Antisense transcription from stress-responsive transcription factors fine-tunes the cold response in Arabidopsis

Abstract Transcription of antisense long noncoding RNAs (lncRNAs) occurs pervasively across eukaryotic genomes. Only a few antisense lncRNAs have been characterized and shown to control biological processes, albeit with idiosyncratic regulatory mechanisms. Thus, we largely lack knowledge about the general role of antisense transcription in eukaryotic organisms. Here, we characterized genes with antisense transcription initiating close to the poly(A) signal of genes (PAS genes) in Arabidopsis (Arabidopsis thaliana). We compared plant native elongation transcript sequencing (plaNET-seq) with RNA sequencing during short-term cold exposure and detected massive differences between the response in active transcription and steady-state levels of PAS gene-derived mRNAs. The cold-induced expression of transcription factors B-BOX DOMAIN PROTEIN28 (BBX28) and C2H2-TYPE ZINC FINGER FAMILY PROTEIN5 (ZAT5) was detected by plaNET-seq, while their steady-state level was only slightly altered due to high mRNA turnover. Knockdown of BBX28 and ZAT5 or of their respective antisense transcripts severely compromised plant freezing tolerance. Decreased antisense transcript expression levels resulted in a reduced cold response of BBX28 and ZAT5, revealing a positive regulatory role of both antisense transcripts. This study expands the known repertoire of noncoding transcripts. It highlights that native transcription approaches can complement steady-state RNA techniques to identify biologically relevant players in stress responses.


Introduction
Widespread long noncoding transcription from the complementary DNA (antisense) strand is present at thousands of proteincoding gene loci in eukaryotic organisms.Recent research in plants has evidenced antisense transcription for 30% of expressed genes in the model plant Arabidopsis (Arabidopsis thaliana) and for 60% of the genes in rice (Chen et al. 2019;Kindgren et al. 2019).Our current understanding is that the majority of antisense transcripts, also called cis-natural antisense transcripts (cis-NATs), are long noncoding transcripts [>200 nucleotides in length, long noncoding RNAs (lncRNAs)], but their general functional significance remains elusive (Reis and Poirier 2021).Complementarity between cis-NATs and the sense transcript, coupled with the overlapping spatiotemporal expression of sense-antisense pairs, endows them with the potential to engage in the formation of double-stranded RNA (dsRNA).dsRNA could undergo subsequent detection by the factors of RNA silencing machinery such as Dicer or Dicer-like (DCL) and Argonaute (AGO) family proteins.However, there is weak evidence that endogenous small interfering RNAs are derived from NAT-sense pairs (Henz et al. 2007;Reis and Poirier 2021).Only a handful of studies from different plant species describe a negative role of NATs over the sense transcription (Borsani et al. 2005;Katiyar-Agarwal et al. 2006;Swiezewski et al. 2007;Held et al. 2008;Wan et al. 2016).
So far, few cis-NATs in plants have been experimentally characterized with mechanistic insights, thereby limiting our understanding of their general function (Wierzbicki et al. 2021).Thus, we only have a rudimentary knowledge of the roles of antisense transcription, and its widespread prevalence is certainly an enigma in modern plant research.
Our global numerical comprehension of noncoding transcription has been mostly based on sequencing technologies that fundamentally use the steady-state RNA detection principle.For example, in Arabidopsis, 70% of mRNAs were postulated to form sense-antisense pairs based on RNA-sequencing (RNA-seq) experiments (Wang et al. 2014).However, noncoding transcription is far more pervasive and complex than visualized by canonical steady-state analyses.A major challenge in the field is the detection of antisense and other noncoding transcription due to the low abundance and high turnover rate (Mayer et al. 2015;Kindgren et al. 2019;Thieffry et al. 2020).Classical steady-state level detection methods, such as RNA-seq, are ill-suited to investigate noncoding transcription.In contrast, a technique that investigates active transcription (i.e.native elongation transcript sequencing [NET-seq]) is better suited for detecting rapidly degraded RNA species (Mayer et al. 2015).NET-seq in plants (plaNET-seq or pNET-seq) captures actively transcribing RNA Polymerase II (RNAPII) complexes and enables strand-specific sequencing of the RNA molecules associated with the captured RNAPII complexes before any degradation can occur, thereby uncovering all transcription events genome-wide (Zhu et al. 2018;Kindgren et al. 2019).Case-in-point, plaNET-seq detected ∼8,000 unannotated long noncoding transcripts, many of them being antisense transcripts initiating from the host gene's 3′-end (poly(A) antisense genes, PAS genes) (Kindgren et al. 2019).
Emerging evidence indicates that antisense transcription is essential for plant development and stress responses (Matsui et al. 2008;Thieffry et al. 2020).However, the underlying molecular mechanism(s) involved in sense/antisense transcriptional crosstalk often seem idiosyncratic with few common components involved (reviewed in Lucero et al. 2021).Nevertheless, antisense RNAs have been characterized to control seed germination (Fedak et al. 2016), phosphate starvation (Jabnoune et al. 2013), flowering control (Swiezewski et al. 2007;Henriques et al. 2017), hormonal regulation (Ariel et al. 2020), and cold acclimation (Kindgren et al. 2018).Cold acclimation, initiated by transcriptional processes, allows plants to adapt and eventually withstand freezing.At present, cold stress is the sole abiotic stress monitored by plaNET-seq (Kindgren et al. 2019).Remarkably, in plaNET-seq datasets, antisense transcription at PAS genes responds rapidly to cold with a global downregulation, suggesting a putative role in the cold response (Kindgren et al. 2019).Thus, exposure to cold temperature is an excellent environmental cue for studying antisense transcription's role in Arabidopsis.
In this study, we use plaNET-seq and RNA-seq data to show an earlier hidden layer of genes involved in cold acclimation.We show that RNA-seq data poorly captures the cold stress-induced transcriptional changes of protein-coding genes.On the contrary, plaNET-seq efficiently uncovers these transcriptional changes since transcription is captured prior to their degradation.Among those, we identified 2 cold-responsive transcription factors (TFs) that responded natively as crucial for cold acclimation in Arabidopsis.Importantly, experimental validation demonstrated that their antisense transcription is positively correlated to the stress-responsiveness of the sense transcription, suggesting a role for antisense transcription in assisting certain TFs' stress responses, such as cold exposure.

PAS genes are enriched in stress-responsive TFs with high active transcription
Genes that host PAS transcription (PAS genes) represent over 3,000 genes in Arabidopsis and were recently detected by plaNET-seq (Kindgren et al. 2019).PAS genes are defined as protein-coding genes with antisense transcription initiating in the 3′-half or 20% of the gene length downstream of the sense gene's PAS (Fig. 1A and Supplementary Data Set 1).Characterization of PAS genes has not been done rigorously in Arabidopsis, so we first investigated which biological processes and molecular functions were overrepresented in the list of all PAS genes (Fig. 1, B and C and Supplementary Data Set 2).Interestingly, there was a clear enrichment of stress-responsive genes that are involved in DNA binding [in particular transcription factors (TFs)].Therefore, for our further analysis, we decided to divide the PAS genes into PAS TFs (n = 294) and PAS non-TFs.We also included TFs without PAS (TFs non-PAS) as a control group.An example of a known stress-responsive gene encoding a TF with an antisense transcript that responds natively to cold temperature was WRKY48 (Fig. 1D; Xing et al. 2008).
Overall, PAS genes tended to be shorter than other expressed genes and TFs non-PAS (Fig. 2A) and the steady-state levels of PAS non-TFs or PAS TFs gene mRNA were not significantly different from those of all expressed genes (Fig. 2B).Surprisingly, PAS gene mRNA had a greater turnover rate compared with all expressed genes when measured by RNA-seq after treatment with the transcription inhibitor, cordycepin (alpha decay) (Fig. 2C) (Sorenson et al. 2018).TFs non-PAS also showed a significantly higher turnover rate, suggesting that the rapid degradation of mRNA in these classes of genes is independent of antisense transcription.We confirmed these results with data from transcription inhibition with another inhibitor, actinomycin D (Supplementary Fig. S1A) (Narsai et al. 2007).Both PAS non-TFs and PAS TFs genes displayed high active RNAPII transcription at the genome-wide level compared with expressed genes and TFs non-PAS (Fig. 2D and Supplementary Fig. S1B).In addition, PAS TFs showed a higher RNAPII occupancy compared with PAS non-TFs (Supplementary Fig. S1B).Thus, the higher expression level of PAS genes and high turnover rate explained the small effects on RNA steady-state levels.Taken together, PAS genes are enriched in stress-responsive TFs with high transcriptional and post-transcriptional regulation.Moreover, the high turnover rate seems to be a general feature of TF mRNAs and not a consequence of PAS TF antisense transcription.

The expression levels of cold-responsive genes detected by plaNET-seq correlate poorly with RNA-seq
The high transcription activity of PAS genes and rapid turnover of their mRNA made us interested in comparing the genome-wide steady-state levels of mRNA to how genes are actively transcribed (plaNET-seq).Thus, we performed RNA-seq using samples from seedlings grown similarly to the plaNET-seq experiment (10 d in long-day conditions at 22 °C and cold stress samples from 3 h at 4 °C and 12 h at 4 °C).In our cold-treated RNA-seq dataset, many mRNAs drastically changed their steady-state level (Fig. 3A and Supplementary Data Set 3).When we compared differentially expressed (DE) genes (fold change) between our RNA-seq and plaNET-seq datasets, we saw that RNA-seq poorly reflected the genes differentially expressed determined by plaNET-seq (Kindgren et al. 2019).The overlap was only 616 of the upregulated (UP) genes and 270 of the downregulated (DOWN) genes in the same direction after 3 h at 4 °C (Fig. 3A).After 12 h at 4 °C, we observed increased overlap, 1,819 of UP genes and 1,674 of DOWN genes (Fig. 3A).In addition, we saw a low correlation between the DE gene fold change at both time points (Fig. 3, B and C).This suggests that there is a clear discrepancy between the active transcriptional changes (plaNET-seq) and changes to the steady-state levels of mRNA (RNA-seq) occurring during cold stress, especially early in the cold response.
Focusing on the PAS genes, we saw a similar pattern as in all DE genes (Fig. 3, D to F). Astoundingly, we found that 2,214 (71%) PAS genes changed their expression significantly to cold (either in plaNET-seq or RNA-seq data) at one or both cold time points, corroborating their responsiveness to stress (Supplementary Data Set 4).Out of these genes, 229 were PAS TFs (78% were responsive to cold, Supplementary Data Set 4).To further confirm that PAS genes are highly responsive to cold, we compared PAS genes UP after 3 h 4 °C with UP genes without antisense transcription (determined by plaNET-seq).At 22 °C, PAS genes showed an increased RNAPII stalling around the +1 nucleosome, a hallmark for stress-responsiveness compared with other UP and non-DE genes (Fig. 4A) and an overall high active transcription (Fig. 4B).After 3 h 4 °C, PAS UP genes showed a more extreme response to cold temperature compared with other UP genes (Fig. 4, C and D), suggesting that, indeed, PAS genes have an enhanced responsiveness to cold compared with other genes.In contrast, antisense transcription for UP PAS genes remained similar throughout the cold response (Supplementary Fig. S2).Taken together, these data strengthen the hypothesis that PAS genes are inclined to respond to cold temperatures, but there is no correlation between PAS-host gene expression and antisense expression during the initial cold response.In addition, the discrepancy between plaNET-seq and RNA-seq argues that some TFs might have been overlooked as cold-responsive genes and highlight the use of combining nascent RNA methods with steady-state levels of mRNA to fully understand the cold response in an organism.

The discrepancy between active transcription and steady-state levels can partly be explained by mRNA turnover rates
Next, we focused on the differentially expressed TFs with antisense transcription.We aimed to identify additional biologically important TFs in the cold response.A special emphasis was put on the expression pattern in plaNET-seq as the distinct dynamics of PAS TFs mRNAs (Fig. 2) might have masked cold-responsive TFs in earlier studies.We reasoned that rapidly responsive TFs had a high probability of being involved in cold acclimation since this expression pattern mirrors that of known assigned TFs in the cold response, such as C-REPEAT BINDING FACTOR2 (CBF2, At4g25470) (Fig. 5A).CBF2 is massively induced and transiently peaks after 3 h at 4 °C.Out of the 242 PAS TFs that responded to cold, 25 showed a similar plaNET-seq expression pattern to CBF2 (UP after 3 h and DOWN between 3 and 12 h) (Supplementary Data Set 5).Only 9 of the 25 genes showed upregulation by RNA-seq after 3 h (Supplementary Data Set 5).An example of a gene that was DE in plaNET-seq but not in RNA-seq was HY5-HOMOLOG (HYH, At3g17609) (Fig. 5B).Due to the RNA degradation characteristics of PAS TFs, we first tested the mRNA turnover rates (by cordycepin incubation) for 5 randomly chosen genes from the 25 genes with convincing expression patterns in plaNET-seq and no or slight upregulation by RNA-seq.MYB DOMAIN PROTEIN-47 (MYB47, At1g18710) has a potential role in drought and hormone signaling (Ding et al. 2014;Marquis et al. 2022).CONSTANS-LIKE 7 (COL7, At1g73870) is involved in the shade avoidance response (Wang et al. 2013).The C2H2-type zinc finger family proteins, C2H2-TYPE ZINC FINGER FAMILY PROTEIN-5 (ZAT5, At2g28200) and SCARECROW-LIKE-8 (SCL8, At5g52510), are uncharacterized proteins and B-BOX DOMAIN PROTEIN-28 (BBX28, At4g27310) has been characterized for its involvement in flowering and photomorphogenesis (Song et al. 2020;Cao et al. 2022).MYB47, SCL8, and BBX28 showed a slight upregulation in the RNA-seq data.However, neither BBX28, SCL8, MYB47, or any of the other chosen proteins have a known function in the cold response of Arabidopsis.Our assay control EUKARYOTIC TRANSLATION INITIATION FACTOR-4A1 (EIF4A1), a stable mRNA at 22 °C, showed similar stability at 4 °C as compared with 22 °C (Fig. 5C).In addition, a recent study showed increased stability in cold for a key TF in the cold response, CBF1 (Zacharaki et al. 2023).In contrast, all the PAS TF candidate genes' mRNAs, except SCL8, showed a significantly decreased transcript stability at 4 °C compared with 22 °C (Fig. 5, D to H).Thus, induced transcriptional activity at gene loci responding to cold may remain partly undetected by steady-state methods due to rapid mRNA decay at 4 °C.Again, our results highlight that nascent transcription methods can complement RNA .PAS genes are highly expressed, but their mRNA is degraded rapidly.A) Violin plot of the length of PAS non-TF genes, TFs non-PAS, PAS TF genes, and all expressed genes.Centerline, median; box limits, upper and lower quartiles; whiskers, 1.5× interquartile range.P-value was calculated with Mann-Whitney U test and P < 0.05 was considered significant.B) Violin plot of the steady-state level of PAS non-TF genes, TFs non-PAS, PAS TF genes, and all expressed genes.Centerline, median; box limits, upper and lower quartiles; whiskers, 1.5× interquartile range.P-value was calculated with Mann-Whitney U test and P < 0.05 was considered significant.C) Violin plot of the decay rate of PAS non-TF genes, TFs non-PAS, PAS TF genes, and all expressed genes after transcriptional inhibition by cordycepin.Centerline, median; box limits, upper and lower quartiles; whiskers, 1.5× interquartile range.P-value was calculated with Mann-Whitney U test and P < 0.05 was considered significant.D) Metagene analysis of plaNET-seq data of PAS non-TF genes, TFs non-PAS, PAS TF genes, and all expressed genes.The shaded area shows a 95% confidence interval for the mean.steady-state level methods to identify stress-responsive genes, in particular those with highly dynamic regulation and mRNA turnover.

ZAT5 and BBX28 are involved in cold acclimation
Can mRNAs with minor differences in steady-state levels during stress have an important biological role in the stress adaptation for the plant?To answer this, we focused on 2 candidate genes for our continued analysis, ZAT5 and BBX28.Both genes responded rapidly to cold temperatures, as detected by plaNET-seq (Fig. 6, A and B).We could not detect any significantly increased steady-state level of ZAT5 after exposure to 4 °C with RT-qPCR, although a positive trend was observed early in the cold response (Fig. 7A).We did see a significant downregulation from control levels starting at 8 h of 4 °C (Fig. 7A).For BBX28, we could not detect any significant upregulation or downregulation, but a positive trend early in the cold response was observed (Fig. 7B).The discrepancy between our RT-qPCR and RNA-seq results may suggest that the upregulation of BBX28 and ZAT5 is slight and not always consistently statistically significant depending on separate cold treatments and the variation between replicates.However, since both BBX28 and ZAT5 have been found to be diurnally expressed (Romanowski et al. 2020), we also checked their steady-state levels with controls (seedlings kept at 22 °C) taken at the same time points as the cold samples (3 h and 12 at 4 °C).Here, we did see a small but significant increase of both BBX28 and ZAT5 after 3 h (Supplementary Fig. S3, A and B).
To test their biological importance in the cold acclimation process, we isolated 2 independent T-DNA lines disrupting the 2 genes and subjected these lines to a freezing test together with wild type (WT; Fig. 7, C and D).In the freezing test, leaf discs of non-acclimated A) Metagene analysis of the plaNET-seq signal (at 22 °C) in a 500 bp window centered at the +1 nucleosome.PAS genes that will be UP after 3 h at 4 °C, genes without antisense transcription but UP after 3 h at 4 °C, and non-DE genes are shown.The shaded area shows a 95% confidence interval for the mean.B) Metagene analysis of plaNET-seq data (at 22 °C).PAS genes that will be UP after 3 h at 4 °C, genes without antisense transcription but UP after 3 h at 4 °C, and non-DE genes are shown.The shaded area shows a 95% confidence interval for the mean.C) Metagene analysis of the plaNET-seq signal (after 3 h at 4 °C) in a 500 bp window centered at the +1 nucleosome.PAS UP genes after 3 h at 4 °C, genes without antisense transcription but UP after 3 h at 4 °C, and non-DE genes are shown.The shaded area shows a 95% confidence interval for the mean.D) Metagene analysis of plaNET-seq data (after 3 h at 4 °C).PAS UP genes after 3 h at 4 °C, genes without antisense transcription but UP after 3 h at 4 °C, and non-DE genes are shown.The shaded area shows a 95% confidence interval for the mean.
and cold-acclimated plants (4 d at 4 °C) are in contact with water and exposed to decreasing freezing temperatures and measured for plasma membrane disruption (i.e.leakage of electrolytes).Therefore, measuring electrolyte leakage is a measurement of how well the cells can survive freezing temperatures.Indeed, we found that both mutant lines were impaired in their acclimation to cold.For zat5-1, the freezing tolerance was significantly lower (−3.7 ± 1.7 °C, P < 0.0001) compared with −6.1 ± 0.2 °C for WT.The freezing tolerance for bbx28-1 was also significantly lower (−5.7 ± 0.2 °C, P = 0.0022) compared with −6.5 ± 0.2 °C for WT.We could not detect any difference in non-acclimated plants (Supplementary Fig. S3, C and D), suggesting that both ZAT5 and BBX28 have a specific role in the cold acclimation process in Arabidopsis.Overall, these results indicate that changes to the transcription activity with minor changes to the mRNA steady-state levels can have a significant biological role.Next, we turned to our second question; what is the role of antisense transcription along the ZAT5 and BBX28 gene body?Our plaNET-seq data revealed that both ZAT5 and BBX28 had antisense transcription affected by cold temperature, although the differences were small compared with sense expression (Fig. 8, A and B, upper panel).To identify the 5´ and 3′-end of asZAT5 and asBBX28, we used available Cap Analysis of Gene Expression sequencing (CAGE) (Thieffry et al. 2020) and Direct RNA Sequencing (DRS) data (Schurch et al. 2014) (Fig. 8, A and B, middle and lower panels).Interestingly, both antisense transcripts are targets of the nuclear exosome (see CAGE data for the exosome mutants hen2-2 and rrp4-2).Consequently, 5′-ends are more prominent in the exosome mutants, suggesting that the transcripts are degraded rapidly after their synthesis.We could see that there was no precise start or end to the antisense transcription, but rather a window at both ends.Antisense transcription starts well beyond the poly(A)-site of their host genes and navigates until at least 1 kb upstream of the host gene's transcription start site (TSS).
To investigate the role of the antisense transcription at the translational level, firefly luciferase (LUC) reporter constructs for ZAT5 were generated, using GreenGate cloning system (Lampropoulos et al. 2013), with and without the 1,403 bp DNA sequence that harbors ZAT5 3′ UTR, associated PAS antisense transcript and putative promoter, i.e.ProZAT5:ZAT5-LUC-UTR-ASProZAT5 and ProZAT5:ZAT5-LUC-tNOS.The Agrobacterium tumefaciens cultures carrying these plasmids were used for infiltration of Nicotiana benthamiana leaves for transient expression assay (Supplementary Fig. S4, A and B).The first construct included the endogenous promoter and cDNA fused to the LUC gene.Downstream of the LUC gene was the endogenous untranslated 5′-region of ZAT5 and the antisense (AS) promoter.In the second construct, the ZAT5 promoter and cDNA were fused to LUC followed by the strong tNOS terminator.The tNOS terminator diminishes any antisense transcription over the LUC and ZAT5 gene body (Kindgren et al. 2018).The endogenous construct (construct 1 in Supplementary Fig. S4) showed an induction after cold exposure, confirming that the ZAT5 promoter is cold-responsive and corresponds with positive transcriptional regulation of ZAT5 by its antisense (Supplementary Fig. S4C).At both 22 °C and 4 ° C, the tNOS construct showed lower LUC activity compared with the full-length construct (Supplementary Fig. S4, D and E).The overall cold-responsiveness decreased from 2.5-fold in the fulllength construct to 2.0-fold in the tNOS construct, suggesting that the antisense transcription over ZAT5 has a positive role in the transcription level and stress-responsiveness of the gene.
To further elaborate their role, the promoter region of the antisense transcription was targeted with the CRISPR-Cas9 approach to minimally alter and knockout only parts of their regulatory sequence and/or the 5′-end of the TSS window.That way, the direct interference with sense transcription would be minimized.For ZAT5, we were able to retrieve a mutant with a 283 bp deletion, i.e. 392 bp from the start of the TSS window of asZAT5 and 710 bp from the poly(A) site for ZAT5 (Fig. 9A).We named this mutant aszat5-1.For BBX28, we aimed to delete the 5′-end of the TSS window of asBBX28.We retrieved a 392 bp deletion that included a deletion of 220 bp of the TSS window (asbbx28-1, Fig. 9A).The deletion is 193 bp from the poly(A) site of sense BBX28.In both mutants, we did not alter the stability of the respective mRNA, suggesting that we did not interfere with transcription termination (Supplementary Fig. S5, A and B).
In aszat5-1, there was a significant downregulation of the nascent transcription of asZAT5 at 22 °C but we could not detect any difference after 3 h 4 °C (Fig. 9B).For ZAT5, the mutant showed lower nascent transcription at both time points (Fig. 9B).The steady-state levels of ZAT5 showed a decreased level after cold treatment (Fig. 9B).These results corroborate the results from the ZAT5 constructs and suggest 2 key regulatory aspects, antisense transcription over the ZAT5 locus is required for proper cold induction and that CRISPR-Cas9 targeted asZAT5 promoter ZAT5 sequences contain regulatory elements in the antisense promoter to regulate the initiation of asZAT5 transcription.To show a biological role for asZAT5, we performed a cold acclimation and freezing test in the aszat5-1 mutant (Fig. 9C).We found a significantly lower freezing tolerance for the aszat5-1 (−4.8 ± 0.2 °C, P = 0.0021) compared with WT (−5.9 ± 0.2 °C).In asbbx28-1, we detected a significant downregulation of the nascent transcription of asBBX28 at 22 °C and lower active transcription of BBX28 at both 22 °C and 3 h 4 °C (Fig. 9D).The steady-state level of the sense BBX28 transcript was also lower in this mutant (Fig. 9D), corroborating that antisense transcription had a positive role on sense BBX28 transcription as it had on ZAT5 transcription.Additionally, we could see a significant decrease in the freezing tolerance in the mutant (−4.0 ± 1.1 °C, P = 0.0015 for asbbx28-1 and −5.9 ± 0.2 °C for WT, Fig. 9E).
Our results suggest that the antisense transcription of ZAT5 and BBX28 has a positive role in priming the sense transcription for stress response and that the characterized antisense transcription has an important biological role in the cold acclimation process.All in all, our study shows that antisense transcription can play a crucial role in priming certain plant stress-responsive TFs.

Discussion
A remarkable finding in our study is that the nascent transcriptional response to cold differs from the one detected by steadystate level measurements (Fig. 3).Additionally, we show that plaNET-seq complements steady-state methods (RNA-seq) to detect genome-wide transcriptional changes in the cold response (Fig. 3).A similar concept has earlier been proposed for heat stress in Arabidopsis (Liu et al. 2021).Thus, these reports highlight the importance of taking into account both active transcription and steady-state levels of mRNA to fully understand the response to stress (Sidaway-Lee et al. 2014).Changes in the active transcription of a gene measured by plaNET-seq have biological relevance, even though the steady-state level of the gene's mRNA is only slightly different to control conditions after cold exposure (Figs. 4 to 5).This report takes active transcription into account to find cold-responsive genes in Arabidopsis.
This study characterizes genes with antisense transcription that initiates from the 3′-end of genes in Arabidopsis (Figs. 1 to 2).Notably, we find a stark contrast in how the Arabidopsis genome implicates antisense transcription compared with other eukaryotes.In human cells, antisense transcription is most prevalent from early exon-intron junctions of genes (Brown et al. 2018).This type of antisense transcription is generally associated with low active transcription (Mayer et al. 2015), albeit with higher stability of the sense transcript (Brown et al. 2018).In Arabidopsis, we see almost the opposite scenario.Antisense transcription is most prevalent from the 3′-end of genes and their host genes are associated with high active transcription and fast turnover rates (Fig. 2) (Kindgren et al. 2019).In agreement, accelerated transcript degradation is suggested as an advantageous evolutionary strategy to facilitate genome-wide swift responses during cold stress in Arabidopsis (Chiba et al. 2012).It is likely that plants, being sessile organisms, have evolved distinct ways of gene regulation compared with other eukaryotes, especially when responding to biotic and abiotic stresses.This hypothesis is corroborated by the fact that many genes with antisense transcription in Arabidopsis are stress-regulated TFs (Fig. 1), proteins that are required to kick-start stress responses.

Exposure time 4°C (h)
Our data paint a picture of how plants keep their stressresponsive TFs in a constant "on" mode to prime their response to stress situations.A tempting hypothesis would be that antisense transcription is involved in sense transcript degradation, a generalized regulatory role that was widely postulated for several thousands of NATs in Arabidopsis and other plant species (Borsani et al. 2005;Held et al. 2008).However, our data do not endorse the concept of "universal gene silencing roles of NATs" but rather indicates that antisense transcription could have a more positive regulatory role on sense transcription (Reis and Poirier 2021).When antisense transcription is downregulated after cold exposure (Fig. 6 and Supplementary Fig. S2), sense transcript stability is lower compared with 22 °C (Fig. 5) and in our CRISPR-Cas9 deletion lines that exhibit reduced antisense transcription, we detected decreased sense mRNA levels at 4 °C (Fig. 9).In addition, our data from ZAT5 and BBX28 highlight that even a marginal reduction of their antisense transcription can impair the cold-responsiveness and ability of plants to acclimate to cold temperatures (Fig. 9).
Thus, an outstanding question and an important avenue for future research is how antisense transcription could relay a positive role to mRNA steady-state levels.A possible mechanism could be that the antisense transcription increases the stability of the sense transcript and assists in the translation of the sense transcript, as shown for the rice lncRNA, cis-NAT PHOSPHATE1;2 (cis-NAT PHO1;2) (Jabnoune et al. 2013).However, this is unlikely to be a general mechanism since most antisense transcripts are short-lived and degraded soon after their synthesis (Fig. 8) (Kindgren et al. 2019;Thieffry et al. 2020) and in our CRISPR lines, we did not see any effect of the mRNA stability (Supplementary Fig. S5).Another, more likely mechanism, could be antisense-promoted changes to the local chromatin environment which could be important in priming the optimal transcriptional response, resembling the example of NUCLEAR ENRICHED ABUNDANT TRANSCRIPT 1 (NEAT1) lncRNA in animals where the act of transcription itself at the NEAT1 locus was shown to be sufficient for biological function (Mao et al. 2011).It is possible that sense-antisense promoters can act autonomously both during developmental transitions and in stress conditions if uncoupled from the genomic context of each other, as shown for e.g.asDOG1, COOLAIR, and SVALKA (Swiezewski et al. 2009;Fedak et al. 2016;Kindgren et al. 2018).In contrast, the strong influence of native antisense transcription in cis over ZAT5 and BBX28 suggests that genomic proximity of the sense-antisense pairs could be crucial for the cold acclimation process (Fig. 9).In fact, genome-wide native transcriptional analysis in Saccharomyces cerevisiae reinforces the idea that the dynamic chromatin structures could be central in determining the landscape of eukaryotic sense-antisense transcription (Murray et al. 2015).
In Arabidopsis, the H3K27 demethylase RELATIVE OF EARLY FLOWERING-6 (REF6) has been proposed to recruit a chromatin remodeling complex that includes BRAHMA to regulate antisense transcription (Li et al. 2016;Archacki et al. 2017).Other repressing chromatin mechanisms have been described for COOLAIR, AGAMOUS INTRONIC RNA-4 (AG-incRNA4), and AUXIN PROMOTER REGULATED LOOP (APOLO), albeit their mechanisms of action are distinct (Ariel et al. 2014(Ariel et al. , 2020;;Csorba et al. 2014;Wu et al. 2018).In a parallel manner, it is possible for lncRNAs to mediate the deposition of activating marks in histones by recruiting other sets of modifiers, as has been shown for the antisense transcript of MADS AFFECTING FLOWERING 4 (MAF4) and MARNERAL SILENCING (MARS) in Arabidopsis (Zhao et al. 2018;Roulé et al. 2022) Detection of cold-responsive transcription factors | 3477 modification depositions at the sense promoters playing a positive role in cold tolerance in Chrysanthemum morifolium (Li et al. 2022).It will be central to focus on different histone marks and histone variants in future studies of antisense genes to elucidate noncoding transcription and its broader role(s) in plants.
Our discoveries take the first step to a broader role of antisense transcription in plants and support the notion that transcription from the complementary strand modulates the responsiveness of stress genes.Furthermore, our study moves away from the paradigm that antisense transcription has a prominent silencing role in plants instead supporting a more positive regulatory function.

Plant growth, mutants and CRISPR-Cas9 mutant generation
For the WT background Arabidopsis (A. thaliana) Col-0 or Columbia accession was employed.For the growth of plants, seeds were surface sterilized and stratified for 2 to 4 d at 4° C in the dark and either transferred to soil directly or plated on ½ Murashige and Skoog (MS) basal medium supplemented with 1% (w/v) sucrose.Plants were grown in long-day conditions [16 h light, 8 dark, ∼100 µE, SciWhite LEDs (Percival)] for 10 d.Biological replicates in all experiments represent approximately 20 to 30 seedlings grown on separate plates.Cold treatment (4° C, ∼25 µE) was initiated at ZT4 to replicate the conditions set by the plaNET-seq dataset.T-DNA insertional mutant lines, viz zat5-1 (SALK_041934), bbx28-1 (SAIL_412_A09), and hen2-2 (GABI_774H07) (Lange et al. 2014) were genotyped and confirmed for homozygosity by PCR.For the CRISPR-Cas9 mutants, guide RNAs (gRNAs) were designed using the CHOPCHOP webserver (http://chopchop.cbu.uib.no/) and a 2gRNA fragment was amplified using DT1T2 plasmid (Xing et al. 2014) template using Phusion DNA polymerase (Thermo Fisher Scientific).Oligos used can be found in Supplementary Data Set 6.The PCR product was electrophoresed, and gel purified followed by GreenGate reaction into a modified pHSE401 binary vector as described before (Xing et al. 2014).In the modified pHSE401 vector, the hygromycin resistance has been replaced by a GFP seed coat expression cassette for faster screening.Final plasmids were verified by sequencing and transformed into WT Col-0 plants by A. tumefaciens (GV3101) floral dip.T1 seeds were first selected by visual screening for GFP expression followed by PCR genotyping.A further selection of the T2 lines was performed by picking seeds lacking the GFP signal, and then homozygous plants were confirmed by PCR for the genomic deletion and the absence of the Cas9 construct.Seeds from homozygous plants were used in experiments.

Generation of reporter constructs
We used GreenGate cloning system for generation of firefly luciferase (LUC) gene reporter constructs (Lampropoulos et al. 2013).To construct ProZAT5:ZAT5-LUC-UTR-ASProZAT5 and ProZAT5:ZAT5-LUC-tNOS, 2,438 bp fragment upstream (1,408 bp promoter and 1,030 CDS of ZAT5 without stop codon) was PCR amplified from genomic DNA.Separate PCR amplification was carried out for 1,402 bp long UTR-AS ZAT5 prom fragment using proofreading Phusion DNA polymerase using a genomic DNA template.LUC and tNOS terminator (tNOS) were also separately PCR amplified in a similar manner.Subsequently, different PCR products were cloned into respective GreenGate entry modules by employing BsaI/T4 DNA ligase and using reaction conditions as described earlier (Lampropoulos et al. 2013).All GreenGate entry plasmids were confirmed by restriction digestion and DNA sequencing.Finally, the GreenGate reaction was performed for the assembly of 6 entry modules to create final destination plasmids, i.e.ProZAT5:ZAT5-LUC-UTR-ASProZAT5 and ProZAT5:ZAT5-LUC-tNOS using pGGZ003 as backbone vector according to the protocol reported earlier (Lampropoulos et al. 2013).Oligos used can be found in Supplementary Data Set 6.

Transient agroinfiltration and luciferase assay
Agrobacterium strain GV3101 was separately transformed with plasmids harboring prom ZAT5::ZAT5-LUC-UTR-ASZAT5 prom and prom ZAT5::ZAT5-LUC-tNOS constructs and plated on LB media containing 10 µg/ml rifampicin, 25 µg/ml gentamicin, and 75 µg/ml spectinomycin.After 48 h of growth positive colonies were selected and grown in 5 ml of LB medium supplemented with antibiotics.Additionally, the presence of constructs in Agrobacteria was confirmed by colony PCR.After overnight growth at 30 °C, 30 ml of fresh induction media (LB supplemented with 10 mM MES pH 5.6, 20 μM acetosyringone, 25 µg/ml, gentamicin, 75 µg/ ml spectinomycin, and 10 µg / ml rifampicin) inoculated.When OD 600 reached 0.5, cells were harvested by centrifugation and resuspended in infiltration media (LB supplemented with 10 mM MES pH 5.6, 10 mM MgCl 2 , and 150 μM acetosyringone without antibiotics) to obtain an OD 600 value of 1 followed by incubation of bacterial culture at room temperature for a minimum of 3 h.Finally, leaves from 6-wk-old N. benthamiana plants were infiltrated with bacterial suspension and the area was delimited and marked with a marker.
Seventy-two hours post agroinfiltration, half of the randomly selected N. benthamiana plants (a minimum of 5 independent plants each with 2 to 3 infiltrated leaves for individual constructs) were subjected to cold stress treatment (3 h cold 4 °C) at ZT4.Control plants were maintained at 22 °C.Immediately after 3 h of cold stress, previously infiltrated leaves with marked areas from cold-treated and control plants were carefully and quickly re-infiltrated with 5 mM working solution of D-Luciferin (GoldBio).The ratio of 5 mM D-Luciferin in 0.01% (v/v) Triton X-100 and sterile water was kept at 1:3 during re-infiltration of leaves.Further, several 1 cm leaf discs were prepared from marked re-infiltrated area and subjected for luminescence measurement after 5 to 10 min of incubation at respective temperatures by GloMax Navigator Microplate Luminometer.Data were extracted and analyzed using Excel and GraphPad software.

Electrolyte leakage assay
Electrolyte leakage measurements were carried out according to a previous report (Kindgren et al. 2015).In short, plants were grown in short days (8 h light /16 h dark cycle) for 4 w.For the cold acclimation experiments, WT and mutant plants were transferred to a cold chamber set at 4 °C for 4 d without changing photoperiodic conditions.Randomized leaf discs of 1 cm diameter, for each genotype in triplicates from several similar-sized leaves, were prepared using a cork borer for acclimated or non-acclimated plants and carefully placed horizontally in a manner to avoid floating in clean glass tubes filled with 200 µl deionized distilled water.The tubes containing 2 leaf discs were then transferred to a programmable freezing bath (FP51, Julabo, Germany) set at −2° C.After 45 min, icing was induced manually in each tube with the help of liquid N 2 and a metallic stick.Temperature decrease occurred at the rate of −1 °C per 30 min, and samples were taken out at designated temperature point(s) followed by incubation on ice for at least 1 h in the cold room (4 °C).Soon after the collection and 1 h ice-incubation of tubes, 1.3 ml of water was added to each tube and placed on a shaker overnight at 4° C and conductivity was measured using a conductivity cell (CDM210, Radiometer, Denmark) on the next day.Finally, all tubes were subjected to flash freeze using liquid N 2 and left on a shaker overnight at room temperature.To obtain the total electrolyte content from leaf discs, conductivity was measured again, and the % of electrolyte leakage was calculated using the formula -(conductivity before flash freeze/conductivity after flash freeze) * 100.Data were fitted into a sigmoidal dose-response curve using GraphPad Prism software and significant differences in the fit were determined with an extra sum-of-squares F test.

RNA extraction, cDNA synthesis, and RT-qPCR
Total RNA extraction from plant material was carried out using RNeasy Plant Mini Kit (QIAGEN) as per suppliers' instructions.The extracted RNA was additionally treated with dsDNase (Thermo Fisher Scientific) for the elimination of genomic DNA contamination.Successively, cDNA synthesis was carried out using Superscript IV (Invitrogen) reverse transcriptase as per manufacturer's instructions using strand-specific RT primers carrying a sequence tag (GACTGGAGCACGAGGACACT) at 5′ end (Parent et al. 2015;Kindgren et al. 2018) together with a reference gene.Quantitative real-time PCR (RT-qPCR) was performed on CFX96 and CFX384 Real-Time PCR detection systems (Bio-Rad) using SYBR premix (Bio-Rad), cDNA, reverse primer (aligning to tag sequence), and appropriate forward primers at the concentration of 10 pmol/µl with the PCR cycler following initial denaturation at 95 ° C for 30 s, standard 40 cycles of 94 ° C for 10 s and 60 ° C for 30 s.The specificity of RT-qPCR products was assessed from the single peak melt curves.For the data analysis, the C q values from a minimum of 3 biological replicates with 2 to 3 technical replicates were averaged and △C q was obtained as C q (gene of interest) − C q (reference gene).Final calculations were performed by 2 (−Δcq) or 2 (−ΔΔcq) , adjusted to experimentally determined primer efficiency for determination of fold change in gene expression levels.Statistical significant differences were calculated with Student's t-test.Primers used are listed in Supplementary Data Set 6.

Figure 1 .
Figure 1.PAS genes are overrepresented by stress-responsive TFs.A) Graphical representation of the definition of PAS genes.TSS, transcription start site; asTSS, antisense TSS.B) GO term enrichment of PAS genes (molecular function).The number in bars indicates the found number of genes/ expected number of genes.C) GO term enrichment of PAS genes (biological processes).The number in bars indicates the found number of genes/ expected number of found genes.D) Example of a PAS gene (WRKY48, At5g49520).Screenshots showing plaNET-seq expression profile from datasets at 22° C, 3 h, and 12 h post cold treatment.Elevated transcriptional activity is indicated by higher peak density and amplitude.
Figure2.PAS genes are highly expressed, but their mRNA is degraded rapidly.A) Violin plot of the length of PAS non-TF genes, TFs non-PAS, PAS TF genes, and all expressed genes.Centerline, median; box limits, upper and lower quartiles; whiskers, 1.5× interquartile range.P-value was calculated with Mann-Whitney U test and P < 0.05 was considered significant.B) Violin plot of the steady-state level of PAS non-TF genes, TFs non-PAS, PAS TF genes, and all expressed genes.Centerline, median; box limits, upper and lower quartiles; whiskers, 1.5× interquartile range.P-value was calculated with Mann-Whitney U test and P < 0.05 was considered significant.C) Violin plot of the decay rate of PAS non-TF genes, TFs non-PAS, PAS TF genes, and all expressed genes after transcriptional inhibition by cordycepin.Centerline, median; box limits, upper and lower quartiles; whiskers, 1.5× interquartile range.P-value was calculated with Mann-Whitney U test and P < 0.05 was considered significant.D) Metagene analysis of plaNET-seq data of PAS non-TF genes, TFs non-PAS, PAS TF genes, and all expressed genes.The shaded area shows a 95% confidence interval for the mean.

Figure 3 .
Figure3.Discrepancy between plaNET-seq and RNA-Seq.A) Number of total genes differentially expressed in RNA-seq and plaNET-seq after 3 and 12 h of 4 °C exposure.The overlap between the 2 sequencing techniques is shown in darker gray.B) and C) Correlation plots between differentially expressed genes in RNA-seq and plaNET-seq (UP: squares, DOWN: diamonds) after B) 3 h of 4 °C and C) 12 h of 4 °C.The genes plotted come from the overlap seen in A).D) Number of PAS genes differentially expressed in RNA-seq and plaNET-seq after 3 and 12 h of 4 °C exposure.The overlap between the 2 sequencing techniques is shown in blue.E) and F) Correlation plots between differentially expressed PAS genes in RNA-seq and plaNET-seq (UP: squares, DOWN: diamonds) after E) 3 h of 4 °C and F) 12 h of 4 °C.The genes plotted come from the overlap seen in D).

Figure 4 .
Figure 4. PAS genes have enhanced cold-responsiveness.A) Metagene analysis of the plaNET-seq signal (at 22 °C) in a 500 bp window centered at the +1 nucleosome.PAS genes that will be UP after 3 h at 4 °C, genes without antisense transcription but UP after 3 h at 4 °C, and non-DE genes are shown.The shaded area shows a 95% confidence interval for the mean.B) Metagene analysis of plaNET-seq data (at 22 °C).PAS genes that will be UP after 3 h at 4 °C, genes without antisense transcription but UP after 3 h at 4 °C, and non-DE genes are shown.The shaded area shows a 95% confidence interval for the mean.C) Metagene analysis of the plaNET-seq signal (after 3 h at 4 °C) in a 500 bp window centered at the +1 nucleosome.PAS UP genes after 3 h at 4 °C, genes without antisense transcription but UP after 3 h at 4 °C, and non-DE genes are shown.The shaded area shows a 95% confidence interval for the mean.D) Metagene analysis of plaNET-seq data (after 3 h at 4 °C).PAS UP genes after 3 h at 4 °C, genes without antisense transcription but UP after 3 h at 4 °C, and non-DE genes are shown.The shaded area shows a 95% confidence interval for the mean.

Figure 5 .
Figure5.ZAT5 and BBX28 mRNA show decreased stability at 4 °C.A) At4g254970 (CBF2).Screenshots are from plaNET-seq and RNA-seq datasets.Elevated transcriptional activity is indicated by higher peak density and amplitude.B) At3g17609 (HYH).Screenshots are from plaNET-seq and RNA-seq datasets.Elevated transcriptional activity is indicated by higher peak density and amplitude.C) to H) Transcript stability assays for C) EIF4A1, D) ZAT5, E) BBX28, F) COL7, G) MYB47, and H) SCL8 after transcriptional inhibition with cordycepin at 22 °C and 4 °C.Half-life (t 1/2 ) was determined from the slope of degradation curves that were obtained after RT-qPCR analysis of cordycepin-treated seedlings at indicated time points.Each data point is the mean of 3 biological replicates.Error bars represent ± SD.

Figure 6 .
Figure 6.Examples of cold-responsive PAS genes.A) AT2G28200 (ZAT5).Screenshots are from plaNET-seq and RNA-seq datasets.Elevated transcriptional activity is indicated by higher peak density and amplitude.B) AT4G27310 (BBX28).Screenshots are from plaNET-seq and RNA-seq datasets.Elevated transcriptional activity is indicated by higher peak density and amplitude.

Figure 7 .
Figure 7. ZAT5 and BBX28 are important for cold acclimation.A) The relative steady-state level of ZAT5 measured by RT-qPCR following cold exposure (4 °C).Steady-state levels were normalized to WT levels at 22 °C.The mean values are from 3 biological replicates.Error bars represent ± SEM.Statistical significance was calculated with Student's t-test (*P < 0.05).B) The relative steady-state level of BBX28 measured by RT-qPCR following cold exposure (4 °C).Steady-state levels were normalized to WT levels at 22 °C.The mean values are from 3 biological replicates.Error bars represent ± SEM.Statistical significance was calculated with Student's t-test (*P < 0.05).C) Electrolyte leakage in WT and zat5-1 of cold-acclimated (4 d of 4 °C) plants.Each data point represents the mean from at least 3 biological replicates (±SEM).The dashed line represents the threshold value, 50%.The dotted lines represent the curve fit.Statistical significance was calculated with an extra sum-of-squares F-test and the P-value is shown in the figure.D) Electrolyte leakage in WT and bbx28-1 of cold-acclimated (4 d of 4 °C) plants.Each data point represents the mean from at least 3 biological replicates (±SEM).The dashed line represents the threshold value, 50%.The dotted lines represent the curve fit.Statistical significance was calculated with an extra sum-of-squares F-test and the P-value is shown in the figure.

Figure 8 .
Figure 8. Characterization of asZAT5 and asBBX28.Screenshots from plaNET-seq (upper panel), CAGE (middle panel), and DRS-seq (lower panel) for A) ZAT5 and B) BBX28.0 indicates the TSS of the sense transcript.Elevated transcriptional activity is indicated by higher peak density and amplitude.

Figure 9 .
Figure 9. asZAT5 and asBBX28 are important for the proper regulation of their host gene.A) Graphical representation of the ZAT5 and the BBX28 loci showing the location of sequence targeted by CRISPR-Cas9 to generate an asZAT5 knockdown line (aszat5-1) and an asBBX28 knockdown line (asbbx28-1).Antisense transcription start site (asTSS) window and distances of knocked out the genomic sequence from 3′-end are marked with dotted lines.The location of RT-qPCR probes for sense and antisense are shown.B) The relative nascent and steady-state level of asZAT5 and ZAT5 in WT and aszat5-1 measured by RT-qPCR at 22 °C and following 3 h of cold exposure (4 °C).All levels were normalized to WT levels at 22 °C.The mean values are from 3 biological replicates.Error bars represent ± SEM.Statistical significance was calculated with Student's t-test (*P < 0.05).C) Electrolyte leakage in WT and aszat5-1 of cold-acclimated (5 d of 4 °C) plants.Each data point represents the mean from at least 3 biological replicates (±SEM).The dashed line represents the threshold value, 50%.The dotted lines represent the curve fit.Statistical significance was calculated with an extra sum-of-squares F-test and the P-value is shown in the figure.D) The relative steady-state level of asBBX28 and BBX28 in WT and asbbx28-1 measured by RT-qPCR at 22 °C and following 3 h of cold exposure (4 °C).All levels were normalized to WT levels at 22 °C.The mean values are from 3 biological replicates.Error bars represent ± SEM.Statistical significance was calculated with Student's t-test (*P < 0.05).E) Electrolyte leakage in WT and asbbx28-1 of cold-acclimated (5 d of 4 °C) plants.Each data point represents the mean from at least 3 biological replicates (±SEM).The dashed line represents the threshold value, 50%.The dotted lines represent the curve fit.Statistical significance was calculated with an extra sum-of-squares F-test and the P-value is shown in the figure.