Arabidopsis PAI Gene Arrangements, Cytosine Methylation and Expression

Corresponding author: Judith Bender, Department of Biochemistry, Johns Hopkins University School of Public Health, 615 N. Wolfe St., Baltimore, MD 21205., jbender{at}welchlink.welch.jhu.edu (E-mail)

Communicating editor: J. A. BIRCHLER

	ABSTRACT

TOP ABSTRACT MATERIALS AND METHODS RESULTS DISCUSSION LITERATURE CITED

Previous analysis of the PAI tryptophan biosynthetic gene family in Arabidopsis thaliana revealed that the Wassilewskija (WS) ecotype has four PAI genes at three unlinked sites: a tail-to-tail inverted repeat at one locus (PAI1-PAI4) plus singlet genes at two other loci (PAI2 and PAI3). The four WS PAI genes are densely cytosine methylated over their regions of DNA identity. In contrast, the Columbia (Col) ecotype has three singlet PAI genes at the analogous loci (PAI1, PAI2, and PAI3) and no cytosine methylation. To understand the mechanism of PAI gene duplication at the polymorphic PAI1 locus, and to investigate the relationship between PAI gene arrangement and PAI gene methylation, we analyzed 39 additional ecotypes of Arabidopsis. Six ecotypes had PAI arrangements similar to WS, with an inverted repeat and dense PAI methylation. All other ecotypes had PAI arrangements similar to Col, with no PAI methylation. The novel PAI-methylated ecotypes provide insights into the mechanisms underlying PAI gene duplication and methylation, as well as the relationship between methylation and gene expression.

THE model higher plant Arabidopsis thaliana has a small genome size of ~10⁸ bp. Despite the relative simplicity of the Arabidopsis genome, many functions in this plant are encoded by gene families rather than by a single gene. For example, many of the enzymes in the tryptophan biosynthetic pathway are encoded by two- (LAST et al. 1991 ; NIYOGI and FINK 1992 ) or three-member (NIYOGI et al. 1993 ; LI et al. 1995 ) gene families. Important questions pertaining to these gene families are how their gene duplications arose and how they are stably maintained in the genome.

In particular, the gene family encoding the enzyme that catalyzes the third step of the tryptophan biosynthetic pathway, phosphoribosylanthranilate isomerase (PAI), displays a number of intriguing features. The PAI gene family has been characterized previously in detail in two standard laboratory strains of Arabidopsis, Columbia (Col), and Wassilewskija (WS; BENDER and FINK 1995 ; LI et al. 1995 ). These analyses revealed that in both strains there is an unusually high degree of sequence identity among the PAI family members, including untranslated portions of the genes, such as intron and promoter sequences. In Col, the PAI family is encoded by three unlinked genes: PAI1 on the upper arm of chromosome 1, PAI2 on the upper arm of chromosome 5, and PAI3 in the middle of chromosome 1. The PAI1 and PAI2 genes are almost perfectly identical to each other over the 2307 bp extending from 355 bp upstream of the ATG translational start codon to 470 bp downstream of the TAA translational stop codon, including five exons and four introns. Within this region, the Col PAI1 and PAI2 genes differ by only 26 scattered single-base differences (99% identity). In contrast, the PAI3 gene is 90% identical to PAI1 and PAI2 due to many scattered base differences. The base differences in the translated parts of the PAI3 gene are predicted to yield a protein with 18 amino acid differences from Col PAI1.

The high degree of sequence identity across PAI gene family members is unusual in comparison to other Arabidopsis tryptophan gene families. For example, the duplicated tryptophan synthase ß-subunit genes TSB1 and TSB2 are 85% identical to each other in their exon sequences (exclusive of presumed chloroplast target sequences), but they are highly divergent in their transcribed untranslated sequences (LAST et al. 1991 ). Furthermore, the duplicated anthranilate synthase -subunit genes ASA1 and ASA2 are 67% identical to each other in their predicted amino acid sequences, but they are divergent in their exon and intron nucleic acid sequences and in their intron structures (NIYOGI and FINK 1992 ). Therefore, the high degree of identity among the Col PAI genes, particularly between PAI1 and PAI2, suggests that they evolved relatively recently.

In WS, there are two more unusual features of the PAI gene family. First, whereas WS carries PAI2 and PAI3 genes that are almost identical to the Col PAI2 and PAI3 genes, at the PAI1 locus, WS carries an inverted-repeat duplication of the PAI1 gene PAI1-PAI4 (BENDER and FINK 1995 ). The PAI1-PAI4 inverted repeat is flanked by ~2.9 kb of perfect direct-repeat sequences. Furthermore, the PAI1-proximal direct repeat extends into the first 731 bp of a duplicated PAI4 gene pai4 5' (BENDER and FINK 1995 ; Figure 3).

View larger version (56K): In this window In a new window Download PPT slide   Figure 1. The PAI1 locus is polymorphic. Genomic DNA prepared from the indicated ecotypes was cleaved with HindIII and probed with a PAI-internal cDNA fragment. HindIII cleaves each gene twice 170 bp apart in the first and second introns; both sites lie upstream of the probe fragment so that each singlet gene yields a single hybridizing band. PAI inverted repeats also yield a single hybridizing band because HindIII does not cleave between PAI1 and PAI4. In the Cvi-0 PAI1 gene, the second intron HindIII site is missing because of a single-base polymorphism, so this gene yields a fragment with an additional 170 bp relative to HindIII digests of other PAI1 genes. Note that in Cvi-0, the partial direct-repeat PAI4* duplication (Figure 3) yields a PAI-hybridizing HindIII band of the same size as the full PAI1-PAI4 inverted repeat so that the two bands are superimposed on this blot. P1 is Col PAI1, P1-P4 is WS PAI1-PAI4, P2 is PAI2, and P3 is PAI3. The molecular weight of each band is indicated along the right margin in kilobases. The molecular weight of the PAI3 band is estimated relative to size standards. The molecular weights of all other bands are based on genomic sequences.

View larger version (55K): In this window In a new window Download PPT slide   Figure 2. Ecotypes with novel PAI gene structures and cytosine methylation. Genomic DNA prepared from the indicated Arabidopsis ecotypes was cleaved with either HpaII (H) or MspI (M) and probed with a PAI-internal cDNA fragment. HpaII/MspI restriction maps and the region covered by the probe are shown in Figure 3. HpaII and MspI are both blocked from cleaving by cytosine methylation of their recognition sequence, 5' CCGG 3', although HpaII is sensitive to methylation of either cytosine in the sequence and MspI is sensitive only to methylation of the outer cytosine. Bands diagnostic of methylation are marked in the margin with asterisks. P1 is Col PAI1, P1-P4 is WS PAI1-PAI4, P2 is PAI2, and P3 is PAI3. The molecular weights of these bands are indicated along the right margin in kilobases. The molecular weights of the PAI3 bands are estimated relative to size standards. The molecular weights of all other bands are based on genomic sequences.

View larger version (28K): In this window In a new window Download PPT slide   Figure 3. Gene structures and restriction maps for the Arabidopsis PAI genes. The XhoI (X) and MspI (M) restriction maps for the PAI1 loci from the indicated ecotypes are shown with maps for the Col PAI2 and PAI3 loci. Thin gray arrows indicate the direct-repeat sequences flanking each PAI1 locus. Wide colored arrows represent genes, with the start of the arrow at the ATG translational start codon and the tip of the arrowhead at the translational stop codon: PAI genes are in black, S15a is in blue, histone H2B is in red, and FAD8 is in green. For the PAI genes, the additional upstream and downstream regions of shared sequence identity are indicated by underlying hatched black lines. In the regions between inverted-repeat PAI genes for WS, Kas-1, and C24, the junctions of PAI1 and PAI4 downstream sequences are marked with a vertical line. For Kas-1, C24, Ita-0, and Cvi-0, central sequences that include the FAD8 coding region are marked in green. In Ita-0, the unique central sequences are shown in yellow, and in Cvi-0, the central sequences derived from PAI sequences are shown in black. See Figure 4 for a close-up view of the structures between inverted-repeat PAI1 and PAI4 genes. PAI1-PAI4 duplications are oriented with the PAI1 gene on the left and the PAI4 gene on the right. In Cvi-0, the leftmost black arrow represents the duplicated PAI4* gene, and in WS, the short leftmost black box represents the partial pai4 5' duplication. The white box upstream of the PAI1 gene in most ecotypes indicates the 944 bp of heterologous sequence relative to WS pai4 5'. Black arrows drawn under short duplicated sequences in Cvi-0, Ita-0, C24, and WS indicate the relative orientations and extent of the duplications. Asterisks indicate methylation. The methylated MspI sites around the Cvi-0 PAI4* duplication could not be determined precisely because of the density of sites flanking this region; the marked sites around this locus are therefore an approximation. Sequences for all loci except Col PAI3 are available from GenBank: Col PAI1, AF130878; Col PAI2, AB005241; WS PAI1-PAI4 (updated), U34757; Cvi-0 PAI1-PAI4, AF130876; Cvi-0 PAI4*, AF130877; Ita-0 PAI1-PAI4, AF130875; C24 PAI1-PAI4, AF130874; and Kas-1 PAI1-PAI4, AF130873.

The second unusual feature of the WS PAI genes is that all four full-length genes and the partial pai4 5' gene are densely cytosine methylated (BENDER and FINK 1995 ; LUFF et al. 1999 ). This methylation covers the regions of the genes that have sequence identity to each other, but it does not spread more than a few hundred bases beyond the boundaries of PAI sequence identity. In contrast, the three Col PAI genes do not display detectable methylation (BENDER and FINK 1995 ). PAI methylation in WS correlates with silencing of at least the PAI2 gene, but there is sufficient total PAI expression in this strain for a normal plant phenotype (BENDER and FINK 1995 ; JEDDELOH et al. 1998 ).

Because the PAI1 locus is polymorphic between WS and Col, we reasoned that an analysis of PAI gene copy number, arrangements, and methylation in other Arabidopsis ecotypes might provide new insights into the correlation between gene structure and the onset of cytosine methylation, as well as provide new genetic tools for understanding methylation and its relationship to gene expression. In addition, the PAI structures observed in other ecotypes could elucidate the mechanism of PAI gene duplication. To investigate this possibility, we screened PAI gene structure and methylation by Southern blot analysis for 39 additional ecotypes of Arabidopsis isolated from around the world. We found that whereas most ecotypes had PAI gene structures identical to that observed for Col (three unmethylated genes) and two ecotypes had PAI gene structures nearly identical to that observed for WS (four methylated genes), four ecotypes had novel PAI arrangements, with a variation of the inverted-repeat PAI gene structure at the PAI1 locus and PAI cytosine methylation. On the basis of our results, we discuss possible models for the differential methylation of PAI gene arrangements, the evolution of the polymorphic PAI gene family, and the effects of methylation on PAI gene expression.

	MATERIALS AND METHODS

TOP ABSTRACT MATERIALS AND METHODS RESULTS DISCUSSION LITERATURE CITED

Arabidopsis strains and growth conditions:
Arabidopsis ecotypes were obtained from the Arabidopsis Biological Resource Center at Ohio State University, with the exception of C24, which was obtained from Patrick Masson (University of Wisconsin, Madison, WI). Plants were grown under continuous light in Fafard Growing Mix 2 (Griffin Greenhouse Supplies).

Plant genomic DNA preparation and Southern blot analysis:
Plant genomic DNA was prepared as follows. Four-week-old plants were frozen in liquid nitrogen and ground into a fine powder with a mortar and pestle. Ground tissue (~10 g) was thawed in 20 ml lysis buffer (100 mM Tris-OH, pH 9.5, 1.4 M NaCl, 20 mM EDTA, pH 8.0, 2% hexadeclytrimethylammonium bromide, 1% polyethylene glycol, M_r 8000), plus 50 µl 2-mercaptoethanol and then heated at 74° for 20 min. The lysed tissue was cooled to room temperature and extracted with an equal volume of CHCl₃. The aqueous phase was transferred to a fresh tube and precipitated with an equal volume of isopropanol at room temperature for 30 min, followed by centrifugation at 5000 x g for 20 min. The pellet was resuspended in 1 ml 0.75 M NaCl, incubated with 5 µl 10 mg/ml RNase A at 37° for 30 min, mixed with 0.25 ml of water and 0.75 ml of JetStar (Genomed) column equilibration buffer E4, and centrifuged for 10 min at 5000 x g to pellet insoluble material. The DNA sample supernatant was loaded on an equilibrated JetStar minicolumn and washed, eluted, and isopropanol precipitated as described by the manufacturer. DNA pellets were resuspended to a final concentration of ~0.5 µg/µl in water. For Southern blot analysis, ~1 µg of genomic DNA was digested with the restriction endonuclease of interest, electrophoresed through an 0.7% TBE agarose gel, transferred to a Hybond-N (Amersham, Arlington Heights, IL) membrane, and fixed by UV cross-linking. Purified probe DNA fragments were labeled using the MegaPrime kit (Amersham) as described by the manufacturer and hybridized to Southern blots in Church buffer (CHURCH and GILBERT 1984 ) at 65°, followed by several washes with 0.2x SSC/0.1% SDS at 65°. The PAI1 cDNA probe is an internal 0.7-kb PstI fragment (BENDER and FINK 1995 ; LI et al. 1995 ). The direct-repeat probe is a 2.3-kb HincII-to-XhoI fragment derived from the sequences upstream of the WS PAI4 gene.

Construction and screening of genomic DNA libraries:
Genomic DNA isolated from Kas-1, C24, Ita-0, or Cvi-0 was partially digested with Sau3AI to an average fragment size of 15–20 kb, ligated with DASH (Stratagene, La Jolla, CA) arms, and packaged into phage particles using Gigapack Gold (Stratagene) in vitro phage packaging extracts as described by the manufacturer. Each library contained 100,000–200,000 primary clones. PAI-positive plaques were isolated by transferring to Hybond-N membranes and hybridizing with a PAI cDNA or direct-repeat probe as described for Southern blot analysis above.

DNA sequencing:
Genomic DNA fragments were subcloned from library isolates into pBluescript II KS+ (Stratagene) and were sequenced either from standard T3 and T7 primer sites in the vector or from custom-designed internal primers by the Johns Hopkins University Department of Biological Chemistry Biosynthesis and Sequencing Facility.

PCR analysis of sequences flanking the PAI1-proximal direct repeat:
PCR primers based on sequences lying just outside the 731-bp pai4 5' duplication in WS and the heterologous 944-bp sequence present at this region in Col (see Figure 3) were used to amplify the analogous regions from Kas-1, C24, and Ita-0 genomic DNA using standard PCR reagents and conditions. These three ecotypes gave fragments that were identical in size and in PstI or HincII restriction patterns to the Col fragment. Also, genomic clones of this region from Kas-1 and Cvi-0 were explicitly sequenced and found to be nearly identical to Col.

Isolation and analysis of PAI cDNAs:
Standard Col (ELLEDGE et al. 1991 ), Landsberg erecta (Ler, MINET et al. 1992 ), and WS (gift of L. Castle and D. Meinke, obtained from the Arabidopsis Biological Resource Center) cDNA libraries were screened by hybridization with the PAI1 cDNA probe to identify PAI clones. Inserts were subcloned into pBluescript II KS+ and sequenced to determine their 5' and 3' end structures and to distinguish from which PAI gene they arose. PAI function was tested by determining whether cDNAs expressed from the lac promoter on pBluescript II KS+ in a pai-deficient Escherichia coli strain W3110 trpC9830 could complement the mutant grown on minimal M9 medium, as described previously (BENDER and FINK 1995 ; LI et al. 1995 ). Site-directed mutageneses to remove the 25-bp intron insertion in the PAI3 cDNA, to introduce the WS PAI4 9-bp deletion into a WS PAI1 cDNA, or to introduce the C24 PAI4 6-bp deletion into a WS PAI1 cDNA were performed using standard methods (KUNKEL et al. 1987 ).

Reverse transcriptase PCR analysis of PAI cDNAs:
Total Arabidopsis RNA isolated as described (NAGY et al. 1988 ) was treated with RNase-free DNase (Promega, Madison, WI) and used as a template for murine Maloney leukemia virus reverse transcriptase (M MLVRT, GIBCO-BRL, Bethesda, MD). Specifically, 20-µl reactions containing 2.0 µg total RNA, 0.5 mM of each dNTP, the appropriate reverse primer, and buffer conditions recommended by the manufacturer were incubated at 65° for 5 min, then chilled to 4°, at which point 200 units of M MLVRT was added. Reactions were then incubated at 42° for 1 hr, heated to 95° for 5 min, and chilled to 4°. For each RT reaction, 0.5 µl was used as a template for PCR amplification in a 100-µl volume using standard reagents and an amplification program of 40 cycles of (94° 30 sec, 55° 30 sec, 72° 1 min). Detailed information about primer sequences used in this analysis is available upon request from J. Bender. RT-PCR products <300 bp were resolved by electrophoresis through 3% agarose/1% NuSieve (FMC, Rockland, ME) TBE gels, and RT-PCR products >300 bp were resolved on 1.5% agarose TBE gels.

Northern blot analysis:
RNA was prepared from whole 4-wk-old plants, electrophoresed, transferred to Hybond-N (Amersham) membranes using standard procedures (AUSUBEL et al. 1989 ), and hybridized with a PAI1 cDNA probe as described for Southern blot analysis above. Blots were subsequently stripped and reprobed with an -tubulin probe to normalize for differences in loading.

	RESULTS

TOP ABSTRACT MATERIALS AND METHODS RESULTS DISCUSSION LITERATURE CITED

Determination of PAI gene structures and methylation in Arabidopsis ecotypes:
Because the PAI1 locus is structurally polymorphic between the previously characterized Col and WS Arabidopsis ecotypes (BENDER and FINK 1995 ), we screened 39 new ecotypes for PAI gene structures using Southern blot assays (Table 1, Figure 1 and Figure 2). This analysis revealed that most ecotypes had PAI structures similar to Col, with a singlet gene at the PAI1 locus, but six ecotypes displayed band patterns diagnostic of novel arrangements at the PAI1 locus. Furthermore, because PAI sequences are not cytosine methylated in the Col ecotype but are densely methylated in the WS ecotype, we also screened the 39 new ecotypes for evidence of PAI gene methylation using a Southern blot assay. This analysis revealed that the ecotypes with a singlet PAI1 gene structure at the PAI1 locus (the Col arrangement) displayed no detectable PAI methylation. In contrast, like WS, the six ecotypes with novel structures at the PAI1 locus displayed band patterns diagnostic of dense PAI methylation (Table 1, Figure 2 and Figure 3). These observations suggest that the acquisition of an unusual structure at the PAI1 locus causes PAI gene methylation.

As a preliminary screen of both PAI structure and methylation, we digested genomic DNA from 39 new ecotypes with the methylation-sensitive restriction endonuclease isoschizomers HpaII and MspI, followed by Southern blot analysis with a PAI1 cDNA probe. Each PAI locus carries a single conserved HpaII/MspI site in the second PAI intron with different flanking HpaII/MspI sites (Figure 3). Therefore, each PAI locus yields distinct HpaII/MspI fragments, diagnostic of either fully cleaved unmethylated or partially cleaved methylated sequences. PAI methylation detected by HpaII/MspI digestion was previously shown to correlate with PAI methylation across the rest of the WS PAI genes detected by several other restriction endonucleases and by a genomic methylation sequencing technique (BENDER and FINK 1995 ; JEDDELOH et al. 1998 ; LUFF et al. 1999 ).

Out of the 39 new ecotypes examined, 31 had HpaII/MspI PAI patterns identical to those observed in Col, diagnostic of three unmethylated genes (Table 1, Figure 2). Two other ecotypes had Col-related patterns. Lu-1 had PAI1 and PAI2 patterns identical to those observed for Col, but a different PAI3 pattern. This pattern was most likely caused by a HpaII/MspI PAI3 polymorphism because when Lu-1 DNA was cleaved with XhoI, it yielded a band pattern identical to Col (data not shown). Ll-0 had PAI2 and PAI3 patterns identical to those observed for Col, but no detectable PAI1 fragment. This pattern was verified with a XhoI digest (data not shown).

Two ecotypes had band patterns related to those observed in WS, diagnostic of a complex PAI arrangement at the PAI1 locus plus dense cytosine methylation across all three PAI loci (Table 1). One PAI-methylated ecotype, Bur-0, had PAI patterns identical to those observed in WS. Another PAI-methylated ecotype, Nd-0, had PAI1-PAI4 and PAI2 patterns identical to those observed in WS, but a different PAI3 pattern. This pattern was most likely caused by a HpaII/MspI PAI3 polymorphism because when Nd-0 DNA was cleaved with XhoI, it yielded a band pattern identical to WS (data not shown). Four other PAI-methylated ecotypes, Kas-1, C24, Ita-0, and Cvi-0, displayed band patterns consistent with novel complex structures at the PAI1 locus (Figure 1 and Figure 2).

Southern blot analysis of the four novel PAI-methylated ecotypes using the enzyme HindIII, which is relatively insensitive to cytosine methylation, confirmed that there were only three PAI loci in each of these strains (Figure 1). The PAI2 and PAI3 loci were structurally invariant relative to Col and WS, whereas the PAI1 locus was polymorphic. The sizes of the PAI1 locus bands were consistent with either singlet genes or inverted-repeat genes spaced further apart than in WS. Southern blot analysis with the enzyme XhoI (restriction map shown in Figure 3), and subsequent cloning and sequencing of the PAI1 locus from each ecotype (see below), showed that each of the novel PAI-methylated ecotypes in fact carried a variation of the PAI1-PAI4 inverted-repeat structure.

The PAI1-PAI4 locus in WS is flanked by ~3 kb of direct-repeat sequences (Figure 3; BENDER and FINK 1995 ). Analysis of the novel PAI-methylated ecotypes with a direct-repeat probe indicated that like WS, all four ecotypes carried direct-repeat sequences flanking the inverted-repeat PAI genes (data not shown). Subsequent cloning and sequencing confirmed this result.

Sequence analysis of PAI1 locus structures:
To obtain a detailed understanding of the novel PAI1 loci detected by Southern blot in Kas-1, C24, Ita-0, and Cvi-0, we constructed genomic libraries from these strains and screened for PAI1 locus clones using direct-repeat and/or PAI1 cDNA probes. Relevant portions of the clones were subcloned and sequenced. We also extended the existing sequences around the cloned WS and Col PAI1 loci. These analyses, as well as the complete sequence of a P1 clone carrying the PAI2 region of the genome (SATO et al. 1997 ), revealed a series of structural relationships among the six different PAI arrangements for WS, Kas-1, C24, Ita-0, Cvi-0, and Col, as well as a structural relationship between the unlinked PAI2 gene and the PAI1-PAI4 duplication (Figure 3).

Sequencing of the WS PAI1-PAI4 locus over a continuous region of ~11 kb (Figure 3) showed that the direct repeats flanking PAI1-PAI4 were nearly identical to each other: the PAI1-proximal repeat was 2880 bp long, and the PAI4-proximal repeat was 2905 bp long because of the presence of a 25-bp insertion near the end furthest away from PAI4. Each of the repeats contained a complete open reading frame for a gene with identity to histone H2B, although this gene was not represented in the Arabidopsis Expressed Sequence Tag (EST) database. Immediately upstream of the PAI1-proximal direct repeat was a 731-bp partial pai4 5' duplication followed by unique sequences with no significant identity to the sequences in the database. The unique sequences immediately beyond the PAI4-proximal direct repeat encoded ribosomal protein S15a (BONHAM-SMITH and MOLONEY 1994 ). Based on an EST sequence for a Col S15a cDNA (GenBank accession no. R31305), the promoter, transcriptional start, and translational start of the S15a gene are duplicated in both copies of the direct repeat so that the S15a promoter lies upstream of both PAI1 and S15a (Figure 3). Furthermore, a comparison of the S15a genomic sequence to the cDNA sequence indicated that there is an intron in the upstream untranslated sequence that is included in the direct repeats.

Previous sequence analysis of the Col PAI1 locus indicated that this locus carries a deleted direct repeat of 813 bp relative to the WS sequence downstream of PAI1, followed by the S15a gene (BONHAM-SMITH and MOLONEY 1994 ; BENDER and FINK 1995 ; LI et al. 1995 ; Figure 3). Southern blot analysis was used to determine that upstream of PAI1 there is at least a partial direct-repeat sequence but no pai4 5' duplication (BENDER and FINK 1995 ). To better understand the structure in this upstream region, we sequenced through the upstream direct repeat into unique flanking sequences. This analysis revealed that Col carried an almost-complete upstream direct repeat relative to WS, missing only the 5 bp most distal to the PAI1 gene. Therefore, Col and WS have almost identical PAI1 promoter sequences over a region of ~3.2 kb upstream of the PAI1 translational start. Beyond this point, Col carried 944 bp of the heterologous sequence relative to the WS 731-bp pai4 5' duplication (Figure 3). Further upstream of this heterologous region, however, both ecotypes had identical sequences. The 944-bp region did not have any significant identity to other sequences in the database.

The novel methylated ecotypes all had overall structures similar to that of WS at the PAI1 locus, with two direct repeats flanking a PAI1-PAI4 inverted repeat. However, each ecotype had unique variations on this basic pattern (Figure 3). For example, the Kas-1 and C24 ecotypes had the same structure as Col upstream of PAI1. The Ita-0 ecotype also had a Col-like sequence upstream of PAI1, except that substituting for the 33 bp of direct-repeat sequence most proximal to PAI1 was 300 bp of the PAI1-distal end of the direct-repeat sequence in an inverted orientation. In Cvi-0, the region upstream of PAI1 had a more complex structure than any of the other ecotypes examined. Beyond the PAI1-proximal direct repeat, there was a duplication consisting of a full-length PAI gene oriented in the same direction as PAI4 (PAI4*) plus the last 207 bp of a direct-repeat sequence. Beyond this deleted direct-repeat PAI4* duplication was the same upstream sequence found in Col, Kas-1, C24, and Ita-0.

In the region upstream of PAI4 the ecotypes Kas-1, Ita-0, and Cvi-0 each had a full direct repeat followed by S15a sequences similar to WS (Figure 3). In C24, however, 96 bp around the junction between PAI4 and its flanking direct repeat was substituted with 929 bp of heterologous sequences. These sequences consisted of a 911-bp duplication of the sequences found immediately adjacent to the upstream end of the PAI1-proximal direct repeat, followed by 18 bp of unique sequence at the junction with PAI4. Beyond this duplication, C24 carried direct-repeat and S15a sequences analogous to those found in the other ecotypes (Figure 3).

Another focus of our analysis was the sequence between the PAI1-PAI4 inverted repeats, representing the sequences 3' of each PAI gene. In Col, the PAI1 3' sequence extended for 470 bp downstream of the translational stop codon with almost perfect identity to the Col PAI2 gene 3' sequence, followed by 10 bp of heterologous sequence and then 813 bp of flanking direct-repeat sequence (Figure 3 and Figure 4). In the ecotypes WS, Kas-1, and C24, the PAI1-PAI4 inverted-repeat central sequences consisted of deleted palindromes of the 470-bp downstream sequence (Figure 4). In Ita-0 and Cvi-0, PAI1 and PAI4 both carried the full 470-bp downstream sequence with a short novel sequence sandwiched in between. In Ita-0, this central sequence carried 24-bp inverted-repeat ends plus another repeat of the 24-bp sequence and a repeat of the last 43 bp of the PAI2-identical downstream sequences in the middle. In Cvi-0,the central sequence carried a short duplication of a PAI fifth-exon sequence flanked by novel sequences. There was no identity between Ita-0 and Cvi-0 central sequences.

View larger version (17K): In this window In a new window Download PPT slide   Figure 4. Structure of central regions in PAI1-PAI4 loci. The regions downstream of the PAI1 and PAI4 translational stop codons (TAA) are shown for the indicated ecotypes, with PAI1 on the left and PAI4 on the right. Black arrows represent the 470 bp of conserved PAI 3' sequence. White arrowheads indicate the termini of deletions within this sequence. Vertical zigzag lines indicate deletion junctions. The double black arrowheads near the end of the Cvi-0 PAI4* sequence represent a 21-bp tandem duplication in this region. The divergently transcribed FAD8 coding sequence (2014 bp from translational start to translational stop) downstream of PAI2 is shown as an open box, with hatched lines indicating that this region is not drawn to scale. The boxed regions at the ends of some PAI1 and PAI4 3' sequences indicate FAD8 duplications (shown in green in Figure 3). The light gray arrow downstream of Col PAI1 represents the 813-bp flanking direct-repeat sequence, with hatched lines indicating that this region is not drawn to scale. Similarly, the light gray line downstream of Cvi-0 PAI4* represents the 207-bp deleted direct-repeat sequence. The partial PAI fifth-exon duplication in Cvi-0 is indicated by a black box, and the flanking heterologous sequence is indicated by a gray box. The novel sequence between the Ita-0 PAI1 and PAI4 genes is indicated by a white box. The black arrowhead in this box represents the partial duplication of PAI 3' sequences, and the gray arrowheads in this box represent novel 24-bp duplications. The table at the right indicates the number of base pairs of sequence included in the PAI1 and PAI4 3' ends and additional center material relative to the conserved 470-bp sequence.

With a combination of sequencing and restriction mapping analysis, we were able to construct HpaII/MspI restriction maps for the novel PAI1 loci. This information plus the observed fragment sizes on a HpaII/MspI Southern blot (Figure 2) allowed a determination of which sites in each PAI1-PAI4 structure were cytosine methylated (marked with asterisks in Figure 3). Methylation was almost entirely contained within the PAI-identical sequences, with little or no spread into flanking direct repeats.

PAI cDNA abundance and function studies:
Using a deletion mutant derivative of WS that lacks the PAI1-PAI4 inverted-repeat genes, we previously showed that cytosine methylation correlates with a loss of expression for the PAI2 gene (BENDER and FINK 1995 ; JEDDELOH et al. 1998 ). Genetic backcross experiments indicated that PAI2 is also silenced in the parental WS strain (BENDER and FINK 1995 ). Nonetheless, parental WS is phenotypically normal, with levels of total PAI transcripts and PAI enzyme activity comparable to levels measured in Col, suggesting that there is a relatively high level of PAI expression from the methylated PAI1 and/or PAI4 genes (BENDER and FINK 1995 ).

To better understand which PAI genes are expressed in PAI-methylated ecotypes, we analyzed PAI transcripts in WS by cloning and sequencing PAI cDNAs from a standard WS library. This approach revealed that PAI1 was the only detectable species (Table 2). In contrast, screening of Col and Ler cDNA libraries revealed that PAI1 and PAI2 were equally abundant in either of these unmethylated PAI ecotypes, with PAI3 approximately fourfold less abundant than either of its sister genes. In both Ler and WS, rare PAI1 cDNAs were found that had upstream direct-repeat sequences spliced to PAI1 first-exon sequences (Table 2). These rare cDNAs are best explained as transcripts that initiate from upstream S15a sequences in the direct-repeat rather than more proximal PAI1 sequences. The material that is deleted between the rare transcript start sites and the PAI1 first-exon site corresponds to predicted intron material in the 5' untranslated region of the S15a transcript on the basis of a comparison between an S15a cDNA sequence (GenBank accession no. R31305) and the genomic sequence. In the Ler 5' spliced transcript, the 3' junction of the S15a first upstream exon is spliced directly to a PAI1 upstream site. In the WS 5' spliced transcript, the 3' junction of the S15a first upstream exon is correctly spliced to the second S15a exon, and then a cryptic site in the second exon is spliced to the same PAI1 upstream site used in the Ler transcript (Table 2).

cDNA analysis also suggested that PAI4 and PAI3, besides being poorly expressed, do not encode functional PAI enzymes. WS PAI4 contains a 9-bp deletion in the fifth exon relative to PAI1 and PAI2 (BENDER and FINK 1995 ). We explicitly mutagenized a WS PAI1 cDNA in an E. coli expression vector to generate this deletion and then tested the ability of the PAI1 parental construct vs. the 9-bp-deleted PAI1 construct to complement the tryptophan auxotrophy of an E. coli PAI-deficient mutant. The WS PAI1 cDNA complemented the mutant, as reported previously for Col PAI1 and PAI2 cDNAs (Table 2; BENDER and FINK 1995 ; LI et al. 1995 ). However, the 9-bp deletion rendered the cDNA unable to complement the mutant. Therefore, WS PAI4 is not likely to produce active PAI enzyme because of this deletion.

Sequencing of PAI3 cDNAs isolated from Col and Ler libraries indicated that the PAI3 transcript was incorrectly spliced to yield a 25-bp insertion of intron sequences between the fourth and fifth exons (Table 2). The insertion is most likely caused by a single-base polymorphism that changes the consensus 3' intron site in the fourth PAI intron from AG to TG. PAI3 cDNAs are predicted to encode a protein with 12 novel amino acids downstream of the fourth exon followed by a premature termination codon. When expressed in a PAI-deficient E. coli strain, the Col PAI3 cDNA failed to complement tryptophan auxotrophy. Therefore, the PAI3 gene is not likely to produce PAI activity in any ecotype that carries the PAI3 splice site polymorphism, including Col, WS, C24, Kas-1, Ita-0, and Cvi-0 (as assessed by sequencing cloned PAI3 genes and/or by monitoring a PstI polymorphism created by the PAI3 splice junction mutation).

RT-PCR analysis of PAI gene expression:
As for WS, the novel PAI-methylated ecotypes Kas-1, C24, Ita-0, and Cvi-0 all had steady-state levels of total PAI transcripts slightly higher than the levels measured in Col (Figure 5A). Each PAI-methylated ecotype also displayed low levels of higher-molecular-weight PAI transcripts, as previously observed in WS (BENDER and FINK 1995 ). The higher-molecular-weight species were largest in Ita-0 and Cvi-0, consistent with these transcripts arising by readthrough of the inverted repeat. PAI polymorphisms in the PAI-methylated ecotypes allowed us to determine via RT-PCR which transcripts are the predominant species detected by Northern blot. In every case, PAI1 proved to be the only abundantly expressed PAI gene (see below).

View larger version (37K): In this window In a new window Download PPT slide   Figure 5. PAI1-methylated ecotypes express PAI transcripts from the inverted repeat. (A) Total PAI steady-state transcript levels in the indicated ecotypes were determined by Northern blot with the PAI-internal cDNA probe. To control for loading differences, the blot was stripped and reprobed with an -tubulin (TUB) probe. The position of the high-molecular-weight PAI RNA species in each PAI-methylated ecotype is marked with an asterisk. (B) The proportion of PAI transcripts carrying second-exon SacI sites was determined by RT-PCR analysis. Total RNA from the indicated ecotypes was reverse transcribed and then amplified by PCR using primers to common sequences in all PAI gene first and fourth exons flanking the polymorphic SacI site. These primers yield a 548-bp fragment that is cleaved by SacI into 435 and 113 bp (not shown) fragments in DNA products that contain the site. Equal amounts of uncut (-) and SacI-cut (+) RT-PCR product for each ecotype were resolved on a 1.5% agarose gel and visualized by ethidium bromide staining.

PAI transcripts expressed from the PAI1-PAI4 inverted-repeat genes can be distinguished from transcripts expressed from the singlet PAI2 and PAI3 genes in most PAI-methylated ecotypes by a restriction site polymorphism. Specifically, the PAI1 and PAI4 genes in the ecotypes WS, Kas-1, C24, and Ita-0 lack a conserved second exon SacI site, whereas the PAI2 and PAI3 genes carry this site. We performed RT-PCR on total RNA from these ecotypes using primers to common PAI sequences in the first and fourth exons that flank the polymorphic SacI site and then cleaved the PCR product with SacI. This analysis showed that in WS, Kas-1, C24, and Ita-0, none of the PAI RT-PCR product cleaved with SacI (Figure 5B), indicating that PAI2 and PAI3 are not expressed at significant levels in these ecotypes (estimated to be <10% of total transcripts) and that the bulk of expression comes from PAI1 and/or PAI4. As a control for SacI cleavage, we performed the analogous RT-PCR reaction on RNA from the Col and Cvi-0 ecotypes, where all the PAI genes carry the SacI site. We found that in these ecotypes, the PCR product was completely cleaved.

In the ecotypes with the PAI SacI polymorphism, we determined that only PAI1 was being significantly expressed, with additional RT-PCR experiments to distinguish between PAI1 and PAI4 transcripts. For WS, PAI1 and PAI4 can be distinguished by the 9-bp deletion in the PAI4 fifth exon. RT-PCR of this region with flanking primers to common PAI sequences showed that there is no detectable PAI4-sized transcript (Figure 6A). Therefore, this experiment confirms the result of the WS cDNA abundance analysis, that PAI1 is the abundant transcript. The primer set used in this analysis also flanks the PAI3 25-bp intron insertion region between the fourth and fifth exons, but no PAI3-sized transcript was detected in either WS or Col. Because cDNA analysis shows that PAI3 can be transcribed in Col (Table 2), our failure to detect this species via RT-PCR of total Col RNA suggests that our assay conditions were not sensitive enough to detect this weakly expressed gene. In particular, the PAI3 transcript might have been discriminated against during PCR amplification because it is longer than the PAI1 or PAI4 species. Nonetheless, this RT-PCR experiment suggests that PAI3 is not a major transcript in either WS or Col.

View larger version (61K): In this window In a new window Download PPT slide   Figure 6. PAI1 is the predominant transcript in PAI-methylated ecotypes. (A) Total WS RNA was reverse transcribed and amplified by PCR with fourth- and fifth-exon primers to conserved sequences that flank a PAI4 fifth-exon 9-bp deletion. Control fragments for PAI1 (P1), PAI3 (P3), and WS PAI4 (P4WS) were amplified from cloned cDNAs. (B) The same 548-bp Kas-1 RT-PCR product used for SacI analysis (Figure 5B) was cleaved with PstI, which will generate 341- and 207-bp fragments from PAI RT-PCR products containing this site. (C) Total C24 RNA was reverse transcribed and amplified by PCR with third- and fourth-exon primers to conserved sequences that flank a PAI4 third-exon 6-bp deletion (138 bp in PAI1 vs. 132 bp in PAI4). Control fragments for PAI1 (P1) and C24 PAI4 (P4C24) were amplified from cloned cDNAs. (D) The same Ita-0 RT-PCR product used for SacI analysis was cloned into pBluescript II KS+. A representative Ita-0 subclone is shown, together with an analogous clone of Cvi-0 PAI1, cleaved with DdeI. In this assay, if the polymorphic DdeI site is missing (diagnostic of Ita-0 PAI1), then a 1309-bp band will be generated. If the polymorphic DdeI site is present (diagnostic of all other PAI genes), however, then a 1168-bp band will be generated. In all panels, DNA was visualized by ethidium bromide staining.

In the ecotype Kas-1, PAI1 can be distinguished from PAI4 by a polymorphism that destroys a conserved PstI site in the third exon of PAI1. RT-PCR analysis of Kas-1 RNA with the same primer set used for the SacI analysis, followed by cleavage of the PCR product with PstI, showed that none of the Kas-1 product was cleaved (Figure 6B). As a control, the same product amplified from Col RNA, where all three PAI genes contain the PstI site, was completely cleaved. Therefore, like WS, Kas-1 expresses transcripts primarily from PAI1. Furthermore, the Kas-1 PAI4 gene has a deletion of 7 bp in the second exon, which is predicted to alter the reading frame so that the protein is terminated near the end of the putative chloroplast transit sequence (LI et al. 1995 ). Thus, regardless of expression, no functional PAI enzyme would be produced from the Kas-1 PAI4 gene.

In the ecotype C24, PAI1 can be distinguished from PAI4 by a 6-bp deletion in the PAI4 third exon. RT-PCR of this region with flanking primers to common PAI sequences showed that there is no detectable PAI4-sized transcript (Figure 6C). Furthermore, the C24 PAI4 gene carries a 91-bp deletion extending from the middle of the fourth intron into the middle of the fifth exon, which is predicted to disrupt the correct splicing of the fourth-intron and fifth-exon coding sequences. Thus, regardless of expression, no functional PAI enzyme would be produced from the C24 PAI4 gene.

In the ecotype Ita-0, PAI1 can be distinguished from PAI4 by a polymorphism that destroys a conserved DdeI site just upstream of the translational start in the first exon. This polymorphic site is contained in the RT-PCR product from the SacI analysis shown in Figure 5B. However, it was difficult to visualize the polymorphism by direct cleavage of the RT-PCR fragment with DdeI followed by gel electrophoresis because the relevant cleavage products were obscured by background nonspecific PCR species. We therefore used the alternative strategy of subcloning the Ita-0 SacI analysis RT-PCR products into a pBluescript II KS+ plasmid vector and analyzing individual clones by DdeI digest (Figure 6D). An analogous subclone of a Cvi-0 PAI1 RT-PCR product was used as a control for the restriction pattern given by a PAI gene carrying the upstream DdeI site. All of 24 independent Ita-0 subclones tested in this way lacked the upstream DdeI site, indicating that in Ita-0, PAI1 is the only abundant transcript species.

To determine which PAI transcripts are expressed in Cvi-0, we cloned the RT-PCR fragments generated in the SacI analysis (Figure 5B) and sequenced 10 independent clones, using single-base polymorphisms unique to each PAI gene as a means of identification. All 10 Cvi-0 clones were PAI1. Therefore, in all PAI-methylated ecotypes, PAI1 is the only significantly expressed PAI gene.

	DISCUSSION

TOP ABSTRACT MATERIALS AND METHODS RESULTS DISCUSSION LITERATURE CITED

Our studies of cytosine methylation, structural polymorphisms, and gene expression indicate that an inverted repeat in the Arabidopsis genome displays a number of unusual behaviors. The PAI1-PAI4 inverted-repeat structures analyzed here consist of ~2 kb of nearly perfect mirror image sequence separated by not more than 247 bp and not less than 90 bp of nonpalindromic sequence (Figure 4). The inverted repeats are densely cytosine methylated over their regions of mirror image identity, without a significant spread into neighboring sequences (Figure 2 and Figure 3). Moreover, the inverted repeats are associated with dense methylation of the unlinked identical sequences PAI2 and PAI3 (Figure 2). These observations suggest that inverted repeats provide uniquely favorable substrates for methylation. The wide variation in PAI inverted-repeat structures observed across ecotypes (Figure 3 and Figure 4) suggests that these structures are unusually unstable, perhaps because they are difficult to replicate accurately and/or because they are recombinationally very active. Finally, in contrast to the singlet PAI2 gene, the inverted-repeat PAI1 gene is not silenced by cytosine methylation, suggesting that it is relatively exposed to the transcription machinery.

Inverted repeats trigger cytosine methylation:
Our original observation that the PAI genes are methylated in the WS ecotype but not in the Col ecotype could be explained by several models, including the difference in PAI gene arrangement and copy number between the two strains or strain-to-strain variation in the efficiency of the methylation machinery. The observations reported here argue that PAI gene arrangements, rather than variations in other loci, determine PAI cytosine methylation. Specifically, none of the 34 ecotypes with two or three unlinked singlet PAI genes displays PAI methylation, whereas all 7 ecotypes with a PAI1-PAI4 inverted repeat at the PAI1 locus display dense PAI methylation (Table 1). Consistent with the model that the inverted repeat locus triggers cytosine methylation, we found that when the WS inverted repeat is combined with the unmethylated Col PAI genes in WS x Col hybrid plants, the Col PAI genes become methylated de novo within a few generations of inbreeding (LUFF et al. 1999 ).

The inverted repeat could promote cytosine methylation via DNA/DNA interactions, RNA/DNA interactions, or both. Evidence suggesting that PAI methylation is triggered by DNA/DNA interactions comes from two observations. First, methylation is coextensive with PAI DNA sequence identity, including intron and promoter sequences (Figure 3; BENDER and FINK 1995 ; LUFF et al. 1999 ). Second, a promoterless pai1-pai4 transgene, when inserted in single copy in either the Col or the WS genome, becomes self-methylated just over its regions of mirror-image sequence within a few generations of inbreeding (LUFF et al. 1999 ). We have proposed that the inverted repeat is uniquely prone to methylation because of its ability to form unusual structures, such as a four-stranded DNA "hairpin," that serve as substrates for de novo methylation (BENDER 1998 ).

An alternative model is that the inverted repeat gives rise to unusual hairpin RNA molecules as a result of readthrough transcription, and that these molecules promote PAI cytosine methylation and silencing, perhaps because they are converted into double-stranded RNA molecules (MONTGOMERY and FIRE 1998 ; WATERHOUSE et al. 1998 ; METTE et al. 1999 ). Indeed, high-molecular-weight PAI transcripts that become proportionally longer with the length of the inverted-repeat region are observed on Northern blots (Figure 5A), consistent with the production of readthrough species that could lead to RNA-directed DNA methylation. Ultimately, the isolation of mutations that alter PAI methylation and silencing should identify the important mechanistic components.

Genesis of the PAI gene family:
The sequence divergence between PAI3 and its sister PAI genes suggests that the two classes of genes are relatively evolutionarily distant from each other. However, the close sequence identity among PAI1, PAI2, and PAI4 argues that they arose more recently from a common progenitor. Transposon-mediated rearrangements provide a mechanism for horizontal transfer of sequences within a genome and for generation of tandem-sequence duplications. For example, transposon-generated, inverted-repeat sequence duplications have been characterized previously at the nivea locus in Antirrhinum majus (BOLLMANN et al. 1991 ) and at the amylose extender1 locus in maize (STINARD et al. 1993 ). It is attractive to speculate that the transmission of a common progenitor PAI sequence into the PAI2 and PAI1-PAI4 genes in Arabidopsis was similarly transposon generated. However, sequence analysis of the regions around these genes in several ecotypes has failed to reveal any obvious transposon-like sequence. Therefore, if the PAI1-PAI4 and/or the PAI2 duplicate genes were transposon-generated, the transposon sequences that mediated the process most likely excised after the duplication event. Because little is known about families of transposons in Arabidopsis (BHATT et al. 1998 ), we cannot determine whether short sequences in the PAI1-PAI4 or PAI2 regions are characteristic of transposon excision "footprints" or of transposon deletion derivatives.

Regardless of the mechanism(s) of PAI sequence duplication, it is most likely that PAI2 on chromosome 5 was the progenitor gene that was copied to produce PAI1 and/or PAI1-PAI4 on chromosome 1. In particular, the 3' sequences that are duplicated in PAI1, PAI2, and PAI4 (but not PAI3) include the 3' end of the FAD8 gene (GIBSON et al. 1994 ) that lies just downstream of PAI2 on chromosome 5 (Figure 3). The simplest explanation of this partial FAD8 duplication is that a region extending from the PAI2 promoter through the 3' end of FAD8 was the basic sequence unit that was transmitted to the PAI1 locus on chromosome 1. In this case, the flanking direct-repeat sequence duplication might have happened concurrently with the rearrangement event that transmitted PAI sequences to the PAI1 locus.

Our analysis of structural variants of the PAI1 locus across six ecotypes of Arabidopsis suggests that all the variants are related and arise from a common progenitor structure. Our data support either of two models for the genesis of the PAI1 locus. One model is that the structure found in Col and 32 other ecotypes (Table 1) is the progenitor structure and that this structure underwent a duplication and rearrangement event to generate an inverted repeat of PAI genes flanked by full direct-repeat sequences in one unusual lineage. This initial inverted-repeat structure then underwent further deletions and rearrangements to generate the variety of inverted-repeat structures observed in PAI-methylated ecotypes today. Presumably, the high degree of variation from the progenitor inverted-repeat structure was due to both the instability of the inverted repeat (HENDERSON and PETES 1993 ; RUSKIN and FINK 1993 ) and the ability of the duplicated sequences in the structure to undergo intramolecular recombination or unequal crossing over. Molecular evolutionary studies of Arabidopsis ecotypes (HANFSTINGL et al. 1994 ; INNAN et al. 1997 ) do not indicate clustering among the PAI-methylated strains, as might be predicted by this model. However, a detailed study of sequences in the PAI1 region of the genome across many ecotypes is needed before this point can be determined definitively. It is also possible, but unlikely, that a progenitor Col-like structure rearranged five independent times to give the various inverted-repeat structures found in Cvi-0, Ita-0, Kas-1, C24, and WS.

An alternative model is that the progenitor structure of all ecotypes was an inverted repeat of two PAI genes flanked by full-length direct repeats. This common structure could have given rise to the Col structure by deletion of one PAI gene and part of a flanking direct repeat, and to the inverted-repeat ecotype structures by other rearrangement events (Figure 3). In this scenario, the predominance of the Col structure over methylated inverted-repeat structures in the wild population (Table 1) might reflect a greater fitness of the Col structure. Because Col expresses PAI enzyme from two unlinked genes, PAI1 and PAI2 (Table 2), this redundancy protects Col from deleterious consequences of PAI gene mutations. In contrast, the PAI-methylated ecotypes express PAI enzyme only from the PAI1 gene (Table 2, Figure 5 and Figure 6), making them vulnerable to tryptophan auxotrophy via PAI gene mutation. This vulnerability might therefore account for the underrepresentation of PAI-methylated ecotypes in the wild population.

Although Col and the inverted-repeat ecotypes all have the potential to yield a deletion at the PAI1 locus because of homologous recombination between the flanking direct-repeat sequences (BENDER and FINK 1995 ), this type of rearrangement was observed only in one ecotype, Ll-0 (Table 1). The lack of PAI1-deleted strains in the wild population suggests that the homologous recombination event that generates this structure might be rare. However, given that five independent isolates of a PAI1-PAI4-deleted version of WS were isolated from a T-DNA-transformed population (BENDER and FINK 1995 ), and that similar deletions occur at a high frequency in EMS-mutagenized WS populations (J. BENDER, unpublished results), homologous recombination between two full-length direct repeats can happen at a relatively high frequency, at least in plants that have undergone mutagenic treatments. Alternatively, PAI1-deleted strains might be selected against in a wild population because they are dependent on a single gene, PAI2, for PAI activity; like the PAI-methylated ecotypes, PAI1-deleted variants might be underrepresented in the wild because of their vulnerability to PAI gene mutation. Furthermore, the reduced PAI expression in PAI1-deleted strains, although sufficient for normal plant morphology under laboratory conditions (BENDER and FINK 1995 ), might not be sufficient for optimal survival in the wild.

In the PAI-methylated ecotypes Cvi-0, Ita-0, C24, Kas-1, and WS, the major structural difference is the sequence between the inverted-repeat PAI genes (Figure 3 and Figure 4). Cvi-0 and Ita-0 are the only ecotypes that carry extra sequences in this central region relative to the sequences downstream of Col PAI1 and PAI2 (Figure 4). Other structural differences unique to particular ecotypes can best be explained as secondary events (Figure 3). For example, the WS pai4 5' duplication could have been generated by pairing between the direct-repeat sequences, followed by gene conversion of the sequences adjacent to the PAI1-proximal repeat to sequences adjacent to the PAI4-proximal repeat. Similarly, the C24 PAI4 promoter rearrangement could have been generated by pairing between the direct-repeat sequences, followed by gene conversion of the sequences adjacent to the PAI4-proximal repeat to sequences adjacent to the PAI1-proximal repeat. The Ita-0 PAI1 promoter rearrangement could have been generated by pairing between the two inverted-repeat PAI genes, with gene conversion of the PAI1 promoter sequences to the PAI4 promoter sequences. The Cvi-0 PAI4* duplication is most simply explained by unequal crossing over between direct repeats to amplify a structure consisting of direct-repeat 1-PAI1*–PAI4*-direct-repeat 2-PAI1-PAI4-direct-repeat 3, followed by a deletion of most of direct-repeat 1 and PAI1*.

Cytosine methylation has been shown to suppress homologous recombination (MALOISEL and ROSSIGNOL 1998 ). Nonetheless, at least some of the structural polymorphisms in the methylated PAI1-PAI4 regions studied here are likely to have arisen via recombination mechanisms. Perhaps even with the negative effects of cytosine methylation, the PAI inverted repeats can pair with each other so readily that some recombination still occurs between them. Alternatively, the inverted repeat might be immune to accumulating the factors that would normally block homologous recombination on methylated sequences because of unique structural or sequence characteristics.

Cytosine methylation and PAI gene expression:
Cytosine methylation is usually correlated with a loss of transcription from methylated sequences (KASS et al. 1997 ). Consistent with this general rule, the singlet PAI2 gene in WS is silenced by cytosine methylation. Specifically, in a derivative of WS lacking the PAI1-PAI4 inverted-repeat genes, PAI2 expression is inversely correlated with the density of methylation on the PAI2 gene (BENDER and FINK 1995 ; JEDDELOH et al. 1998 ). Genetic backcross experiments show that PAI2 is also silenced in parental WS (BENDER and FINK 1995 ). However, parental WS is phenotypically normal and expresses levels of PAI transcripts and PAI enzyme activity comparable to those measured in Col. Therefore, PAI1 and/or PAI4 must account for the bulk of PAI activity in WS despite being densely methylated. In fact, cDNA abundance (Table 2) and RT-PCR analyses (Figure 5 and Figure 6) indicate that PAI1 is the only detectable PAI transcript in WS. Furthermore, PAI1 is the only detectable transcript in the other PAI-methylated ecotypes (Figure 5 and Figure 6).

Why is the methylated PAI1 gene expressed while the methylated PAI2 gene is silenced? Detailed analysis of cytosine methylation patterns for WS PAI1 and PAI2 have revealed no significant differences in the extent or density of methylation (LUFF et al. 1999 ). Therefore, the different effects of methylation on PAI gene expression might be caused by differences in the genomic contexts (cis-acting sequences, DNA structure, and/or chromosomal domains) that determine different chromatin assembly and expression states. Interestingly, the PAI4 gene immediately adjacent to PAI1 is not expressed in PAI-methylated ecotypes (Figure 6) despite ~250 bp (or in the case of Ita-0, ~550 bp) of promoter identity. The lack of PAI4 expression could result from the absence of critical upstream promoter sequences and/or from silencing by methylation. The later case would suggest that the transcriptionally active state at the PAI1 promoter does not propagate as far as the PAI4 promoter, 4 kb away.

	ACKNOWLEDGMENTS

We thank Laura Pawlowski, Gromoslaw Smolen, and Tomoko Hamma for technical assistance. This work was supported by a Basil O'Connor Starter Scholar Award 5-FY98-0535 from the March of Dimes, a National Institute of Environmental Health Sciences Training Grant to S.M. (ES 07141), and a Searle Scholars Award 97-E-103 to J.B.

Manuscript received April 1, 1999; Accepted for publication May 10, 1999.

	LITERATURE CITED

TOP ABSTRACT MATERIALS AND METHODS RESULTS DISCUSSION LITERATURE CITED

AUSUBEL, F. M., R. BRENT, R. E. KINGSTON, D. D. MOORE, J. G. SEIDMAN et al., 1989 Current Protocols in Molecular Biology. Greene Publishing Associates and Wiley-Interscience, New York.

BENDER, J., 1998  Cytosine methylation of repeated sequences in eukaryotes: the role of DNA pairing. Trends Biochem. Sci. 23:252-256[Medline].

BENDER, J. and G. R. FINK, 1995  Epigenetic control of an endogenous gene family is revealed by a novel blue fluorescent mutant of Arabidopsis. Cell 83:725-734[Medline].

BHATT, A. M., C. LISTER, N. CRAWFORD, and C. DEAN, 1998  The transposition frequency of Tag1 elements is increased in transgenic Arabidopsis lines. Plant Cell 10:427-434[Abstract/Free Full Text].

BOLLMANN, J., R. CARPENTER, and E. S. COEN, 1991  Allelic interactions at the nivea locus of Antirrhinum.. Plant Cell 3:1327-1336[Abstract/Free Full Text].

BONHAM-SMITH, P. C. and M. M. MOLONEY, 1994  Nucleotide and protein sequences of a cytoplasmic ribosomal protein S15a gene from Arabidopsis thaliana.. Plant Physiol. 106:401-402[Medline].

CHURCH, G. M. and W. GILBERT, 1984  Genomic sequencing. Proc. Natl. Acad. Sci. USA 81:1991-1995[Abstract/Free Full Text].

ELLEDGE, S. J., J. T. MULLIGAN, S. W. RAMER, M. SPOTTSWOOD, and R. D. DAVIS, 1991  YES: a multifunctional cDNA expression vector for the isolation of genes by complementation of yeast and Escherichia coli mutations. Proc. Natl. Acad. Sci. USA 88:1731-1735[Abstract/Free Full Text].

GIBSON, S., V. ARONDEL, K. IBA, and C. SOMERVILLE, 1994  Cloning of a temperature-regulated gene encoding a chloroplast omega-3 desaturase from Arabidopsis thaliana.. Plant Physiol. 106:1615-1621[Abstract].

HANFSTINGL, U., A. BERRY, E. A. KELLOGG, J. T. COSTA, III, and W. RUDIGER et al., 1994  Haplotypic divergence coupled with lack of diversity at the Arabidopsis thaliana alcohol dehydrogenase locus: roles for both balancing and directional selection? Genetics 138:811-828[Abstract].

HENDERSON, S. T. and T. D. PETES, 1993  Instability of a plasmid-borne inverted repeat in Saccharomyces cerevisiae.. Genetics 134:57-62[Abstract].

INNAN, H., R. TERAUCHI, and N. T. MIYASHITA, 1997  Microsatellite polymorphism in natural populations of the wild plant Arabidopsis thaliana.. Genetics 146:1441-1452[Abstract].

JEDDELOH, J. A., J. BENDER, and E. J. RICHARDS, 1998  The DNA methylation locus DDM1 is required for maintenance of gene silencing in Arabidopsis.. Genes Dev. 12:1714-1725[Abstract/Free Full Text].

KASS, S. U., D. PRUSS, and A. P. WOLFFE, 1997  How does DNA methylation repress transcription? Trends Genet. 13:444-449[Medline].

KUNKEL, T. A., J. D. ROBERTS, and R. A. ZAKOUR, 1987  Rapid and efficient site-specific mutagenesis without phenotypic selection. Methods Enzymol. 154:367-382[Medline].

LAST, R. L., P. H. BISSINGER, D. J. MAHONEY, E. R. RADWANSKI, and G. R. FINK, 1991  Tryptophan mutants in Arabidopsis: the consequences of duplicated tryptophan synthase ß genes. Plant Cell 3:345-358[Abstract/Free Full Text].

LI, J., J. ZHAO, A. B. ROSE, R. SCHMIDT, and R. L. LAST, 1995  Arabidopsis phosphoribosylanthranilate isomerase: molecular genetic analysis of triplicate tryptophan pathway genes. Plant Cell 7:447-461[Abstract].

LUFF, B., L. PAWLOWSKI, and J. BENDER, 1999  An inverted repeat triggers cytosine methylation of identical sequences in Arabidopsis.. Mol. Cell 3:505-511[Medline].

MALOISEL, L. and J.-L. ROSSIGNOL, 1998  Suppression of crossing-over by DNA methylation in Ascobolus.. Genes Dev. 12:1381-1389[Abstract/Free Full Text].

METTE, M. F., J. VAN DER WINDEN, M. A. MATZKE, and A. J. M. MATZKE, 1999  Production of aberrant promoter transcripts contributes to methylation and silencing of unlinked homologous promoters in trans.. EMBO J. 18:241-248[Medline].

MINET, M., M.-E. DUFOUR, and F. LACROUTE, 1992  Complementation of Saccharomyces cerevisiae auxotrophic mutants by Arabidopsis thaliana cDNAs. Plant J. 2:417-422[Medline].

MONTGOMERY, M. K. and A. FIRE, 1998  Double-stranded RNA as a mediator in sequence-specific genetic silencing and co-suppression. Trends Genet. 14:255-258[Medline].

NAGY, F., S. A. KAY and N.-H. CHUA, 1988 Analysis of gene expression in transgenic plants, pp. B4/11–B4/12 in Plant Molecular Biology Manual, edited by S. B. GELVIN and R. A. SCHILPEROORT. Kluwer Academic Publishers, Dordrecht, The Netherlands.

NIYOGI, K. K. and G. R. FINK, 1992  Two anthranilate synthase genes in Arabidopsis: defense-related regulation of the tryptophan pathway. Plant Cell 4:721-733[Abstract/Free Full Text].

NIYOGI, K. K., R. L. LAST, G. R. FINK, and B. KEITH, 1993  Suppressors of trp1 fluorescence identify a new Arabidopsis gene, TRP4, encoding the anthranilate synthase ß subunit. Plant Cell 5:1011-1027[Abstract/Free Full Text].

RUSKIN, B. and G. R. FINK, 1993  Mutations in POL1 increase the mitotic instability of tandem inverted repeats in Saccharomyces cerevisiae.. Genetics 134:43-56[Abstract].

SATO, S., H. KOTANI, Y. NAKAMURA, T. KANEKO, and E. ASAMIZU et al., 1997  Structural analysis of Arabidopsis thaliana chromosome 5. I. Sequence features of the 1.6 Mb regions covered by twenty physically assigned P1 clones. DNA Res. 4:215-230[Abstract].

STINARD, P. S., D. S. ROBERTSON, and P. S. SCHNABLE, 1993  Genetic isolation, cloning, and analysis of a Mutator-induced, dominant antimorph of the maize amylose extender1 locus. Plant Cell 5:1555-1566[Abstract].

WATERHOUSE, P. M., M. W. GRAHAM, and M.-B. WANG, 1998  Virus resistance and gene silencing in plants can be induced by simultaneous expression of sense and antisense RNA. Proc. Natl. Acad. Sci. USA 95:13959-13964[Abstract/Free</