What can cold‐induced transcriptomes of Arctic Brassicaceae tell us about the evolution of cold tolerance?

Abstract Little is known about the evolution of cold tolerance in polar plant species and how they differ from temperate relatives. To gain insight into their biology and the evolution of cold tolerance, we compared the molecular basis of cold response in three Arctic Brassicaceae species. We conducted a comparative time series experiment to examine transcriptional responses to low temperature. RNA was sampled at 22°C, and after 3, 6, and 24 at 2°C. We then identified sets of genes that were differentially expressed in response to cold and compared them between species, as well as to published data from the temperate Arabidopsis thaliana. Most differentially expressed genes were species‐specific, but a significant portion of the cold response was also shared among species. Among thousands of differentially expressed genes, ~200 were shared among the three Arctic species and A. thaliana, while ~100 were exclusively shared among the three Arctic species. Our results show that cold response differs markedly between Arctic Brassicaceae species, but probably builds on a conserved basis found across the family. They also confirm that highly polygenic traits such as cold tolerance may show little repeatability in their patterns of adaptation.

in summer, which is important for timing of plant development and circadian rhythm. In the Arctic, the average temperature of the warmest summer month is not more than 10°C (Elvebakk, 1999), and plants must always be prepared for occasional frost. By studying how Arctic plants cope with low temperatures, we can gain insights into how plants of temperate origins acquire additional cold tolerance, and if there are general trends in plant adaptation to extreme polar environments.
Plants face many challenges upon the transition to colder environments. Low temperatures can affect nearly all aspects of plant cell biochemistry, including protein properties, photosynthesis reactions, cell membrane fluidity, and the creation of gametes (Bomblies et al., 2015;Shi et al., 2018). Ice formation comes with its own set of challenges and is deadly if ice forms within the cell (Körner, 2003).
Some plants can tolerate freezing of the apoplast (the space between the cells), but this may draw water out of the cell and lead to severe dehydration, as well as increase the level of salts and toxic solutes (Körner, 2003;Steponkus, 1984;Wisniewski et al., 2004;Wisniewski & Fuller, 1999). In temperate areas, from where most Arctic plants are probably derived (Abbott & Brochmann, 2003), plants prepare for cold periods via cold acclimation, that is, increasing freezing tolerance in response to low nonfreezing temperatures (Thomashow, 2010). Exposing temperate plants to low temperatures typically results in complete reorganization of the transcriptome, ultimately leading to increased freezing tolerance (Kilian et al., 2007;Kreps et al., 2002;Thomashow, 2010).
The CBF transcription factors (C-repeat-binding factors) are among the main "regulatory hubs" of the plant cold response, and have been isolated in many different plant species (Park et al., 2015;Shi et al., 2018;Thomashow, 2010). The CBFs are induced shortly after exposure to cold stress, and control the expression of >100 cold-regulated (COR) genes downstream (the CBF pathway ;Park et al., 2015). Arabidopsis thaliana exhibits three cold-induced CBFs (CBF1, CBF2 and CBF3, also called DREB1b, DREB1c and DREB1a;Jia et al., 2016), but these transcription factors are not always found in other species (Zhao et al., 2012). It is becoming increasingly clear that the CBF regulon involves extensive coregulation by other, lesser-known transcription factors, and that the low temperature regulatory network is complex (Park et al., 2015).
One could envision that polar plant species are in less need of a cold acclimation period than temperate species, as temperatures are low year-round and the risk of sudden summer frost is high (e.g., their transcriptomes could be less responsive to a drop in temperature), or that their cold response is somehow more complex (e.g., involving more genes) or more effective (e.g., faster, or involving fewer genes). These hypotheses are relatively unexplored, as there are few in-depth studies of cold-induced transcriptomes in polar plants. Archambault and Strömvik (2011) have studied the Arctic Oxytropis, Wang et al. (2017) the temperate-subarctic Eutrema (Thellungiella) salsugineum, and Lee et al. (2013) the Antarctic Deschampsia antarctica. These studies give valuable species-specific information on cold response, but only limited insight into how polar plant species differ from temperate relatives. In this study, we therefore perform a whole transcriptome investigation of cold response in plant species independently adapted to the high Arctic and compare their response to previously published data from a temperate relative.
The focal species selected for this study, Cardamine bellidifolia, Cochlearia groenlandica, and Draba nivalis, belong to different temperate lineages that colonized the Arctic independently (Carlsen et al., 2009;Jordon-Thaden et al., 2010;Koch, 2012) and represent three of the main clades of the Brassicaceae family (clade A, B, and C; divergence time ~30 Ma; Huang et al., 2016). They are ideal model species for studying cold response of Arctic plants for four main reasons: (a) all have their main distribution above the Arctic Circle, (b) they are found in all Arctic bioclimatic subzones (even in polar desert; Elven et al., 2011), (c) they belong to the family in which cold response has been most extensively studied, as it includes the model species A. thaliana and various economically important crop species (Kilian et al., 2007;Park et al., 2015;Shi et al., 2018), and (d) we recently found evidence of positive selection on genes associated with cold stress in all three species (Birkeland et al., 2020).
Different genes belonging to similar stress response pathways seem to be under positive selection in different species, suggesting that the three species represent independent adaptation to the Arctic (Birkeland et al., 2020). However, our previous study was limited to protein coding regions, and theory predicts that there could be a higher chance of convergence in their expression profiles (e.g., Sackton et al., 2019;Stern, 2013). The reason is that mutations in cisregulatory regions should have fewer pleiotropic effects than mutations in coding regions, as protein function may be affected only in a subset of the full expression domain (Gompel & Prud'homme, 2009;Stern, 2013). This is tied to the fact that a gene can have several cis-domains and that one cis-domain may bind several different transcription factors.
To assess the degree of similarity between Arctic Brassicaceae expression profiles, we subjected the three species to a simulated summer cold shock and identified differentially expressed genes (DEGs) after 3, 6 and 24 h with cold treatment. We aimed to (i) characterize the cold-induced transcriptomes of C. bellidifolia, C. groenlandica, and D. nivalis, (ii) describe how their cold response differs from that of their temperate relative A. thaliana, and (iii) identify potential convergent expression patterns in the three Arctic species.

| Plant material
For each species, we sowed seeds from a single selfed parent derived from wild Arctic populations from Alaska (Table S1). Several seeds were sown in each of six pots per species, and each pot (with several germinated seeds) was treated as one replicate. Because the plants were siblings derived from selfed parents, and because selfing is assumed to be the predominant mode of reproduction in these species in the wild (Brochmann & Steen, 1999), we expected the six replicates to be close to genetically identical. The plants were grown in a natural daylight room in the phytotron at the University of Oslo from

| Cold shock treatment
Eight weeks after sowing, the plants were given a 24-h cold shock to simulate a sudden drop in temperature during a typical Arctic summer. At 1:00 PM (to minimize correlation with circadian change), we transferred the pots from 22°C in the daylight room to 2°C in a cooling room with artificial light from 250 W high-intensity discharge lamps (140μmol m −2 s −1 measured at plant height) to mimic light conditions in the room where the plants were raised. Leaf tissue was sampled at four time points; just before they were transferred (0 h; control), and after 3, 6, and 24 h. We sampled all six pots for RNA extractions at each time point, but only used the four best RNA extracts per time point for sequencing (i.e., in terms of RNA quality and quantity). These were used as four biological replicates per time point per species. RNA was immediately extracted from fresh, fully expanded leaves as described below. were quality-checked with FastQC version 0.11.8 (Andrews, 2010), and a single FastQC report per species was generated with MultiQC version 1.7 (Ewels et al., 2016).

| Transcriptome assembly and annotation
As there were no available genome assemblies at the start of this study, we assembled a reference transcriptome de novo for each species using Trinity version 2.8.5 (Grabherr et al., 2011) based on all acquired RNA samples. Trinity was run with the integrated Trimmomatic option (Bolger et al., 2014), strand-specificity, and a minimum assembled contig length of 300 bp. The transcriptomes were filtered and annotated with eukaryotic non-model transcriptome annotation pipeline (EnTAP; Hart et al., 2020) in two rounds: first to apply the EnTAP filtering option on the raw transcriptome (in order to reduce inflated transcript estimates), and then to annotate the highest expressed isoform and filter out contaminants (used for the annotation of DEGs; see below). For expression filtering, an alignment file was generated with bowtie2 (Langmead & Salzberg, 2012) in combination with RSEM (Li & Dewey, 2011) using default options in the "align_and_estimate_abundance.pl" script provided with the Trinity software suite. Numbers of complete and fragmented benchmarking universal single-copy orthologues (BUSCOs) in the filtered transcriptomes were estimated with BUSCO version 4.0.6 (Simão et al., 2015). The filtered transcriptomes were used as the final reference in the differential expression analyses.

| Differential expression analyses
The Trimmomatic filtered reads were mapped to the reference transcriptomes using the alignment free mapper Salmon with a GC content bias correction (Patro et al., 2017). Genes that were differentially expressed after 3, 6 and 24 h of cold treatment were identified with DESeq2 version 1.22.1 (Love et al., 2014), using a design formula controlling for the effect of pot number (design = ~pot number + time). This means that we tested for the effect of time with 2°C treatment, while controlling for the individual effects of the sampled pots. A generalized linear model was fitted to each gene and a Wald test (Love et al., 2014) applied to test if the 3, 6 and 24 model coefficients differed significantly from zero when contrasted to the 0 h model coefficient. A gene was considered as differentially expressed if the transcript level exhibited ≥ twofold change in response to the cold treatment at the different time points (log2 fold change ≥1). p-values were adjusted using the Benjamini-Hochberg procedure (Benjamini & Hochberg, 1995) to control the false discovery rate (FDR) and we considered FDR 0.05 as a cutoff for significance.
Heatmaps of the top 30 differentially expressed genes with the lowest false discovery rate were generated with the pheatmap package in R (Kolde, 2019) using the regularized log function (rld) on original count data. The mean expression value of a gene was subtracted from each observation prior to heatmap generation.

| Comparison of DEG sets among Arctic species and A. thaliana
To enable comparison of DEGs among the Arctic species, and among the Arctic species and A. thaliana, we used published data on DEGs in wild-type A. thaliana in response to 24 h chilling treatment (Table S1 in Park et al., 2015). In this experiment, A. thaliana wild type plants were grown at 22°C and constant illumination, then exposed to a 4°C chilling treatment for 24 h (Ws-2; see Park et al., 2015 for details). We used two different approaches to compare the 24 h DEG sets among species.
First, we ran OrthoFinder version 2.3.12 (Emms & Kelly, 2019) to identify groups of orthologues among the four species (orthogroups), using the assembled transcriptomes (filtered based on highest expressed isoform) and the Araport11 peptide file of A. thaliana downloaded from www.arabi dopsis.org. Second, we used the BLASTX (proteinprotein) search of BLAST+ version 2.12.0 (Camacho et al., 2009) to identify putative A. thaliana homologues in the three Arctic species, using the Araport11 peptide file as database and each of the Arctic transcriptome files as query (e-value < 1e-10; choosing the output with the highest bit score). This second approach enabled us to compare gene identity and function more closely among species based on gene pairs. The significance of the overlaps among differentially expressed orthogroups at 24 h was evaluated using the supertest function in SuperExactTest version 1.0.7 (Wang et al., 2015). We also visualized potential unique overlaps among differentially expressed orthogroups at 24 h and among differentially expressed A. thaliana homologues at 24 h, using UpSetR version 1.4.0 (Conway et al., 2017). To compare transcription factor composition, we annotated the A. thaliana 24 h DEG set with EnTAP (as above).

| Gene function enrichment analyses
To functionally characterize sets of upregulated and downregu-  Gentleman et al., 2004). The elim algorithm works by traversing the GO-bottom-up and discarding genes that already have been annotated to significant child terms, and this is the recommended algorithm by the creators of topGO due to its simplicity (Alexa et al., 2006). For the enrichment analyses, we used the gene annotations of the transcriptomes as background gene sets in each test (using the GO annotations acquired with EnTAP). For A. thaliana, we used the org.At.tair.db R package version 3.7.0 to annotate the 24 h DEG set of Park et al. (2015), and for creating a background gene set used in the enrichment tests (Carlson, 2018). A GO term was considered significantly enriched if p < .05. We did not correct for multiple testing as the enrichment-tests were not independent.
Overlaps among enriched GO terms in similar DEG sets (i.e., up-and downregulated genes at similar time points) were compared among species using UpSetR as above.

| Comparison of DEGs with data set on positively selected genes
In a previous paper, we identified convergent substitutions and tested for positive selection in the same three Arctic Brassicaceae species (see Birkeland et al., 2020 for details). We were thus able to check for potential overlaps between positively selected genes/ convergent genes and the 24 h DEG sets. We blasted the newly assembled transcriptomes against the transcriptomes of our previous study using BLASTX with an e-value cutoff of <1e-10 and choosing the result with the highest bit score.

| Gene coexpression network analyses
To identify coexpressed gene modules, we performed weighted correlation network analysis for each species using the R package WGCNA . The gene expression matrix was prepared by first filtering out genes with consistent low counts (row sums ≤ 10), and then applying a variance stabilizing transformation within DESeq2. We also filtered out genes with low expression variance by only maintaining genes with variance ranked above the 25th percentile in each data set. A gene adjacency matrix was constructed using default settings and raised to a soft thresholding power of 18 (signed network type). The scale-free topology model fit was only 0.70 for C. bellidifolia, 0.79 for C. groenlandica, and 0.56 for D. nivalis (Figures S1-S3). However, this is not uncommon when subsets of samples that are globally different from the rest (for instance cold treated vs. noncold treated samples) lead to high correlation among large groups of genes (see Langfelder & Horvath, 2017). The adjacency matrix was translated into a topological overlap matrix (TOM) and coexpressed gene modules were identified by average linkage hierarchical clustering and the dynamic tree-cut algorithm (minimum module size of 30, deepSplit of 2, and pamRespectsDendro as FALSE; . Closely adjacent modules were merged based on the correlation of their eigengenes (correlation threshold of 0.75). We identified hub genes within each module as the genes with the top 10 % eigengene-based connectivity (module membership). We also calculated the correlation between each module eigengene and a binary cold trait (i.e., cold treated vs. nontreated; 1/0) as well as a time-based trait (0, 3, 6, and 24 h with cold). Gene function enrichment analyses for each module were performed with topGO as described above. To check for modules overlapping between species, we made UpSet plots based on orthogroup identity. For these plots we included only modules that had a positive correlation ≥0.70 with the binary cold trait or 3, 6, or 24 h with cold.

| Transcriptome assemblies and differentially expressed genes (DEGs)
The three de novo assemblies contained ~22,000-24,000 (Trinity) genes and were highly complete in terms of BUSCOs (>90% complete; Table 1). We identified a gradual increase in the number of DEGs with time at 2°C from about 700-1000 DEGs after 3 h to 2500-3000 DEGs after 24 h (Tables 2 and S2-S4). After 3 h at 2°C, most DEGs were upregulated, but after 24 h we found similar numbers of down-and upregulated DEGs in all species (Table 2). Many 3 and 6 h DEGs were also found in the 24 h set, but some DEGs were unique for each time point (Tables S5-S7).
Most 24 h DEGs were species-specific, meaning that they did not belong to shared orthogroups. However, 212 DEGs were shared by the three Arctic species and A. thaliana, and 106 were shared by the Arctic species but not by A. thaliana (based on orthogroup membership; Figure 1). When testing for significant overlaps between the four 24 h DEG sets, we found that all species combinations had larger overlaps than expected by chance (based on orthogroup identity; all p < .01 based on the supertest; Table S8). Thus, our main findings were that (a) the shared portion of the cold response was relatively small, but larger than expected by chance, and (b) more cold responsive genes were shared by the three Arctic species and A. thaliana than by the Arctic species alone.

| Comparison of DEG annotations among species
Transcription factors made up 9%-14% of all Arctic DEGs (Tables 3   and S2- We performed gene function enrichment analyses to functionally characterize the DEG sets and found that most of the significantly enriched GO terms within the BP, CC, and MF domains were species-specific ( Figure 3, Tables 4 and S9-S18). However, we found that 20 GO terms were significantly enriched in all 24 h upregulated DEG sets (i.e., in the three Arctic species and A. thaliana), while 6 GO-terms were significantly enriched exclusively in the Arctic 24 h upregulated DEG sets (BP, CC, and MF; Table S18). Among the GO terms shared among the Arctic species and A. thaliana were many BP terms associated with cold and freezing (e.g., "response to salt stress", "response to water deprivation", and "response to oxidative stress" ; Figures 3-4). Shared terms also included those that were associated with the hormones abscisic acid and ethylene, karrikin, and possible cryoprotectants (flavonoids and sucrose; BP, Figure 4). The GO terms that were significantly enriched exclusively in the Arctic species included terms such as "spermidine biosynthetic process" (BP) and "arginine decarboxylase activity" (MF). All 24 h downregulated DEGs in the Arctic species and A. thaliana were enriched for genes associated with growth-related GO terms such as "phototropism" (BP; all species) and "auxin-activated signalling pathway" (BP; only Arctic species).
We examined the annotations of the DEGs that were shared exclusively among the three Arctic species, and among the three Arctic species and A. thaliana (based on putative A. thaliana homologues; Tables S19-S23). The exclusively shared Arctic DEGs included several transcription factors (e.g., ddf2, ABI5, REVEILLE 2, RAP2.2, PCL1, and HY5) and MAPK related genes (MAPK7, MAPKKK14, and MAPKKK18), while DEGs shared among the Arctic species and A. thaliana included known cold induced genes such as LEA14, COR15B, TCF1, COR27 and COR28 (upregulated in all species; Tables S19-S20). Many DEGs that were exclusively shared among the Arctic species, as well as among the Arctic species and A. thaliana, were found among the topmost differentially expressed genes with the lowest false discovery rate in the Arctic focal species ( Figure S4).  ; Tables S24-S28). Through the same approach we also identified several cold responsive genes previously found to contain convergent substitutions between C. bellidifolia and C. groenlandica, C. bellidifolia and D. nivalis, and/or C. groenlandica and D. nivalis (Tables S25-S28; Supporting Information Text 1).

| Gene coexpression modules
The gene coexpression network analyses resulted in 13 coexpressed modules in C. bellidifolia, 23 in C. groenlandica and 14 in D. nivalis  in each species stood out as highly positively correlated with the binary cold trait (meaning that the eigengene increased continuously with cold; Pearson correlation coefficient >0.90 and p < .001; Figure 5). These were the light yellow module of C. bellidifolia, the light cyan module of C. groenlandica, and the dark orange 2 module of D. nivalis. They shared 42 coexpressed orthogroups ( Figure S6) and were significantly enriched for 16 of the same GO terms, including "response to cold", "response to abscisic acid", and "regulation of transcription" (Tables S29-S38). However, most GO terms were module or species-specific. The three modules mostly had different hub genes (i.e., genes with top 10% highest kME; Tables S39-S41), but MAPK7 was a hub gene in the light yellow and dark orange 2 modules (belonging to C. bellidifolia and D. nivalis, respectively), while CONSTANS-like 4 was a hub gene in the light cyan and dark orange 2 modules (belonging to C. groenlandica and D. nivalis, respectively).
In addition, several related genes such as REVEILLE1, 2 and 6, and CONSTANS-like 4, 9, 10 and 13 had high kME in all modules, but the exact genes differed among species. The light yellow, light cyan and dark orange 2 modules also included many well-known coldregulated genes, such as COR78 and COR47 (light yellow; C. bellidifolia), CBF1 (light cyan; C. groenlandica) and COR27 (dark orange 2;

D. nivalis).
Three modules that were positively correlated with 24 h of cold stood out as highly overlapping among species ( Figure S7-S8). The dark magenta module in C. bellidifolia, the thistle 2 module in C.
groenlandica, and the sky blue 3 module in D. nivalis shared 282 coexpressed orthogroups. In general, the hub genes in these modules varied from species to species, but RPOT1 was a hub gene in all modules, while EMB2742, PPC1, and GPPS were hub genes in both dark magenta (C. bellidifolia) and skyblue 3 (D. nivalis). MAPK7 (a hub gene in light yellow and dark orange 2 above) was hub gene in thistle 2 in C. groenlandica. The three modules shared 20 significantly enriched GO terms (domains BP, CC, MF; Table S42), but most GO terms were module or species-specific (Tables S30-S38). The shared GO terms were related to metabolism and general cell processes, as well as different cell parts like "nucleus", "cytosol" and "vacuolar membrane".

| The cold response in Arctic Brassicaceae is highly species-specific
Our main finding was that the cold response in Arctic Brassicaceae was highly species-specific, but also that the shared portion of the cold response was higher than expected by chance. This suggests that the evolution of cold response in Arctic lineages of this family mainly followed independent genetic trajectories, but with some conserved components. The results broadly agree with our previous study of protein sequence evolution in the same three species, where we found very little overlap in positively selected genes and only a few genes with convergent substitutions (Birkeland et al., 2020). We did expect a higher degree of convergence in cold induced expression profiles, but our results agree well with those of a similar study of cold acclimation in the temperate grass subfamily Pooideae where phylogenetically diverse species of grasses showed widespread species-specific transcriptomic responses to low temperatures, but with some conserved aspects (Schubert et al., 2019)   these two cases, adaptation may end up taking the same routes in the presence of pleiotropic constraints or limited standing genetic variation (as discussed in Gould & Stinchcombe, 2017). Our finding of mainly independent evolution of cold tolerance suggests that such constraints have been of little importance in these Arctic lineages.

| Conserved aspects of the cold response in Brassicaceae
Another major finding in our study was that the cold response in the three Arctic species seems to have more in common with their temperate relative A. thaliana than with each other. This shared aspect of the cold response may represent conserved parts of the CBF pathway, which is present also in other plant families (Jaglo et al., 2001;Shi et al., 2018;Vergara et al., 2022). In our study, genes shared between A. thaliana and the Arctic species included those tied to circadian regulation and regulation of freezing tolerance (e.g., COR27 and COR28; Li et al., 2016). This fits well with the fact that the CBF pathway is gated by the circadian clock (Dong et al., 2011;Fowler et al., 2005). The continuous increase in DEGs with time at 2°C also indicates that Arctic plants respond to low temperatures in a way similar to temperate plants (see e.g., Calixto et al., 2018;Kilian et al., 2007). This suggests that Arctic plants are not completely hard-wired to tolerate low temperatures, but that they need a cold acclimation period to develop cold tolerance.
The cold response of our Arctic species included biological processes that have been demonstrated in the cold response of A. F I G U R E 3 UpSet plot of significantly enriched gene ontology (GO) terms in the 24 h upregulated gene sets (biological process domain only). The plot in the left corner shows total numbers of significantly enriched GO terms, and the main plot shows the number of unique significantly enriched GO terms, followed by the GO terms that were significantly enriched in more than one species (connected dots)  thaliana, other Brassicaceae species, and species from other plant families (e.g., Buti et al., 2018;Lee et al., 2013;Zhao et al., 2012), including stress responses associated with low temperatures, salt stress, and water deprivation. Stress response pathways associated with cold have been shown to be partially homologous with those of dehydration and salt tolerance (Bartels & Souer, 2004;Ingram & Bartels, 1996;Shamustakimova et al., 2017). In addition, ice formation in the apoplast draws water out of the cells and increases the concentration of salts and toxic solutes (Körner, 2003;Wisniewski & Fuller, 1999), leading to cell dehydration (Shi et al., 2018). We also found that low temperatures triggered responses to oxidative stress, which accompanies other abiotic stresses such as cold stress, and especially high light intensity in combination with low temperatures (Heino & Palva, 2004;Kilian et al., 2007;Lütz, 2010).
We found that genes associated with the hormones abscisic acid (ABA) and ethylene were upregulated in response to cold in all species. ABA is important in plant stress signalling and has been shown to increase in abundance during cold acclimation (Heino & Palva, 2004;Tuteja, 2007). Exogenous application of ABA even increases freezing tolerance in A. thaliana and other plants (Thomashow, 1999).
Ethylene is also reported to influence freezing tolerance and regulates the CBF pathway itself (Kazan, 2015). In most plant species, this entails a positive regulation of freezing tolerance (e.g., in tomato and tobacco; Zhang & Huang, 2010), although contradicting results have been found in A. thaliana (Kazan, 2015). The shared trends outlined by this study indicate that cold response is built upon a similar scaffold in Brassicaceae and support the claim that cold tolerance is a complex trait that is difficult to evolve (Donoghue, 2008). Our results suggest that the last common ancestor of the core Brassicaceae possessed some kind of cold tolerance machinery. The major clades of the Brassicaceae are thought to have radiated in response to the change to a colder and drier climate at the Eocene-Oligocene transition (~33.9 Ma; Huang et al., 2016), suggesting that evolution of a basic cold tolerance machinery contributed to its successful colonization of cold and dry regions.
Nevertheless, the highly species-specific cold response we found in our Arctic species also suggests that there is great evolutionary flexibility in cold coping strategies once there is a basis to build upon.

| The Arctic cold response
The ~100 genes we identified as exclusively shared among the three Arctic species included many transcription factors that potentially may have large effects. Among these, ddf2 may be an important candidate for adaptation to cold conditions as it is closely related to CBF1-3 (genes with key regulatory roles in the CBF path-      is upregulated in response to salinity stress (Medina et al., 2011).
Another interesting transcription factor in relation to Arctic cold regulation is RAP2.2, which is related to DEAR1 -a transcription factor known to mediate freezing stress responses in plants (Tsutsui et al., 2009). We also found several shared transcription factors related to circadian rhythm, such as REVEILLE 2, PCL1 and HY5. We cannot rule out that our experimental light regime may have provoked the expression of these genes, but the interplay between the circadian clock and the CBF pathway is known to be important for balancing freezing tolerance and plant growth (e.g., Dong et al., 2011;Shi et al., 2018). In addition, REVEILLE 1, 2 and 6 had high eigengene-based connectivity in cold correlated coexpression modules in all species. One could therefore speculate that the Arctic light regime has triggered the evolution of new links between the CBF pathway and the circadian clock. Our finding of exclusively shared transcription factors among the Arctic species fits well with theory predicting that hub genes may be important hotspots for convergent evolution (Martin & Orgogozo, 2013;Yeaman et al., 2016). Intriguingly, two of the shared Arctic transcription factors (ddf2 and ELF4) are even sister genes of already known evolutionary hotspots for convergent evolution in plants (Martin & Orgogozo, 2013).
The exclusively shared Arctic DEGs further contained traces of a mitogen activated protein kinase cascade (MAPK7, MAPKKK14, MAPKKK18). Such cascades are known to be important in regulating the CBF pathway (Shi et al., 2018;Teige et al., 2004). A potential role in cold regulation is supported by our finding of MAPK7 as a hub gene in several coexpression modules highly correlated with cold. In the temperate A. thaliana, it has been shown that the MKK2 pathway regulates cold stress signalling (Teige et al., 2004), and that the sister gene MAPK6 is involved in releasing inhibitory effects on CBF gene expression (Kim et al., 2017). The shared MAPK and MAPKKKs may therefore have important roles in Arctic cold responses.
Although we found that the Arctic DEG sets were functionally similar to that of the temperate A. thaliana, sperimidine related genes were only overrepresented in the Arctic upregulated DEG sets.
Spermidine is a polyamine that may be important for maintaining photosynthesis rates during low temperatures as cucumber plants pretreated with spermidine show less decline in photosynthesis rates during chilling than non-treated plants (He et al., 2002). Sustaining photosynthesis at low temperatures is especially important for Arctic plants as temperatures are low in summer, and they have optimum photosynthesis rates at lower temperatures than other plants (Chapin, 1983). The significant enrichment we found was caused by only a few cold induced genes related to spermidine. One of these, ADC1, which has been found to be involved in acquiring freezing tolerance in wild potato (Kou et al., 2018), was also responsible for the significant enrichment of arginine decarboxylase activity found in the Arctic species. This makes it an interesting candidate for possible adaptation to cold conditions in Arctic species.
We also observed that several of the genes we previously identified as being under positive selection or containing convergent substitutions in the Arctic species also are cold responsive (Birkeland et al., 2020). These included genes such as CSDP1 and COR15B in D. nivalis, two transmembrane proteins in C. bellidifolia, and LEA4-5 in C. groenlandica. The convergent EMB2742 and MAPKKK14 were upregulated in response to cold in all three species, which indicates that they might have an important function in cold tolerance. Since our previous analyses of positive selection and convergent molecular evolution were not based on cold induced transcriptomes, it is possible that additional positively selected and convergent genes may exist among our cold induced DEGs.

| CON CLUS I ON S , LIMITATI ON S , AND FUTURE PER S PEC TIVE S
We have presented the first comparative study of Arctic coldinduced transcriptomes, providing new insight into the evolution of cold response in the Brassicaceae. We found that the cold response in different species of Arctic Brassicaceae has more in common with their temperate relative A. thaliana than they have with each other.
Certain transcription factors, a potential MAPK cascade, and the ability to perform photosynthesis under low temperatures, may nevertheless represent important shared adaptations to the Arctic biome.
Here we only included published data from one temperate species, and it is therefore possible that some of the ~100 genes we found to be exclusively shared among the Arctic species can be found in Our comparative approach gave valuable insights into the basis of a conserved cold response possibly found throughout the Brassicaceae family. This shared cold response includes well known genes in the CBF pathway and may be coregulated by abscisic acid and ethylene. It also included upregulation of genes associated with sucrose, flavonoids, and karrikin. Our results are based on F I G U R E 5 Heatmaps showing Pearson correlation between each coexpression module (module eigengenes) and temperature: 0, 3, 6, and 24 h with cold treatment, and a binary measure of cold (cold treatment/no cold treatment). Each row corresponds to a coexpression module and each column to a temperature trait. The number of genes in each module is indicated in brackets after the module name. Cells show the correlation coefficient and corresponding p-value (in parentheses). From the top: (a) Cardamine bellidifolia, (b) Cochlearia groenlandica and (c) Draba nivalis comparisons between our new data for the Arctic species and previously published data for A. thaliana, and we note that there might be methodological differences that may affect the number of DEGs considered as significant and thus the number of shared genes. Our de novo assembled transcriptomes may contain some inaccuracies in the delimitation of genes (e.g., isoforms of the same gene mistakenly being delineated as different genes). However, we used a stringed filtering scheme to reduce inflated transcript numbers, and such differences should not affect the overall results.
Further studies are needed to verify the expression of cold regulated genes shared among the Arctic species, as well as among A.
thaliana and the Arctic species, and to obtain a better understanding of species-specific gene function. Future investigations should address the effects of shared transcription factors and MAPK genes on cold and freezing tolerance. This introductory study has provided novel insights into how Arctic plants may respond to a summer cold shock, but more investigations are needed to learn how they cope with long-term cold stress, as well as cold stress in combination with the light regime typical of Arctic environments.

CO N FLI C T O F I NTE R E S T
The authors declare that they have no conflict of interest.

B EN EFIT-S H A R I N G S TATEM ENT
The research described in the publication complies with relevant national laws implementing the Convention on Biological Diversity and Nagoya Protocol agreements. Benefits from this research accrue from the sharing of our data and results on public databases as described above.