translational start, the open reading frame of gHL1Ler encodes a polypeptide of 310 amino acids whereas the cDNA HL1 codes for 312 amino acids (see below).
Transgenic complementation of fus3 mutations
For unknown reasons repeated attempts to transform fus3 mutants with genomic fragments spanning FUS3 and neighbouring genes have been unsuccessful thus far. In order to prove the identity of the FUS3 cDNA, an expression construct containing the HL1 sequence driven by the CaMV 35S promoter was transformed into the homozygous fus3–3 background. T2 seeds were allowed to desiccate completely. Drought-tolerant seeds were germinated on kanamycin-containing medium, transferred to soil, and grown in the greenhouse. Selfing progeny of transgenic plants was characterized with regard to the segregation of green and purple seeds prior to desiccation; additionally, anthocyanin in the seeds was quantified. As summarized in Table 1 the largest proportion of T3 seeds from transgenic fus3–3 plants displays a reduction of anthocyanin content, mostly to wild-type levels. The results prove that the cDNA HL1 encodes a functional gene product which is able to complement the defective Fus3– phenotype.
Table 1. Complementation of the Fus3– phenotype by expression of the cDNA HL1. Suppression of anthocyanin accumulation in homozygous fus3 seeds
T3 seeds counted (total)
The values are given in parentheses and represent arbitrary units, normalized to the number of seeds used in each sample.
In the T2 generation, Fus3– rescuing activity was strictly correlated with resistance to kanamycin, but various segregation ratios of mutant and reverted T3 seeds were observed among transgenic lines. On selective medium T3 seeds containing anthocyanin most often were sensitive to kanamycin, but cases of resistance were also observed which might be attributed to gene silencing.
Variations among wild-type FUS3 alleles
Using a set of primers deduced from the HL1 sequence, the corresponding genomic fragments of Di-G and Col-0 were amplified by PCR (polymerase chain reaction), cloned and sequenced. Variations in the nucleotide sequence are shown in Fig. 3. At nucleotide position 208 an A to C transversion created a Sty I cleavage site in the Col-0-derived cDNA. As already evidenced by the RFLP mapping data, the Sty I site was polymorphic and absent in gHL1Di-G whereas it was present in gHL1Col-0. Thus, the Sty I polymorphism is an intragenic marker co-segregating with the Fus3 locus. In addition, the amino acid on position 9 was changed from asparagine in gHL1Ler and gHL1Di-G to lysine in gHL1Col-0.
Another transversion from A to C at nucleotide position 401 in gHL1Col-0 and gHL1 Di-G changes the amino acid at position 74 from threonine (in gHL1Ler) to proline. Moreover, following an asparagine-rich stretch within the C-terminal portion, the polypeptide deduced from gHL1Col-0 has an insertion of two amino acids, serine and asparagine. This is caused by an additional direct repeat of the hexanucleotide AGCAAC in the last exon.
Defective exon/intron structures in fus3 mutants
The coding region of the FUS3 gene is interrupted by 5 introns ( Fig. 3). In each of these the terminal two nucleotides correspond to the 5′ GT . . . AG 3′ consensus ( Brown 1996). To determine the EMS-induced fus3 mutations, allele-specific gHL1fus3 fragments were amplified by PCR using DNA from homozygous fus3–1, fus3–2 (both Di-G) and fus3–3 (Col) plants. Comparison of the sequences to those of the wild-type strains revealed three allele-specific G to A transitions residing at exon-intron boundaries, at nucleotide positions 782 and 884. In fus3–2 the AG 3′ site of the second intron mutated to AA in fus3–1, and in fus3–3 the 5′ GT terminus of the third intron was changed to AT. The latter two alleles represent independent mutational events because the ecotype-specific polymorphisms found in the respective wild-type alleles reappeared consistently in the nucleotide sequences of gHL1fus3–1 and gHL1fus3–3.
Aberrant allele-specific fus3 transcripts
From homozygous mutant plants, siliques covering the developmental stages between pollination and the onset of seed desiccation were harvested and RNA was isolated. On Northern blots of size-fractioned total RNA, abnormal splicing products were detected ( Fig. 2c). Two classes of fus3 RNAs, on the one hand in the size range of the wild-type transcript and on the other of distinctly smaller size, were detected when the entire cDNA HL1 was used as a probe. Small fus3 RNAs were found to be truncated 5′ segments as they failed to hybridize to a 3′-specific subfragment of HL1. Mutant HL1fus3–2 and HL1fus3–1 cDNAs were amplified from preparations of total silique RNA harvested about 12 d.a.p. Two rounds of amplification (see Experimental procedures) produced DNA fragments of distinct sizes (not shown) which were excised from agarose gels, cloned and sequenced. The results indicated several aberrant pathways for post-transcriptional processing of fus3 transcripts ( Fig. 4).
Two cDNAs, HL1fus3–2A retaining the second intron and HL1fus3–2B in which intron two was excised improperly, correlate with the fus3–2 double Northern band of which the upper part migrated slightly more slowly than the wild-type sized RNA. In HL1fus3–2B, the AG dinucleotide four bases downstream of the mutated 3′ intron terminus was used as a cryptic splice site. In a third cDNA, HL1fus3–2C, the second intron was spliced similarly but exons four and five were skipped causing the connection of exon three to exon six. A transcript with a corresponding size was not detected by Northern analysis indicating a very low abundance of HL1fus3–2C. All three fus3–2 cDNAs predicted a premature translational stop shortly after the last amino acid encoded by exon two, either by a continuation of the reading frame into intron two or by a frame shift. No cDNA representing truncated fus3–2 RNA could be amplified using the combination of an oligo-dT primer and a primer which matched the beginning of the coding region. However, in the case of fus3–1 the same primer combination allowed the amplification of cDNAs reflecting truncated RNAs of the fast migrating fus3–1 Northern band. The sequences of HL1fus3–1A and -A′ both indicate that due to the mutation at the 5′ terminus of intron three transcription stopped prematurely within the intron, after nucleotide 966 and 987, respectively, and was followed by polyadenylation. The continuation of the reading frame into the intron sequence predicted a translational stop shortly after the last amino acid encoded by the preceding exon. Corresponding to the wild-type-sized fus3–1 RNA on Northern filters, HL1fus3–1B was found to have a deletion of only six nucleotides caused by improper excision of intron three using the GT dinucleotide at position 878 as a cryptic intronic 5′ terminus. Thus, HL1fus3–1B lacked two consecutive codons, for tyrosine and arginine at positions 143 and 144.
As a control, the Northern filter was subsequently probed with the ABI3 cDNA ( Giraudat et al. 1992 ) which detected a single transcript in the wild-type and fus3 alleles. Remarkably, ABI3 bands were always sharp whereas fus3 bands, in particular, but also FUS3 bands were always somewhat fuzzy ( Fig. 2b,c). An additional control hybridization experiment with rDNA or several different cDNAs of Myb genes (not shown) as probes produced the same result. Slight variations in transcript lengths were observed in HL1fus3–1A and -A′ as a consequence of several polyadenylation signals in intron three. The DNA sequence at the 3′ end of the gene could also theoretically provide more than one polyadenylation signal ( Joshi 1987b). If so, this would explain the fuzzy Northern bands of the larger size in fus3 mutants and the wild-type.
FUS3 and VP1/ABI3-like proteins share the conserved B3 domain
The protein encoded by the FUS3 cDNA (Col-0) consists of 312 amino acids with a predicted molecular weight of about 35 kDa and an isoelectric point of 5.3. The amino acid sequence was used for a computer-based similarity search ( Altschul et al. 1990 ). It turned out that over a continuous stretch of more than 100 residues ( Fig. 3, underlined) FUS3 has high sequence similarity to the so-called basic region 3 or B3 domain which is conserved among VP1/ABI3-like proteins ( Giraudat et al. 1992 ;McCarty 1995). An exemplary alignment of this region from FUS3, ABI3, VP1 ( McCarty et al. 1991 ) and ARF1 (Auxin Responsive Factor 1 from Arabidopsis;Ulmasov et al. 1997 ) in Fig. 5 shows that the ARF1 sequence shares fewer similarities, while those of ABI3, VP1 and FUS3 can be grouped separately. For the latter, the B3 alignment is devoid of gaps and more than 50% of the amino acid positions are identical. Notably, the small internal deletion in HLfus3–1B ( Fig. 4) affects two of these strongly conserved amino acids, and one of them is also conserved in ARF1 ( Fig. 5a).
On the DNA level, sequence similarity is very limited but the exon/intron structures of VP1 and FUS3 are highly conserved. In both cases, the first intron always closely precedes the beginning of the B3 encoding sequence and the last intron is inserted shortly upstream of the end of that sequence. With regard to the location of codons corresponding to identically conserved amino acids, the positions of three introns (the second, fourth and fifth), were found to reside at identical positions. The insertion sites of intron one differed by three and those of the third intron by only two nucleotides.
This report documents the isolation of the FUS3 gene of Arabidopsis by positional cloning. Having defined the FUS3 region by a YAC walk on chromosome 3 the gene itself was mapped at high resolution on a contig of genomic cosmid and phage library clones spanning approximately 80 kb. The FUS3 cDNA was identified by library screening with a clone which showed absolute linkage. Its length of 1241 bp reflects the size of a low-abundant transcript which is seen as a single band on Northern blots of wild-type RNA. The gene product expressed by the FUS3 cDNA under the control of the CaMV 35S-promoter corrects the Fus3– phenotype in transgenic homozygous fus3–3 plants. In wild-type plants the FUS3 transcript is expressed mainly in siliques during embryogenesis but also in other tissues at much lower levels.
The FUS3 gene contains five introns of which the terminal nucleotides at the splice sites are of the consensus 5′ GT . . . AG 3′. Sequence analysis of PCR-amplified genomic DNA from three different fus3 alleles demonstrates G to A transitions at two splice sites, at the 3′-terminus of intron two and the 5′ terminus of intron three. In line with previous studies (summarized by Brown 1996) alterations of this kind disrupt the splicing process leading to several classes of defective transcripts. Northern analysis, as well as sequencing of cDNAs reflecting fus3 transcripts from homozygous mutants, demonstrates not only aberrant splicing but also the possibility of altered co-transcriptional regulation. The latter can be observed in the case of the mutated fus3–1 5′ terminus of intron three when, in the absence of splicing, transcription terminates prematurely within the intron and is followed by polyadenylation. Intron three is very AT-rich and the finding of minor size variation among small fus3–1 transcripts indicates several polyadenylation signals ( Joshi 1987b). Thus, in the wild-type an active splicing complex probably masks premature termination at this region.
The damaged 3′ site of the second intron in fus3–2 may cause the activation of a cryptic splice site at the AG dinucleotide four bases downstream of the original site. During subsequent splicing recognition of the borders of exons four and five can fail, leading to a direct connection of exon three to the last exon six. The integrity of the 3′ site of the second intron, as well as its surrounding sequence or its precise distance to the intron borders further downstream, thus appear to be important for a proper mRNA maturation process.
With one exception all fus3 transcripts analyzed at the sequence level predict largely truncated gene products. However, one fus3–1 transcript results from splicing of intron three at a cryptic 5′ splice site six bases upsteam of the original site. As a consequence, the reading frame of the mRNA is altered by a deletion of precisely two codons. While this deletion is associated with the Fus3– phenotype, insertion/deletion of two amino acids at a position located more towards the C-terminus is found as a neutral variation in FUS3 alleles. Although no data are available showing the expression of the mutant HLfus3–1B gene product, the deletion might predict a discrete region of the FUS3 protein which is critical for its function.
It is known that the B3 domain of VP1, when taken out of its native context, binds in vitro to the DNA element TCCATGCAT ( Suzuki et al. 1997 ). It is likely, therefore, that the B3 domain of FUS3 similarly mediates binding to specific promoter sequences ( Bäumlein et al. 1994 ). The same applies to ABI3 ( Nambara et al. 1995 ;Parcy et al. 1994 ). During embryogenesis the FUS3 transcript becomes detectable 2 d.a.p. paralleled by the appearance of the ABI3 mRNA. The finding of co-expression at differing levels, however, is consistent with the fact that FUS3 and ABI3, in a combinatorial and synergistic fashion, control overlapping yet not identical aspects of Arabidopsis seed maturation ( Parcy et al. 1997 ). Future studies will have to elucidate to what extent structural similarities of FUS3 and ABI3 contribute to this synergism.
So far, FUS3 is the smallest member of the protein family characterized by VP1/ABI3-like B3 domains ( Fig. 6). VP1 and ABI3, as well as their presumed orthologs in bean ( Bobb et al. 1995 ) or rice ( Hattori et al. 1994 ), have two additional basic domains, B1 and B2 ( Giraudat et al. 1992 ;Hill et al. 1996 ), preceding B3. In FUS3, B1 is absent but, to a small extent, there is sequence similarity to B2 ( Fig. 5b). Moreover, the C-terminal portion of FUS3 downstream of the B3 domain includes an acidic stretch and an amide sequence typically found in transcriptional activation domains. Such properties have indeed been shown for the N-terminal acidic portions of VP1 and the bean VP1/ABI3 ortholog ( Bobb et al. 1995 ;McCarty et al. 1991 ). Provided that these features are the basis for discrete functions of FUS3, an allelic series including a true null allele will have to be analyzed.
In conclusion, the new insights gained from the analysis of FUS3 pave the way for an increasingly specific approach towards understanding both the molecular mechanisms underlying the control of late embryonic development and seed maturation, as well as the evolution of the factors involved.
Plants were grown in a greenhouse at temperatures between 15°C and 24°C or at 22°C in a growth room with 6000 lux of white light for 16 h daily. All mutant alleles of FUS3 used in this work were induced by EMS;fus3–1 and fus3–2 are in the Dijon-G (Di-G) background while fus3–3 is in the homozygous gl1 background of the Columbia ecotype. The molecular mapping experiments were based on the cross fus3–2/fus3–2 (Di-G) ×FUS3/FUS3 (Col-0). The selfing progeny of F1 plants (F2) was harvested after complete desiccation of seeds which caused lethality to all fus3 homozygotes among the F2. The genotypes of the remaining F2, either FUS3 homozygotes or fus3/FUS3 heterozygotes, were identified after selfing by inspection of F3 seeds in immature siliques. Genotyped F3 families were harvested separately and used for DNA analysis. At least 50 plants of each F3 family were taken for the isolation of genomic DNA according to Ausubel et al. (1987) .
For the analysis of FUS3 transcription in the course of embryonic development RNA was isolated from silique tissue. Pollen release by the anthers and pollination was monitored by visual inspection of selfing flowers. Siliques were harvested every other day and in each sample about 10 embryos were staged using cleared preparations of whole seeds as described in Jürgens & Mayer (1994). Samples cover embryonic development starting from the day of pollination; 2 days after pollination (d.a.p.) globular embryos, 4 d.a.p. heart stage and 6 d.a.p. torpedo stage embryos were observed. Subsequent samples reflected the progressive increase of embryo size and, subsequently, the onset of seed maturation. The beginning of seed desiccation was observed 18 d.a.p.
Unless stated otherwise, standard methods were used ( Sambrook et al. 1989 ). For RFLP analysis, 20 μg of genomic DNA was digested with a given restriction enzyme under the conditions recommended by the supplier, size-fractioned on 0.8% agarose gels containing 1 μg ml–1 ethidium bromide in TBE buffer, and blotted onto nylon filters (Hybond+, Amersham). Filters were probed with random-primed 32P-labeled DNA (Megaprime DNA labeling kit, Amersham). The segregation of RFLP markers was analyzed on autoradiographs ( Chang et al. 1988 ;Nam et al. 1989 ). In the case of the GAP-A marker, PCR and segregation analysis were performed as described previously ( Konieczny & Ausubel 1993).
The libraries of Arabidopsis DNA used in this work were the following: (a) SuperCOS1, Col-0 DNA (P. Morris, personal communication); (b) pBIC20, genomic Ler DNA ( Meyer et al. 1994 ); (c) λGEM11, genomic Col-0 DNA (J.T. Mulligan and R.W. Davis, personal communication); (d) EG, genomic YAC Library, Col-0 ( Grill & Somerville 1991); (e) λZapII, silique-specific Col-0 cDNA library ( Giraudat et al. 1992 ). YAC techniques and chromosomal walking were performed essentially as described by Gibson & Somerville (1992).
For complementation experiments, the cDNA HL1 was excised as an EcoRI fragment ( Giraudat et al. 1992 ) and inserted between the 35S promoter and the OCS terminator of pBinAR-33 ( Höfgen & Willmitzer 1990) which carries the NPT II gene as selectable marker. Recombinants were cloned in the Escherichia coli DH5α strain, and then mobilized into the Agrobacterium tumefaciens C58C1Rif(pGV2260) strain by triparental mating using E. coli K514 as a helper strain. The integrity of the transconjugate was controlled by gel blot analysis of total Agrobacterium DNA. Plants homozygous for fus3–3 and gl1 were transformed with the construct using the in planta technique ( Katavic et al. 1994 ). T2 seeds from selfed primary transformants (T1) were dried at least for 4 weeks before selection. From any T1 plant which produced progeny resistant to 50 μg ml–1 kanamycin, a single transgenic line was established. Segregation of wild-type (green) and fus3 (purple) seeds in transgenic gl1/gl1 T2 plants was determined by inspection of immature siliques. Shortly before the onset of seed coat coloration, quantification of anthocyanins was undertaken as described by Parcy et al. (1997) , with the exception that the measurements were taken at 525 nm. For each value at least 100 seeds were analyzed.
For Northern analysis total RNA isolated according to Heim et al. (1993) was size-fractioned on 1.2% agarose gels containing 1 m formaldehyde and blotted onto nylon filters (Hybond+, Amersham). Filters were hybridized overnight at 65°C with random-primed 32P-labeled gene-specific fragments, washed twice in 2× SSPE, 0.1% SDS and twice in 1× SSPE, 0.1% SDS, both at 65°C. The probe specific for the 3′ half of the FUS3 cDNA is defined by the cDNA sequence downstream from the BglII site (A′GATCT) at nucleotide position 1411 in Fig. 3. A tomato rDNA was used for control hybridizations. Autoradiographs were processed using Fuji Bio Imaging plates type BASIII, a Fuji Bio Imaging analyzer and the BAS2000 software package (Fuji Photo Film Co., Tokyo, Japan) together with the TINA software package v2.08 beta (Raytest, Sprockhövel, Germany).
PCR amplifications of gHL1 DNA fragments were performed using the primers Hal1u 5′ ACTAGGCAGTGTTAACCAATTGAG 3′ and H1uni 5′ AATTCTTATATATATTATAAGG 3′ under the following conditions: preheating at 94°C for 3 min; 35 cycles, each comprising 1 min at 56°C, 2 min at 72°C, 1 min at 94°C; post-treatment 72°C for 10 min.
Synthesis of cDNA and amplification for 3′-RACE were conducted as described by Frohman et al. (1988) . For 3′ end amplification of truncated HL1fus3–1 cDNAs, the primer H1rev 5′ TGATGGTTGATGAAAATGTGG 3′ was used. Conditions for the PCR reaction were: 4 min pre-treatment at 94°C; 40 cycles, each comprising 1 min at 94°C, 30 sec at 55°C, 1 min at 72°C. RT–PCR of longer HL1fus3–1 and HL1fus3–2 cDNAs was performed with 2 μg of total RNA isolated from siliques using the Titan- RT–PCR system (Boehringer) under the conditions recommended by the manufacturer. The following primers were used: first round H1uni (see above) and HAL1uni 5′ ACTAGGCAGTGTTAACCAATTGAG 3′; second round H1rev (see above) and Hntrl2 5′ CTTGTTAAGTTTGTGTAAACGTCG 3′. PCR conditions were the same as for 3′-RACE.
For sequencing, DNA fragments were subcloned in the pUC19 vector; PCR fragments were cloned in the TA vector (Invitrogene). DNA sequences were determined in both directions using an A.L.F. DNA sequencer (Pharmacia LKB) and the Autoread Sequencing kit. When PCR-generated DNA fragments were sequenced, experiments were repeated three times independently with identical results. Computer-based similarity searches were conducted using the BLAST algorithm ( Altschul et al. 1990 ) at URL http://www.ncbi. nlm.nih.gov/, multiple alignments were made using the algorithm described by Corpet (1988) and the MULTALIN tool at URL http://www.toulouse.inra.fr/multalin.html.
We are indebted to H. Bäumlein, M. Ganal, J. Giraudat, G. Jürgens and U. Wobus for encouragement and many fruitful discussions. The skilful technical assistance of B. Kettig, K. Blaschek, B. Faßhauer and M. Ohle is gratefully acknowledged. We would like to thank M. Koornneef, H. Bäumlein, U. Wobus and two anonymous reviewers for critical comments on the manuscript. We thank E. Grill, J. Giraudat, P. Morris, R.W. Davis and J.T. Mulligan for making available various Arabidopsis DNA libraries, as well as U. Sonnewald for the pBinAR-33 vector. YAC libraries and RFLP markers have been obtained from the DFG-sponsored Arabidopsis Resource Center in Cologne. Seeds of the fus3–3 line have kindly been provided by K. Keith and P. McCourt. Arabidopsis wild-type strains have been provided by the Genbank at the IPK in Gatersleben. This work has been funded by the Institut für Pflanzengenetik und Kulturpflanzenforschung (IPK), BMFT grant no. 0316301B/19 and DFG grant Mi-512/1–1.