Plants overwinter before flowering in Arabidopsis through FRI-dependent upregulation of FLC. This mechanism requires FRIGIDA-LIKE 1 (FRL1), FRIGIDA-ESSENTIAL 1 (FES1), SUPPRESSOR OF FRIGIDA 4 (SUF4) and FLC EXPRESSOR (FLX), which associate in a FRI complex8. In laboratory conditions of constant cold, FLC is transcriptionally repressed in around two to three weeks5,6,11. However, in natural autumnal fluctuating temperatures in field conditions, FLC transcriptional shutdown took many months3,4. We therefore investigated whether and how FRI function might change in response to fluctuating temperature.

FRI condensates accumulate in the cold

First, we analysed how cold influences FRI protein interactions. A line carrying a translational FRI-GFP fusion, expressed at the same level as endogenous FRI, and fully complementing the fri early flowering phenotype (Extended Data Fig. 1), was used for immunoprecipitation in combination with mass spectrometry (IP–MS)10 (Supplementary Table 1). FRI–GFP did not enrich any of the FRI complex components from warm-grown plant extracts, but, counterintuitively, FRI–GFP accumulated in plants that underwent two weeks of cold exposure, and in these conditions FRL1 and FRL2—but not FLX, SUF4 or FES1—were enriched8 (Extended Data Table 1). FRI interacted with subunits of the Mediator complex12,13, WDR5a and ATX2 (which promote H3K4me3)7,9,14, the PAF1 complex15, general transcription factors13, RNA-polymerase-II-associated proteins, and many RNA splicing factors and uridine-rich small nuclear ribonucleoproteins (snRNPs)16,17,18, which suggests that FRI has a role in co-transcriptional regulation (Extended Data Table 1). The higher prevalence of these interactors in extracts from cold-grown plants may reflect the cold-induced accumulation of FRI (Extended Data Table 1).

Second, we analysed the localization of FRI in vivo. Like many other co-transcriptional regulators19,20 we found that FRI–GFP forms nuclear condensates, which were increased in size and number after cold exposure (Fig. 1a, Extended Data Fig. 2a–d). A FRI–Myc fusion showed similar condensate formation (Extended Data Fig. 2e, f). Fluorescence recovery after photobleaching (FRAP) revealed relatively slow FRI–GFP dynamics (Fig. 1b, c), like other biomolecular condensates20, but different to FCA–GFP, which have previously been shown to form liquid-like foci (condensates) with a FRAP recovery time of seconds21. Formation of the FCA–GFP condensates was not enhanced by cold (Extended Data Fig. 2g–i); thus, FRI and FCA condensates have different biophysical properties.

Fig. 1: Cold-promoted FRI nuclear condensates are linked to FLC transcriptional shutdown.
figure 1

a, Confocal microscopic images of FRI–GFP nuclear condensates in root cells in the indicated conditions. For quantitative analysis, see Extended Data Fig. 2c, d. b, c, Images (b) and quantification (c) of FRAP of FRI–GFP nuclear condensates. Time 0 indicates the time of the photobleaching pulse. Red arrows indicate the bleached condensates. Mean ± s.e.m.; n = 10 condensates in 10 cells. d, Confocal analysis of subnuclear colocalization of FRI with the co-expressed proteins in tobacco leaf nuclei. Data represent three independent experiments. e, f, Representative images (e, in the warm; f, in the cold) of nuclei expressing FRI–GFP (green) sequentially hybridized with intronic smFISH probes for FLC (red). DNA was labelled with DAPI (blue). Three independent experiments gave the same conclusion. g, h, Frequency distribution of FRI–GFP condensates (left), non-spliced FLC transcript signals (middle) and their colocalization per nucleus (right) in root cells in the warm (g) and in the cold (h). Numbers of analysed nuclei are as indicated. Scale bars, 5 μm (a, b, d); 10 μm (e, f).

Source data

We then analysed whether the FRI interactors co-associate with FRI in the condensates. FRL1–mScarlet I colocalized with FRI–GFP in nuclear condensates after transient co-expression in tobacco leaves (Fig. 1d) but did not form nuclear condensates when transfected alone (Extended Data Fig. 3a–c). Consistently, loss of FRL1 reduced the cold-induced enhancement of the size and number of FRI condensates (Extended Data Fig. 3d–f). By contrast, FRI condensates were less affected in flx-2 and suf4 mutants (Extended Data Fig. 3d–f), consistent with FRL1, but not FLX or SUF4, immunoprecipitating with FRI (Extended Data Table 1). Other FRI interactors, GFP–ELF7 and TAF15b–GFP, colocalized with FRI–mScarlet I in condensates (Fig. 1d, Extended Data Fig. 3g) and condensates containing the Cajal body marker protein U2B′′ (ref. 18) frequently colocalized with FRI–GFP condensates after cold (Extended Data Fig. 3h–j). FRI condensates may therefore share components with Cajal bodies, similar to PML bodies in mammalian cells22. Overall, these data support the notion that multiple protein interactions enhance the formation of FRI condensates.

We further investigated the domains of FRI that are required for nuclear condensate formation. Consistent with disordered domains having important roles in biomolecular condensation19,21, a GFP fusion protein with a version of FRI in which the C-terminal disordered domain was deleted no longer formed nuclear condensates (Extended Data Fig. 4). This supports the C terminus being required for FRI function23. Deletion of the FRI coiled-coil domains10 also led to loss of condensate formation (Extended Data Fig. 4b, c). A naturally occurring 16-base-pair deletion that is found in the loss-of-function FRI allele in Col-010,23, which would produce a protein without the C-terminal disordered and coiled-coil domains, prevented condensate formation (Extended Data Fig. 4b, c). Thus, both the C-terminal disordered domain and two coiled-coil domains are required for FRI condensation in vivo.

FRI is more stable in the cold

Increased protein concentration is known to promote biomolecular condensation19,20 so we further examined the cold-induced accumulation of FRI. FRI–GFP accumulated after two to four weeks of cold (Extended Data Fig. 5a–c), with no corresponding change in the levels of FRI mRNA (Extended Data Fig. 1a). Analysis of nuclear FRI–TAP showed that this stability was not a consequence of the GFP fusion (Extended Data Fig. 5d). The FRI half-life was measured as less than 24 h in the warm, but more than 24 h in the cold (Extended Data Fig. 5e–h); thus, the cold enhancement of FRI nuclear condensates appears to be a consequence of increased FRI protein stability.

Overexpression using 35S:FRI-GFP (Extended Data Fig. 6a, b) led to a small accumulation of FRI–GFP protein after two weeks of cold exposure and a decrease after four weeks (Extended Data Fig. 6c–e), although FRI is less stable in warm than in cold conditions (Extended Data Fig. 6f). A previous report showed that FRI abundance decreased in the cold24, which may be a consequence of the medium as FRI protein is highly induced by glucose in warm conditions (Extended Data Fig. 6f). Overexpression of FRI–GFP is likely to influence nuclear condensate dynamics25, but one to four weeks of cold still enhanced FRI–GFP condensation even though more FRI condensates were found in warm-grown 35S:FRI-GFP lines compared to FRI-GFP (Extended Data Fig. 6g–i).

Levels of FRI–GFP protein decreased in frl1-1 and suf4 mutants, with no concomitant change in FRI-GFP mRNA8 (Extended Data Fig. 6j–l). By contrast, FRI–GFP protein levels in flx-2 did not change (Extended Data Fig. 6k, l), despite the disruption to FRI condensation (Extended Data Fig. 3d–f), supporting the notion that increased protein concentration is not sufficient for cold enhancement of FRI condensate formation. Together, these data show that cold stabilization of FRI and increased interaction with multiple factors reciprocally enhance the formation of FRI condensates.

Condensates sequester FRI away from FLC

We further addressed the functional consequences of FRI nuclear condensate formation. FRI expanded the zone of FLC expression in warm-grown plants and this was antagonized by cold exposure26 (Extended Data Fig. 7a, b). Association of FRI–GFP with the FLC 5′ region in warm conditions8,9 decreased after two weeks of cold exposure (Extended Data Fig. 7c–f), paralleling the decrease in FLC transcription5,11,26 (Extended Data Fig. 7g) and raising the possibility that condensates are linked to FLC transcriptional repression. We used single-molecule RNA fluorescence in situ hybridization (smRNA FISH) and FLC intron 1 probes27 to identify nascent FLC transcripts that mark transcriptionally active FLC loci and investigated whether FRI nuclear condensates associated with an active FLC locus. The nascent FLC transcripts and the FRI–GFP condensates never colocalized with each other (Fig. 1e–h). Notably, FRI association with the COOLAIR promoter increased after two weeks of cold exposure (Extended Data Fig. 7e), consistent with COOLAIR upregulation5,11,26. However, despite more cells transcribing COOLAIR in the cold26 and more FRI condensates, we did not detect any colocalization between non-spliced COOLAIR clouds and FRI–GFP condensates (Extended Data Fig. 7h). FRI and its associated factors may therefore be sequestered into nuclear condensates away from the FLC locus, with nascent transcription dissolving any condensates at the locus28.

Supporting this hypothesis, in the 35S:FRI-GFP line, in which more FRI nuclear condensates form (Extended Data Fig. 6g–i), FLC transcript levels are lower (Extended Data Fig. 7i, j) and a reduction of FRI nuclear condensates in the frl1-1 mutant (Extended Data Fig. 3d–f) correlates with a slower FLC transcriptional shutdown (Extended Data Fig. 7k–m). The low levels of FLC transcript in flx-2 and suf4 suggest a different mechanism (Extended Data Fig. 7k–m). FLX and SUF4 appear to promote the ability of FRI to transcriptionally activate, but they have less of a role in cold-induced FRI condensation (Extended Data Fig. 3d–f).

FRI condensates track temperature shifts

FLC transcriptional shutdown occurs during autumn as temperatures fluctuate widely over daily and weekly timescales3,4, so we tested whether the cold-induced sequestration of FRI has an important role in fluctuating temperatures. FRI–GFP nuclear condensate dynamics were analysed during a 12-h transient cold exposure (mimicking a cool night in autumn)3. We found that FRI–GFP condensates gradually increased in size and number (Fig. 2a, Extended Data Fig. 8a, b), through a process requiring protein synthesis (Extended Data Fig. 5e–h). The accumulation of FRI condensates correlated with downregulation of FLC to its lowest expression 12 h after transfer to cold (Fig. 2b), which was further repressed after 2 weeks of cold exposure (Extended Data Fig. 8a–c). Thus, FRI condensation increases rapidly in response to decreasing temperature, correlating with FLC transcriptional shutdown.

Fig. 2: Short-term temperature fluctuations influence the formation of FRI nuclear condensates and FLC transcription.
figure 2

a, c, Confocal microscopic images of FRI–GFP nuclear condensates in wild-type (WT) (top) and TEX (bottom) root cells after 0, 6 and 12 h of cold treatment (a) and in wild-type plants after they were returned to the warm for 0, 6, 12 and 24 h after a 2-week cold treatment (c). Scale bars, 5 μm. For quantitative analysis, see Extended Data Fig. 8a, b, d, e. b, d, Relative transcript level of unspliced FLC in the indicated plants within the same time course of changed temperatures in a, c by quantitative PCR with reverse transcription (RT–qPCR). Mean ± s.e.m.; n = 4 (b) and 3 (d) biologically independent experiments.

Source data

It has previously been observed that a spike of high temperature in autumn slows FLC shutdown3,4; thus, we asked whether the cold-induced accumulation of FRI condensates is reversible by warmth. FRI–GFP condensates, monitored at 6-h intervals over the first 24 h after plants were returned to warm temperature, were significantly reduced within the first 6 h and did not recover in either root or leaf cells (Fig. 2c, Extended Data Fig. 8d–f). This process was not fully blocked by inhibiting proteasome-mediated protein degradation (Extended Data Fig. 8g–j). FLC transcription was upregulated in parallel (Fig. 2d) and showed further reactivation after transfer to warm conditions for 10 days (Extended Data Fig. 8k). Therefore, the cold-induced condensation of FRI is easily reversed by warmth, which suggests that the condensates serve as reservoirs allowing a rapid response to warm temperature spikes. These rapid dynamics would buffer the shutdown of FLC transcription in the fluctuating temperatures of autumn, contributing to a requirement for the absence of warmth for FLC silencing3.

To gain further insight into the dynamics of FRI nuclear condensates in response to temperature shifts, we performed a time-lapse experiment using a temperature-controlled microscope stage. This showed that FRI–GFP nuclear condensates undergo dynamic changes in response to changing temperature, disappearing within five hours of a return to warm conditions (Extended Data Fig. 8l, Supplementary Video 1), but being rescued by cold after a three-hour warm spike (Extended Data Fig. 8m, Supplementary Video 2). In response to the temperature shifts, the FRI condensates fluctuated in number as they grew and fused (Extended Data Fig. 8n, o, Supplementary Video 2)—behaviour typical of biomolecular condensates21,28,29. This induction of FRI condensates did not occur in transient assays in tobacco leaves exposed to cold temperatures (Extended Data Fig. 8p–r).

COOLAIR promotes FRI condensates in the cold

To investigate which other cold-specific factors might contribute to the heterotypic interactions required for condensate formation we tested a role for COOLAIR. One COOLAIR isoform is differentially induced by the FRI complex after two weeks of cold (Extended Data Fig. 9a–f): class II.ii, a distally polyadenylated transcript that includes an additional exon5,11 (Fig. 3a). This change in splicing may involve FRI interacting with splicing factors (Extended Data Fig. 3h–j, Extended Data Table 1) that have previously been shown to have a role in cold-responsive gene regulation16,17. RNA immunoprecipitation (RNA-IP) showed that FRI–GFP specifically enriched COOLAIR class II.ii after two weeks of cold (Fig. 3b, Extended Data Fig. 9g, h). In frl1-1, in which the FRI condensation is severely attenuated (Extended Data Fig. 3d–f) but COOLAIR expression is still relatively high (Extended Data Fig. 9a–c), this enrichment was reduced (Extended Data Fig. 9i). Therefore, the FRI–class II.ii interaction is tightly connected with the cold induction of FRI condensation—a conclusion supported by the rapid changes in class II.ii in response to temperature shifts (Fig. 3c, d) and further induction after two weeks of cold (Extended Data Fig. 9f).

Fig. 3: COOLAIR promotes cold induction of FRI–GFP nuclear condensates and sequestration of FRI from the FLC promoter.
figure 3

a, Schematic of FLC and COOLAIR transcripts at the FLC locus. Untranslated regions are indicated by grey boxes and exons by black boxes. kb, kilobase; TSS, transcription start site. b, RNA-IP assay of spliced COOLAIR enrichment by FRI–GFP with UBC as control. Mean ± s.d.; n = 4 replicates over 2 biologically independent experiments. Two-tailed t-test. c, d, Relative transcript level of COOLAIR class II.ii in the indicated plants within the same time course of changed temperatures as in Fig. 2 by RT–qPCR. Mean ± s.e.m.; n = 4 (c) and 3 (d) biologically independent experiments. e, Confocal images of wild-type and TEX root tip nuclei expressing FRI–GFP. Scale bars, 5 μm. For quantitative analysis, see Extended Data Fig. 9j, k. f, FRI–GFP occupancy on FLC promoter region in WT and TEX plants by CHIP. Mean ± s.e.m.; n = 3 biologically independent experiments. The exact distance from TSS referred to a. Two-way ANOVA adjusted by Sidak’s multiple comparisons test. NS, no significance. g, A working model for temperature-controlled FRI nuclear condensation in FLC transcriptional regulation. CC, coiled-coil domain; CR, co-transcriptional regulators; DD, disordered domain.

Source data

We next crossed FRI–GFP to a previously described FLC terminator exchange (TEX) line5,26,27. The FLC terminator–COOLAIR promoter is exchanged for an RBCS3B terminator, which prevents the cold induction of COOLAIR transcription5,26 (Fig. 3c). There was no increase in the size of FRI condensates within the first 12 h of plants experiencing cold (Fig. 2a, Extended Data Fig. 8a, b), and both the size and the number of FRI condensates were significantly reduced after two weeks of cold (Fig. 3e, Extended Data Fig. 9j, k). This is linked to decreased levels of FRI protein in TEX, which might reflect negative feedback from reduction of COOLAIR expression (Extended Data Fig. 9l, m). Similarly, the lower cold induction of FRI condensation in flx-2 and suf4 mutants (Extended Data Fig. 3d–f) may relate to reduced COOLAIR expression (Extended Data Fig. 9a–f). In addition, chromatin immunoprecipitation (ChIP) experiments showed that the cold-induced reduction of FRI occupancy at the FLC promoter was less in the TEX line compared to the wild type (Fig. 3f), which could account for the inefficient cold-induced FLC transcriptional shutdown in TEX5,26 (Fig. 2b). This mechanism could explain the COOLAIR-facilitated removal of H3K36me3 at the FLC locus in the early vernalization phase5 (Fig. 3g).


Together, our experiments point to a temperature-controlled condensation mechanism that modulates FRI activation of FLC transcription; such a mechanism would facilitate FLC shutdown in natural fluctuating temperatures. FRI associates with transcriptional co-activators and recruits histone modifiers to the FLC promoter, establishing an active transcriptional state at FLC in the warm (Fig. 3g). The instability of FRI in the warm enables a fast turnover of these transcriptional complexes (Fig. 3g). After transfer to the cold, FRI protein is stabilized and changes in protein or COOLAIR interactions result in the accumulation of nuclear condensates that sequester FRI away from the FLC promoter (Fig. 3g). This contrasts with mechanisms in which condensates promote the dwell time of transcriptional regulators at specific genomic loci19. The combinatorial effect of temperature-specific splicing, and splicing-specific nuclear condensation, gives a wide range of possibilities for FLC regulation in different temperature regimes. Given the recognized importance of biological condensates in gene regulation, this type of mechanism may have widespread roles in the plasticity of plant responses to varying environments29,30,31.


Plant material and growth conditions

All the mutants and transgenic lines were in Columbia-0 (Col-0), except for FCA-GFP and FLC-GUS, which are in Landsberg erecta (Ler). TEX FRI (ref. 5), FCA-GFP (ref. 21), frl1-1 (ref. 32), flx-2 (ref. 33), suf4 (ref. 34) and FLC-GUS with or without FRIJU223 (ref. 35) have previously been described. FRIJU223 is a functional transgenic FRI allele as described previously10. FRI is Col-0 with an introgressive active Sf2 FRI allele and was previously described36. frl1-1, flx-2 and suf4 were crossed into the FRI background to generate frl1-1 FRI, flx-2 FRI and suf4 FRI.

Seeds were surface-sterilized and sown on Murashige and Skoog (MS) agar plates without glucose or sucrose (unless otherwise stated) and kept at 4 °C in the dark for 3 days. For non-vernalization (NV), seedlings were grown in warm conditions (16 h light, 8 h darkness with constant 20 °C) for 10 days unless otherwise stated. For cold treatment, seedlings were pre-grown in warm conditions (16 h light, 8 h darkness with constant 20 °C) for 10 days unless otherwise stated, and then transferred to cold (8 h light, 16 h darkness with constant 5 °C) for a long-term vernalization, such as 2 weeks and 4 weeks (2WV and 4WV). As there is no difference between FRI and fri in response to photoperiod34 and photoperiod influences flowering in parallel with FLC37,38, we use this short-day photoperiod for long-term cold treatment to match natural conditions. For short-term cold treatment (0–12 h), the light conditions were kept the same as in warm conditions. For cold-to-warm transfer experiments, after 1 or 2 weeks of cold treatment, seedlings were moved to warm conditions (16 h light, 8 h darkness with constant 20 °C) for another specified duration, such as 6 h. Plants grow faster in the warm than in the cold (one day growth in the warm is approximately equivalent to seven days in the cold)39, therefore for warm–cold comparisons plants were harvested at the same developmental stage rather than after the same time period.

Plasmid construction and generation of transgenic lines

To generate pFRI:FRI-GFP and pFRI:FRI-Myc constructs, a 4.56-kb PmII/HpaI fragment containing genomic FRI sequence from H51 (ref. 10) was cloned into pBluescript KS (+). This clone spans 0.84 kb upstream of the FRI start codon to 1.4 kb downstream of the FRI stop codon. GFP sequence was amplified and inserted into the XbaI site over the FRI stop codon. The FRI-GFP fragment was then subcloned into SLJ7551540 and transformed into Agrobacterium tumefaciens C58 (pGV2260) by triparental mating36. Myc sequence was inserted into the XbaI site 6 bp downstream of the FRI start codon. The FRI-Myc fragment was then subcloned into the EcoRI site of pGreenII 0229 and transformed into A. tumefaciens pGV3101 (pMP90). All of the XbaI sites were introduced by site-directed mutagenesis by the oligos listed in Supplementary Table 2. Transgenic FRI-GFP and FRI-Myc plants in the Col-0 background were generated by floral dipping with A. tumefaciens. Transgenic lines with a single insertion were identified in the T2 generation and segregated 3:1 for Basta resistance. Independent FRI-GFP lines showed the same cold-induced condensation.

For the pFRI:FRI-TAP and p35S:FRI-GFP constructs, FRI cDNA was fused with the TAP and GFP sequences, respectively, and FRI-TAP was subcloned into pBluescript KS (+) with the FRI promoter, whereas FRI-GFP was driven by a CaMV35S promoter. pFRI:FRI-TAP and p35S:FRI-GFP in the binary vectors were then transferred into A. tumefaciens. Transgenic plants were also in the Col-0 background.

The 35S:GFP-ELF7 (ref. 41), pTAF15b:TAF15b-GFP (ref. 42) and pCsV (Cassava vein mosaic virus):TAF15b-GFP (ref. 42) constructs were previously described.

To generate pFRI:FRI-mScarlet I, the same FRI genomic sequence from H51 (ref. 10) was used as a template and mScarlet I sequence was inserted to the XbaI site over FRI stop codon in pBluescript KS (+). The FRI-mScarlet I fragment was then subcloned into SLJ699940. For pFRI:FRI-Col-0-GFP, an equivalent FRI genomic sequence from Col-0 was amplified10 and the GFP sequence was fused before the premature stop codon (after 1,335 bp downstream from ATG, equivalent to 314 amino acids). For pFRI:FRI-DD-GFP, the C-terminal disordered domain (DD) from 1,833 bp downstream from ATG to the stop codon (equivalent to 451 to 609 amino acids) was deleted. For pFRI:FRI-CC-GFP, two fragments encoding the coiled-coil domains (CC) are deleted: the first spans 118 bp to 300 bp downstream from ATG (equivalent to 40 to 100 amino acids) and the second is from 1,680 bp to 1,832 bp downstream from ATG (equivalent to 400 to 451 amino acids). For pFRL1:FRL1-mScarlet I, a 3,972-bp fragment containing FRL1 genomic sequence from Col-0 was amplified. It includes 1,502 bp upstream from FRL1 ATG and 1,057 bp downstream from FRL1 TAG. mScarlet I sequence was inserted before the TAG translational stop codon. FRI-Col-0-GFP, FRI-DD-GFP, FRI-CC-GFP and FRL1-mScarlet I DNA fragments were then cloned into the SLJ7551540 destination vector by Gateway Cloning. Plasmids were transformed into A. tumefaciens C58 (pGV2260) for infiltration of Nicotiana benthamiana21. All primers used for construction are listed in Supplementary Table 2.

IP–MS analyses

The IP–MS analyses were performed as described previously21,36. For the first experiment (IP1) 2.5 g of seedlings was cross-linked in 1% formaldehyde as previously described43. For the second (IP2) and third experiment (IP3), the amount of plant material used was doubled. Cross-linked plants were ground into fine powder and lysed in 50 ml of cell lysis buffer (20 mM Tris HCl, pH 7.5, 250 mM sucrose, 25% glycerol, 20 mM KCl, 2.5 mM MgCl2, 0.1% NP-40, 5 mM DTT). The lysate was filtered through two layers of Miracloth (Merck, D00172956) and pelleted by centrifugation. The pellets were washed twice with 10 ml of nuclear wash buffer (20 mM Tris HCl, pH 7.5, 2.5 mM MgCl2, 25% glycerol, 0.3% Triton X-100, 5 mM DTT) and resuspended with RIPA buffer (1× PBS, 1% NP-40, 0.5% sodium deoxycholate, 0.1% SDS). All the buffers were supplemented with 0.1 U μl−1 RNaseOUT (Invitrogen, 10777019), 1 mM PMSF and Roche Complete tablets to keep the integrity of any RNA–protein and protein–protein complex. Nuclear lysate was sonicated three times (Diagenode Bioruptor; M level, 30 s ON–30 s OFF, 5 min each time) and nuclear supernatant was obtained after centrifuge. GFP–Trap magnetic agarose beads (25 μl) (Chromotek, GTMA-20) were pre-blocked with 1 mg ml−1 BSA and 1 mg ml−1 yeast total RNA (Sigma) in RIPA buffer for 1 h at 4 °C and incubated with the nuclear supernatant for 2 h at 4 °C. Beads were sequentially washed twice with high-salt wash buffer and low-salt wash buffer (20 mM Tris HCl, pH 7.5, 500 mM (high-salt) or 150 mM NaCl (low-salt), 0.5 mM EDTA, 0.1% SDS and 1% Triton X-100). The beads were finally resuspended in 1× SDS loading buffer and boiled at 95 °C for 15 min to reverse cross-linking. The protein samples were purified from 10% SDS–PAGE gels followed by trypsin digestion. For LCMS analysis, an Orbitrap Fusion Tribrid mass spectrometer (Thermo Fisher Scientific) equipped with an UltiMate 3000 RSLCnano LC system (Thermo Fisher Scientific) was used. Data were searched using Mascot server 2.7 (Matrix Science). Results were imported and evaluated in Scaffold 4.10.0 software (Proteome Software).

RNA expression analysis

RNA was extracted with the hot phenol method5,36. Genomic DNA contamination was removed by TURBO DNA-free kit (Invitrogen, AM1907) following the manufacturer’s guidelines. cDNA was synthesized by the SuperScript IV reverse transcriptase (Invitrogen, 18090050) using gene-specific reverse primers. All primers are listed in Supplementary Table 2. The TAIR (The Arabidopsis Information Resource) accession numbers for the genes analysed in this study are FRI (AT4G00650), FLC (AT5G10140) and COOLAIR (AT5G01675). Standard reference genes PP2A (AT1G13320) and UBC (AT5G25760) for gene expression were used for normalization4. Data were analysed using Microsoft Excel (v.2102, 64-bit) and GraphPad Prism 7.

Microscopy and image quantification

The seedlings used for imaging were grown in the following conditions: NV, plants were grown in warm conditions for 7 days; 1WV, plants were grown in warm conditions for 6 days before being transferred to cold conditions for 1 week; 2WV or 4WV, plants were grown in cold conditions for 2 or 4 weeks with a 5-day pre-growth in warm conditions; before a short cold treatment (12 h or shorter), plants were grown in warm conditions for 7 days, so that all of the seedlings imaged are developmentally equivalent to a 7-day-old warm-grown seedling. The warm and cold conditions are the same as described in ‘Plant material and growth conditions’ unless otherwise stated.

The subcellular localization of FRI–GFP in root tips and first true leaves (Extended Data Figs. 2a, b, 8f) was imaged on a Zeiss LSM780 confocal microscope using a 40×/1.2 water objective and GAsP spectral detector array. The seedlings were kept intact during imaging. GFP was excited at 488 nm (Argon ion laser, laser power: 5.0 %) and detected at wavelengths of 491–695 nm in lambda mode. Linear unmixing projection in ZEN Black (2012) software was applied and the autofluorescence detected at 526–695 nm was unmixed and labelled as background (blue). Images were exported by ZEN Blue (2012).

For quantitative analysis of FRI–GFP and FCA–GFP nuclear condensates, imaging was performed on a Zeiss LSM880 confocal microscope using a 40×/1.1 water objective and GAsP spectral detector array. GFP was excited at 488 nm (Argon ion laser, laser power: 5.0 %) and detected at wavelengths of 499–525 nm in lambda mode. An optimal z-step size of 0.61 μm was used over a total depth of 10.37 μm from the upper surface (18 z-slices) and maximum intensity projection was applied for the z-stack. A binary image of just the spots was then created through threshold in Fiji (ImageJ) and the spot area was subsequently measured by Analyze Particles in Fiji (ImageJ). As all images were obtained with a 0.11 μm pixel size, the area of a single pixel is 0.0121 μm2. To exclude single pixels, 0.02 μm2 was set as the minimum size during the analysis. The spot number (with area larger than 0.02 μm2) per nucleus was gained through displaying the spot outline generated by Analyze Particles in the original image. Violin plots reflecting the data distribution were produced with PlotsOfData44. Statistical evaluations with multiple comparisons tests were performed using the GraphPad Prism 7 software.

To compare the fluorescence intensity between roots, imaging was performed with a Zeiss LSM780 confocal microscope using an EC Plan-Neofluar 20×/0.50 objective. To allow comparison between treatments, the same settings were used for all images. Measurement of the fluorescence intensity was conducted with Fiji (ImageJ) with the intensity normalized to the background for each image. To collect the total intensity of each root tip, sum slices projections from a z-stack of 16 steps with a step size of 2 μm were applied before analysis.


FRAP of FRI–GFP nuclear condensates was performed as described45 with a Zeiss LSM780 confocal microscope coupled with a Linkam Heating and Cooling Stage (Meyer instruments). A chamber was created on slides using Grace Bio-Labs Secure Seal adhesive sheets (Sigma, GBL620001) and filled with MS medium. The 2WV FRI-GFP seedlings were carefully transferred into the chamber with stage temperature kept at 4 °C. Using a 40×/1.2 water objective, the region of an FRI–GFP nuclear condensate was bleached using a laser intensity of 100% at 488 nm. Recovery was recorded for every 2 min for a total of 20 min after bleaching. At each time point, maximum intensity projections from a z-stack of 12 steps with a step size of 1 μm were applied. Analysis of the recovery curves was carried out with Fiji (ImageJ), Microsoft Excel (v.2102, 64-bit) and GraphPad Prism 7.

Time-lapse imaging

The time-lapse microscopy of FRI–GFP nuclear condensates was performed with a Zeiss LSM780 confocal microscope coupled with a Linkam Heating and Cooling Stage (Meyer instruments). For Supplementary Video 1, 1WV seedlings were transferred into a same chamber as in FRAP experiment and observed under a 40×/1.2 water objective. GFP was excited at 488 nm (Argon ion laser, laser power: 5.0 %) and detected at wavelengths of 490–551 nm. The temperature was first set at 4 °C and images were acquired every 15 or 20 min for 5 h after the temperature rose to 22 °C. At each time point, maximum intensity projections from a z-stack of 12 steps with a step size of 1 μm were applied. For Supplementary Video 2, 1WV seedlings were imaged after being grown in warm conditions (20 °C) for 3 h with the stage temperature kept at 4 °C, and images were acquired every 20 min for 6 h. The same imaging settings were used as in Supplementary Video 1. Images were processed and FRI–GFP nuclear condensates were quantified with Fiji (ImageJ).

Cycloheximide and MG132 treatment

Cycloheximide (CHX) and MG132 treatment were according to a previous study39. Before treatment, plants were grown on MS agar plates as described above. For treatment, seedlings were carefully transferred to new MS agar plates supplemented with either 100 μM CHX (C1988, Sigma-Aldrich) or 100 μM MG132 (474787, Sigma-Aldrich) growing for 24 h. In parallel, the same seedlings were transferred to new MS agar plates supplemented with the same amount of solvent (ethanol for CHX and dimethyl sulfoxide for MG132) as a control. Root tips were then imaged with a Zeiss LSM780 confocal microscope using a 40×/1.2 water objective. GFP was excited at 488 nm (Argon ion laser, laser power: 20.0 %) and detected at wavelengths of 489–530 nm. An optimal z-step size of 0.435 μm was used over a total depth of 11.75 μm from the upper surface (28 z-slices). Images were processed and measurements were conducted with Fiji (ImageJ). Sum slices projection was applied for the z-stack. The whole nuclear intensity was measured and normalized to background in each image before comparison. All images were obtained with a 0.1-μm pixel size and 0.02 μm2 was set as the minimum size during the analysis of FRI–GFP nuclear condensates. Statistical evaluations with multiple comparisons tests were performed using the GraphPad Prism software.

smFISH and immunofluorescence

smFISH and immunofluorescence was performed as previously described26,46. Plants were grown in warm conditions for seven days (NV) or in cold conditions for two weeks with a five-day pre-growth in warm conditions (2WV). For FRI–GFP fluorescence microscopy with sequential smFISH, root tips were fixed with 4% paraformaldehyde (PFA) before being squashed. After permeabilization in 70% ethanol for 1 h, the subnuclear localization of FRI–GFP was imaged and the stage positions were saved at the microscope. Next, cover slips were carefully unmounted and FLC intron 1 or COOLAIR intron probes labelled with Quasar 570 were hybridized at 37 °C overnight. The probes were the same as previously described26,27. The probe signals were detected using the same stage positions as FRI–GFP. The microscope used was Zeiss Elyra PS with a 100×/1.46 oil-immersion objective and a cooled electron multiplying CCD (charge-coupled device) Andor iXon 897 camera. GFP was exited at 488 nm and detected at wavelengths of 495–550 nm; probes labelled with Quasar 570 were excited at 561 nm and detected at 570–620 nm; DAPI was excited at 405 nm and detected at 420–480 nm. An optimal z-step size of 0.2 μm was used over a total depth of 2.4 μm (23 z-slices) and maximum intensity projection was applied for the z-stack using Fiji (ImageJ).

For FRI–Myc immunofluorescence and colocalization of FRI–GFP and U2B′′, squashed root cells were immersed in 70% ethanol overnight at 4 °C for permeabilization. Cell walls were digested with 0.2% Driselase (Sigma, D9515) and 0.15% Macerozyme R-10 (Duchefa Biochemie, M8002) in 1× PBS at 37 °C for 40 min. After being washed three times with 1× PBST, cells were blocked with 2.5% BSA (Thermo Fisher Scientific, AM2616). Primary antibodies: 1: 125 diluted anti-c-Myc (Sigma, M5546), 1: 500 diluted anti-GFP (Abcam, ab290) and 1: 20 diluted anti-U2B′′ (4G3, a gift from P. Shaw) were incubated at 37 °C for 3–4 h. Alexa Fluor 488 anti-rabbit secondary antibody (Thermo Fisher Scientific, A-11008, dilution: 1:200) and Alexa Fluor 555 anti-mouse secondary antibody (Thermo Fisher Scientific, A-21424, dilution: 1:200) were incubated for 1 h. After DAPI staining, slides were mounted in Vectashield (Vector Lab, H-1000). Images were acquired on a Zeiss LSM780 confocal microscope using a 63×/1.40 oil objective. For FRI–Myc immunofluorescence, DAPI was excited at 405 nm and detected at 410–513 nm; FRI–Myc-Alexa Fluor 555 was excited at 514 nm and detected at 545–697 nm. For FRI–GFP and U2B′′ colocalization, DAPI was excited at 405 nm and detected at 407–489 nm; FRI–GFP–Alexa Fluor 488 was excited at 488 nm and detected at 491–561 nm; U2B′′–Alexa Fluor 555 was excited at 561 nm and detected at 562–634 nm.

GUS staining

FLC–GUS staining was performed as previously described35. Whole seedlings were vacuum-infiltrated with GUS staining buffer (1 mM X-gluc (5-bromo-4-chloro-3-indolyl glucuronide), 100 mM phosphate buffer, pH 7.0, 10 mM EDTA, 0.1% Triton X-100, 0.5 mM ferricyanide and ferrocyanide, pH was finally adjusted to 7.0) followed by an overnight or shorter incubation at 37 °C. Whole seedlings and roots were immediately imaged after de-staining with 95% ethanol.

ChIP and quantitative PCR

FRI–GFP ChIP was performed as previously outlined with some modifications1,36. Five grams of NV seedlings and 2.5 g of 2WV seedlings were cross-linked in 1% formaldehyde and nuclei were purified and lysed as in the IP–MS experiment. Anti-GFP (Abcam, ab290, dilution: 1:400) and protein A agarose beads containing salmon sperm DNA (Millipore, 16-157) were used in the immunoprecipitation. The enriched DNA was purified using the ChIP DNA Clean & Concentrator Kit (Zymo Research, D5205)43. All ChIP experiments were quantified by quantitative PCR (qPCR) with appropriate primers (Supplementary Table 2). The enrichment levels of FRI–GFP at ACTIN (ACT) and SHOOT MERISTEMLESS (STM) were used as controls. A transgene carrying a wild-type FLC crossed with FRI–GFP was used as control for FRI–GFP ChIP in TEX.

Western blot analysis

For checking FRI–GFP enrichment in the ChIP experiment, beads were directly boiled in 1×SDS loading buffer at 95 °C for 15 min after immunoprecipitation. For detecting FRI–GFP and FRI–TAP protein levels in NV, 2WV and 4WV plants, nuclei were purified with the same protocol of IP–MS but from 0.3 g of non-cross-linked seedlings. Whole seedlings were grounded and lysed to get total protein extraction for detecting FRI–GFP protein levels in 35S:FRI-GFP plants. ChIP elution and the extracted nuclear and total proteins were separated on NuPAGE 4–12% Bis-Tris protein gel (Invitrogen, NP0321BOX) and transferred to a PVDF membrane (GE Healthcare Life Sciences). Antibodies against GFP (Roche, 11814460001, dilution: 1:2,000), TAP (Thermo Fisher Scientific, CAB1001, dilution: 1:1,000), H3 (Abcam, ab1791, dilution: 1:5,000) and tubulin (Merck Sigma-Aldrich, T5168, dilution: 1:4,000) were used as primary antibodies. Horseradish peroxidase (HRP)-conjugated secondary antibodies: anti-mouse IgG (GE, NA931, dilution: 1:20,000) and anti-rabbit IgG (GE, NA934, dilution: 1:20,000) were used for protein detection with chemiluminescent substrate (Thermo Fisher Scientific, 34095). Western blot signal was captured in ImageQuant LAS 500.

RNA-IP assay

Five grams of NV seedlings and 2.5 g of 2WV seedlings were used for each immunoprecipitation. Nuclei were extracted and purified from cross-linked plant material with the same procedure as in the IP–MS experiment. The pellet was resuspended in 1 ml of nuclear lysis buffer (50 mM Tris HCl, pH 7.5, 100 mM NaCl, 1% Triton X-100, 1 mM MgCl2 and 0.1 mM CaCl2) and incubated at 37 °C for 10 min with Turbo DNase (Invitrogen, AM1907)47. Then SDS was added to a final concentration of 0.1% and NaCl to 150 mM. Following another hour incubation at 4 °C, nuclear lysate was cleared by centrifugation. A 100-μm quantity of the supernatant was saved as the input. Twenty-five microlitres of pre-blocked GFP–Trap magnetic agarose beads (Chromotek, GTMA-20) were incubated with the nuclear supernatant following the same immunoprecipitation and wash procedure as in IP–MS experiment. All the buffers used were supplemented with 0.1 U μl−1 RNaseOUT (Invitrogen, 10777019), 1 mM PMSF and Roche Complete tablets. Then RNA was eluted and purified as previously described21. After another DNA digestion with Turbo DNase, reverse transcription with gene specific primers was performed by SuperScript IV reverse transcriptase. Primers for reverse transcription and qPCR are listed in Supplementary Table 2. RNA enrichment was analysed in Microsoft Excel (v.2102, 64-bit) and GraphPad Prism 7. RNA enriched in NV FRI-GFP was normalized to NV FRI whereas 2WV FRI-GFP was normalized to 2WV FRI to reduce any possible influence from COOLAIR expression variation.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this paper.