A gene loop containing the floral repressor FLC is disrupted in the early phase of vernalization

Abstract

Gene activation in eukaryotes frequently involves interactions between chromosomal regions. We have investigated whether higher-order chromatin structures are involved in the regulation of the Arabidopsis floral repressor gene FLC, a target of several chromatin regulatory pathways. Here, we identify a gene loop involving the physical interaction of the 5′ and 3′ flanking regions of the FLC locus using chromosome conformation capture. The FLC loop is unaffected by mutations disrupting conserved chromatin regulatory pathways leading to very different expression states. However, the loop is disrupted during vernalization, the cold-induced, Polycomb-dependent epigenetic silencing of FLC. Loop disruption parallels timing of the cold-induced FLC transcriptional shut-down and upregulation of FLC antisense transcripts, but does not need a cold-induced PHD protein required for the epigenetic silencing. We suggest that gene loop disruption is an early step in the switch from an expressed to a Polycomb-silenced state.

Introduction

The complexity of gene regulation in the chromatin environment of eukaryotic cells is enormous and frequently involves chromosomal interactions between regulatory sequences and their target genes (Li et al, 2007; Palstra, 2009). Chromosomal loops that bring distant regulatory elements to gene promoters (Tolhuis et al, 2002; Louwers et al, 2009a; Kagey et al, 2010) and co-localization of silenced genes in Polycomb bodies (Lanzuolo et al, 2007; Bantignies et al, 2011) are examples of functionally important long-range interactions. Short-range chromatin interactions also occur and are exemplified by gene loops where the promoter and terminator regions of individual genes are juxtaposed when transcribed (O'Sullivan et al, 2004; Ansari and Hampsey, 2005). Gene loops were first described in yeast but have now also been observed in higher eukaryotes (O’Reilly and Greaves, 2007; Perkins et al, 2008; Tan-Wong et al, 2008; Larkin et al, 2012). The formation of these gene loops relies on initial rounds of transcription and requires interaction of TFIIB, a general RNA polymerase II transcription factor, with 3′ end processing factors (Singh and Hampsey, 2007; Perkins et al, 2008; Medler et al, 2011). In yeast, the promoter–terminator loop conformation is proposed to promote recycling of RNA polymerase from the 3′ terminator to the promoter (Ansari and Hampsey, 2005; Lykke-Andersen et al, 2011). Gene looping has also been proposed to be involved in intron-mediated enhancement of transcription (Moabbi et al, 2012) and plays a role in maintaining ‘transcriptional memory’ in yeast (Laine et al, 2009; Tan-Wong et al, 2009). Recently, a comprehensive study using yeast and mammalian cells has proposed that transcriptional directionality is enhanced by loop formation (Tan-Wong et al, 2012). However, gene loop formation is not always associated with enhanced gene activity in higher eukaryotic cells (Tan-Wong et al, 2008; Larkin et al, 2012), suggesting that gene looping may have different functions in different biological scenarios.

We were interested to see if gene looping may play a role in the complex transcriptional regulation of the Arabidopsis thaliana floral repressor gene FLOWERING LOCUS C (FLC). This gene is a target of several regulatory pathways involving chromatin remodelling, co-transcriptional RNA processing and Polycomb silencing (Baurle and Dean, 2006; Crevillen and Dean, 2010; Ietswaart et al, 2012; Song et al, 2012). These pathways significantly influence the timing of the transition to reproductive development in Arabidopsis, a trait important for adaptation to different climates. In ambient temperatures, FLC expression is upregulated by FRIGIDA (FRI), a plant-specific protein that elevates FLC mRNA levels by a co-transcriptional mechanism involving direct physical interaction with the nuclear cap-binding complex (Geraldo et al, 2009; Crevillen and Dean, 2010). FRI function requires conserved transcriptional components such as the Paf complex and chromatin remodelling activities (Crevillen and Dean, 2010; Choi et al, 2011). High FLC expression also requires the Arabidopsis Trithorax ATX1 and ATX2 histone methyltransferases (Pien et al, 2008), and the SWR1 complex subunit ARP6 (Choi et al, 2005; Deal et al, 2005; Martin-Trillo et al, 2006). The autonomous pathway antagonizes these activation functions (Baurle and Dean, 2006) in a mechanism involving the RNA-binding proteins FCA and FPA, the 3′ processing factor FY and the histone H3 lysine 4 demethylase FLD (Liu et al, 2007). The RNA-binding proteins and 3′ processing factors alter processing of FLC antisense transcripts (Hornyik et al, 2010; Liu et al, 2010; Ietswaart et al, 2012), resulting in changed histone methylation levels and reduced expression of FLC (Liu et al, 2007).

The upregulation of FLC levels by FRI or mutations in the autonomous pathway is suppressed by vernalization, the acceleration of flowering in response to winter. This process is an adaptation used by many plants to align flowering with favourable conditions of spring. In Arabidopsis, vernalization involves a cold-induced Polycomb-mediated epigenetic silencing of FLC (recently reviewed by Song et al, 2012). After 2–3 weeks of cold, FLC transcription is downregulated concomitantly with increased accumulation of antisense transcripts, named COOLAIR (Swiezewski et al, 2009). Cold also induces the quantitative accumulation of Polycomb silencing. This involves activation of the PHD protein VIN3 (Sung and Amasino, 2004), heterodimerization with a homologue VRN5 (Sung et al, 2006; Greb et al, 2007) and association with a pre-loaded polycomb repressive complex 2 (PRC2) at an internal site in FLC (De Lucia et al, 2008). The PHD–PRC2 complex causes localized and progressive increases in histone H3 lysine 27 trimethylation (H3K27me3) during the cold (Finnegan and Dennis, 2007; Angel et al, 2011). Upon transfer to the warm, the PHD–PRC2 complex spreads across the gene resulting in high H3K27me3 across the locus, necessary for epigenetic stability through the rest of development (Finnegan and Dennis, 2007; De Lucia et al, 2008; Angel et al, 2011). A sense non-coding RNA (ncRNA) FLC transcript, COLDAIR, is induced by cold but later than COOLAIR and is involved in recruitment of the PHD–PRC2 complex (Heo and Sung, 2010).

To address whether higher-order chromatin structure is involved in any of these regulatory pathways, we used quantitative chromosome conformation capture (3C) (Dekker et al, 2002) in a range of Arabidopsis genotypes. We detected a robust gene loop at the FLC locus reflecting interaction between sequences in the FLC 5′ flanking region with sequences in the 3′ flanking region. Analysis of loop formation in a range of mutants suggested that loop formation was not dependent on FLC expression level. However, it was efficiently disrupted within the first 2 weeks of cold exposure during vernalization and did not reform after transfer of plants back to warm conditions. Disruption did not require the cold-induced PHD protein VIN3, important for nucleation of the epigenetic silencing of FLC. Similar results were observed with an FLC transgene, indicating that gene loop formation and disruption are independent of the genomic context of the locus. Thus, as in many other higher eukaryotes, gene loop formation occurs in plant genes perhaps aiding forms of transcriptional regulation. We propose that for FLC and maybe other Polycomb regulated genes, loop disruption is an early step during the switch to an epigenetically silent state.

Results

The 5′ and 3′ flanking regions of FLC interact creating a gene loop

We performed 3C experiments to explore whether a higher-order chromatin structure may contribute to the complexity of FLC regulation. This technique relies on formaldehyde crosslinking to detect interacting chromatin fragments in intact cells (Dekker et al, 2002; Hagege et al, 2007; Louwers et al, 2009b). We divided the FLC locus into different fragments using double digest with BglII and BamHI restriction enzymes, schematically shown in Figure 1, and searched for chromosome interactions. Tandem primers were designed to ensure specific amplification of 3C ligated products only and chromatin interactions were estimated by calculating the ligation frequencies in the 3C DNA preparations by real-time quantitative PCR (Q-PCR) (see Materials and methods and Hagege et al, 2007). Fragment I, hereinafter referred to as FI, which includes the FLC promoter, first exon and key regulatory elements in intron 1 (Finnegan and Dennis, 2007; De Lucia et al, 2008; Angel et al, 2011), was used as an anchor region. Using young Columbia (Col) wild-type (WT) seedlings high interaction frequencies were found between the anchor region and FV (Figure 2A), a region ∼355–970 bp downstream of the FLC poly (A) sites (see Figure 1). The interaction frequencies observed between FI and FV were higher than expected by random collisions, indicating that the 5′ and 3′ flanking regions of the FLC locus are frequently in physical contact in vivo (Dekker et al, 2002; Louwers et al, 2009b). We also found significant interaction frequencies between FI and FIV (Figure 2A), a fragment starting downstream of the FLC poly (A) sites (see Figure 1). This loop seemed similar to promoter–terminator loops described in yeast, but in the case of FLC the gene loop extends beyond the poly (A) site. FIV is a region we previously identified with homology to small RNAs (Swiezewski et al, 2007) and contains the major transcriptional start site of the COOLAIR antisense transcripts. FV region is further downstream of FLC and probably contains promoter elements necessary for COOLAIR antisense RNA transcription (Swiezewski et al, 2009), although these have still to be defined. The FLC loop was also detected when the 3′ regions FIV and FV were used as anchor fragments in the 3C experiments (see Figure 5). There was no PCR amplification when plant material was not crosslinked or ligase was not included in the 3C reactions, indicating that the observed chromosomal interactions are not PCR artefacts. The discovery of a chromosomal loop containing the FLC gene raises the question of whether the loop plays a role in the different regulatory pathways that target the FLC locus.

Figure 1.

Schematic representation of the FLC locus and the 5′ region of the downstream gene (At5g10130) showing the regions analysed in this study. BamHI and BglII restriction sites are indicated with vertical dotted lines. Primer positions are represented with open arrows. Cold-induced ncRNA transcripts COLDAIR (Heo and Sung, 2010) and COOLAIR (Swiezewski et al, 2009) are represented in grey. COOLAIR has alternatively processed transcripts but only the two main spliced forms present in warm-grown tissue are represented. The H3K27me3 nucleation region is also represented (Angel et al, 2011).

Figure 2.

FLC loop formation in Columbia seedlings and genotypes with high FLC expression. (A) Quantitative 3C of the FLC locus using FI as the anchor region in 10-day-old WT Col, Col-FRI and fcafpa seedlings. Relative interaction frequencies were calculated as described in Materials and methods. The data are the average of four biological replicates each with two technical replicates. The error bars indicate s.e.m. (n=8). In the graph, BamHI and BglII restriction sites are indicated with vertical dotted lines, the analysed FLC regions are numbered with Roman numerals and the anchor region is highlighted with a grey-shadowed area. A schematic representation of the FLC locus is shown above. (B) RNA expression analysis of 10-day non-vernalized seedlings from different genotypes. The graph represents the ratio FLC/UBC normalized to Col WT. The error bars indicate s.e.m. (n=3).

High FLC expression levels do not enhance loop formation

The FRI complex is associated with increased FLC transcriptional activation and 5′ mRNA capping (Geraldo et al, 2009; Crevillen and Dean, 2010; Choi et al, 2011). Common laboratory accessions like Columbia (Col) carry naturally mutated versions of FRI, so we tested whether the FLC gene loop was influenced by the FRI-dependent increase in expression using Col-FRI seedlings—Col plants into which an active FRI Sant Feliu allele had been introgressed (Michaels and Amasino, 1999). Col-FRI has 30- and 25-fold increase in FLC RNA unspliced and mRNA levels, respectively (Figure 2B) but no major difference in chromosome interaction was found compared with Col (Figure 2A).

In yeast, gene loop formation requires an interaction between general transcription factors and 3′ RNA processing factors (Ansari and Hampsey, 2005; Medler et al, 2011; Tan-Wong et al, 2012). However, mutations in Arabidopsis polyadenylation factors FCA and FPA did not disrupt the FI–FV interaction (Figure 2A) despite increasing FLC transcript levels 20-fold (Figure 2B). As well as influencing polyadenylation of the FLC antisense transcript FCA and FPA promote the use of proximal poly (A) sites in many other transcripts in the Arabidopsis genome (Sonmez et al, 2011). We found that loss of FCA and FPA leads to a small but significant reduction in use of the FLC major sense poly A sites (Feng et al, 2011; Supplementary Figure 1). A reduced interaction frequency between FI and FIV was observed in fcafpa compared with Col and Col-FRI (Figure 2A), suggesting the increased transcriptional read-through influenced the FI–FIV interaction.

FLC loop is not disrupted in mutants with low FLC expression levels

We asked if mutations in core chromatin remodelling components that impair FLC expression (Figure 2B) would perturb FLC loop formation. The deposition of histone H3 lysine 4 methylation at FLC promoter by the Arabidopsis Trithorax proteins ATX1 and ATX2 is required for high FLC expression (Pien et al, 2008). 3C analysis on seedlings of a Col-FRI atx1-2 atx2-1 (atx1atx2) double mutant (Pien et al, 2008) showed that the FLC loop was not reduced compared with Col-FRI seedlings (Figure 3A). Mutations in Arabidopsis ARP6 greatly reduce FLC transcript levels by affecting H2A.Z variant deposition at promoter and terminator regions of the gene (Choi et al, 2005; Deal et al, 2005; Martin-Trillo et al, 2006; Deal et al, 2007). We found that FLC RNA levels in arp6-1 were reduced more than 5- and 100-fold compared with Col and Col-FRI, respectively (Figure 2B). Nevertheless, the interaction between the FLC 5′ and 3′ flanking regions was only moderately reduced in arp6-1 seedlings (Figure 3B).

Figure 3.

FLC loop formation in mutants with low FLC expression. Quantitative 3C of the FLC locus fragment FI as the anchor region. Relative interaction frequencies were calculated as described in Materials and methods. (A) Chromatin interactions in 10-day-old Col-FRI (WT) and atx1atx2-FRI seedlings. The data are the average of two biological replicates each with two technical replicates. The error bars indicate s.e.m. (n=4). (B) Chromatin interactions in Col (WT) compared with arp6-1 mutant 10-day-old seedlings. The data are the average of two biological replicates each with two technical replicates. The bars in the graphs indicate s.e.m. (n=4). In the graphs, BamHI and BglII restriction sites are indicated with vertical dotted lines, the analysed FLC regions are numbered with Roman numerals and the anchor region is highlighted with a grey-shadowed area. A schematic representation of the FLC locus is shown above each graph.

The lack of an association of loop formation with the expression state of the FLC locus is striking when one compares the expression differences of the genotypes used (Figures 2 and 3). There is a >10-fold increase in fcafpa or FRI compared with Col in FLC mRNA and unspliced RNA levels, the latter used as a proxy for transcriptional level; and a >5-fold reduction in atx1,atx2-FRI or arp6 compared with Col-FRI or Col, respectively (Figure 2B).

FLC loop is disrupted as an early step during vernalization

We then investigated the relationship between the physical and transcriptional state of the FLC locus during vernalization, where FLC expression is downregulated by prolonged cold and then epigenetically silenced by the Polycomb machinery (Song et al, 2012). In Arabidopsis, this epigenetic silencing overrides the high FLC levels induced by FRI activity or fcafpa mutations (Baurle and Dean, 2006). 3C experiments were undertaken on Col-FRI seedlings given 2 weeks of cold and then either harvested immediately (2WT0) when FLC transcription is shut-down (Swiezewski et al, 2009) and a PHD–Polycomb complex accumulates at the nucleation site in FLC; or 7 days later after further growth in the warm (2WT7) when H3K27me3 and the PHD–Polycomb complex spreads across FLC and the locus is epigenetically silenced (Finnegan and Dennis, 2007; De Lucia et al, 2008; Angel et al, 2011; Song et al, 2012). When 3C experiments were performed on Col-FRI vernalized seedlings, we found that the FI interaction with the 5′ flanking region of FLC locus was substantially reduced in plants grown for 2 weeks in the cold (Figure 4A). Interestingly, this non-looped conformation was maintained in the absence of cold during the subsequent growth of the plant in warm when FLC expression remains epigenetically silenced (Figure 4A). FLC gene loop disruption is not an immediate response to chilling temperatures as we could not detect reduced interaction frequencies after 5 days in the cold (Supplementary Figure 2). It is also not a genome-wide response to the cold because intragenic interactions detected within the housekeeping UBC locus (At5g25760; Czechowski et al, 2005) are not disrupted during the cold (Supplementary Figure 3). Interestingly, loop disruption parallels the timing of downregulation of FLC expression in the cold (Figure 4B).

Figure 4.

FLC loop is disrupted during and after cold exposure. (A) Quantitative 3C of the FLC the locus using FI as the anchor region in Col-FRI seedlings harvested after 10 days of growth in standard conditions (NV), immediately after 2 weeks in the cold (2WT0) and after 2 weeks in the cold and 7 days of further growth in the warm (2WT7). Relative interaction frequencies were calculated as described in Materials and methods. The data are the average of three biological replicates each with two technical replicates. The error bars indicate s.e.m. (n=6). In the graph, BamHI and BglII restriction sites are indicated with vertical dotted lines, the analysed FLC regions are numbered with Roman numerals and the anchor region is highlighted with a grey-shadowed area. A schematic representation of the FLC locus is shown above. (B) RNA expression analysis of Col-FRI during vernalization. The graph represents the ratio FLC/UBC or COOLAIR/UBC normalized to Col-FRI non-vernalized. The error bars indicate s.e.m. (n=3).

Transcription of ncRNA can result in novel chromatin interactions as recently shown for the Hox genes and Igf2/H19 locus in animal cells (Court et al, 2011; Wang et al, 2011). We considered the possibility that the induction of ncRNAs at FLC during the cold could generate alternative loops. At least two ncRNA are produced from FLC: COLDAIR (Heo and Sung, 2010) and COOLAIR (Swiezewski et al, 2009). COLDAIR sense ncRNA production does not overlap with either FI or FV and is induced after longer vernalization periods (∼3 weeks) (Heo and Sung, 2010), but FV likely contains COOLAIR promoter elements, so we reasoned that during the cold it could promote an alternative loop disrupting the FI–FV loop. We tested this model by looking for the appearance of novel chromosomal interactions during the cold of FV and FIV with FIII, FII or downstream FLC promoter (F0; see Figure 1). These fragments had very high interaction frequencies with nearby regions (Figure 5), as expected, due to random collisions of such fragments (Dekker et al, 2002). FV and FIV also showed high interaction frequency with FI confirming the previously detected FLC loop (Figures 2, 3, 4). We also confirmed that the FI–FIV and FI–FV interactions were disrupted by vernalization but no additional non-random interactions were detected between the fragments analysed (Figure 5). Thus, COOLAIR induction does not appear to promote new chromatin interactions that could disrupt the FLC gene loop formed in warm conditions.

Figure 5.

Search for alternative FLC gene loops induced by vernalization. (A) Quantitative 3C analysis using FIV as the anchor region. (B) Quantitative 3C analysis using FV as the anchor region. Col-FRI seedlings were harvested after 10 days of growth in standard conditions (NV), immediately after 2 weeks in the cold (2WT0) and after 2 weeks in the cold and 7 days of further growth in the warm (2WT7). Relative interaction frequencies were calculated as described in Materials and methods. The data are the average of two biological replicates. In the graphs, BamHI and BglII restriction sites are indicated with vertical dotted lines, the analysed FLC regions are numbered with Roman numerals and the anchor region is highlighted with a grey-shadowed area. A schematic representation of the FLC locus is shown above each graph.

FLC loop disruption is independent of Polycomb repression

Cold causes downregulation of FLC transcription independently of the cold-induced accumulation of VIN3 (Swiezewski et al, 2009). The dynamics of these cold-induced processes vary in different Arabidopsis accessions but in Col-FRI seedlings 2 weeks cold is sufficient to trigger both processes (Swiezewski et al, 2009). In order to analyse which cold-induced process was involved in loop disruption, we undertook 3C experiments on the vin3 mutant. Cold-induced transcriptional downregulation occurs in this mutant but epigenetic silencing is impaired (Sung and Amasino, 2004; Greb et al, 2007). The FLC loop was similarly disrupted by 2 weeks vernalization in Col-FRI and vin3-4 FRI mutant seedlings (Figure 6A). These data indicate that loop disruption is independent of VIN3 function, suggesting that loop disruption is associated with the early transcriptional downregulation of FLC.

Figure 6.

FLC loop is disrupted independently of Polycomb complexes or the genomic context of the locus. Seedlings were harvested after 10 days of growth in normal conditions (NV) and immediately after 2 weeks in the cold (2WT0). (A) Quantitative 3C of the FLC locus using FI as the anchor region in Col-FRI (WT) and vin3-FRI (B) Quantitative 3C of the FLC::LUC transgene in flc-2 background. Relative interaction frequencies were calculated as described in Materials and methods. The data are the average of two biological replicates each with two technical replicates. The error bars indicate s.e.m. (n=4). In the graph, BamHI and BglII restriction sites are indicated with vertical dotted lines, the analysed FLC regions are numbered with Roman numerals and the anchor region is highlighted with a grey-shadowed area. A schematic representation of the FLC locus is shown above each graph.

Altering genomic location does not influence loop formation or disruption

We wondered if the genomic context of the endogenous FLC locus was important for loop formation. To address this we used an FLC::LUC transgenic line (Mylne et al, 2004; Greb et al, 2007), with the transgene inserted on chromosome 3 rather than the native position on chromosome 5, in a FRIGIDA flc-2 genetic background (Michaels and Amasino, 1999). The FLC::LUC transgene does not introduce any new restrictions site or affect the primer sets used in our study (Mylne et al, 2004), and flc-2 carries a large deletion that removes a large section of the endogenous FLC gene (Michaels and Amasino, 1999). The data in Figure 6B show that the FLC::LUC transgene forms a gene loop and that, as detected with the endogenous FLC locus, the loop is disrupted by vernalization. Thus, formation and disruption of the FLC loop does not depend on the chromosomal position.

Discussion

We have demonstrated a gene loop involving the 5′ and 3′ flanking regions of the FLC locus in Arabidopsis seedlings. Thus, we can now extend the concept of gene looping, previously described for yeast and mammalian genes, to plant gene expression. We found that FLC loop formation is independent of expression level but is disrupted in the early phase of vernalization (Figure 4A). This coincides with downregulation of transcription and accumulation of antisense transcripts to FLC (Figure 4B). Loop disruption occurs independently of VIN3, the cold-induced PHD protein involved in triggering the cold-induced epigenetic silencing at FLC. We suggest that gene loop disruption occurs as an early step associated with cold-induced transcriptional shut-down of FLC but preceding and independently of the Polycomb-mediated accumulation of epigenetic silencing.

Gene looping is tightly associated with transcriptional regulation in yeast. In the case of FLC, we cannot rule out that the formation of the loop requires initial rounds of transcription but the presence of the loop is not affected by mutations that significantly increase or reduce FLC gene expression (Figures 2 and 3). All these data suggest that gene looping and FLC transcription have different dynamics. In yeast, gene looping plays a role in ‘transcriptional memory’—where rapid gene reactivation depends on the persistence of the loop in non-induced conditions (Laine et al, 2009; Tan-Wong et al, 2009). Loop function in transcriptional memory in yeast has been associated with locus-specific H2A.Z-mediated localization at the nuclear periphery (Brickner et al, 2007; Tan-Wong et al, 2009). However, FLC gene loop is still present in arp6 mutant (Figure 3A), which argues against this association for Arabidopsis FLC.

The genomic region defined by the FI–FV FLC gene loop (Figure 1) corresponds exactly to the genomic region accumulating very high levels of H3K27me3 during vernalization (Angel et al, 2011). This feature and the relatively high frequency of interaction of FI–FV suggest that the loop could delimit a functional chromosomal domain. Factors defining such chromosomal domains are unclear but topologically associated domains have been proposed to delimit genomic regions in which genes are co-regulated and distinctly regulated from adjacent domains (Splinter et al, 2011). Given the complex regulation of FLC and the relatively compact Arabidopsis genome, some kind of physical isolation of this important developmental regulator may be functionally important. This domain is independent of the genomic context of FLC as loop formation and normal vernalization response are observed when FLC is integrated into random locations in the genome (Figure 6B).

We asked whether alternative chromatin interactions would form at different phases of the cold-silencing process, perhaps stimulated by expression of the COOLAIR antisense transcripts. However, no alternative loops were identified during COOLAIR antisense induction (Figure 5). This is similar to what has been reported for the yeast GAL10 locus where a gene loop is disrupted despite the presence of antisense transcription (Laine et al, 2009; Murray et al, 2012). A recent report has shown gene loops prevent divergent transcription from otherwise bidirectional promoters (Tan-Wong et al, 2012). We have not observed any divergent transcription from either FLC or COOLAIR promoters in the cold based on our custom tiling array analysis (Swiezewski et al, 2009; Angel and Dean, unpublished results); however, these analyses did not include Arabidopsis genotypes deficient in exosome function.

It has been reported that multiple Polycomb-dependent long chromosomal loops maintain human GATA-4 gene locus in a silenced but inducible state (Tiwari et al, 2008). To our knowledge there are no reports linking gene looping and Polycomb silencing. The independence of the FLC gene loop from VIN3 suggests that it is different from changes in chromosomal conformation previously described for non-plant Polycomb targets (Delest et al, 2012). The gene-specific looping detected here does not exclude additional long-range chromosomal interactions occurring before, during or after vernalization. Polycomb components modulate three-dimensional genome architecture through formation of long-range loops and clustering of targets at discrete nuclear foci called Polycomb bodies (Lanzuolo et al, 2007; Bantignies et al, 2011; Delest et al, 2012). It is therefore possible that FLC loop disruption is just one of many chromosome conformation changes during vernalization (Delest et al, 2012; Moissiard et al, 2012).

From all our data we propose that the FLC gene loop provides a transcriptional memory that is independent of actual FLC mRNA expression levels. As an early step in vernalization, cold induces disruption of the loop coincident with the transcriptional downregulation. We favour a model where this loop disruption reveals COOLAIR promoter elements, thus contributing to higher antisense transcription that aids shut-down of FLC expression. The observed time course of events supports this; loop disruption happens very early in the cold (Supplementary Figure 2) before antisense transcription reaches its maximum levels (Swiezewski et al, 2009; Heo and Sung, 2010). However, more detailed analyses will be required to fully elaborate cause and effect. After further cold, PHD–Polycomb complexes form at the internal nucleation site and H3K27me3 quantitatively accumulates. Upon transfer back to warm conditions, H3K27me3 and PHD–Polycomb complexes spread over FLC locus that flips the locus into an epigenetically repressed state. This state would impede further loop formation, maybe because Polycomb silencing impairs specific chromosomal interactions or because FLC is sequestered into a different nuclear compartment. Our work reveals an important role for gene loops in plant gene regulation in response to environmental signals and raises the possibility that loop disruption is a general early step during the switch to an epigenetically silent state.

Materials and methods

Plant material and growth conditions

Seeds were sown on GM media plates, stratified for 2 days and grown in long-day conditions for 10 days (16-h light at 20°C, 8-h darkness at 16°C). To vernalize, seeds were pre-grown for 7 days at standard warm-growing conditions (16-h light at 20°C, 8-h darkness at 16°C) before being transferred to cold (8-h light and 16-h darkness at 5°C) for 2 weeks, and then returned to warm conditions for 7 days.

Chromosome Conformation Capture (3C)

3C assays were performed according to Louwers et al (2009b) with some modifications. Arabidopsis seedlings (2 g) were crosslinked with 2% formaldehyde PBS buffer at room temperature for 20 min. Nuclei were purified and treated with SDS 0.3% at 65°C for 20 min. SDS was sequestered with 1% Triton X-100. Digestions were performed overnight at 37°C with 600U BamHI and BglII. Restriction enzymes were inactivated by addition of 1.6% SDS and incubation at 65°C for 10 min. After that 2% Triton X-100 was added to sequester SDS. Ligations were performed at 16°C for 5 h in 5-ml volume using 50 U of T4 DNA ligase. Reverse crosslinking was performed by overnight treatment at 65°C. DNA was recovered after Proteinase K treatment by phenol/chloroform extraction and ethanol precipitation.

3C quantification, normalization and controls

Relative interaction frequencies (Hagege et al, 2007) were calculated by Q-PCR on a Roche LightCycler 480 system using SYBR Green I Master mix from the same supplier. Differences in DNA concentration between samples were normalized using a loading control (LC), an FLC primer set that does not span any restriction site. To compensate for different primer efficiencies during PCR, we normalized to a control template (CT) DNA including all possible ligation products in equimolar amounts. The CT DNA was generated by digesting purified FLC-15 plasmid with BamHI and BglII, and subsequent random ligation without dilution. FLC-15 carries a 20-kb genomic DNA fragment containing the whole FLC locus (∼6 kb). In the case of UBC locus, the CT was generated by mixing equimolar quantities of purified PCR products. All figures in this work are the average of at least two biologically independent samples; each 3C DNA preparation was quantified twice by independent Q–PCR experiments; and each Q–PCR experiment included triplicates of all DNA samples. The values represented on each graph are relative interaction frequencies relative to the value of FI–FV in Col-FRI or Col. The full list of primers used in this work and an example of the calculations performed can be found in the Supplementary Information.

The quality of each experiment was assessed using two controls. First, we monitored the restriction efficiency using Q–PCR on chromatin aliquots taken before and after digestion. Under our experimental conditions, we found that restriction of crosslinked chromatin was between 70–90% of total genomic DNA. Second, genomic BglII and BamHI digestion resulted in fragments of the control locus UBC, which is unrelated to FLC and is routinely used to normalize mRNA expression data in our laboratory (Czechowski et al, 2005). The interaction frequencies of two fragments of UBC were quantified by quantitative 3C using the primers UBC_b2F and UBC_d2F. Experiments in which UBC 3C values were lower than the average indicated a failure in the experimental procedure (nuclei isolation, ligation, etc), and were subsequently discarded and not considered in the calculations.

RNA expression analysis

Hot-phenol total RNA extraction, DNAse I treatment, cDNA synthesis and Q-PCR analyses was performed as described by Sonmez et al (2011) using TURBO DNA-free (Life Technologies), Superscript III reverse transcriptase (Life Technologies) and SYBR Green I Master mix (Roche). Each data point is based on nine PCR reactions from three biological replicates.