The non-coding RNA SVALKA locus produces a cis-natural antisense transcript that negatively regulates the expression of CBF1 and biomass production at normal temperatures

Non-coding transcription is present in all eukaryotic genomes, but we lack fundamental knowledge about its importance for an organism’s ability to develop properly. In plants, emerging evidence highlights the essential biological role of non-coding transcription in the regulation of coding transcription. However, we have few molecular insights into this regulation. Here, we show that a long isoform of the long non-coding RNA SVALKA-L (SVK-L) forms a natural antisense transcript to the host gene CBF1 and negatively regulates CBF1 mRNA levels at normal temperatures in the model plant Arabidopsis thaliana. Furthermore, we show detailed evidence for the specific mode of action of SVK-L. This pathway includes the formation of double-stranded RNA that is recognized by the DICER proteins and subsequent downregulation of CBF1 mRNA levels. Thus, the CBF1-SVK regulatory circuit is not only important for its previously known role in cold temperature acclimation but also for biomass production at normal temperatures. Our study characterizes the developmental role of SVK-L and offers mechanistic insight into how biologically important overlapping natural antisense transcripts can act on and fine-tune the steady-state levels of their host gene’s mRNA.


INTRODUCTION
Climate change is increasing global temperatures, and future winters are thus predicted to be milder. This change is harmful to plants, as it disrupts their ability to acclimate to cold temperatures (Shepherd, 2016). The key players in the plant cold acclimation process are the C-repeat binding factors (CBFs) (Thomashow, 1999). Expression of the neighboring genes CBF1-3 is transiently upregulated within minutes of cold exposure, and their products are responsible for activating the early steps of cold acclimation (Medina et al., 1999). CBF1, one of the three cold-induced CBFs, has been extensively studied for its role in cold acclimation in a wide range of plant species (Jaglo-Ottosen et al., 1998;Benedict et al., 2006;Ito et al., 2006;Badawi et al., 2007;Zhang et al., 2011).
The role of CBF1 in cold protection is well established. However, part of its regulation at cold temperatures has only recently been described (Kindgren et al., 2018). The long non-coding RNA (lncRNA) SVALKA (SVK) is transcribed downstream and from the antisense strand of CBF1. SVK has the potential to be a cisnatural antisense transcript (NAT) if transcription proceeds over the CBF1 gene body. NATs can form double-stranded RNA with the coding mRNA when both RNAs are present in the same cell. This can be recognized by the RNA silencing machinery and processed into small interfering RNAs (siRNAs) (Bologna and Voinnet, 2014). cis-NATs are a common feature of Arabidopsis genes. A recent study that adapted native elongation transcript sequencing (plaNET-seq) to Arabidopsis found that over 30% of all expressed genes have an associated long non-coding cis-NAT (Kindgren et al., 2019). Remarkably, considering that cis-NATs have been known for a long time, we know little about their general function (Bologna and Voinnet, 2014). Studied examples of cis-NATs have often found idiosyncratic mechanisms with few overlapping components (Zhang et al., 2012(Zhang et al., , 2013Yuan et al., 2015;Reis and Poirier, 2021). Still, a common working hypothesis is that cis-NATs are especially important in regulating developmental and environmental decisions by the plant (Zhang et al., 2013).
Indeed, at cold temperatures, SVK negatively fine-tunes the transcriptional output of CBF1 via an RNA Polymerase II (RNAPII) collision mechanism due to extreme cold-induced transcriptional activity over the CBF1 gene body (Kindgren et al., 2018). Collisions occur when CBF1 expression has reached its peak after 2 to 4 h of cold exposure, resulting in premature termination and rapid degradation of CBF1 transcripts (Kindgren et al., 2018). This intricate lncRNA-mediated regulation highlights the importance of maintaining output from the CBF1 gene at a controlled level throughout the cold response.
Interestingly, CBF1 regulation is not restricted to cold acclimation. CBF1 is regulated under normal growth conditions by other environmental stimuli, including light, plastid, and circadian signals (Zarka et al., 2003;Fowler et al., 2005;Noré n et al., 2016;Jiang et al., 2020). Moreover, the CBF1 mRNA was shown to be posttranscriptionally regulated and to have the highest turnover of any mRNA in Arabidopsis at 22 C (Sorenson et al., 2018). A possible hint about the importance of maintaining finely tuned CBF1 expression levels comes from the fact that both overexpression and knockdown of CBF1 result in the same stunted growth phenotype (Jaglo-Ottosen et al., 1998;Dong et al., 2020). Thus, earlier findings pinpoint CBF1 as a key player under both normal and cold temperatures, although why plants invest considerable amounts of energy maintaining CBF1 mRNA levels under such tight surveillance is an outstanding question.
In this study, we expand our knowledge of CBF1 regulation mediated by non-coding SVK transcription and its role in plant development and growth. We show that the regulatory cascade of SVK antisense transcription on CBF1 sense transcription is a dynamic process, active at normal temperatures, with a mechanism distinct from that at cold temperatures. At 22 C, SVK transcription navigates the antisense strand of CBF1 to form doublestranded RNA (dsRNA) with the sense mRNA, and this decreases the half-life of the CBF1 mRNA. Impaired SVK expression has a biological role and leads to increased biomass production at 22 C due to mis-regulation of CBF1. Our study identifies important players in the cis-NAT-host gene regulatory pathway and demonstrates that CBF1 is a major plant growth regulator.

SVK transcription traverses the CBF1 gene body at normal temperatures
To understand the mechanisms behind the rapid turnover of CBF1 at 22 C, we first corroborated the degradation kinetics of the CBF1 mRNA with an orthogonal technique. We used plant Native Elongation Transcript Sequencing (plaNET-seq) data (Kindgren et al., 2019) over CBF1 and its neighboring gene CBF3 coupled with direct RNA-sequencing data (DRS). DRS detects stable polyadenylated RNAs (Supplemental Figure 1A), whereas plaNET-seq detects nascent actively transcribed RNAs. plaNET-seq showed that CBF1 and CBF3 were actively transcribed at similar levels. However, DRS-seq showed that CBF3 had a prominent peak at the predicted poly(A) (polyadenylation) site, whereas CBF1 barely had any detectable peak. This suggests that CBF1 is highly transcribed, but its mRNA is rapidly degraded at 22 C compared with the CBF3 mRNA (Supplemental Figure 1A). A possible explanation is the presence of an antisense transcript derived from transcription of the lncRNA SVALKA (SVK). Indeed, we observed extensive SVK transcription throughout the CBF1 gene body at 22 C using the plaNET-seq data ( Figure 1A). Low transcriptional activity in the region would decrease the probability of RNAPII collisions at 22 C compared with 4 C (Kindgren et al., 2018). Thus, if SVK has any regulatory potential to fine tune the expression levels of CBF1 at 22 C, it is likely to involve a mechanism distinct from that of CBF1-SVK regulation at 4 C.
To investigate the role of SVK at 22 C, we isolated a line that harbors a transfer DNA (T-DNA) insertion in the CBF1 promoter, which limits CBF1 expression without affecting steady-state levels of SVK (cbf1-3, Figure 1B-1D, Supplemental Figure 1B).
In T-DNA lines, large inserts of DNA disrupt the genomic context of the insertion site (Alonso et al., 2003). We also included the uns-1 T-DNA mutant that uncouples the expression of CBF1 and SVK, and the svk-1 T-DNA mutant that knocks out SVK (Figure 1B-1D) (Kindgren et al., 2018). Interestingly, CBF1 steady-state levels increased in the uns-1 and svk-1 mutants ( Figure 1C), indicating that SVK may indeed have a regulatory role toward CBF1 at 22 C. SVK steady-state levels were unaffected in cbf1-3 and increased in uns-1 ( Figure 1D). The common feature of svk-1 and uns-1 is that SVK transcription cannot proceed into the CBF1 gene body in both mutants (Kindgren et al., 2018). To further confirm a cis role for SVK at 22 C, we used firefly LUCIFERASE (LUC) reporter constructs. In one construct ( p CBF1::CBF1-LUC-T NOS ), LUC was fused between the coding sequence of CBF1 and the T NOS terminator. The T NOS terminator was used to abolish any antisense transcription (Kindgren et al., 2018). In the other construct ( p CBF1::CBF1-LUC-SVK), LUC was fused between CBF1 and the endogenous SVK sequence, enabling SVK transcription along the LUC gene body (Kindgren et al., 2018). We used homozygous stable transgenic lines from two independent transformation events for each construct. LUC activity was lower in lines that included SVK than in those with the T NOS terminator (Supplemental Figure 1C). Thus, these results strongly suggest that SVK transcription along the CBF1 gene body negatively regulates the CBF1 gene output at 22 C.
The main 5 0 end of SVK at 22 C is situated 1669 base pairs (bp) downstream of the transcription start site of CBF1 (Kindgren et al., 2018). To further characterize the long isoform of SVK (SVK-L) at 22 C, we used DRS-seq to detect the 3 0 end of the transcript (Supplemental Figure 1D). We found two poly(A)signal intervals, one proximal for the short SVK isoform (SVK-S) that is dominant at 4 C (Kindgren et al., 2018) and a longer distal poly(A) interval that ends close to the 5 0 end of CBF1 (À432 to +77 with respect to the TSS of CBF1, Supplemental Figure 1D and 1E). Based on nascent transcription data, we estimated that about 75% of SVK transcription extends to the distal poly(A) signal at 22 C (Supplemental Figure 1F), a result that is consistent with the level of poly(A) transcripts detected by DRS-seq. By contrast, after 12 h at 4 C, nascent transcription levels indicated that only about 3% of SVK transcription extends to the distal poly(A) signal (Supplemental Figure 1G and 1H). This corroborates earlier results from 4 C (Kindgren et al., 2018) and indicates that SVK-L is the dominant isoform of SVK at 22 C.
Our results demonstrate that SVK transcription negatively regulates CBF1 mRNA levels and that SVK-L is predominant at 22 C and may be involved in CBF1 regulation.

SVK-L modulates CBF1 mRNA stability
At 4 C, readthrough transcription from the dominant proximal poly(A) site of SVK is quickly degraded by the nuclear exosome (Kindgren et al., 2018). To investigate whether SVK-L is under nuclear exosome surveillance at 22 C, we measured the steadystate level of SVK with two qPCR probes in the wild type and the rrp4-2 mutant. (Figure 2A). RRP4 is an exosome core subunit and is present in both nuclear and cytosolic exosome complexes (Lange et al., 2014). Both probes showed an increased SVK level in rrp4-2, indicating that it is an exosome target at 22 C ( Figure 2B and 2C). Interestingly, and in contrast to 4 C, SVK is not a HEN2 target at 22 C (Supplemental Figure 2A-2C). HEN2 is an exosome subunit that associates only with the complex in the nucleoplasm (Lange et al., 2014). This suggests that RNA products of SVK transcription along the CBF1 gene body are degraded by exosome complexes with distinct compositions in a temperature-dependent manner. The intriguing degradation kinetics of CBF1 and SVK led us to investigate their half-lives with a cordycepin assay (Fedak et al., 2016). Cordycepin inhibits transcription, and after its incorporation during RNA synthesis, it is possible to follow the degradation of specific transcripts. In wild-type seedlings, CBF1 displayed a half-life of 97.5 ± 2.5 min and SVK of 66.7 ± 3.8 min ( Figure 2D). These values fall between the half-life of the stable mRNA of EUKARYOTIC INITIATION FACTOR 4A-1 (EIF4A1) (>120 min) and the highly unstable transcript of EXPANSIN L1 (EXPL1, 51 ± 9.6 min), which were used as assay controls. Our cbf1-3 and svk-1 mutants allowed us to further test our hypothesis that CBF1 degradation depends on the presence of SVK and vice versa. Indeed, blocking SVK transcription (svk-1) resulted in higher stability of CBF1 ( Figure 2E). Conversely, downregulation of CBF1 (cbf1-3) resulted in a more stable SVK ( Figure 2F). We could also confirm that SVK-L is an RRP4 target at 22 C ( Figure 2F). Interestingly, CBF1 stability is increased at 4 C, suggesting that SVK is not responsible for degradation of fulllength CBF1 mRNA at cold temperatures (Supplemental Figure 2D). These results are compelling evidence that the presence of SVK-L cis-NAT is responsible for degrading CBF1 mRNA at 22 C but not at 4 C. Furthermore, they show that the abundances of both transcripts depend on each other's presence and absence, supporting the idea that SVK-L and CBF1 transcripts may undergo hybridization in planta, leading to the formation of double-stranded RNA (dsRNA) at 22 C.

CBF1-SVK forms dsRNA at 22 C
According to a consensus model, long dsRNA templates formed by a sense mRNA-cis-NAT are recognized as substrates by the endonuclease DICER-LIKE (DCL) proteins, leading to generation of short dsRNA fragments as a result of cleavage activity. The cleavage product(s) are stabilized by HUA ENHANCER 1 (HEN1)-mediated methylation ( Figure 3A) (Bologna and Voinnet, 2014), and one of the guide RNA strands from the short dsRNA can be loaded onto ARGONAUTE1 (AGO1) and/or AGO4 to form RNA-induced silencing complex (RISC). RISC may eventually establish a self-perpetuating silencing loop that can potentially amplify the extent of downregulation of the target transcript (Bologna and Voinnet, 2014). Among mutants of the four DCL proteins present in Arabidopsis, only dcl2dcl4 displayed higher CBF1 expression at 22 C ( Figure 3B). Similarly, hen1 showed increased CBF1 mRNA levels, implicating several components in the regulation of CBF1 at 22 C ( Figure 3C). We also observed increased CBF1 mRNA levels in two separate ago1 mutants but not in an ago4 mutant ( Figure 3D), suggesting specific involvement of AGO1 in the regulation of CBF1 expression at 22 C. In line with our previous results, none of the mutants that had increased CBF1 expression levels at 22 C showed any mis-regulation of CBF1 after cold induction, suggesting that the DCL pathway is not active at 4 C when CBF1 needs to be rapidly induced (Supplemental Figure 3A). To further investigate whether the RNA-directed DNA methylation (RdDM) pathway is active in CBF1 transcriptional gene silencing via small RNAs (sRNAs), we measured CBF1 mRNA levels in mutants of the plant-specific RNA polymerases Pol IV and Pol V. However, CBF1 expression in both pol IV and pol V mutants was similar to wild-type levels at 22 C (Supplemental Figure 3B), reinforcing the plaNET-seq data and pointing to the conclusion that CBF1 downregulation is most likely exclusively dependent on RNAPII transcription. To determine whether cleavage products generated after DICER activity from the CBF1:SVK-L hybrid are subject to sequential amplification by RNA DEPENDENT RNA POLYMERASE (RDR) 1, 2, or 6 at 22 C, we quantified transcript abundance of CBF1 in the rdr 1, 2, 6 triple mutant (rdr1-1, rdr2-1, rdr6-15). Our results suggested that RDRs are not part of this regulation, based on wild-type levels of CBF1 in the mutants (Supplemental Figure 3B). In addition, we mined the Arabidopsis Small RNA Database (Feng et al., 2019) to see whether there was an accumulation of sRNAs in the CBF1 region. There was a small accumulation of sRNAs 21 nucleotides (nt) and 24 nt in length, indicating that sRNAs are present on the CBF1 locus but at a low level (Supplemental Figure 2C). Together, these results suggest that the CBF1::SVK-L dsRNA does not lead to anticipated gene silencing amplification from other RNA polymerases, supporting the presence of a stringent regulatory landscape for CBF1 expression that allows only its fine-tuning rather than complete silencing, making the cleavage products inherently challenging to detect with currently available methods.
Nevertheless, to further corroborate our hypothesis, we performed RNase protection assays that detect local dsRNA (Okano et al., 2014). We used three probes on the SVK-L transcription unit ( Figure 4A), and we could confirm an increased CBF1 level in dcl2dcl4 without RNase treatment ( Figure 4B). After RNase treatment, we could detect a signal for dsRNA only at probe b in the 3 0 end of CBF1 in the wild type ( Figure 4C and 4D). The signal was significantly higher in the dcl2dcl4 mutant, indicating that dsRNAs were formed in the region and recognized as substrates for DCL2 and/or DCL4, leading to their degradation. Supporting these results, we also detected a longer half-life for the CBF1 mRNA in both dcl2dcl4 and ago1-45 mutants, further suggesting their direct involvement in CBF1 degradation (Supplemental Figure 3D). To confirm the direct involvement of AGO1, we used AGO1-IP RNA data to identify potential AGO1-bound RNAs. Several RNAs were found in the CBF1 region (Supplemental Figure 3E), Supplemental Table 1). Intriguingly, we could detect three 21-nt AGO1-bound RNAs in the 3 0 half of CBF1, corresponding to the region where we detected the dsRNAs. All in all, these results demonstrate that several components are directly involved in degradation of CBF1::SVK-L dsRNA hybrids at 22 C, confirming the regulation of CBF1 by SVK during normal development. The dsRNA formation also strongly suggests that SVK-L achieves this downregulation in a cis manner.

SVK-CBF1 regulation governs biomass production at 22 C
To better understand the regulation of CBF1 mRNA by SVK-L and its biological role, we monitored CBF1 steady-state levels during the light period in our cbf1 and svk mutants under standard shortday conditions ( Figure 5A). In line with earlier reports, CBF1 levels peaked shortly after the middle of the day (Noré n et al., 2016). Steady-state CBF1 levels in svk-1 and uns-1 plants increased at time points close to the end of the day ( Figure 5A). As expected, steady-state levels of CBF1 were lower in the cbf1-3 mutant throughout the day ( Figure 5A). Intriguingly, the SVK steady-state level peaked at the end of the light period, showing a similar anti-correlation to CBF1 expression during cold exposure ( Figure 5B) (Kindgren et al., 2018). Both cbf1-3 and uns-1 displayed wild-type levels of SVK, whereas svk-1 had low levels of SVK throughout the day ( Figure 5B). To confirm that we could also detect dsRNA under these conditions, we performed dsRNA-IP-qPCR using an antibody specific to dsRNA (Supplemental Figure 4A). Here, dcl2dcl4 again showed a higher level of dsRNA accumulation compared with the wild type. The cbf1-3 mutant showed lower levels of dsRNA, providing further evidence that dsRNA was formed by CBF1 and SVK-L transcripts and that DCL2 and/or DCL4 are involved in its cleavage. Importantly, we observed a notable difference in biomass production of the mutants, indicating a physiological role for CBF1-SVK-L regulation at 22 C ( Figure 5C and 5D). In particular, cbf1-3 showed decreased biomass compared with wild-type plants, whereas svk-1 and uns-1 had increased biomass. By contrast, despite their high sequence similarity, we did not observe a similar phenotype for mutants of CBF1  neighboring genes CBF3 and CBF2, highlighting the absence of full redundancy between the CBFs and suggesting a potential CBF1-specific role in biomass control (Supplemental Figure 4B). Furthermore, overexpression of SVK in cis (Kindgren et al., 2018) did not alter biomass production (Supplemental Figure 4C), indicating that SVK transcription in wild-type plants is not the limiting factor in this regulation.
In conclusion, our data indicate that fine-tuning the steady-state levels of CBF1 mRNA is important for regulating biomass production and, in this regulation, the lncRNA SVK-L has an intricate cis-NAT role in maintaining the proper CBF1 mRNA level.

DISCUSSION
Our findings from this study can be summarized in a working model of CBF1 regulation by SVK (Figure 6). At normal growing temperatures, SVK-L is the dominant isoform of SVK, and it functions as a cis-NAT that interacts with the host gene's mRNA to form dsRNA. CBF1-SVK-L dsRNA is recognized by DCL2 and/or DCL4 and cleaved. The cleaved product is loaded onto AGO1 for further cleavage of CBF1 mRNA. By contrast, at 4 C, increased transcriptional activity on the CBF1 gene body limits the presence of SVK-L. In addition, a switch to a proximal poly(A) site makes SVK-S the major isoform at 4 C. SVK transcription is also induced a few hours after CBF1 during cold exposure, which results in RNAPII collisions and prematurely terminated CBF1 transcription. Thus, the lncRNA SVK negatively downregulates CBF1 with two distinct and temperature-specific mechanisms.
The transcription factor CBF1 has been extensively studied in plant tolerance to cold and freezing temperatures (Jaglo-Ottosen et al., 1998;Benedict et al., 2006;Ito et al., 2006;Badawi et al., 2007;Zhang et al., 2011). Here, we expand the functional repertoire of CBF1 by demonstrating its significant role in controlling biomass production in a standard growing regimen. A modest increase in CBF1 mRNA levels results in larger plants (Figure 5), whereas highly increased (Gilmour et al., 2004) or strongly decreased CBF1 expression leads to lower biomass ( Figure 5). Thus, a slightly elevated level of CBF1 mRNA is beneficial for the plant in a controlled environment, both in terms of biomass production ( Figure 5) and cold acclimation (Kindgren et al., 2018). Our results recognize CBF1 as a converging component in plant development and stress response. Earlier reports place this phenotype in a different context: light quality regulates CBF1 and its downstream targets PHYTOCHROME INTERACTING FACTOR 4 (PIF4) and PIF5, which form a regulatory hub that controls many processes during plant development and growth (Nieto et al., 2015;Dong et al., 2020). Moreover, plants overexpressing PIF4 show striking phenotypic similarities to CBF1-overexpressing plants (Gangappa et al., 2017). CBF1 is also known to regulate levels of the plant hormone gibberellin. Gibberellin controls plant growth by regulating DELLA proteins (Achard et al., 2008). All these data together highlight the need for tight regulation of CBF1 mRNA levels in plants to achieve a proper regulatory balance between maximizing biomass production and priming the low-temperature stress responses.
Regulation of CBF1 involves two distinct temperature-dependent cis mechanisms, in which transcription of the lncRNA SVK (Kindgren et al., 2018) has a central role. Within minutes of exposure to cold temperatures, CBF1 expression is quickly induced by transcriptional activators such as ICE1 and CAMTA3 (Zarka et al., 2003;Doherty et al., 2009), which in turn dramatically increases the transcription and steady-state level of CBF1 to initiate the cold acclimation process (Medina et al., 1999). Massive transcriptional activity on CBF1 poses a steric hindrance to any RNAPII complex transcribing the antisense strand. However, after 2-3 h of cold treatment, SVK transcription is induced, which results in increased accumulation of RNAPII at the 3 0 end of CBF1 (Kindgren et al., 2018). Therefore, the RNAPII collisions that occur during cold exposure require massive transcriptional activity on both strands. By contrast, at normal growing temperatures (22 C), SVK transcription does not face the same steric hindrance from sense CBF1 transcription and gives rise to antisense transcription on the CBF1 locus (Figure 1). SVK-L forms a cisnatural antisense transcript (cis-NAT) to CBF1 and triggers the rapid degradation of the CBF1 mRNA (Figures 2, 3, and 4).
NATs are an important class of lncRNAs with a broad range of molecular functions in eukaryotes (Lapidot and Pilpel, 2006;Britto-Kido et al., 2013). In Arabidopsis, the cis-NAT FLORE is involved in circadian regulation of its host gene, CYCLING DOF FACTOR 5 (CDF5), albeit through an unknown mechanism (Henriques et al., 2017). Another example is the cis-NAT PHO1;2 in rice, which promotes translation of its host gene, PHOSPHATE1:2 (Jabnoune et al., 2013). Recent advances suggest that up to 70% of all mRNAs in Arabidopsis have an NAT (Wang et al., 2014), and many genes with an NAT are responsive to stress or play developmental roles (Wunderlich et al., 2014;Thieffry et al., 2020;Meena et al., 2021). This suggests an important role for non-coding transcription in plant survival under unfavorable conditions. The sense mRNA and NAT can form dsRNA owing to their sequence complementarity, similar to the first described report in the case of salt stress response in Arabidopsis (Borsani et al., 2005). We confirmed that CBF1 and SVK-L also form dsRNA at the 3 0 end of CBF1 (Figure 4, Supplemental Figure 4A). Plants have a highly evolved surveillance system that recognizes dsRNA, which includes four specialized Dicer-like RNase-III endonucleases (DCL1-4) (Bologna and Voinnet, 2014). By contrast, humans (Homo sapiens) and worms (Caenorhabditis elegans) only have one DCL protein. The CBF1-SVK-L dsRNA is recognized by DCL2 and/or DCL4 (Figures 3 and 4, Supplemental Figure 4A), two proteins that have been extensively studied during virus infection (Bouché et al., 2006;Deleris et al., 2006). The products of DCL2/4 cleavage are modified (a methyl group is added to the 2 0 OH of the RNA) by HEN1 (Figure 3). DCL2/4 and HEN1 are considered to be part of the cis-NAT degradation pathway (Bologna and Voinnet, 2014). However, CBF1-SVK-L dsRNA regulation is not amplified by any RNA DEPENDENT RNA POLYMERASES (RDRs) (Supplemental Figure 3B). One of the RNA strands from the produced short dsRNA is subsequently loaded onto AGO1 for degradation of the CBF1 mRNA ( Figure 3D). Moreover, the CBF1-SVK-L pathway seems to be specific for AGO1 and does not include AGO4 ( Figure 3D). Thus, how specificity is achieved for the DCL and AGO proteins in different cis-NAT pathways is an interesting question for future research. The process of cis-NAT regulation is complex, and each case often seems idiosyncratic, but our results provide a potential candidate for further studies to unravel this complexity. An important point is that CBF1-SVK-L regulation does not depend on regulatory enhancement by polymerases other than RNAPII. This may enable careful fine-tuning of CBF1 in contrast to pathways that include POLIV/POLV and/or RDRs, which would require additional levels of regulation. As a result, DICER generated sRNAs (Supplemental Figure 3C) may be scarce, explaining the finely tuned regulatory window and the difficulty of detecting sRNA products from cis-NATs (Wang et al., 2005;Zhang et al., 2012;Li et al., 2013). Another interesting aspect of this study is the promotion of a dominant poly(A) site for SVK approximately 100 bp downstream of the poly(A) site of CBF1 at 4 C compared with the distal poly(A) site close to the CBF1 TSS at 22 C (Figure 1) (Kindgren et al., 2018). Alternative polyadenylation (APA) is a common mechanism in eukaryotes for increasing the number of active isoforms derived from one gene (Tian and Manley, 2017). Interestingly, APA also seems to be important for the functional role of lncRNAs. Another temperature switch in poly(A) site preference has been described for the lncRNA COOLAIR and its regulation of flowering (Swiezewski et al., 2009). The preferred poly(A) site for COOLAIR is controlled by specific RNA 3 0 -end processing factors (Liu et al., 2010), which could also be the case for SVK. However, there is also the possibility that the switch between the main poly(A) site for SVK is governed by the increased termination process at the end of the CBF1 gene body after cold treatment and the fact that SVK and CBF1 are transcribed at the same locus (a requisite for RNAPII collisions), in contrast to FLC and COOLAIR that are transcribed at mutually exclusive loci (Rosa et al., 2016). Polyadenylation was shown to be promoted in Arabidopsis by formation of liquid-liquid phase compartments at the 3 0 end of genes (Fang et al., 2019). The distance between the CBF1 and SVK-S poly(A) sites is within the size of these compartments (Fang et al., 2019). Massive transcriptional activity on the CBF1 gene body early in the cold response would result in the presence of high concentrations of polyadenylation-associated factors in the liquid-liquid phase compartments at the 3 0 end of CBF1, which would encompass the SVK-S poly(A) site. Thus, this presents an elegant way to switch poly(A) preference without the need for additional components. However, future experiments are required to fully understand the poly(A) signal preference we detect for SVK.
In this study, we found a connection between 3 0 -5 0 exosomemediated RNA surveillance and CBF1-SVK regulation. The polyadenylated SVK transcripts are targeted by the exosome, as seen by a higher transcript level in the rrp4-2 mutant ( Figure 2B and 2C). SVK transcripts degraded by the exosome are most likely RNA species that are not part of dsRNA formation or RNA molecules left over after dsRNA cleavage. However, SVK transcripts prematurely terminated by RNAPII collisions at 4 C undergo a more rapid degradation (Kindgren et al., 2018). Although all SVK transcripts are targeted by the exosome core subunit RRP4, only the prematurely terminated and readthrough transcripts are targeted by exosome complexes that also contain HEN2. In contrast to RRP4, which localizes in both nucleoplasm and cytoplasm, HEN2 is only present in the nucleoplasmic exosome, where active transcription occurs (Lange et al., 2014). This indicates that SVK transcripts are degraded by the exosome at spatially separated sites. Arabidopsis has two RNA helicases that activate the nuclear exosome, HEN2 and the rRNA processing helicase AtMTR4 (Lange et al., 2014). HEN2 is a plant-specific protein, whereas MTR4 is an essential RNA helicase for all nuclear exosome activities in yeast and humans (Bernstein et al., 2008;Lubas et al., 2011). Our results suggest that prematurely terminated transcripts are degraded in a HEN2-dependent manner close to the active RNA synthesis site compared with transcripts that have reached their poly(A) signal. This is true for both coding and non-coding transcripts. Interestingly, the involvement of HEN2 in the CBF1-SVK circuit might suggest that sessile plants have evolved specialized nuclear exosome RNA helicases to properly respond to stress stimuli.

Transcript stability assay
The half-life (t 1/2 ) of transcripts was determined as described previously (Fedak et al., 2016). In brief, Col-0 (wild-type) seedlings were grown vertically on plates containing 1 / 2 MS medium for 10 days and then transferred to 1 / 2 MS liquid medium and maintained at 22 C or shifted to 4 C for 4 h under the same light conditions. Seedlings from both temperatures were transferred into 12-well plates and incubated in buffer A (1 mM Model of downregulation of CBF1 by SVK at 22 C and 4 C. CBF1 and SVK form dsRNA that is recognized by DCL2 and/or DCL4. The resulting sRNA is stabilized by HEN1 and loaded onto AGO1 for further downregulation of CBF1 mRNA. At 4 C, increased transcriptional activity of CBF1 and SVK leads to RNAPII collisions. PIPES at pH 6.25, 1 mM trisodium citrate, 1 mM KCl, and 15 mM sucrose) at 22 C or 4 C, respectively. After 30 min of incubation, 150 mg/L cordycepin (3 0 -deoxyadenosine, Sigma Aldrich) was added and vacuum-infiltrated twice for 5 min. Samples were collected after 0, 15, 30, 60, 120, and 300 min in triplicate (15 seedlings per replicate), followed by total RNA extraction and RT-qPCR analysis using gene-specific primers (Supplemental Table 1). EIF4A1 and EXPL1 were used as assay controls (Perea-Resa et al., 2012;Fedak et al., 2016). C t values were normalized with the C t value at 0 min [Ct(n) = (ln(C t /C t (0)) 3 (À10)] and the slope to determine the half-life of the transcripts was calculated as follows [t 1/2 = (ln 2 )/slope] of transcript.
Luciferase activity assay CBF1 reporter constructs with or without the SVK gene unit at the 3 0 end of CBF1 were used as described previously to analyze the effect of SVK transcription on CBF1 transcriptional output (Kindgren et al., 2018). In brief, two independent CBF1 reporter constructs were used in the analysis. The p CBF1::CBF1:Tnos construct carried the CBF1 promoter) and CBF1 gene body (À1903 bp to +639 bp relative to the translation start site) fused to the firefly luciferase gene (LUC) and the Tnos terminator. In the second construct, p CBF1::CBF1:LUC:SVK, Tnos was replaced by the SVK promoter and transcription unit (+643 bp to +3410 bp relative to the CBF1 translation start site) after removal of the stop codon in the CBF1 ORF.
To perform the bioluminescent firefly luciferase assay, T3 homozygous seeds from two independent reporter lines for each construct were sterilized with bleach and stratified at 4 C for 2 days, followed by growth on 1 / 2 MS medium in vertical plates under long-day conditions (16 h light/8 h dark) with 22 C/18 C day/night temperatures. Sufficient numbers (15-20) of 12-day-old individual seedlings in replicates were carefully placed and submerged in 96-well plates containing 200 mL luciferase assay reagent (5 mM luciferin in 0.01% Triton X-100). After a short 30-min incubation, plates were read using a GloMax Navigator Microplate Luminometer (Promega), and levels of bioluminescence produced per well were recorded. Data were exported, normalized for background bioluminescence noise from the average signal intensity of at least 4-5 wild-type Col-0 seedlings with no LUC gene, and further analyzed to obtain the average LUC intensity per genotype.
dsRNA protection assay and dsRNA-IP-qPCR The dsRNA protection assay was performed as described previously (Okano et al., 2014) with a few modifications. In brief, extracted total RNA (5 mg) was treated with DNase I and incubated with the indicated concentrations of RNase One Ribonuclease (Promega) for 1 h at 37 C. RNase-treated RNA was used as a template for reverse transcription using random primers. The resulting cDNA was used in a qPCR with oligos found in Supplemental Table 2. The data were further analyzed in Microsoft Excel, and relative steady-state levels were calculated by comparison with the level of ACT2 in samples without RNase treatment. For dsRNA-IP-qPCR, nuclei from 3 g of 12-day-old seedlings were isolated with Honda buffer (0.44 M sucrose, 1.25% Ficoll, 2.5% Dextran T40, 20 mM Tris-HCl [pH 7.4], 10 mM MgCl 2 , and 0.5% Triton-X). After addition of lysis buffer (0.3 M NaCl, 20 mM Tris-HCl [pH 7.4], 5 mM MgCl 2 , 5 mM BME, 0.5% Tween, and RNase and protease inhibitors) and centrifugation, the supernatant was mixed with a dsRNA-specific antibody (J2, Jena Biosciences), and IP was performed as described previously (Gao et al., 2020). The isolated RNA was used as a template in a room temperature reaction with Superscript IV (Invitrogen) and genespecific oligos (Supplemental Table 2) and then used in qPCR reactions. IP samples were compared to input samples for further analysis.

Available sequencing data
Sequencing data from Kindgren et al. (plaNET-seq) (2019) and Schurch et al. (DRS-seq) (2014) were used in this study. Raw data were analyzed according to Kindgren et al. (2019). BedGraph files were opened in IGV (Broad Institute) and exported for further analysis. AGO1-interacting sRNAs were identified from published datasets (Liu et al., 2018;Annacondia and Martinez, 2021

AUTHOR CONTRIBUTIONS
V.Z., S.K., and P.K. conceived of and designed the study. V.Z., S.K., and P.K. undertook data collection. V.Z., S.K., and P.K. analyzed the data. All co-authors contributed to the writing of the manuscript.