COOLAIR and PRC2 function in parallel to silence FLC during vernalization

Contributed by Caroline Dean; received July 13, 2023; accepted December 11, 2023; reviewed by Martin Crespi and James C. Locke
January 18, 2024
121 (4) e2311474121


The role of noncoding transcription in chromatin regulation is still controversial, extending to the role of transcription of antisense transcripts called COOLAIR in the Polycomb-mediated epigenetic silencing of Arabidopsis FLC (FLOWERING LOCUS C), a key step during vernalization. Here, we show that COOLAIR transcription and PRC2 (Polycomb Repressive Complex 2) silence FLC in parallel pathways: an antisense-mediated transcriptional repression capable of fast response and a slow PRC2 epigenetic silencing, both of which are affected by growth dynamics and temperature fluctuations. These features explain the varied importance of COOLAIR transcription in cold-induced FLC epigenetic silencing seen in various studies using different conditions. The parallel repressive inputs and extensive feedbacks make the mechanism counterintuitive but provide great flexibility to the plant.


Noncoding transcription induces chromatin changes that can mediate environmental responsiveness, but the causes and consequences of these mechanisms are still unclear. Here, we investigate how antisense transcription (termed COOLAIR) interfaces with Polycomb Repressive Complex 2 (PRC2) silencing during winter-induced epigenetic regulation of Arabidopsis FLOWERING LOCUS C (FLC). We use genetic and chromatin analyses on lines ineffective or hyperactive for the antisense pathway in combination with computational modeling to define the mechanisms underlying FLC repression. Our results show that FLC is silenced through pathways that function with different dynamics: a COOLAIR transcription-mediated pathway capable of fast response and in parallel a slow PRC2 switching mechanism that maintains each allele in an epigenetically silenced state. Components of both the COOLAIR and PRC2 pathways are regulated by a common transcriptional regulator (NTL8), which accumulates by reduced dilution due to slow growth at low temperature. The parallel activities of the regulatory steps, and their control by temperature-dependent growth dynamics, create a flexible system for registering widely fluctuating natural temperature conditions that change year on year, and yet ensure robust epigenetic silencing of FLC.
Noncoding transcription has emerged as an important mechanism in environmentally responsive gene regulation. In some cases, noncoding transcription induces chromatin changes that are lost if the environmental signal is removed (1, 2). In other cases, chromatin changes, particularly those involving the Polycomb mark H3K27me3, are epigenetically maintained providing a memory of the inductive signal. One well-characterized example of the latter is the winter-induced epigenetic silencing of the Arabidopsis floral repressor gene, FLC (FLOWERING LOCUS C) (3, 4). This underpins the vernalization process, the acceleration of flowering by winter exposure. The process includes early induction of a series of antisense transcripts, called COOLAIR (5); a slow epigenetic switch from an active chromatin environment (marked by H3K36me3) to a silenced chromatin state (marked by H3K27me3) at an internal three nucleosome region (6); and spreading of the H3K27me3 Polycomb silencing over the whole locus (7, 8). The switching mechanism involves canonical Polycomb Repressive Complex 2 (PRC2) and Arabidopsis PRC2 accessory proteins VIN3 and VRN5. VIN3 is slowly induced by cold exposure (9), interacts with PRC2 at the nucleation region downstream of the FLC transcription start site (TSS), and has a functionally important head-to-tail polymerization domain (10).
The timing of early antisense transcription and later VIN3 expression led to the view that antisense transcription was a prerequisite for PRC2 silencing. Consistent with this, single-molecule FISH experiments revealed that COOLAIR expression was mutually exclusive with FLC sense transcription at each allele (11). This sequence of events was initially tested through T-DNA insertions into the COOLAIR promoter. These had little effect on long-term vernalization (12). Similarly, FLC silencing was unaffected in studies using a CRISPR deletion of the COOLAIR promoter or mutation of CBF factors, known to facilitate cold induction of COOLAIR (13). However, replacement of COOLAIR 5′ sequences (TEX1 line) attenuated FLC transcriptional silencing and disrupted the coordinated changes in H3K36me3 and H3K27me3 occurring at the FLC nucleation region (14).
COOLAIR had much stronger effects in experiments analyzing FLC silencing in natural field conditions. COOLAIR expression was strongly induced on the first freezing night of autumn (15, 16), a result recreated in controlled environment cabinets (15). In these experiments, one freezing night was sufficient to induce COOLAIR, but several freezing nights were required to silence FLC, with silencing attenuated by disruption of antisense transcription. These data are reminiscent of many Saccharomyces cerevisiae loci, where noncoding transcription plays an important role in environmental responsiveness (1, 17, 18). However, extensive feedback mechanisms between chromatin, transcription and cotranscriptional processes make functions of noncoding transcription difficult to elucidate. In particular, buffering between transcription and RNA stability leads to changed transcriptional dynamics with no change in steady state RNA (2).
To clarify the regulatory mechanism at Arabidopsis FLC, we have undertaken a series of genetic, molecular, and computational analyses to investigate the role of COOLAIR in cold-induced FLC silencing. Here, we show that FLC is silenced through parallel pathways. COOLAIR transcription can limit sense transcription, and this is associated with reduction in levels of the active histone mark H3K36me3; this mechanism involves disruption of a 5′-3′ FLC gene loop (19). In parallel, PRC2 silencing switches each allele from an epigenetically ON to an OFF state; this involves nucleation of H3K27me3 and subsequent spreading over the locus during subsequent growth, associated with further reduction in H3K36me3 (6). The nucleated and spread states differentially influence FLC transcription, which is still modulated by COOLAIR transcription. While FLC silencing by the PRC2 pathway operates on a slow timescale, the rapid induction capability of COOLAIR transcription, as seen in freezing conditions (15), enables this pathway in these conditions to silence FLC transcription on fast timescales. Components of both pathways are also regulated by their common transcriptional regulator NTL8 (20), which accumulates based on reduced dilution dependent on growth dynamics in the different cold phases. We integrate these parallel regulatory activities into a mathematical model that predicts FLC chromatin dynamics and transcription in different conditions. We argue that parallel activities converging onto a common target provides great flexibility in gene regulation, providing responsiveness to a wide variety of conditions. There are extensive similarities between how antisense transcription modulates FLC and how it alters sense transcription dynamics in yeast (2).


COOLAIR Rather than PRC2 Nucleation is the Major Contributor to FLC Repression in ntl8-D3.

Two independent genetic screens in different genotypes had identified dominant mutations that revealed NTL8 regulates VIN3 and COOLAIR (15, 20). Ectopic COOLAIR expression leads to very low FLC levels in warm grown plants (15). We therefore confirmed that VIN3 and COOLAIR are both misregulated in the dominant mutant ntl8-D3 (Fig. 1A), and then used it to genetically activate both pathways simultaneously, independently of cold. FLC transcriptional output, histone modifications and chromatin topology were analyzed. Paralleling cold effects on wild-type plants the ectopic COOLAIR expression in ntl8-D3 resulted in a clear decrease in H3K36me3, as compared to ColFRI, at the FLC TSS and over the gene body (Fig. 1B). The high COOLAIR transcription in ntl8-D3 led to accumulation of H3K36me3 at the COOLAIR promoter (Fig. 1B), matching the cold-induced transient increase of H3K36me3 in ColFRI at the same position. The decrease in H3K36me3 was not accompanied by an increase in H3K27me3 observed during vernalization (Fig. 1C). Likewise, H2Aub, another histone modification that accumulates at FLC during early vernalization did not accumulate ectopically in ntl8-D3 (Fig. 1D). The lack of accumulation of H3K27me3 and H2Aub in ntl8-D3 compared to ColFRI in the absence of cold supports the view that VIN3 expression itself is not sufficient to cause Polycomb mediated silencing of FLC. These data indicate that antisense-mediated suppression rather than VIN3-mediated nucleation of H3K27me3 is the major factor causing FLC repression in ntl8-D3. Repression of sense FLC transcription in ntl8-D3 is almost completely suppressed when COOLAIR transcription is blocked, giving further support to this conclusion (15).
Fig. 1.
ntl8-D3 mimics cold exposure, except for the accumulation of H3K27me3 and H2Aub. (A) Expression of total COOLAIR and VIN3 in ntl8-D3 FRI and ColFRI in nonvernalized conditions (NV). Data are presented as the mean ± SEM. Each open circle represents a biological replicate. (BD) Enrichment of (B) H3K36me3, (C) H3K27me3, and (D) H2Aub across FLC measured by ChIP in wild-type ColFRI and ntl8-D3 at NV conditions. H3K36me3 data are shown relative to H3 and actin. H3K27me3 data are shown relative to H3 and STM. H2Aub data are shown relative to H3. Error bars represent SEM (n ≥ 3 biological replicates). (E) VIN3-eGFP ChIP-qPCR enrichment at FLC at NV. Data are shown as the percentage input. Nontransgenic ColFRI plants were used as a negative control sample. Error bars represent SEM (n = 3 biological replicates). (F) Enrichment of H3K27me2 across FLC measured by ChIP in wild-type ColFRI and ntl8-D3 at NV conditions. Data are expressed relative to H3. Error bars represent SEM (n = 3 biological replicates). (G) Quantitative 3C-qPCR over the FLC locus in 10-d-old ColFRI and ntl8-D3 FRI seedlings. A schematic representation of the FLC locus is shown above. BamHI and BglII restriction sites are indicated with dotted lines, and the respective regions are numbered with Roman numerals. Red arrows indicate the location of the primers used for 3C-qPCR. The region around the FLC TSS was used as the anchor region in the 3C analysis. The data below show the relative interaction frequencies (RIF).

Ectopically Expressed VIN3 Localizes to FLC but Fails to Induce H3K27me3 Nucleation.

To understand what prevents the accumulation of H3K27me3 in ntl8-D3 despite ectopic VIN3 expression, we tested whether other epigenetic factors are misexpressed in ntl8-D3. Only one of the tested genes changed slightly in expression (SI Appendix, Fig. S1). We then analyzed association of VIN3 at the nucleation region in ntl8-D3. Despite the lack of H3K27me3 accumulation in ntl8-D3, we found VIN3-eGFP accumulated at the FLC nucleation region in warm conditions, mimicking the accumulation during vernalization (Fig. 1E). Thus, VIN3 accumulation at the nucleation region does not result in stable nucleation of H3K27me3. To distinguish VIN3 intrinsic binding to the FLC nucleation region, independently of COOLAIR transcriptional induction, we expressed VIN3-eGFP under the promoter of VRN5 (SI Appendix, Fig. S2A). This resulted in expression levels in nonvernalized plants that paralleled VIN3 induction after 6 wk cold (6WT0) (SI Appendix, Fig. S2B). In this line VIN3-eGFP was enriched at the FLC locus in NV conditions (SI Appendix, Fig. S2C), showing VIN3 can remain associated with the nucleation region even when FLC is strongly expressed. We found that VIN3 association in ntl8-D3 led to H3K27me2 enrichment despite no accumulation of H3K27me3 (Fig. 1F). Thus, cold-induced features, possibly influencing residence time, are required to enable VIN3 functionality to deliver H3K27me3 to the nucleation region.

Ectopic Induction of COOLAIR Correlates with Chromatin Topology Changes.

A cold-induced feature at FLC is disruption of a gene loop conformation that links the TSS and the transcription termination site (19). In ntl8-D3, we found that the gene loop was ectopically disrupted, mimicking vernalization (Fig. 1G). This suggests that gene loop disruption is linked with antisense-mediated reduction in FLC transcription. We also found that the TEX2.0 transgene, where a nos terminator promotes early COOLAIR termination, reduces gene loop formation (SI Appendix, Fig. S3), consistent with earlier reports using a similar, but not identical transgene (21). This result suggests a role for the activity of the antisense promoter/TSS, rather than antisense transcription per se, as being important for gene loop disruption.

Disrupting COOLAIR Transcription Perturbs H3K27me3 Dynamics Before and during Cold, but Not Postcold H3K27me3 Levels.

We further investigated the fact that ectopic expression of antisense transcription is enough to cause lower H3K36me3 around the FLC sense TSS and in the gene body, even in the absence of cold. Antisense transcription could lower H3K36me3 levels, either through direct removal mediated by antisense transcription or indirectly by limiting sense transcription, thus preventing the cotranscriptional addition of H3K36 methylation. To dissect the interplay of H3K36me3 and H3K27me3, we studied the dynamic changes in these modifications using a vernalization time course.
Our previous analyses of TEX transgenes were in an flc-2 background, where part of the endogenous FLC genomic sequence remains (4). This limited the regions where the chromatin modifications on the transgene could be studied (14). To overcome this limitation, we generated a FRI + FLC null (flclean) where the entire FLC genomic sequence had been deleted using CRISPR (SI Appendix, Fig. S4) and introduced the previously described TEX1.0 (replacement of the COOLAIR promoter) and TEX2.0 (insertion of a nos terminator to truncate COOLAIR transcription) transgenes. We also included a FRI FLCΔCOOLAIR CRISPR line, which deletes the COOLAIR promoter at the endogenous locus (22). Using these multiple defective COOLAIR lines and respective controls, we undertook a detailed time course of histone modifications during vernalization, including multiple time points postcold (Fig. 2 A and B).
Fig. 2.
Cold-induced chromatin and RNA dynamics in COOLAIR defective lines. (A and B) Enrichment of H3K27me3 (A) and H3K36me3 (B) across FLC measured by ChIP in wild-type ColFRI and the three defective COOLAIR lines, TEX1, TEX2, and FLCΔCOOLAIR, before, during, and after vernalization. Data are expressed relative to H3 and STM. Error bars represent SEM (n = 3 biological replicates). (C and D) Average levels of H3K27me3 (C) and H3K36me3 (D) in the nucleation region during vernalization. The averages were calculated by averaging the ChIP enrichment over three primers in the FLC nucleation region during vernalization in ColFRI and each of the defective COOLAIR lines. (E and F) FLC expression during a vernalization time course in ColFRI and the three defective COOLAIR lines, unspliced (E) and spliced RNA (F), was measured and is shown relative to UBC and NV levels. Error bars represent SEM (n = 3 biological replicates).
The rate of accumulation of H3K27me3 during cold exposure was not reduced in the COOLAIR defective genotypes compared to the wild type, and at some timepoints was even accelerated (Fig. 2C and SI Appendix, Fig. S5 A and B), consistent with our previous data (14). By 6WT0, wild-type and COOLAIR defective genotypes show similar H3K27me3 levels in the nucleation region (Fig. 2A and SI Appendix, Fig. S5C). However, there were clear differences in the starting levels of H3K27me3, being significantly lower in the nucleation region in all defective COOLAIR genotypes (Fig. 2A and SI Appendix, Fig. S5D). Consistent with the differences in starting H3K27me3 levels, and supporting a role for antisense transcription in establishment of the initial FLC chromatin state (23), the defective COOLAIR genotypes showed a consistent trend of higher FLC RNA before cold exposure (SI Appendix, Fig. S5E), although the differences were small. The similar trend in TEX1, TEX2 and FLCΔCOOLAIR argues against this being a specific TEX transgene effect. We interpret the H3K27me3 level in ColFRI before cold as representing a fraction of FLC alleles that have switched to a stable Polycomb silenced state. Thus, higher H3K27me3 levels in ColFRI compared to the COOLAIR defective genotypes may reflect the COOLAIR role in developmentally regulated PRC2 silencing of FLC (24, 25). After cold there was no significant difference in H3K27me3 levels in the nucleation region between ColFRI and any of the COOLAIR defective genotypes (Fig. 2A and SI Appendix, Fig. S5C). Spreading of H3K27me3 was also unaffected in the COOLAIR defective genotypes, as seen from the similar levels in the gene body at 6WT10 and 6WT20 (Fig. 2A). Overall, we find that H3K27me3 dynamics before and during cold are perturbed by COOLAIR, but that postcold H3K27me3 levels are not.

Disrupting COOLAIR Transcription Attenuates H3K36me3 Removal during Vernalization.

H3K36me3 levels were similar in all genotypes before vernalization (Fig. 2B and SI Appendix, Fig. S5F) but decreased at different rates during cold exposure (Fig. 2D). This contrasts with the clear NV differences in H3K27me3 levels. However, this is consistent with the NV H3K27me3 levels coming from a small fraction of silenced alleles, while most alleles are transcriptionally active and contribute to the observed H3K36me3 levels, a scenario that generates bigger fold changes in H3K27me3 than in H3K36me3, as we observe (SI Appendix, Fig. S5G). H3K36me3 levels reduced more slowly in all defective COOLAIR genotypes at 6WT0 (Fig. 2B and SI Appendix, Fig. S5G), but after 2 wk cold H3K36me3 levels increased in the gene body compared to NV (Fig. 2B). There were no differences in H3K36me3 levels between COOLAIR defective genotypes and wild-type ColFRI after transfer back to warm (Fig. 2B). In ntl8-D3, where VIN3 and COOLAIR are both overexpressed, faster reduction of H3K36me3 in the cold was observed, while H3K27me3 was less affected (SI Appendix, Figs. S6 AD and S9 B and C). Together our results demonstrate that the Polycomb pathway is effective enough to completely silence the FLC locus, despite either an ineffective or hyperactive antisense pathway. The COOLAIR-mediated pathway mediates not only the removal of H3K36me3 but also H3K4me1 through the activity of the demethylase complex FLD-LD-SDG26 (23). H3K4me1, like H3K36me3, has been shown to be added cotranscriptionally in plants (26). Consistently, we found that in the COOLAIR defective lines, H3K4me1 reduction during vernalization was attenuated (SI Appendix, Fig. S7) showing the same trend as H3K36me3, including the increase at 2WT0. Overall, we find that COOLAIR defective genotypes have reduced rates of H3K36me3 removal, but after cold, any differences in H3K36me3 levels disappear. The relative changes of the unspliced FLC RNA levels did not match the corresponding H3K36me3 levels in the COOLAIR defective genotypes and effects on spliced FLC levels were different to unspliced (Fig. 2 E and F). This suggests a similar interconnected mechanism linking chromatin modification to transcript stability as found in yeast, with unspliced and spliced transcripts affected in different ways (2).

H3K27me3 Accumulation Is Not Necessary for COOLAIR-Mediated Transcriptional Downregulation.

A mutation in the core PRC2 component Su(z)12 (VRN2) only partially disrupted FLC repression (SI Appendix, Fig. S8A) (6), while H3K36me3 fold reduction at FLC during the cold was hardly changed (SI Appendix, Fig. S8B), despite accumulation of H3K27me3 being abolished (6). Thus, H3K36me3 reduction and FLC RNA downregulation do not rely on H3K27me3 nucleation. Analysis of a vrn5-TEX1.0 combination, defective in H3K27me3 accumulation and COOLAIR, had shown that the H3K36me3 reduction seen in an H3K27me3 nucleation mutant is mediated by COOLAIR (14). To examine this aspect further, we analyzed changes in the two modifications, H3K36me3 and H3K27me3, in fluctuating cold conditions, where we see the clearest indication of COOLAIR transcription regulating FLC expression (15). Under these conditions, COOLAIR was highly up-regulated, causing significant downregulation of FLC sense transcript (15). While we have previously shown that full-length COOLAIR transcription is essential for the FLC downregulation in these conditions (15), a role for H3K27me3 nucleation had not been investigated. Here, we analyzed H3K36me3 and H3K27me3 levels in ColFRI at 2WT0, under three different cold conditions (as in ref. 15), constant 5 °C (CC, Constant Cold), mild 3 to 9 °C (FM, Fluctuating Mild), and strong fluctuating conditions −1 to 12 °C (FS, Fluctuating Strong). Zhao et al. (15) showed that COOLAIR upregulation and FLC downregulation were greatest in the FS condition. Therefore, we would expect H3K36me3 to show the largest changes in this condition, and indeed this is seen in our data (Fig. 3A and SI Appendix, Fig. S9E). Both mild fluctuating and CC result in smaller changes (Fig. 3A and SI Appendix, Fig. S9E). In contrast, H3K27me3 accumulation showed little difference between the different conditions (Fig. 3B and SI Appendix, Fig. S9D), indicating that H3K27me3 is not the major contributor to the enhanced downregulation under strongly fluctuating conditions. The lack of difference in H3K27me3 accumulation, despite the relatively large change in antisense expression, further highlights the parallel and almost independent nature of these FLC repression pathways. To further confirm that COOLAIR transcription is necessary for the changes in H3K36me3 under fluctuating conditions, we subjected the TEX1.0 defective COOLAIR line to these conditions. As expected, the reduction in H3K36me3 was also attenuated relative to ColFRI (Fig. 3 C and D). This is consistent with the lack of FLC sense transcriptional shutdown under FS conditions in a COOLAIR deletion line as recently described (27). Interestingly, for COOLAIR defective genotypes, the slight increase in H3K36me3 at 2WT0 observed in CC was also recapitulated in the FM conditions. The reduction under FS conditions was significantly attenuated in TEX1.0 (SI Appendix, Fig. S9F), and an analysis to test whether this is also the case in the COOLAIR deletion line is ongoing. Overall, we find that COOLAIR-mediated transcriptional repression does not strongly depend on H3K27me3 nucleation, supporting our earlier results in ntl8-D3. The contrasting dynamics of H3K36me3 and H3K27me3 under 2 wk of fluctuating conditions further highlight the fast-response capability of the antisense-mediated repression. In ColFRI, the changes in nucleation region H3K36me3 after only 2 wk of FS conditions (Fig. 3 A and C) are comparable to H3K36me3 changes after 6 wk in CC (Fig. 2B and SI Appendix, Fig. S5G). The fast response capability of FLC antisense transcription to temperature changes is also supported by field data for A. halleri FLC (28). They showed that H3K4me3 associated with COOLAIR transcription at the 3′ end of the FLC locus responds to temperature changes on a much faster timescale compared to the 5′ end (associated with FLC sense transcription), which responds mainly on a slow timescale.
Fig. 3.
Fluctuating cold and mathematical modeling of the role of COOLAIR in histone modification dynamics. (A and B) Changes in H3K36me3 (A) and H3K27me3 (B) at FLC after 2 wk of CC, FM, or FS conditions, measured by ChIP-qPCR. Data are expressed relative to H3. Error bars represent SEM (n = 3 biological replicates). (C and D) Comparing changes in H3K36me3 at FLC between ColFRI (C) and TEX1 flclean (D) after 2 wk of FM or FS conditions, measured by ChIP-qPCR. Data are expressed relative to H3 and STM. Error bars represent SEM (n = 3 biological replicates). (E) Schematic of the mathematical model showing core components: PRC2-mediated silenced states (non-nucleated, nucleated, and spread) at individual FLC alleles, antisense transcription–mediated repression of FLC transcription, and the contribution of these components to the average population-level H3K36me3 coverage at the FLC locus.

Mathematical Modeling of FLC Regulation Reconciles the Different Effects of Antisense Transcription on Chromatin State.

The dynamics of the two repression pathways are difficult to dissect quantitatively purely through molecular experiments. We therefore turned to mathematical modeling to see how the observed behavior in COOLAIR defective mutants could be reconciled with our existing understanding of FLC repression in the cold. We have previously developed and experimentally validated a mathematical modeling framework describing dynamically changing fractions of active/silenced FLC alleles and their associated histone modifications (29, 30). Here, we built on this framework to develop an augmented model, incorporating an antisense-mediated silencing component. A schematic of the model developed here is shown in Fig. 3E (details in SI Appendix—the MATLAB code used to simulate the ODE model is available at The model was built based on our main conclusion from the above data: namely that two pathways work in parallel to silence FLC, antisense transcription and PRC2 nucleation. We then interrogated the model to see whether it was capable of quantitatively reproducing histone modification dynamics in ColFRI and the various mutants.
The effect of the antisense-mediated pathway on sense transcription was modeled implicitly as a cold-dependent graded modulator of sense initiation/transition to productive elongation. This is consistent with high antisense transcription in ntl8-D3 causing low levels of FLC transcription, independently of H3K27me3 nucleation. This is also consistent with previous data showing that sense and antisense transcription at FLC are anticorrelated in ColFRI, both in warm (31) and in cold conditions (11). This may be through a mutual exclusivity model for the FLC locus, similar to that reported for the CBF1-SVALKA locus, where full-length sense transcription is inhibited by antisense transcription (32). Another key aspect of the model is the cotranscriptional delivery of the H3K36me3 modifications. Changes in Pol II elongation behavior can affect the H3K36me3 profile across mammalian genes (33), with slower Pol II speed allowing a larger window of opportunity for adding H3K36me3 at any given location. Any changes in transcription at FLC may be expected to produce corresponding changes in H3K36me3. However, to explain the increase in gene body H3K36me3 observed in the defective COOLAIR lines, specifically at 2WT0 compared to NV, despite the lack of any increase in transcriptional output over that time period (Fig. 2E), the model includes a cold-induced reduction in Pol II speed in this region, resulting in a longer dwell time. The SDG8 H3K36 methyltransferase, which we have shown cotranscriptionally associates with RNA PolII (34), is likely mediating these H3K36me3 changes. The model also allows for H3K36me3 removal on a timescale consistent with the experimentally observed lifetime of H3K36 methylation in other systems (35), so that its levels at the nucleation region would decay quickly in the absence of sense transcription. The model also describes dynamic changes in FLC mRNA levels as modeled in ref. 29. However, due to the highly variable behavior of spliced FLC RNA observed in the different COOLAIR defective mutants (Fig. 2F), which potentially reflects changes in RNA stability, we do not try to capture these levels using the model.
We also incorporate the PRC2 pathway and how it silences FLC through H3K27me3 accumulation during vernalization. In cells which can have active or nonactive cell cycles, we consider that FLC alleles can be in three different states; non-nucleated (without H3K27me3 nucleation), nucleated, and spread, with the latter only attained in active cycling cells (29). To generate reasonable fits to our data, particularly the higher levels of H3K36me3 observed during cold in COOLAIR defective mutants, we found that an extension to our previous models was needed, where we allow for different levels of FLC transcription in the three states: highest in non-nucleated, much lower in nucleated, and even lower in spread. Satisfactory fits also necessitated that the non-nucleated and nucleated states be capable of further downregulation by antisense transcription. Consistent with the possibility of some limited transcriptional activity in the nucleated state, we, therefore, allowed for potential coexistence of H3K27me3 and H3K36me3 on the same nucleosome. We then fitted the model to capture the qualitative changes in H3K36me3, H3K27me3, and transcription (sense and antisense) observed in the cold for ColFRI and the defective COOLAIR lines. We found that this model could capture all the qualitative features of the data observed in ColFRI and the defective COOLAIR lines (Fig. 4 AD), including both the increase of H3K36me3 at 2WT0 and the subsequent significantly slower reduction in H3K36me3 in the latter (Fig. 4 A and B), as well as the reduction of H3K36me3 in the postcold seen in all the lines (Fig. 4 A and B).
Fig. 4.
Model predictions of the impact of vernalization mutants on histone dynamics and schematic representation of parallel pathways that repress FLC expression. (AD) Mathematical model predicted levels of H3K36me3 (A and B) and H3K27me3 (C and D) over a CC time course in a defective COOLAIR mutant (A and C) and the wild type, ColFRI (B and D). The predictions are compared to the ChIP-qPCR time course data for the different genotypes presented in Fig. 2 (A and B). (E) Model for the parallel pathways that repress FLC. In the warm FLC forms a gene-loop conformation, which mediates a high expression state of FLC. The high expression state is marked by high levels of H3K36me3 around the FLC TSS. After cold exposure, the repressive pathways are activated: 1) the antisense-mediated pathway leading to disruption of the gene loop and removal of H3K36me3 from the TSS of FLC and 2) the Polycomb pathway leading to deposition of H3K27me3 and repression of FLC transcription. The two pathways work in parallel rather than through a linear sequence of causation to give the final FLC expression outcome during vernalization.
We then tested whether the model could capture our previous datasets by simulating other mutants that affect FLC silencing in the cold [see previously published data (6)], including an H3K27me3 nucleation mutant (e.g., vrn2, vin3), a spreading mutant (e.g., lhp1, clf). In all cases, the simulation outputs from the model are qualitatively consistent with data, including the postcold behavior of the two histone modifications (SI Appendix, Fig. S10 AD). Interestingly, in addition to recapitulating the behavior captured by our previous models, the current model can capture the reduction of H3K36me3 in the postcold seen in ColFRI—a trend that could not be previously captured (Fig. 4B). This is because the current model allows for higher levels of transcription (and consequently higher H3K36me3) in a nucleated state relative to a spread state.
The model also incorporates a fast timescale response in the antisense mediated pathway, which can respond to temperature fluctuations (see SI Appendix for details). Briefly, this consists of a simple step increase in antisense mediated repression resulting from temperatures dropping below a threshold, which is incorporated into the slower timescale increase in antisense mediated repression. The fast timescale response allows the model to qualitatively capture the differences in H3K36me3 and FLC spliced RNA between the different cold conditions CC, FM, FS (SI Appendix, Fig. S11). The model predicts that H3K36me3 and FLC mRNA levels respond on a fast timescale, exhibiting oscillations in response to the daily repeated temperature profiles of FM and FS conditions. While the agreement between the model and experiments is overall good, the model did predict reduced H3K27me3 nucleation in FS conditions, which was not observed experimentally. This discrepancy may potentially arise from differences between the field conditions used to parameterize our model for H3K27me3 nucleation (30) and the FS experimental conditions used here. The model predicts that antisense transcription limits H3K36me3 through a graded, analog reduction in FLC transcription rather than by directly mediating H3K36me3 removal. The increased H3K36me3 at 2WT0 in COOLAIR defective lines, is predicted to arise from a combination of higher FLC sense transcription (since antisense mediated repression is disrupted) and cold-induced reduction in Pol II speed in the nucleation region. In a second slower response chromatin pathway involving PRC2, each allele progressively switches from a non-nucleated to H3K27me3 nucleated state during the cold and then to a spread H3K27me3 state during postcold growth. The model indicates that intermediate levels of FLC transcription in the nucleated state, which can be further down-regulated by antisense transcription, can explain how clear differences in H3K36me3 between defective COOLAIR lines and the wild type can emerge in the cold yet subsequently disappear during growth after transfer to warm conditions. In these conditions, all the nucleated FLC alleles convert to the H3K27me3 spread state due to an active cell cycle (6), regardless of H3K36me3 levels and any residual expression. Hence, in the context of vernalization, the COOLAIR repressive pathway is most important during rather than after cold.


Focused dissection of the mechanism underlying winter-induced FLC silencing has established a role for antisense transcription and PRC2 activity in registration of long-term exposure to noisy environmental signals (5, 15, 36, 37). However, the complexity of the mechanism, and its sensitivity to variable temperature and growth parameters, has led to different studies questioning the importance of the antisense transcription in cold-induced FLC silencing. Here, using a combination of experiments and mathematical modeling, we have elucidated the role of antisense transcription and PRC2 activity as parallel pathways, both leading to FLC silencing (Fig. 4B). The antisense-mediated pathway involves the FLC gene loop and represses FLC transcription (38). Two other lncRNAs have been described at FLC, COLDWRAP (39) and COLDAIR (40). We have detected cold-up-regulated FLC transcripts with upstream TSSs including COLDWRAP that influence FLC expression levels but not cold-induced transcriptional silencing (41); we have not found COLDAIR equivalents.
The multiple effects of the COOLAIR-mediated transcriptional pathway on H3K36me3 in the 5′ region of FLC required modeling to deconvolve fully. In the wild type, lower transcription leads to H3K36me3 reduction, but this is partly hidden in defective COOLAIR lines in the cold through a predicted increase in H3K36me3 from slower RNA PolII speed at the 5′ end of FLC. The slow PRC2 switch at each FLC allele from a non-nucleated to a H3K27me3 nucleated and then spread state is associated with decreasing frequencies of FLC transcription, consistent with previous findings of the relationship between H3K27me3 and FLC transcription (6). Both COOLAIR-mediated and PRC2 pathways are affected by the common transcriptional regulator NTL8, which accumulates slowly and variably dependent on reduced dilution by slower growth at low temperatures (20). Our observation that in ntl8-D3 both COOLAIR and VIN3 are ectopically expressed, yet the H3K27me2 modification but not H3K27me3 accumulates, implies a requirement for other cold-induced factors for vernalization (42, 43). These parallel repressive activities with multiple temperature inputs enables modulation of transcriptional silencing independently of robust epigenetic silencing. This gives the plants great flexibility to respond to autumnal conditions that vary in different geographical regions and from year to year yet ensure robust silencing. Indeed, variation in FLC transcriptional silencing has been shown to be an important adaptive determinant in Arabidopsis thaliana accessions (16). It seems likely that similar parallel mechanisms may be involved in other seasonal responses, e.g., seed and bud dormancy and germination, for similar reasons.
The chromatin changes in COOLAIR defective mutants are not directly reflected in steady-state unspliced and spliced FLC levels (Fig. 2 E and F), similar to the situation in yeast (2). Which RNA stability mechanisms are involved remain to be determined, but m6A methylation has been shown to influence FLC regulation (38, 44, 45) and is enriched in the FLC 3′ UTR. This disconnect between chromatin dynamics and steady-state RNA levels is likely to have contributed to the controversy over the role of noncoding RNA in chromatin regulation generally. In addition, the effective combination of parallel pathways hides effects of mutations after saturating induction, e.g., mutations in CBF-binding factors (13). Another debate has been over the use of transgenes to modulate COOLAIR expression (13), but the use here of FLCΔCOOLAIR and ntl8-D3 for under/overexpression of COOLAIR argues against this. However, future studies need to generate a fully antisense null genotype since all defective COOLAIR genotypes so far produced still contain cryptic antisense promoters, which become more active when the endogenous COOLAIR promoter is mutated/deleted (15). The difficulty of completely removing antisense transcription is also seen in other systems (46) and suggests transcription initiation from open chromatin regions rather than specific promoter elements. Such a line would not only help elucidate the role of COOLAIR in the cold-induced silencing of FLC but also in the starting FLC expression upon germination, a key determinant of natural variation underpinning adaptation (16).
The large number of plant chromatin regulators that interact with noncoding RNAs points to an important role of similar cotranscriptional mechanisms in environmental plasticity (47, 48). This is similarly true in yeast (2), where antisense expression has been associated with genes that are environmentally silenced (1). Future work will address the evolutionary parallels and conservation of a mechanism enabling rapid transcriptional changes and switches to epigenetic silencing in response to noisy environmental cues.

Materials and Methods

Detailed descriptions of materials and methods are provided in SI Appendix. A brief summery is provide here.

Plant Materials.

All mutant and transgenic lines were in the FRIsf2 background. ntl8-D3 FRI was described previously (15). Generation of new mutant and transgenic lines is detailed in SI Appendix.

Expression Analysis.

RNA analysis was performed as previously described (49). Total RNA was extracted and genomic DNA contamination removed using TURBO DNase (Invitrogen). cDNA was synthesized with SuperScript IV reverse transcriptase (Invitrogen). qPCR was performed using SYBR Green I Master (Roche) and analyzed on a LightCyler 480 machine (Roche).

Chromatin Immunoprecipitation (ChIP).

ChIP was performed as described in ref. 49. ChIP was performed with antibodies: α-H3 (Abcam, ab1791), α-H3K36me3 (Abcam, ab9050), α-H3K27me3 (Abcam, ab192985), α-H3K27me2 (Upstate, 07-452), α-H2AK119ub (Cell Signaling Technology, #8240), α-H3K4me1 (Abcam, ab8895), and α-GFP (Abcam, ab290).

Chromatin Conformation Capture.

Chromatin conformation capture was performed as described previously described (19) with minor modifications.

Mathematical Modeling.

The models used in this study are constructed within a framework we have previously developed and experimentally validated (29, 30). In addition to what is captured by previous models, our current model also incorporates the dynamics of H3K36me3 and H3K27me3 at FLC, as observed in ColFRI and the COOLAIR defective mutants.

Data, Materials, and Software Availability

Previously published data were used for this work (6). All other data are included in the manuscript and/or SI Appendix.


We would like to thank members of the Dean lab for thoughtful inputs and Shuqin Chen for excellent technical assistance. This work was funded by the European Research Council Advanced Grant (EPISWITCH, 833254), Wellcome Trust (210654/Z/18/Z), and a Royal Society Professorship (RP\R1\180002) to C.D. and BBSRC Institute Strategic Programmes (BB/J004588/1 and BB/P013511/1) and Engineering and Physical Sciences Research Council/Biotechnology and Biological Sciences Research Council (EPSRC/BBSRC) Physics of Life grant (EP/T00214X/1) to M.H. and C.D.

Author contributions

M.H., and C.D. supervised the research; M.N., Y.Z., E.M.-B., P.W., and S.Z. performed research; M.N., G.M., Y.Z., E.M.-B., P.W., and S.Z. analyzed data; G.M. undertook the modelling; and M.N., G.M., M.H., and C.D. wrote the paper.

Competing interests

The authors declare no competing interest.

Supporting Information

Appendix 01 (PDF)
Dataset S01 (XLSX)


Z. Xu et al., Antisense expression increases gene expression variability and locus interdependency. Mol. Syst. Biol. 7, 468 (2011).
T. Brown et al., Antisense transcription-dependent chromatin signature modulates sense transcript dynamics. Mol. Syst. Biol. 14, e8007 (2018).
C. C. Sheldon, D. T. Rouse, E. J. Finnegan, W. J. Peacock, E. S. Dennis, The molecular basis of vernalization: The central role of FLOWERING LOCUS C (FLC). Proc. Natl. Acad. Sci. U.S.A. 97, 3753–3758 (2000).
S. D. Michaels, R. M. Amasino, FLOWERING LOCUS C encodes a novel MADS domain protein that acts as a repressor of flowering. Plant Cell 11, 949–956 (1999).
S. Swiezewski, F. Liu, A. Magusin, C. Dean, Cold-induced silencing by long antisense transcripts of an Arabidopsis Polycomb target. Nature 462, 799–802 (2009).
H. Yang et al., Distinct phases of Polycomb silencing to hold epigenetic memory of cold in Arabidopsis. Science 357, 1142–1145 (2017).
E. J. Finnegan, E. S. Dennis, Vernalization-induced trimethylation of histone H3 lysine 27 at FLC is not maintained in mitotically quiescent cells. Curr. Biol. 17, 1978–1983 (2007).
S. Berry, M. Hartley, T. S. G. Olsson, C. Dean, M. Howard, Local chromatin environment of a Polycomb target gene instructs its own epigenetic inheritance. Elife 4, e07205 (2015).
S. Sung, R. M. Amasino, Vernalization in Arabidopsis thaliana is mediated by the PHD finger protein VIN3. Nature 427, 159–164 (2004).
M. Fiedler et al., Head-to-tail polymerization by VEL proteins underpins cold-induced Polycomb silencing in flowering control. Cell Rep. 41, 111607 (2022).
S. Rosa, S. Duncan, C. Dean, Mutually exclusive sense–antisense transcription at FLC facilitates environmentally induced gene repression. Nat. Commun. 7, 1–7 (2016).
C. A. Helliwell, M. Robertson, E. J. Finnegan, D. M. Buzas, E. S. Dennis, Vernalization-repression of Arabidopsis FLC requires promoter sequences but not antisense transcripts. PLoS One 6, e21513 (2011).
M. Jeon et al., Vernalization-triggered expression of the antisense transcript COOLAIR is mediated by CBF genes. Elife 12, e84594 (2023).
T. Csorba, J. I. Questa, Q. Sun, C. Dean, Antisense COOLAIR mediates the coordinated switching of chromatin states at FLC during vernalization. Proc. Natl. Acad. Sci. U.S.A. 111, 16160–16165 (2014).
Y. Zhao et al., Natural temperature fluctuations promote COOLAIR regulation of FLC. Genes Dev. 35, 888–898 (2021).
J. Hepworth et al., Natural variation in autumn expression is the major adaptive determinant distinguishing Arabidopsis FLC haplotypes. Elife 9, e57671 (2020).
J. A. Martens, L. Laprade, F. Winston, Intergenic transcription is required to repress the Saccharomyces cerevisiae SER3 gene. Nature 429, 571–574 (2004).
T. Nguyen et al., Transcription mediated insulation and interference direct gene cluster expression switches. Elife 3, e03635 (2014).
P. Crevillén, C. Sonmez, Z. Wu, C. Dean, A gene loop containing the floral repressor FLC is disrupted in the early phase of vernalization. EMBO J. 32, 140–148 (2013).
Y. Zhao, R. L. Antoniou-Kourounioti, G. Calder, C. Dean, M. Howard, Temperature-dependent growth contributes to long-term cold sensing. Nature 583, 825–829 (2020).
Z. Li, D. Jiang, Y. He, FRIGIDA establishes a local chromosomal environment for FLOWERING LOCUS C mRNA production. Nat. Plants 4, 836–846 (2018).
X. Luo, T. Chen, X. Zeng, D. He, Y. He, Feedback regulation of FLC by FLOWERING LOCUS T (FT) and FD through a 5’ FLC promoter region in Arabidopsis. Mol. Plant 12, 285–288 (2019).
X. Fang et al., The 3′ processing of antisense RNAs physically links to chromatin-based transcriptional control. Proc. Natl. Acad. Sci. U.S.A. 117, 15316–15321 (2020).
Y. Tian et al., PRC2 recruitment and H3K27me3 deposition at FLC require FCA binding of COOLAIR. Sci. Adv. 5, eaau7246 (2019).
R. L. Antoniou-Kourounioti et al., Integrating analog and digital modes of gene expression at Arabidopsis FLC. Elife 12, e79743 (2023).
S. Oya, M. Takahashi, K. Takashima, T. Kakutani, S. Inagaki, Transcription-coupled and epigenome-encoded mechanisms direct H3K4 methylation. Nat. Commun. 13, 4521 (2022).
P. Zhu, C. Dean, Reply to: Cold induction of nuclear FRIGIDA condensation in Arabidopsis. Nature 619, E33–E37 (2023).
H. Nishio et al., Repressive chromatin modification underpins the long-term expression trend of a perennial flowering gene in nature. Nat. Commun. 11, 2065 (2020).
J. I. Qüesta et al., Noncoding SNPs influence a distinct phase of Polycomb silencing to destabilize long-term epigenetic memory at Arabidopsis FLC. Genes Dev. 34, 446–461 (2020).
R. L. Antoniou-Kourounioti et al., Temperature sensing is distributed throughout the regulatory network that controls FLC epigenetic silencing in vernalization. Cell Syst. 7, 643–655.e9 (2018).
R. Ietswaart, S. Rosa, Z. Wu, C. Dean, M. Howard, Cell-size-dependent transcription of FLC and its antisense long non-coding RNA COOLAIR explain cell-to-cell expression variation. Cell Syst. 4, 622–635.e9 (2017).
P. Kindgren, R. Ard, M. Ivanov, S. Marquardt, Transcriptional read-through of the long non-coding RNA SVALKA governs plant cold acclimation. Nat. Commun. 9, 4561 (2018).
N. Fong, T. Saldi, R. M. Sheridan, M. A. Cortazar, D. L. Bentley, RNA Pol II dynamics modulate co-transcriptional chromatin modification, CTD phosphorylation, and transcriptional direction. Mol. Cell 66, 546–557.e3 (2017).
H. Yang, M. Howard, C. Dean, Antagonistic roles for H3K36me3 and H3K27me3 in the cold-induced epigenetic switch at Arabidopsis FLC. Curr. Biol. 24, 1793–1797 (2014).
B. M. Zee et al., In vivo residue-specific histone methylation dynamics. J. Biol. Chem. 285, 3341–3350 (2010).
A. R. Gendall, Y. Y. Levy, A. Wilson, C. Dean, The VERNALIZATION 2 gene mediates the epigenetic regulation of vernalization in Arabidopsis. Cell 107, 525–535 (2001).
R. Bastow et al., Vernalization requires epigenetic silencing of FLC by histone methylation. Nature 427, 164–167 (2004).
C. Xu et al., R-loop resolution promotes co-transcriptional chromatin silencing. Nat. Commun. 12, 1790 (2021).
D.-H. Kim, S. Sung, Vernalization-triggered intragenic chromatin loop formation by long noncoding RNAs. Dev. Cell 40, 302–312.e4 (2017).
J. B. Heo, S. Sung, Vernalization-mediated epigenetic silencing by a long intronic noncoding RNA. Science 331, 76–79 (2011).
P. Zhu, M. Schon, J. Questa, M. Nodine, C. Dean, Causal role of a promoter polymorphism in natural variation of the Arabidopsis floral repressor gene FLC. Curr. Biol. 33, 4381–4391.e3 (2023).
R. Ye et al., Glucose-driven TOR–FIE–PRC2 signalling controls plant development. Nature 609, 986–993 (2022).
D. J. Gibbs et al., Oxygen-dependent proteolysis regulates the stability of angiosperm polycomb repressive complex 2 subunit VERNALIZATION 2. Nat. Commun. 9, 5438 (2018).
U. Amara, J. Hu, J. Cai, H. Kang, FLK is an mRNA m6A reader that regulates floral transition by modulating the stability and splicing of FLC in Arabidopsis. Mol. Plant 16, 919–929 (2023).
B. Sun et al., FIONA1-mediated methylation of the 3’UTR of FLC affects FLC transcript levels and flowering in Arabidopsis. PLoS Genet. 18, e1010386 (2022).
J. P. Uhler, C. Hertel, J. Q. Svejstrup, A role for noncoding transcription in activation of the yeast PHO5 gene. Proc. Natl. Acad. Sci. U.S.A. 104, 8011–8016 (2007).
F. Bardou et al., Long noncoding RNA modulates alternative splicing regulators in Arabidopsis. Dev. Cell 30, 166–176 (2014).
U. Chorostecki, N. G. Bologna, F. Ariel, The plant noncoding transcriptome: A versatile environmental sensor. EMBO J. 42, e114400 (2023).
E. Franco-Echevarría et al., Distinct accessory roles of Arabidopsis VEL proteins in Polycomb silencing. Genes Dev. 37, 801–817 (2023).

Information & Authors


Published in

Go to Proceedings of the National Academy of Sciences
Go to Proceedings of the National Academy of Sciences
Proceedings of the National Academy of Sciences
Vol. 121 | No. 4
January 23, 2024
PubMed: 38236739


Data, Materials, and Software Availability

Previously published data were used for this work (6). All other data are included in the manuscript and/or SI Appendix.

Submission history

Received: July 13, 2023
Accepted: December 11, 2023
Published online: January 18, 2024
Published in issue: January 23, 2024


  1. antisense transcription
  2. Polycomb Repressive Complex 2
  3. FLC
  4. Arabidopsis


We would like to thank members of the Dean lab for thoughtful inputs and Shuqin Chen for excellent technical assistance. This work was funded by the European Research Council Advanced Grant (EPISWITCH, 833254), Wellcome Trust (210654/Z/18/Z), and a Royal Society Professorship (RP\R1\180002) to C.D. and BBSRC Institute Strategic Programmes (BB/J004588/1 and BB/P013511/1) and Engineering and Physical Sciences Research Council/Biotechnology and Biological Sciences Research Council (EPSRC/BBSRC) Physics of Life grant (EP/T00214X/1) to M.H. and C.D.
Author Contributions
M.H., and C.D. supervised the research; M.N., Y.Z., E.M.-B., P.W., and S.Z. performed research; M.N., G.M., Y.Z., E.M.-B., P.W., and S.Z. analyzed data; G.M. undertook the modelling; and M.N., G.M., M.H., and C.D. wrote the paper.
Competing Interests
The authors declare no competing interest.


Reviewers: M.C., Institute of Plant Sciences Paris Saclay (IPS2); and J.C.L., University of Cambridge Sainsbury Laboratory.



Mathias Nielsen1
Department of Cell and Developmental Biology, John Innes Centre, Norwich NR4 7UH, United Kingdom
Govind Menon1
Computational and Systems Biology, John Innes Centre, Norwich NR4 7UH, United Kingdom
Yusheng Zhao
Department of Cell and Developmental Biology, John Innes Centre, Norwich NR4 7UH, United Kingdom
Eduardo Mateo-Bonmati
Department of Cell and Developmental Biology, John Innes Centre, Norwich NR4 7UH, United Kingdom
Philip Wolff
Department of Cell and Developmental Biology, John Innes Centre, Norwich NR4 7UH, United Kingdom
Shaoli Zhou
Department of Cell and Developmental Biology, John Innes Centre, Norwich NR4 7UH, United Kingdom
Computational and Systems Biology, John Innes Centre, Norwich NR4 7UH, United Kingdom
Department of Cell and Developmental Biology, John Innes Centre, Norwich NR4 7UH, United Kingdom


To whom correspondence may be addressed. Email: [email protected] or [email protected].
M.N. and G.M. contributed equally to this work.

Metrics & Citations


Note: The article usage is presented with a three- to four-day delay and will update daily once available. Due to ths delay, usage data will not appear immediately following publication. Citation information is sourced from Crossref Cited-by service.

Citation statements



If you have the appropriate software installed, you can download article citation data to the citation manager of your choice. Simply select your manager software from the list below and click Download.

View Options

View options

PDF format

Download this article as a PDF file


Get Access

Login options

Check if you have access through your login credentials or your institution to get full access on this article.

Personal login Institutional Login

Recommend to a librarian

Recommend PNAS to a Librarian

Purchase options

Purchase this article to get full access to it.

Single Article Purchase

COOLAIR and PRC2 function in parallel to silence FLC during vernalization
Proceedings of the National Academy of Sciences
  • Vol. 121
  • No. 4







Share article link

Share on social media