Natural temperature fluctuations promote COOLAIR regulation of FLC

In this study, Zhao et al. set out to characterize how plants respond to cold through regulation of FLC expression. Using genetics and genomics approaches, the authors reveal how natural temperature fluctuations promote COOLAIR regulation of FLC, with the first autumn frost acting as a key indicator of autumn/winter arrival.

As sessile organisms, plants have to extract specific temperature cues from fluctuating environments to time their developmental transitions. Knowledge of these mechanisms will be key to understanding the consequences of climate change. In Arabidopsis, a major determinant of seasonal flowering is the floral repressor FLC, which is epigenetically silenced by prolonged cold during vernalization (Michaels and Amasino 1999;Sheldon et al. 2000). Vernalization involves a multiphase cold-dependent silencing, where cold-induced transcriptional silencing precedes a low probability PRC2 epigenetic switching mechanism. The epigenetic switch is promoted by the PRC2-accessory protein VIN3, which is induced slowly, taking weeks to reach maximal levels (Sung and Amasino 2004). VIN3 accumulation requires absence of warm temperature spikes, in addition to long-term cold exposure Zhao et al. 2020). The transcriptional silencing involves cold-induced FLC antisense transcripts, COOLAIR, whose induction correlates with FLC transcriptional shutdown and switching of the chromatin states at FLC (Swiezewski et al. 2009;Csorba et al. 2014;Rosa et al. 2016). COOLAIR also functions to promote rapid cycling in plants in warm conditions in a cotranscriptional chromatin silencing mechanism that links promotion of proximal polyadenylation to FLC histone H3K4me1 demethylation (Costa and Dean 2019;Fang et al. 2020;Wu et al. 2020).
COOLAIR is highly conserved across different species (Castaings et al. 2014;Hawkes et al. 2016;Jiao et al. 2019) and has been shown to contribute to the variation in FLC regulation across natural Arabidopsis accessions (Shindo et al. 2006;Coustham et al. 2012;Li et al. 2014). Notably, a noncoding single-nucleotide polymorphism (SNP) alters COOLAIR splicing, increasing FLC transcription levels (Li et al. 2015). This mechanism involves the activator FRIGIDA protein and distally polyadenylated COOLAIR (Johanson et al. 2000;Geraldo et al. 2009;Li et al. 2018). The natural variation studies implicating a functional role for COOLAIR in FLC regulation have been followed up by transgene experiments designed to further explore the mechanism of COOLAIR action (Csorba et al. 2014;Wang et al. 2014;Rosa et al. 2016). However, in some cases, studies of transgenes aimed at attenuating antisense expression have concluded that COOLAIR expression is not required for vernalization (Helliwell et al. 2011;Li et al. 2018;Luo et al. 2019;Luo and He 2020). This led to the suggestion that COOLAIR functions in FLC regulation at warm temperatures but potentially not in the cold. In order to test this, we investigated the role of COOLAIR in vernalization, in both natural and laboratory conditions and in different accessions. Our study demonstrates the importance of COOLAIR-mediated FLC silencing in natural conditions, with the first seasonal freezing temperatures leading to COOLAIR-mediated FLC transcriptional silencing.

Result
COOLAIR shows variable temperature sensitivity and is highly up-regulated by freezing temperatures Through the analysis of >1000 worldwide Arabidopsis natural accessions, we previously identified five predominant FLC haplotypes defined by noncoding SNPs (Li et al. 2014). According to the varied vernalization responses measured by flowering time after growth at a constant 5°C, these haplotypes were classified into "rapid vernalizing" (RV) and "slow vernalizing" (SV) types (Li et al. 2014). To further investigate their vernalization response in field conditions, we selected representative accessions of these five major haplotypes (RV: Edi-0 and Col FRI SF2 ; SV: Var2-6, , as well as an extra SV accession, Löv-1, collected from the North Swedish field site (Duncan et al. 2015;Qüesta et al. 2020). To allow the comparison among these haplotypes, we generated new nearisogenic lines (NILs) by repeatedly backcrossing each FLC haplotype to Col FRI SF2 , our reference genetic background ("Col FRI") (Duncan et al. 2015;Li et al. 2015). Recent work has shown that these accessions and their associated NILs show variation in response to autumn cold in the field (Hepworth et al. 2020).
Here, we analyzed COOLAIR induction in field conditions to determine whether COOLAIR plays a role in natural conditions in different Arabidopsis accessions and their associated NILs.  Hepworth et al. 2018Hepworth et al. , 2020. Norwich generally has mild winters with few frosts, while in Ullstorp temperatures often fall below freezing, whereas the Ramsta site is usually snow-covered during part of the winter. The experiments ran from late summer/early autumn 2014 until spring 2015 and were repeated in North Sweden from late summer 2016 to the spring of 2017. Sowing times were adjusted to each site (earlier further north), and in the first year, two plantings were performed in North Sweden, a fortnight apart.
Three SV accessions (Löv-1, Ull2-5, Bro1-6) showed higher COOLAIR levels than the RV Col FRI across all the plantings (Supplemental Fig. S2). In contrast, Col-0 (which is a rapid cycler with no vernalization requirement for flowering) had very low levels of total COOLAIR in Norwich and at most time points in Sweden. COOLAIR levels in the RV accession Edi-0 and its NIL were very similar to Col FRI in Norwich, but in Sweden, especially North Sweden, COOLAIR was higher in Edi-0 than Col FRI. This was also the case for Var2-6, with early differences only in Sweden. These data suggest COOLAIR induction in the different accessions shows different temperature sensitivity.
We noticed a strong peak of expression of COOLAIR in all genotypes, even Col-0, in the second week of measurements (the 26th day after sowing) in the first North Sweden planting and the sixth week (the 59th day after sowing) in the South ( Fig. 1A; Supplemental Figs. S1, S2). These unusual peaks occurred when temperatures dipped below 0°C on the morning of the day of measurement (e.g., for South Sweden 2014, see

Recapitulation of the COOLAIR expression spike in temperature-controlled chambers
We confirmed the up-regulation of COOLAIR by freezing by reproducing the temperature profile of the week before the first peak in South Sweden in a growth cabinet. COOLAIR expression rose within an hour of experiencing freezing and peaked ∼8 h after freezing but returned to cool-temperature levels within 24 h (Supplemental Fig.  S3C). There was a small reduction in the level of FLC transcripts immediately after freezing exposure (Supplemental Fig. S3D,E), reflecting the coordination of antisensesense transcriptional circuitry. To further analyze the interrelationship of freezing, COOLAIR expression, and FLC expression, we grew seedlings for 2 wk in matched chambers at an average temperature of 5°C, but given either as a constant temperature or two different fluctuating temperature regimes repeated every 24 h (Fig. 1B). We saw the expected induction of proximal polyadenylation of COOLAIR by constant cold; however, this was significantly enhanced in the chamber where the fluctuations included a below-freezing period ( Fig. 1C; Supplemental Fig. S3F-I). Of note, this hyperinduction of COOLAIR is not just a consequence of a steep temperature drop, as COOLAIR induction was lower after transfer from nonvernalized (NV) 20°C to constant 5°C (Supplemental Fig.  S3J). All the temperature regimes led to a reduction in FLC expression, including the FLC mRNA and the unspliced transcript level (nascent transcript containing intron 2 and intron 3) used as a proxy for transcription (Fig. 1D,E). However, the enhanced COOLAIR induction in the freezing regime caused significantly lower FLC mRNA levels (Fig. 1E). The mechanism behind the different behavior of the unspliced transcripts and mRNA after exposure to freezing is currently unclear.
Genetic up-regulation of COOLAIR is associated with FLC transcriptional shutdown The complex relationship of the FLC down-regulation and the freezing-induced COOLAIR spike was intriguing with respect to the registration of external temperature and FLC regulation. In order to better understand the relationship between COOLAIR up-regulation and FLC repression, we sought to identify a mutant constitutively expressing high levels of COOLAIR. A forward genetic screen using a COOLAIRprom::luciferase reporter (Swiezewski et al. 2009;Sun et al. 2013), identified a dominant mutant, ntl8-D3, that had a high luciferase signal in warm conditions ( Fig. 2A,B). All endogenous COOLAIR transcripts, including both proximal (class I) and distal (class II) polyadenylated forms, showed ectopic up-regulation in ntl8-D3 ( Fig. 2C-E), with a relative increase in proximal polyadenylation as seen in wild-type plants after cold ( Fig.  2F; Supplemental Fig. S3I). Two separate mutant alleles, ntl8-D1 and ntl8-D2, and three overexpression transgenics, ntl8-OE1, ntl8-OE2, and ntl8-OE3, showed the same COOLAIR up-regulation (Supplemental Fig. S4A-F; Zhao et al. 2020). ntl8-D alleles cause a truncation in the NTL8 protein that deletes the C-terminal transmembrane domain and results in enhanced nuclear localization (Supplemental Fig. S4G; Zhao et al. 2020). NTL8 encodes a NAC domain transcription factor that directly binds VIN3 and the COOLAIR promoter ( Fig. 2G; Supplemental Fig. S4H; O'Malley et al. 2016;Xi et al. 2020;Zhao et al. 2020). Interestingly, the ntl8-D mutant alleles are also defective in long-term temperature regulation of VIN3 (Zhao et al. 2020), and the same logic that we have described for VIN3 accumulation-namely, reduced dilution of wild-type NTL8 protein during cold exposuremay be important in promoting long-term COOLAIR expression in autumn conditions. Previous analysis of ntl8-D2 vin3-6 double mutants revealed that the VIN3 overexpression, which also occurs in the ntl8-D mutant, was not necessary for the lower FLC expression before vernalization (Zhao et al. 2020). In order to better understand whether the higher COOLAIR expression was a reflection of more expression from the same cells or a higher fraction of cells expressing COOLAIR, we performed single-molecule RNA fluorescence in situ hybridization (smRNA FISH) on the ntl8-D3 mutant in warm conditions (Rosa et al. 2016). COOLAIR expression was found at high levels in all cells of the root (Fig. 2H,I); thus, ntl8-D3 had expanded the expression zone normally seen for COOLAIR, from just prevasculature cells to all cell types (Rosa et al. 2016). This is consistent with the reduced FLC-Venus signals in the ntl8-D3 mutant in warm conditions (Fig. 2J). This mutually exclusive transcription of COOLAIR and FLC supports the repressive effect of COOLAIR on FLC transcription from the same gene copy, agreeing with the previous study (Rosa et al. 2016).

COOLAIR is required for FLC transcriptional shutdown in natural and laboratory conditions
We then investigated how COOLAIR may mediate FLC shutdown by interrogating transgenic lines carrying FLC The TEX1.0 line was generated in a previous study (Csorba et al. 2014), and the TEX2.0 line was newly generated by inserting a NOS terminator to terminate COOLAIR transcription without disturbing the 3 ′ UTR of FLC (Fig. 3A). COOLAIR transcripts (detected by primers for total, class I, and class II.i) (Csorba et al. 2014) were greatly reduced in both the TEX1.0 and TEX2.0 lines (Fig. 3B). In natural conditions, the unspliced FLC transcript levels in both TEX1.0 and TEX2.0 reduced more slowly than in wildtype controls (transgenic FLC-WT and nontransgenic Col FRI) during vernalization (Fig. 3C), supporting a role for COOLAIR in mediating FLC transcriptional shutdown. Moreover, the FLC mRNA level was accordingly higher in TEX2.0 but not in TEX1.0 (Fig. 3D), likely due to the change in the 3 ′ UTR causing instability of the FLC mRNA in TEX1.0 (Csorba et al. 2014). Consistently, the same FLC expression in TEX1.0 and TEX2.0 and lateflowering phenotypes were also observed in laboratory conditions ( Fig. 3E-G), reinforcing the view that COOL-AIR plays a role in FLC transcriptional shutdown.
Given that the freezing-enhanced COOLAIR induction led to significantly lower FLC mRNA levels ( Fig. 1E), these two TEX lines were also tested in the temperature regimes of Figure 1B. Similar to the observation in Figure  1, COOLAIR was more induced in wild-type plants in the freezing regime compared with the other two temperature regimes (Fig. 3H), and FLC mRNA levels in the freezing regimes were accordingly lower (Fig. 3I). In contrast, such lower FLC mRNA levels associated with freezing temperature were not observed in either of the TEX lines, where COOLAIR induction is greatly reduced (Fig. 3H,I). This supports the view that COOLAIR is responsible for the enhanced decrease in FLC mRNA levels in the freezing regimes. To further determine the causality of COOLAIR up-regulation and FLC down-regulation, we generated a double mutant of ntl8-D3 with TEX2.0, as well as a double mutant of ntl8-OE3 with a COOLAIR promoter deletion line, FLC ΔCOOLAIR (Luo et al. 2019). In both double mutants, there is no longer COOLAIR up-regulation by NTL8 ( Fig. 3J; Supplemental Fig. S4I-L). We found that FLC down-regulation in ntl8-D3 and ntl8-OE3 is substantially suppressed by the COOLAIR knockout (Fig. 3K,L; Supplemental Fig. S4M). Therefore, COOLAIR up-regulation is a major causal factor for the FLC down-regulation in the ntl8-D or ntl8-OE mutants, supporting a direct role of COOLAIR or COOLAIR transcription in FLC transcriptional shutdown.
Transgenes aimed at knocking out COOLAIR promote production of novel convergent antisense transcripts During the analysis of the TEX lines, it became clear that distally polyadenylated COOLAIR was still produced (Fig. 3B). This was unexpected given the presence of the upstream terminator sequence. To understand the origin of the COOLAIR distal amplicon in the TEX1.0 and TEX2.0 lines, 5 ′ RACE (rapid amplification of cDNA ends) was performed (Supplemental Fig. S5A). In wild- type plants, grown without (NV) or with 2 wk of cold treatment (2WV), two major spliced distal COOLAIR forms with a range of transcriptional start sites were identified (Fig. 4A,B), consistent with the previous study (Swiezewski et al. 2009). In addition, a low abundance convergent antisense transcript (hereafter referred to as CAS as previously defined by Kindgren et al. [2020]) pos-sessing a 5 ′ cap was identified with a transcriptional start site within FLC exon 1 (Fig. 4A,B; Supplemental Table S1). In the TEX1.0 line, antisense transcription was found to initiate within the RBCS terminator fragment, producing a distal COOLAIR with the same splice sites as on the endogenous locus (Fig. 4A,B; Supplemental Table S1). In the TEX 2.0 line, no COOLAIR transcripts were detected except for the CAS transcript initiating within FLC exon 1 (Fig. 4A,B). Interestingly, additional CASs containing a 5 ′ cap were identified in both TEX lines originating from different positions inside FLC intron 1 (Fig. 4A,B; Supplemental Table S1). The multiple transcriptional start sites of CAS (Fig. 4B) are likely to contribute to the higher antisense expression in TEX2.0 (Figs. 3B, 4C). The proximally polyadenylated COOLAIR (class I) present at very low levels in both TEX lines is cold induced (Figs. 3B, 4C) and originates from the residual FLC fragment in the flc-2 background, rather than the TEX transgene (Supplemental Fig. S5B). Thus, the major COOLAIR-CAS transcripts in the TEX lines are novel CAS and are not cold induced (Fig. 4C). Given the low frequency of CAS in the wild type, their transcription is likely to be suppressed by COOLAIR transcription from the upstream native promoter. We further investigated three other transgenic lines (FLC + MAF2-T, FLC + NOS-T, and FLC ΔCOOLAIR ) that had been designed to remove COOLAIR from FLC (Li et al. 2018;Luo et al. 2019). The FLC + MAF2-T and the FLC + NOS-T lines were generated by replacing the COOLAIR promoter with the MAF2 terminator and NOS terminator, respectively (Li et al. 2018), while the FLC ΔCOOLAIR line was generated by deleting a 324-bp region in the COOLAIR promoter, downstream from FLC 3 ′ UTR, using the CRISPR method (Luo et al. 2019). The novel CAS transcripts originating from FLC intron 1 described above were detected in all three lines, with an even higher level in FLC + MAF2-T and FLC + NOS-T (Supplemental Fig. S5B-D). In addition, similar to TEX1.0, antisense transcript start sites were detected in the MAF2-T terminator in the FLC + MAF2-T line (Sup-plemental Fig. S5D). Similar findings of cryptic promoter usage after disruption of upstream promoters, and alternative transcripts arising when antisense transcription is perturbed, have been reported in S. cerevisiae (Mayer et al. 2015;Kim et al. 2016). The generation of novel COOLAIR-CAS transcripts from intragenic regions in FLC at each attempt to remove COOLAIR, including the CRISPR deletion of the endogenous COOLAIR promoter (FLC ΔCOOLAIR ), suggests a tight interconnection between antisense transcription and chromatin state at the locus.

Discussion
Plants effectively use fluctuating temperature cues to judge seasonal progression, but the molecular mechanisms underlying this are poorly understood. Through field studies of different Arabidopsis genotypes, we have shown how a temperature dip below freezing hyperinduces transcription of antisense transcripts at the floral repressor locus FLC to facilitate transcriptional silencing (Fig. 5). Since in natural field conditions, colder weather generally follows the first autumn frost, the hyperinduction of COOLAIR by freezing may be one of the many cues used by plants to monitor seasonal progression. The evolutionary significance of this remains to be explored. The ability to respond to acute or ambient cold temperature may provide the plasticity important for adaptation to different climates. This would then have parallels to environmentally regulated gene regulation in yeast, where antisense transcription has been shown to induce faster and higher amplitude changes in the associated gene in response to environmental cues (Xu et al. 2011;Beck et al. 2016;  Figure 3A. Cloutier et al. 2016). Subsequent frosts show weaker and variable effects on COOLAIR expression ( Fig. 1; Supplemental Figs. S1, S2), likely due to the epigenetic silencing of the whole locus by the VIN3-dependent Polycomb switching mechanism that occurs after the cold-induced transcriptional silencing Hepworth et al. 2018).
How such small changes in temperature-from just above to below freezing-cause such a strong up-regulation of COOLAIR expression remains to be determined. An R loop generated by the invasion of COOLAIR into the DNA duplex at the COOLAIR promoter limits further rounds of antisense transcription (Sun et al. 2013). Whether freezing temperature alters the biophysical behavior of this structure to enable strong up-regulation of expression is an interesting possibility (Sun et al. 2007). NTL8 may also interface with the R-loop to influence COOLAIR expression. We have recently found that NTL8 is a direct regulator of VIN3, providing long-term cold information to VIN3 through the mechanism of reduced dilution from slower growth (Zhao et al. 2020). It would seem likely that NTL8 provides long-term cold information to COOLAIR (Fig. 5), but whether NTL8 or its homologs are also involved in the response of COOLAIR to acute freezing temperature remains to be addressed.
How COOLAIR transcription influences the transcriptional output of FLC remains to be resolved, but the ectopic transcription of COOLAIR in the ntl8-D mutant demonstrated causality between up-regulation of COOLAIR and FLC down-regulation at the same gene copy (Fig. 2). Antisense transcription has been shown to alter sense transcription dynamics in a chromatin-dependent manner in yeast and human cells (Murray et al. 2015;Brown et al. 2018). This appears to also be the case for FLC as removal of COOLAIR disrupted the synchronized replacement of H3K36me3 with H3K27me3 at the intragenic FLC nucleation site during the cold (Csorba et al. 2014). It is also possible that COOLAIR transcription reduces functionality of trans factors or gene/intronic loops that promote FLC transcription (Crevillén et al. 2013;Li et al. 2018).
The failed attempts with a range of transgenes to completely remove COOLAIR helps explain the confusion on COOLAIR function. All the independent transgenes analyzed produce new antisense transcripts initiating in the first intron. Consistent with this, recent genome-wide measurements of TSSs showed that extensive alternative intragenic transcriptional initiation occurs in Arabidopsis, and this can be affected by cold (Kindgren et al. 2018), cotranscriptional RNA degradation (Thieffry et al. 2020;Thomas et al. 2020), or mutants disrupted in chromatin signaling (Nielsen et al. 2019;Le et al. 2020). The activation of alternative TSSs significantly influences transcription from nearby TSSs, and thus is important for plant development and adaptation to environmental changes (Kindgren et al. 2018;Nielsen et al. 2019;Le et al. 2020;Thieffry et al. 2020;Thomas et al. 2020). Such CAS transcripts have been found to initiate globally from promoter-proximal exon-intron boundaries across the Arabidopsis genome and are correlated with promoter-proximal RNA Pol II stalling, a checkpoint for transcriptional regulation (Adelman and Lis 2012;Core and Adelman 2019;Kindgren et al. 2020). Interestingly, the major COOLAIR CAS transcripts originate in a region encompassing the alternative splice site of the distal polyadenylated COOLAIR class II.iv. This isoform has higher abundance in the natural accession Var2-6 and is associated with increased FLC expression through a cotranscriptional mechanism involving capping of the FLC nascent transcript (Fig. 5; Li et al. 2015). Moreover, the slower FLC shutdown rate in Var2-6 in the field supports the importance of this checkpoint in FLC transcription regulation (Hepworth et al. 2020). Similar to other systems (Lenstra et al. 2015), these CAS transcripts are likely suppressed at the endogenous locus by COOLAIR transcription from the upstream native promoter. This suppression may be influenced by the cold promotion of proximal 3 ′ processing/polyadenylation of COOLAIR. This would parallel FLC silencing in warm conditions, promoted by the alternative 3 ′ processing of COOLAIR (Liu et al. 2010;Marquardt et al. 2014;Fang et al. 2020). We envision that these CAS transcripts might be differentially expressed when COOLAIR transcription is altered, for example, during the initial freezing-dependent (1) Natural low temperatures (indicated by graded blue) promote NTL8 slow accumulation (graded green arrow), providing the long-term cold information for COOLAIR upregulation and FLC regulation. Frosts function as strong cues to enhance COOLAIR up-regulation, conferring transcriptional plasticity to FLC. Furthermore, proximal polyadenylation of COOLAIR is enhanced by natural low temperatures, including freezing, supporting the mechanism tightly linking the altered 3 ′ processing/polyadenylation of COOLAIR to the transcriptional state at the FLC locus (Liu et al. 2010;Marquardt et al. 2014;Fang et al. 2020).
(2) CAS transcription is inhibited by transcription from the upstream COOLAIR promoter, possibly influenced by the cold-promoted proximal polyadenylation of COOLAIR.
(3) CAS originates from the region (shaded in pink) where the class II.iv exon is alternatively spliced (Li et al. 2015), highlighting its importance in the COOLAIR regulation of FLC transcription.
COOLAIR spike, possibly contributing to the switch of the local chromatin/transcription states. It will be informative in the future to capture the dynamics of these low abundant CAS transcripts during vernalization. Our current understanding of the many ways COOLAIR regulates FLC is shown in Figure 5.
Overall, our work reveals how first frost acts as a seasonal cue to up-regulate COOLAIR and transcriptionally repress FLC. Such a temperature-regulated antisensesense circuitry endows a transcriptional plasticity to the FLC locus to effectively respond to the natural fluctuating temperatures of autumn. On a different timescale, but also cold-induced, the Polycomb nucleation mechanism then locks in the FLC silenced transcriptional state to maintain the epigenetic memory of cold exposure. Dissection of the vernalization response has thus elucidated the mechanisms used by plants to translate natural temperature fluctuations into long-term seasonal information.

Materials and methods
No statistical methods were used to predetermine sample size. Field experiments were randomized in a complete-block design as described (Hepworth et al. 2020), with sample size chosen on the basis of feasibility and to buffer against sample loss. Laboratory experiments were not randomized and investigators were not blinded to allocation during experiments and outcome assessment. Sampling in all cases was performed by collecting material from new plants (not repeated sampling) for replicates and also between time points.
The COOLAIRprom::luciferase reporter line was generated by transforming the COOLAIRprom-luciferase construct, as previously described (Sun et al. 2013), into Col FRI plants. A transgenic line containing a single-copy transgene was selected as the progenitor line (named as CTL) for the forward genetic screen.
Field experiments Details for field experiments have been described previously (for 2016-2017winter, see Antoniou-Kourounioti et al. 2018for winter, see Hepworth et al. 2018for both 2016for both -2017for both and 2014for both -2015for both winters, see Hepworth et al. 2020. Briefly, the experiment site in the north was at Ramsta Seeds were sown on plates with selective antibiotic and MS agar media without glucose. Plants were grown for 10 d postgermination in long days at 20°C and then grown for an additional 3 d before NV sampling or moved to a growth room for vernalization with 8-h light and grown for 2 or 4 wk (2WV or 4WV) at 5°C, before transferring seedlings to soil and growing for 20 d at 22°C/20°C with 16-h light/8-h darkness (4WT20). Three replicates of >15 seedlings (NV, 2WV, and 4WV) or three replicates of three pooled leaf/meristem tissue (4WT20) were screened for expression.
Flowering time was analyzed as previously described (Liu et al. 2010). Briefly, plants were vernalized for 4 wk at 5°C with 8-h light before being transferred to soil at 22°C/20°C with 16-h light/8-h darkness. The days plants took to bolting when flower buds were visible at the shoot apical meristem were counted as a measurement of flowering time. Counting was stopped after 50 d as all plants had flowered for controls and TEX1.0 and nothing for TEX2.0.
Recreation of field freezing conditions in the laboratory (Supplemental Fig. S3C-E) Plants were grown in a nonvernalizing growth chamber in long days for 1 wk at 22°C/20°C 16-h light/8-h darkness and then transferred to a growth chamber (Conviron) with the temperatures and light period matching the week before the first frost (to allow acclimation) and the day following the morning frost. Temperatures and light times are shown in Supplemental Table  S2. Three biological replicates of more than five seedlings were sampled at 17:00 and 20:00 on the seventh day, the night before freezing; at 07:00, 09:00, 11:00, 13:00, 15:00, 17:00, and 20:00 on the eighth day, when freezing occurred at 10:00 and 13:00-14:00; and at 20:00 on the following day. Col FRI seeds were sown onto soil. Nonvernalized plants were grown for 10 d in the nonvernalizing growth chamber and sampled at six time points over 24 h before plants were then moved to 8-h light/16-h darkness growth chambers for vernalization Light was on from 09:00 to 17:00. For vernalization, plants were treated with three different temperature regimes: constant 5°C, daily fluctuating 5°C (3°C-9°C) and daily fluctuating 5°C (−1°C to 12°C) (Fig. 1B). Three biological replicate samples were taken at six time points over 24 h after 2-wk exposure to vernalization. For the NV samples, plants were sampled at 12:00, 16:00, 20:00, and 24:00 and at 08:00 twice 24 h apart. For the vernalization samples, plants were sampled at 12:00, 16:00, 20:00, and 24:00 and at 09:00 twice 24 h apart. Lights came on at midnight for NV and at 09:00 in vernalization treatments. For Figure 3, H and I, experiments were performed as described above, except sampling was performed at the time point (09:00). For Supplemental Figure S3, F-J, experiments were performed following the above procedures, except that Col FRI seeds were sown onto medium in Petri dishes and were sampled at the time point (09:00).
Mutagenesis, genetic screening, and gene cloning of ntl8-D3 mutation Mutagenesis was carried out following the procedures previously described (Liu et al. 2010). Around 20 M2 (mutagenesis generation 2) seeds from each single M1 plant were screened by being sown on MS medium and stratified for 3 d in the cold (5°C). After growing in a growth cabinet for 10 d, the M2 seedlings were assayed for the bioluminescence with 1 μM luciferin (Promega E1603) under a CCD camera (NightOwl). A mutant was identified to also show high VIN3 expression in warm conditions (Supplemental Fig. S4N), similar to what we found in ntl8-D1 and ntl8-D2 mutants (Zhao et al. 2020). Further Sanger sequencing showed that the mutant carries a mutation in AT2G27300 and so was named ntl8-D3 (Supplemental Fig. S4G,O).

RNA extraction and QPCR
Total RNA was extracted as previously described (Box et al. 2011). Genomic DNA was digested with TURBO DNA-free (Ambion Turbo DNase kit AM1907) according to the manufacturer's guidelines, before reverse transcription was performed. The reverse transcription was performed with the SuperScript III reverse transcriptase (ThermoFisher 18080093) following the manufacturer's protocol using gene-specific primers. Relevant primers are listed in Supplemental Tables S3 and S4. For field experiments, RNA extraction and QPCR were performed as described Hepworth et al. 2018Hepworth et al. , 2020. In brief, analysis of qPCR results was performed with LinReg with normalization to the geometric means of the At5g25760 ("PP2A") and At1g13320 ("UBC") control genes. The same analysis was also used in Figure 1, C-E, and Supplemental Figure S3, C-E. The rest results were normalized to single reference gene, UBC. Primers used are described in Supplemental Table S3.

Microscopy
For detecting the fluorescence of FLC-Venus, confocal imaging was performed using a 20×/0.7 NA multi-immersion lens, with water as the immersion fluid on a Leica TCS SP8 X confocal microscope following the procedures in Berry et al. (2015). Roots were immersed in 2 μg/mL propidium iodide (Sigma-Aldrich P4864) to label the cell wall. FLC-Venus was excited with illumination at 514 nm (Argon ion laser). Emissions from Venus were detected between 518 nm and 555 nm using a cooled Leica HyD SMD detector in photon-counting mode. Propidium iodide was detected simultaneously with FLC-Venus by collecting emissions between wavelengths 610 nm and 680 nm. To allow comparison between treatments, the same settings were used for images in both the wild type and ntl8-D3 mutant.
The smRNA FISH experiment was performed following the procedures described previously (Rosa et al. 2016).

′ RACE
Five micrograms of total RNA was first treated with calf intestine alkaline phosphatase (CIAP; Merk Sigma P4978) for 1 h at 37°C and purified with phenol:chloroform. After 5 ′ cap removing for 1 h at 37°C with cap-clip acid pyrophosphatase (CCAP; Cambio C-CC15011H), 1 μL of 5 ′ RACE adapter 0.3 μg/μL; (5 ′ -GCU GAUGGCGAUGAAUGAACACUGCGUUUGCUGGCUUUG AUGAAA-3 ′ ) was ligated to a half volume of the CIAP/CCAPtreated RNA (2.5 μL) by T4 RNA ligase (NEB M0204S) for 2 h at 37°C. The CIAP-treated RNA without following CCAP treatment was used as control. All the 5 ′ RACE adapter-ligated RNA was then reverse transcribed by SuperScript III reverse transcriptase (ThermoFisher 18080093) with a distal COOLAIR-specific primer, RT-R, with TUB6 as a control (Fig. 4B). The cDNA was submitted to nested PCR, and the subsequently purified PCR products were ligated to a T-vector (Supplemental Fig. S5A). At least 48 colonies from each RNA sample were examined by PCR before at least two colonies from each different sized PCR product were sent to Sanger sequencing. Oligos for 5 ′ RACE are listed in Supplemental Table S4.