Experimental design to investigate circadian gating of acute cold responses

We wished to discover whether there is circadian gating of responses to acute cold temperature cues in hexaploid bread wheat (Triticum aestivum). We used low temperature treatments as an experimental model because circadian gating of responses to short cold treatments has been reported for specific Arabidopsis genes [28,50,51]. It is not known whether there is circadian gating of low temperature responses across the entire genome in plants. For this study, we selected a hexaploid spring wheat cultivar (Cadenza) that has some low temperature tolerance, and good genome sequence coverage and germplasm resources. Using spring wheat also avoids introducing compounding effects of vernalization requirements into the data.

We considered circadian gating to be defined as a 24 h oscillation in the sensitivity of a transcript to a stimulus (Fig 1A), such that the magnitude of its response to an identical stimulus, given at a range of times of day, oscillates across the 24 h cycle (Fig 1B and 1C). Therefore, to test whether any given transcript undergoes circadian gating in its response to a stimulus, it is necessary to measure the abundance of the transcript under control conditions, and following an acute stimulus at each timepoint. To test for the occurrence of circadian gating, acute cold treatments were applied to separate batches of wheat seedlings under free running (constant) conditions (Fig 1D). This comprised identical three-hour cold treatments (4°C), at six sequential timepoints, under free running conditions (Fig 1D). This design allows the detection of transcripts that have a response to cold that is restricted to certain times of day, or a 24 h fluctuation in the magnitude of their response to the cold treatment (Fig 1B and 1C). After each treatment, tissue was harvested for RNA isolation from the second leaf of control temperature (22°C) and cold-treated seedlings, with each seedling sampled once only (Fig 1D). Transcript count values (isoforms) were condensed to the gene level, so transcript abundance changes described here refer to changes at the gene level.

The time of day affects the number of genes that respond to cold temperature conditions

The number of differentially expressed genes (DEGs) in response to the cold treatment was different at each at each timepoint (log fold change > 0.5, false discovery rate < 0.05; Fig 2A). This indicates that the time of subjective day of the cold treatment determines the cold-induced transcriptome in wheat (Fig 2A). The greatest number of genes were differentially expressed in response to a cold treatment at zeitgeber time (ZT) 48 (ZT48), with 4350 and 3403 genes significantly up or down-regulated, respectively (Fig 2B; ZT refers to the time relative to dawn, which is ZT0). In contrast, the number of genes that responded to cold at ZT44 was approximately 80% fewer than at ZT48 (Fig 2B), despite a similar total number of transcripts detected at these timepoints (S1 Dataset). There was a large difference in the direction of change of DEGs that responded to a cold treatment at ZT36 and ZT40, with the number of up-regulated genes comparable to those at ZT32 and ZT48, but a smaller number of down-regulated genes that was comparable to the number at ZT28 and ZT44 (Fig 2B). These data suggest the wheat transcriptome is more responsive to a cold treatment around the middle of the subjective day (ZT32), and towards the end of the subjective night (ZT48) (Fig 2B).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Different sets of transcripts respond to cold treatments given at different times of day in bread wheat.

(A) Differentially expressed genes after a 3 h cold treatment given at six timepoints (log fold change > 0.5, FDR < 0.05). Yellow and grey shading indicate subjective day and night, respectively. (B) The number of differentially expressed genes (DEGs) between control temperature conditions and following a 3 h cold treatment at each timepoint. (C) The proportion of DEGs that only responded to a cold treatment at a single timepoint. (D) Summary of the difference between mean normalised control and cold transcript abundance of DEGs across the time course. Orange and teal colours indicate normalized change in cold-induced and cold-repressed transcripts, respectively.

https://doi.org/10.1371/journal.pgen.1010947.g002

Fewer transcripts were downregulated than upregulated by an acute cold treatment given at any timepoint tested, whereas the proportion of cold-downregulated genes that were unique to each timepoint was consistently greater than the proportion upregulated genes that were unique to each timepoint (Fig 2C). This suggests that genes downregulated in response to cold are especially influenced by the time of day. Comparing the dynamics of a representative transcript (TaLHY) using RNA sequencing analysis (S1A–S1C Fig) and RT-qPCR (S1D Fig) provides confidence in our transcriptomic analysis.

The set of cold-responsive genes fluctuates across the 24 h cycle

To provide a foundation for investigating the circadian gating of cold temperature responses, we assessed whether the composition of the cold-responsive transcriptome changes over the 24 h cycle. To achieve this, we compared the identity of the genes that responded to a cold treatment given at each timepoint tested. Four of the largest six groups of DEGs were cold-induced at specific timepoints, so more genes responded to cold at specific (unique) times than were cold responsive at all timepoints (Figs 2B, 2C and S2A). Furthermore, the majority of cold-repressed transcripts were only repressed by cold at a single timepoint (Figs 2C and S2B). Therefore, under free running conditions, the set of genes that responds to an acute cold treatment depends upon the time of the cold treatment (Fig 2D). Some transcripts were only responsive to a cold treatment given at one timepoint, some across part of the 24 h cycle, and some were cold-responsive regardless of the time of the treatment (Figs 2B–2D; and S2A, S2B).

Circadian gating of transcriptional responses to cold is extensive, and not restricted to specific times of day

Our data identified that different subsets of transcripts respond to cold treatments given at different times of day (Fig 2), so we hypothesized that there might be widespread circadian gating of the cold-responsive transcriptome. To test this formally, we obtained a measure of the cold-sensitivity of each transcript, in response to a cold treatment given at each timepoint, by calculating the difference between the mean CPM of the transcript at control temperature and after each cold temperature treatment (Fig 3A). This provided a measure of the magnitude of the response to cold, at each timepoint, that we termed ∆CPM (Fig 3A). Inclusion of the underlying (control temperature) transcript abundance in the calculation removes the effect of any underlying circadian rhythm in transcript abundance from the circadian fluctuations in its cold sensitivity. Using this approach, a 24 h fluctuation in the sensitivity of any given transcript to the time-series of cold treatments will manifest as an oscillation in ∆CPM (Fig 3A, lower panel). We tested whether there was a circadian rhythm in this cold-sensitivity profile for every transcript (meta2d analysis for rhythmicity; threshold of P < 0.05). Circadian rhythms of cold sensitivity (∆CPM) were detected for 4388 transcripts, which we filtered to include only transcripts that were differentially expressed in response to cold at least once across the time course, and exclude transcripts that could not be confidently assigned to a particular time of maximum cold-responsiveness. This yielded 1677 transcripts that we classify as having circadian gating of their responses to cold in wheat (Fig 3B).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. The temporal structure of the cold-responsive transcriptome.

(A) We defined circadian gating of the response of genes to cold as a 24 h cycle of sensitivity of the response of the gene to identical acute (3 h) cold treatments. To identify such genes, we calculated the difference between the mean control and cold-treated CPM (termed ∆CPM). Circadian gating of a cold response manifests as a 24 h oscillation of ∆CPM, which captures cold induction, repression, or a displacement in transcript cycling. (B) The number of cold-responsive differentially-expressed gene transcripts (DEGs; LogFC >0.5, FDR <0.05) that have a rhythmic ∆CPM profile across the time course (meta2d, p <0.05). This represents a set of wheat transcripts with circadian gating of their response to a 3 h cold treatment. (C) Transcripts with circadian gating of their response to a 3 h cold treatment, organized according to their estimated time of maximum cold responsiveness (ETMR). Direction of response (induction, reduction, displacement) was based on the average difference between mean normalised control and cold-treated CPM to normalize differences in ∆CPM profile amplitude across the transcriptome. The direction was formalized as circadian-gated inductions (∆mean.normCPM > 0.5), cold induced displacement (-0.5 < ∆mean.normCPM < 0.5) and circadian gated reductions (∆mean.normCPM > -0.5). ETMR was derived from meta2d phase estimates. Yellow/grey shading = subjective day/night. Pink lines indicate mean ∆mean.normCPM in each directional ETMR group. (D) The number of genes within each directional cold ETMR group. 1533 genes were included, after filtering to remove ∆CPM profiles from ETMR groups < 4 h and > 20 h having peak CPM at an opposing timepoint from the ETMR. Darker colours indicate greater numbers of transcripts.

https://doi.org/10.1371/journal.pgen.1010947.g003

We found that this set of transcripts divided into three groups. There were circadian gated cold-inductions (where mean[∆normalisedCPM] > 0.5), circadian gated cold-reductions (mean[∆normalisedCPM] < -0.5), and a group where the transcript abundance was relatively stable for the duration of the cold treatment (mean[∆normalisedCPM] between -0.5 and 0.5), leading to a temporal displacement relative to the control (Fig 3A). Within these cold response types, we calculated the estimated time of maximum cold responsiveness (ETMR) of each transcript using phase estimates from meta2d (Fig 3C). This approach allowed us to estimate the time of maximum cold responsiveness that fell between sampling timepoints, thereby overcoming some limitations of the sampling frequency. For circadian gated reductions (Fig 3A), the time of maximum cold-responsiveness was the trough of the oscillation.

This analysis identified that over 500 transcripts had circadian gating of their induction by cold (Fig 3C and 3D). Within these transcripts, the most common estimated time of maximum cold response was between 16 h and 20 h (Fig 3D), with a large proportion of the transcripts having a time of greatest cold response between 4 h and 8 h. The largest proportion of cold-repressed transcripts had greatest cold-sensitivity when the cold treatment was given during the subjective night (Fig 3D). The greatest number of transcripts with rhythmic sensitivity to acute cold treatments were those whose abundance was stabilized for the duration of the cold treatment (cold-induced displacement; Fig 3C and 3D). Together, this indicates that there is extensive circadian gating of the cold-responsive transcriptome in bread wheat, with different transcripts having greatest responsiveness to cold treatments given at different times of day. Our analysis might under-estimate the number of circadian gated cold-responsive genes, because a longer and higher resolution time course would provide greater statistical power.

Examples of regulatory genes that undergo circadian gating of their response to cold

To assess how the time of day affects cold sensitivity of key regulatory components, four genes were chosen as examples. These genes in wheat were selected because they are orthologs of Arabidopsis genes that are known to have strong responses to cold, undergo circadian gating of environmental responses, or participate in metabolic regulation that might be associated with circadian rhythms. The hexaploid nature of the bread wheat genome means that many transcripts are present as triads, with gene homoeologs on the A, B and D subgenomes. Therefore, all three homoeologs were examined, to obtain a representative view of how a signalling component responds to a stimulus.

As a first example, we examined the dynamics of a wheat gene from the C-repeat binding factor (CBF) family. CBFs are well-described cold responsive transcription factors that participate in cold acclimation [52,53], with up to 25 CBFs present in wheat [54]. Certain CBF transcripts have a circadian rhythm in barley and wheat [43,55], and in Arabidopsis the cold-induction of CBFs is gated by the circadian clock [28]. Our data indicate that circadian gating of CBF cold-responsiveness is conserved in wheat and can extend across homoeolog triads, such as for TaCBFVc-B14, which is a cold-responsive CBF ortholog with high homology to AtCBF2 (Fig 4A–4C) [24,54]. In our experiments, all three homoeologs of TaCBFVc-B14 were upregulated most strongly by a cold treatment given at the end of the subjective day (Fig 4A–4C).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Circadian gating of the cold response of selected wheat triads.

(A-C) TaCBFVc-B14; triad no. 15598. (D-F) TaBBX17; triad no 9634. (G-I) TaBAM1; triad no 15540. (J-L) TaBAM3; triad no 15539 [49]. Yellow/grey shading = subjective day/night. Solid lines are mean (N = 3 biological replicates; blue/green shading = ± s.e.m). Asterisks indicate times when the transcript was differentially expressed between the control and cold temperature conditions (LogFC >0.5, FDR < 0.05). Each cold treatment lasted 3 h, and ended at the time shown on the x axis. Data points show transcript abundance before (green) and after (blue) the cold treatment. CPM (counts per million) determined by edgeR [82].

https://doi.org/10.1371/journal.pgen.1010947.g004

As a second example, we chose a gene from the B-BOX CONTAINING PROTEIN (BBX) family. In Arabidopsis, members of the BBX gene family participate in responses to environmental cues [56], such as the circadian modulation of AtBBX17 transcript responses to heat [34,41]. In wheat, transcripts from the TaBBX17 7ABD triad, which has closest homology to AtBBX17, undergo rhythmic accumulation under control temperature conditions (Fig 4D–4F). The response to cold varied across the triad, with TaBBX17 7A upregulated by a cold treatment given at the start of the subjective day, but not a cold treatment given during the subjective night (Fig 4D). This contrasts 7B and 7D, which were downregulated by a cold treatment given at the end of the subjective day (Fig 4D–4F). This example suggests that features of circadian gating are not always conserved across triads.

As a final example, we chose two wheat genes that encode proteins associated with the mobilization of transitory starch. In Arabidopsis, the circadian clock regulates starch metabolism, with this thought to explain the reduction in growth that occurs when the circadian clock is misaligned with the environmental cycle [57–59]. A range of related Arabidopsis transcripts encoding starch degradation enzymes are circadian regulated, but this is not always conserved in wheat, where certain orthologs of β-amylase (BAM, e.g. BAM3) have a circadian rhythm, whereas BAM1 does not [43]. Under our conditions, the TaBAM1 triad and TaBAM3 triad were upregulated by cold treatments given during the subjective day (Fig 4G–4L). TaBAM1 was not responsive to cold treatments given during the subjective night, whereas TaBAM3 was (Fig 4G–4L). The cold-induced relative abundance of TaBAM3 1B and TaBAM3 1D was approximately ten-fold greater than the corresponding homoeologs of TaBAM1 (Fig 4H, 4I, 4K and 4L), supporting the notion of a role for BAM3 in cold conditions.

Extent of the subgenome bias of the circadian gating of cold responses

The hexaploid bread wheat genome arose from hybridisation of three diploid ancestors. Therefore, each circadian oscillator component is likely to be encoded by three copies of each clock gene, with a copy residing on each of the A, B, and D subgenomes. No specific subgenome is favoured across the circadian regulated transcriptome of wheat [43], whereas circadian regulated paralogs have differential expression patterns in other crops [39]. Genes that occur as a single copy on each of the A, B and D subgenomes are known as triads [49]. Environmentally-responsive gene triads do not always have expression that is balanced across the subgenomes [49]. Therefore, we wished to determine whether there was subgenome bias in the circadian gating of transcriptomic responses to cold.

For transcripts classified as having a circadian gated cold-induction or a displaced pattern of cycling, the relative proportion of these transcripts originating from A, B and D subgenomes was determined for each ETMR group (Fig 5A). The proportion derived from each subgenome remained relatively constant in ETMR groups representing the subjective day (30–40%; Fig 5A). During the subjective night, the largest proportion of circadian gated cold induced transcripts derived from the D subgenome (ETMR > 12 h) (Fig 5A). This was restricted to cold-induced transcripts, and did not occur for transcripts with displaced cycling. The number of genes with circadian-gated repression by cold was rather small, so the proportional subgenome contribution to cold-repressed transcripts is noisy (S3 Fig).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Subgenome organization of the circadian gating of the cold-responsive transcriptome.

(A) The number and proportion of transcripts from each subgenome that have a circadian gated response to an acute (3 h) cold treatment, divided according to the time of maximum cold responsiveness. (B) The proportion of transcripts that have a circadian gated response to an acute cold treatment and belong to an annotated triad [49], divided according to the time of maximum cold responsiveness. Not all homoeologs within any given triad are gated or present within the same maximum responsiveness group, and this is indicated by the bar colour. (C-E) Worked examples of ternary plots showing the relative contribution of each homoeolog within a triad to the combined transcript accumulation from the triad [49,99]. The position of each point on the plot reflects the proportion of total contribution from the three homoeologs within the triad (subgenome A, yellow axis; subgenome B; green axis, subgenome D, blue axis). (C) A point in the centre of the plot represents a triad where all three homoeologs are equally cold-responsive. (D, E) Points towards corners (shaded areas) represent triads where the majority of the cold response derives from a specific homoeolog. (F) The relative contributions of homoeologs to the cold responsiveness of each triad, organized according to the time of maximum cold responsiveness. Each triad contains at least one homoeolog that is within the ETMR group. Therefore, certain triads are present in only one ETMR (orange points), some in two (brown) and some in three (cyan).

https://doi.org/10.1371/journal.pgen.1010947.g005

Approximately 10% of induced circadian gated transcripts with ETMR < 4 h, 4–8 h, 12–16 h and 16–20 h were detected as complete triads of three homoeologs, but no complete triads were present in the ETMR groups 12–16 h and > 20 h (Fig 5B). Only single transcripts from triads had circadian gated responses to cold in the 8–12 h ETMR group, suggesting differential behaviour within triads (Fig 5B). Furthermore, it appears that the time of maximum cold responsiveness of some triads is spread across multiple phases, with different homoeologs responding most strongly to cold treatments given at different times because the proportion of transcripts detected as complete triads across all the ETMR groups was greater than within any individual ETMR group (Fig 5B).

The contribution of each homoeolog to the overall cold induction of the corresponding triad was assessed [49]. To achieve this, we plotted the proportional contribution of each homoeolog to the cold-responsiveness of the whole triad using ternary plots. In these plots, the relative position of each data point indicates the contribution by each subgenome to the overall expression of the triad. For example, if the point is near the centre of the ternary plot, there is little or no expression bias towards a particular subgenome (Fig 5C). Alternatively, if the point is nearer to a corner of the plot, one subgenome makes a greater contribution to the expression of the triad. For example, Fig 5D and Fig 5E show a strong bias towards the A and B subgenomes, respectively. In general, no subgenome obviously contributed more to the set of circadian-gated cold responsive transcripts, although there was strong subgenome bias present for certain individual transcripts (Fig 5F). Many triads had dynamic responses to cold, in that the contribution from each subgenome to total transcript level was different under control temperature conditions compared with after the cold treatment (S4 Fig). The distribution in the contribution of each subgenome to triad cold responsiveness, in each timing group, suggests that there is no overall bias towards a particular subgenome (Fig 5F).

Genes associated with responses to light are circadian gated to the subjective day

Gene ontology-term analysis identified multiple biological processes associated with responses to light as over-represented in transcripts that had greatest induction by cold treatments during the subjective day (ETMR 0–12 h). Processes such as “cellular responses to blue light”, “cellular responses to UV-A” and “anthocyanin containing compound biosynthesis” were enriched in multiple subjective day ETMR groups (S5 Fig; adjusted p < 0.001). Enrichment of terms such as “photoprotection” and “stress-activated MAPK cascade” suggests the cold stimulus elicited a broad response when delivered in the subjective day, at a time when light and low temperatures may occur simultaneously.

There was a clear split between the GO-terms that were enriched in subjective day ETMR groups and subjective night ETMR groups. “Cold acclimation” is shared across two subjective night groups (S5 Fig; adjusted p < 0.001), suggesting that night-time low temperatures could have a key role in adapting wheat to cold seasonal weather. Also enriched are terms associated with DNA transcription, which can be an indication of preparation for a large shift in gene expression (S5 Fig; adjusted p < 0.001). “Photosystem II” associated processes were enriched in subjective night ETMR group 16–20 h, which might relate to unexpected exposure to continuous light conditions (S5 Fig; adjusted p < 0.001). Whilst GO-term analysis interpretation requires caution, it is possible that restriction of the cold response of certain processes to specific times of day provides an advantage to wheat by ensuring that energy demanding cold-responsive processes are not induced at inappropriate times of day [60].

Acute cold treatments displace the oscillation of certain circadian clock components

We noticed that a variety of transcripts appeared to be maintained at a relatively uniform level for the duration of the cold treatment. Under these circumstances, the transcript abundance after the acute cold treatment was similar to its abundance at the previous timepoint (see example in Fig 6A). Although this appears to be a delay in the cycling of the transcript, it is not possible to determine whether this represents a phase shift because the cycling of transcripts was not monitored following each cold treatment. Instead, this process appears to be a stabilization of the transcript abundance for the duration of the cold treatment.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. Cold temperature displaces the oscillation of a set of rhythmic transcripts.

(A) Dynamics of a transcript with abundance stabilized during the cold treatment (TraesCS4A02G091700), whereby the transcript abundance after an acute cold treatment (blue) is similar to the abundance before cold treatment (green). The difference between cold and control treated transcript abundance (∆CPM) is shown in dark red. (B) Dynamics of a rhythmic transcript (TraesCS7A02G416300; meta3d: ARS p <0.05) that has an abundance that is stabilized during the cold treatment, but has a more linear ∆CPM profile due to its underlying phase profile. (C) Artificial data exemplifying these dynamics. ∆CPM is the difference between the control and cold-treated CPM at each timepoint (dark red arrows), and adjusted ∆CPM is the difference between the control temperature CPM at one timepoint and the cold-treated CPM at the next timepoint (blue arrows). (D) A displacement ratio was calculated from the average adjusted ∆CPM as a proportion of the original average ∆CPM for the transcript, and (E) all transcripts with rhythmic control and cold-treated CPM profiles (meta3d: ARS p <0.05). For more details on the calculation of the displacement ratio, see S6 Fig. In (D, E) the larger circle represents 100% of these transcripts, the middle circle represents transcripts with a displacement ratio <1, and the smaller circle transcripts with a displacement ratio < 0.5. Displacement ratios < 1 suggest transcript abundance was stabilized for the cold treatment duration. (F) Displacement ratios of eleven circadian clock gene triads [43] suggest their abundance was stabilized for cold treatment duration for the majority of these transcripts. (G) Transcript dynamics of TaGI 3A (TraesCS3A02G116300), which had stabilized abundance during the cold treatment, and TaLUX/BOA 3A (TraesCS3A02G526600) where this did not occur. Solid lines are mean (N = 3). Blue/green shading = ± s.e.m. Red line represents ∆CPM. Yellow/grey shading = subjective day/night.

https://doi.org/10.1371/journal.pgen.1010947.g006

We wished to systematically identify the set of transcripts that had this response. We developed an analytical method that used the ∆CPM to consider the relationship between the transcript levels at adjacent timepoints (Figs 6C; and S6; see methods for calculations). This involved calculating an average adjusted ∆CPM for each transcript, by finding the difference between control CPM values and the subsequent cold CPM value in the time course. The average adjusted ∆CPM value was expressed as a proportion of the original average ∆CPM of the time course to give a “displacement ratio” value (Figs 6C; and S6; see methods for calculations). Over three-quarters of the 882 transcripts with rhythmic ∆CPM have a displacement ratio of < 1, supporting the interpretation that the abundance of these transcripts was stabilized during the acute cold treatment (Figs 6A and 6D). A similar proportion of the 3849 transcripts with rhythmic control and cold CPM profiles had a displacement ratio of < 1, suggesting that the abundance of these transcripts was also stabilized during the cold treatment also (Fig 6B and 6E). A possible explanation is that the cold treatment maintained a relatively constant level of these transcripts for the duration of the treatment. 47% of transcripts with circadian gated cold-induction responses had displacement ratios of < 1, suggesting around half of the transcripts with a gated response to cold did not undergo this displacement of cycling during the cold treatment.

The GO-term “circadian rhythm” was enriched within the genes that had a displacement ratio < 0.5 in response to cold, in combination with either rhythmic CPM profiles or rhythmic ∆CPM (adjusted p < 0.001; S7 Fig). This suggests that for the duration of the cold treatment, the abundance of certain circadian oscillator components was stabilized, displacing the oscillation of the transcript (S6 Fig). All homoeologs of six circadian clock genes [43], including morning phased TaLHY and evening phased TaGI, had a displacement ratio < 1 (Fig 6F; see S8 Fig for dynamics of individual clock transcripts). Therefore, the cold treatment displaced the oscillation in the expression of these genes (Fig 6F). Homoeologs of TaELF3, TaELF4-L and TaTOC1 had a mixture of displacement ratios, indicating that the cold treatment did not equally affect the cycling of homoeologs of all circadian clock transcripts (Fig 6F).

Cold treatments did not have a consistent impact upon all circadian oscillator components. For example, cold had little effect upon expression of the TaLUX-Lb and TaLUX/BOA triads, which had lower or similar mean ∆CPM before adjustment for all three homoeologs (Fig 6F; S8 Fig). Therefore, cold does not appear to affect all circadian clock transcripts equally, as exemplified by TaGI having a stabilized and displaced abundance after each cold treatment, whereas TaLUX/BOA did not (Fig 6G). This supports the idea that cold has an unbalanced effect upon circadian clock components, reflecting a complex relationship between the circadian clock and cold.

A dual effect of cold on both the clock and clock-controlled genes

Here, we have identified extensive genome-wide circadian gating of a response to cold temperature in bread wheat. This demonstrates that the circadian gating of responses to cold of single genes [28] extends across the entire genome, and supports reports of the circadian gating of specific genes to high temperature stress in Arabidopsis [32,40]. Given the considerable phylogenetic distance between wheat and Arabidopsis, we reason that the circadian gating of low temperature responses might be conserved across the angiosperms.

Our results demonstrate the importance of circadian regulation in shaping transcriptomic responses to short-term cold. The diversity of temporal responses to cold suggests considerable complexity in the integration of low temperature and circadian signals (Figs 2D and 3). The presence of a circadian rhythm for a transcript under control temperature conditions does not mean that its cold-responsiveness is gated by the circadian clock, because some circadian regulated transcripts are not cold-responsive (Figs 7A and S9A). Likewise, some transcripts are cold-responsive, but this is independent from circadian regulation because these transcripts are cold-induced by a similar magnitude, irrespective of the time of day of the cold treatment (Figs 7B and S9B). In one form of circadian gating, the circadian clock restricts the response to the stimulus to certain times of day, and the transcript is completely unresponsive to cold treatments given at other times (Figs 7C and S9C). This type of circadian gating is bidirectional, because suppression of transcript levels by an acute cold treatment can also be circadian-gated (Fig 3C). In another form of circadian gating, a circadian modulation of sensitivity occurs across the day, such that a response to cold always occurs, but the response to a cold treatment given at certain times is greater compared with other times (e.g. TaBAM3; Fig 4J–4L). The phase of maximum cold responsiveness of some transcripts is sometimes aligned with the phase of the underlying circadian rhythm under control temperature conditions, but in other cases (e.g. TaBAM3) the phase was misaligned (Figs 4J–4L; 7D and S9D). The latter suggests that for some genes, separate circadian signalling components might regulate underlying rhythms and cold responsiveness.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 7. Features of the circadian gating of cold temperature responses.

The interaction between the circadian clock and an acute cold treatment produces a diversity of transcript response profiles across the wheat transcriptome. Diagrams conceptualize how the circadian clock (dark red colour clocks) and cold (blue arrows) can affect underlying control-temperature transcript levels (dotted black line), producing a diversity of cold-responsive transcript profiles (orange line). (A) certain transcripts can be regulated by the circadian clock but transcript levels are unaltered by cold (e.g. TraesCS5D02G447500 and TraesCS1D02G372000). (B) transcripts can respond equally to cold, irrespective of the time that the cold treatment is applied (e.g. TraesCS2D02G284400 and TraesCS3B02G287400). (C) the cold-responsiveness of transcripts can be gated by the circadian clock, resulting in a response to an acute cold treatment at specific times of day, but not other times (e.g. TraesCS3D02G325300). (D) the circadian clock gates the response to cold of the transcript, with the time of maximum responsiveness to cold having a different phase from the phase of the underlying circadian rhythm of the transcript (e.g. TaBAM3). (E) cold affects the circadian oscillator directly (cold blue clock), stabilizing oscillator transcript levels for the duration of the acute cold treatment (e.g. TaGI 3A, TraesCS5B02G271300). (F) the cold-responsiveness of a transcript is restricted to a particular phase, and cold also displaces the oscillation of the circadian clock transcript, apparently by stabilizing its abundance for the duration of the cold treatment. In combination, this upregulates and delays the accumulation of the transcript (e.g. TraesCS1A02G289800 and TraesCS1B02G095000). Experimental transcript data are shown in S9 Fig alongside these concepts.

https://doi.org/10.1371/journal.pgen.1010947.g007

Short cold treatments affect the dynamics of certain wheat circadian clock components, by stabilizing their abundance for approximately the duration of the cold treatment (Fig 7E, 7F; S9E, S9F Fig). This pervades hundreds of other circadian-regulated cold-responsive transcripts (Fig 6D and 6E). It is possible that this would cause a cold-induced delay in the circadian oscillator under free running conditions, leading to comparable delays in the response of downstream circadian-regulated transcripts. Interestingly, this occurs for some circadian clock genes (e.g. TaPRR73 and TaGI) but not others (e.g. LUX orthologues) (Figs 6F and S8). One potential explanation for this bifurcating response is that the stability of certain transcripts is insensitive to cold, so the transcript level continues to change under low temperature conditions. Another explanation could be that different cell types in the leaves of wheat have differential temperature responses which, combined with a predominance of different circadian clock components in different cell types [61], causes bifurcated dynamics of distinct circadian clock components in response to cold. Uneven temperature responses have been reported across the components of the Arabidopsis circadian oscillator [62], so a similar situation might occur in wheat. Finally, an open and speculative question- for which there is not currently evidence- is of whether polyploid genomes, such as the genome of hexaploid wheat, can ever support partial decoupling of subgenome-derived circadian oscillators in single cells.

It will in future be informative to know whether this cold-induced stabilization of oscillator transcript levels can act as a zeitgeber for the wheat circadian clock, in a manner reminiscent of the circadian phase shifts caused in Arabidopsis by moderate temperature reductions [63]. This is important because it could contribute to the phase relationship between circadian-regulated processes and environmental phase under field conditions, with consequences for metabolism or flowering time. It will also be informative to understand conserved and differing aspects of these responses in winter wheat, given its different vernalization and photoperiod requirements compared with spring wheat varieties such as Cadenza, used here. Taken together, these results suggest that fluctuations in environmental temperature might cause a dynamic adjustment of circadian phase in cultivated wheat [64].

Potential cold induction bias towards D subgenome during subjective night

Genes that had greatest responsiveness to a cold treatment during the subjective day had a balanced contribution of transcripts from each subgenome. In contrast, for transcripts with largest induction by a cold treatment given during the subjective night, the greatest proportion of transcript derived from the D subgenome (Fig 5A). However, when comparing the relative contributions from homoeologs that are present as triads, there did not seem to be a bias towards the D subgenome (Fig 5E). Therefore, this greater proportion of transcripts from the D subgenome could arise from genes that are present as diads or singletons. This might relate to the origins of the D subgenome donor, Aegilops tauschii. The D subgenome hybridizations are thought to involve A. tauschii derived from mountainous regions near the Caspian Sea [65,66], so perhaps the greater cold induction from the D subgenome during the subjective night relates to adaptation of A. tauschii to relatively cold conditions.

Relationship between the circadian gating of the cold-responsive transcriptome and circadian clock outputs

Rhythms in transcript abundance do not always lead to rhythms in protein abundance or activity [67–70], although there is circadian gating of the high temperature-responsive translatome [41]. In many cases, the abundance of proteins encoded by rhythmic transcripts remains constant [70–72], whilst in some cases (e.g. BAM3 in Arabidopsis) the protein abundance tracks the transcript abundance [72]. Circadian regulation of protein turnover occurs in eukaryotes [73,74], and it is thought that circadian regulation of gene expression might maintain proteostasis in the presence of rhythmic protein degradation [75]. This might explain why rhythms in protein abundance are less prevalent than rhythms of transcript abundance.

This idea could be relevant to interpreting the physiological significance of circadian gating in wheat. For approximately half of the genes that have a circadian gated response to cold, the phase of maximum cold sensitivity was similar to the phase of the rhythm at control temperatures (Fig 7C). This alignment of the phase of maximum cold sensitivity with the phase of the underlying rhythm likely increases the availability of mRNA at certain times, perhaps in anticipation of a greater rate of protein turnover [76]. This idea is exemplified by the greatest sensitivity to cold treatments given during the subjective day of genes that have light signalling GO terms (S4 Fig). It is possible that this temporal regulation increases transcript levels for these components, in anticipation of cold- and light-induced oxidative damage in the daytime. This interpretation based on GO-term analysis requires caution, because GO-term annotation is biased by the extent of previous investigation of gene function, and because changes in the abundance of transcripts associated with a process does not necessarily cause changes in protein activity or physiology. Conversely, transcripts that have circadian gating of their downregulation by cold might encode proteins that are stabilized in cold- so do not need to be replaced- or contribute to processes that are not preferred in cold conditions. Alternatively, down regulated genes may represent repressive regulators, like many components of the circadian clock [77], so a reduction in their abundance could lift repression of cold response mechanisms.

Conclusions

We identified that the circadian clock plays a key role in shaping the cold responsive transcriptome through the process of circadian gating. This establishes that circadian regulation is central to the temperature responses of a major cereal crop. These gating responses could be modified by photoperiod changes and by vernalization. This represents an important area for future investigation, because it could provide information about seasonal changes in the process of circadian gating. Our study supports the notion that circadian or temporal gating of responses to environmental cues represents a widespread regulatory mechanism in plants, from peach trees and grape vines [78,79] to Arabidopsis [34], and here at a genome-wide scale in hexaploid bread wheat. Our findings open opportunities to understand the contributions of circadian gating to wheat crop performance, and could underpin biotechnological developments that optimize the resilience of cereals to an increasingly unpredictable climate. We also developed a new set of general concepts concerning circadian gating (Fig 7), which are relevant to the circadian regulation of environmental responses across all organisms.

Plant materials and growth conditions

Seeds of Triticum aestivum cv. Cadenza (donated by James Simmonds, John Innes Centre) were stratified on damp filter paper for 3 days at 4°C before germination at 22°C in darkness for two days. Seedlings were cultivated in 24-cell trays for 12 days on a bespoke cereal cultivation mixture (65% peat, 25% loam, 10% horticultural grit, 3 kg m⁻³ dolomitic limestone, 1.3 kg m⁻³ PG Mix fertiliser (Yara), 3 kg m⁻³ osmocote extract) under cycles of 12 h light and 12 h darkness, at approximately 200 μmol m⁻² s⁻¹ of white light at 22°C (S10A Fig).

Circadian time course sampling

14-day old wheat seedlings at Zadok stage GS1.2 were transferred to constant light for 24 h before experimentation commenced, to ensure data were free from transitory effects caused by the final dawn (S10B Fig). After 25 h of constant conditions (zeitgeber time (ZT) 25, where ZT 0 refers to the start of the day), three individuals were transferred to 4°C with identical light conditions for 3 h, whilst control plants remained at 22°C. After the 3 h cold treatment, 3 cm of tissue from the second leaf was collected from each control and treated plant (indicated by red boxes on S10B Fig), immediately frozen in liquid N₂, and stored at -80°C until processing. After sampling, the cold-treated and control plants were not used for subsequent samples. Six such further treatments were conducted, beginning at ZT 25, ZT 29, ZT 33, ZT 37, ZT 41 and ZT 45, and ending at ZT 28, ZT 32, ZT 36, ZT 40, ZT 44 and ZT 48, respectively (Fig 1D). Each set of samples used a different batch of plants.

RNA sequencing data collection, QC and read mapping

RNA was extracted using Macherey-Nagel NucleoSpin RNA Mini Kit with on-column DNase treatment, according to manufacturer’s instructions. RNA was stored at -80°C in nuclease-free water. RNA integrity was assessed using an Agilent 2100 Bioanalyzer System at the John Innes Centre, and again by our sequencing partner Novogene (Cambridge, UK). All samples had RIN scores > 6 and concentrations > 50 ng RNA μl^-1, thereby passing quality control. Libraries were constructed by Novogene using a mRNA polyA enrichment library protocol involving fragmentation, cDNA synthesis, end repair, A-tailing, adapter ligation, size selection, amplification and purification. Libraries were verified on a Bioanalyzer before pooling for sequencing. After verification, 150bp paired-end sequencing was performed on the Illumina NovaSeq 6000 platform to generate a minimum of 15 GB data per sample, generating 675.5 GB total data with an average read depth of 62.5 million reads per sample. All samples had ≥ 98.5% clean reads suitable for alignment.

RNA sequencing analysis

Transcript quantification was conducted using the RNA-seq pseudoaligner kallisto v0.44.0 (Bray et al., 2016) using 31 bp k-mers and default parameters. The index file was constructed from the high confidence CDS sequence from the International Wheat Genome Sequencing Consortium (IWGSC) T. aestivum cv. Chinese Spring RefSeqv1.1 [48].

Differential gene expression analysis

Kallisto transcript count data were condensed to the gene level using R v2022.07.1 [80]. The Bioconductor [81] packages edgeR [82–84] and Limma [85] were used to identify differentially expressed genes between cold-treated and control seedlings at each timepoint. Counts per million (CPM) were determined by edgeR and low-expressed transcripts filtered out of the data (edgeR function filterByExpr, which uses the filtering strategy described in Chen et al. 2016) (S1 Dataset, S2 Dataset). This filtering strategy retains genes with a CPM value above k in n samples, where k is determined from the set minimum count value and sample size, and n determined from the smallest group size within the design matrix [84]. Around 68.5% of transcripts remained after filtering (S3 Dataset). Empirical Bayes moderation was performed for predictions of gene-wise variability [86] and a correction for multiple testing (Benjamini-Hochberg method) was used to obtain the false discovery rate (FDR) of differentially expressed genes. Differentially expressed genes were defined as those with an adjusted p-value < 0.05 and logFC > 0.5 (50% change).

Data processing and visualisation

CPM and logFC values were assembled using base R and the tidyverse package dplyr [87, 88]. Graphs were produced using ggplot2 [89] and radar charts with fmsb function radarchart [90]. Upset plots (S2 Fig) were created with UpSetR [91] and GO-terms presented using code adapted from [43,92]. The sensitivity heatmap was produced using mean normalised CPM values determined by dividing the average CPM at any given timepoint by the mean CPM for each gene across the entire time course under both temperature conditions. The heatmap was created with ComplexHeatmap, using the Ward D.2 method for clustering, with k = 12 [93,94].

Identification of genes having a circadian gated response to cold

We used the difference between the mean control transcript level and mean cold-treated transcript level (∆CPM) at each timepoint as a measure of the sensitivity of gene expression to cold. To test for rhythmicity of cold sensitivity, the ∆CPM profiles were analysed across the entire time course duration using the MetaCycle v1.2.0 function meta2d [95] (Figs 3A; and S6). The minimum and maximum permitted period lengths were 20 h and 28 h, respectively [96]. We considered meta2d to be especially suitable for this analysis because it integrates period, phase and p-values from three algorithms, reducing the limitations of individual algorithms, and is appropriate for analysis of a relatively small number of time-points [96]. A rhythmicity cut-off of meta2d p-value < 0.05 was used. In general, time-series close to this p-value threshold had approximately 24 h rhythmicity (S11 Fig). As an additional filter, only genes with rhythmic ∆CPM that were also differentially expressed at least once in the time-series were considered to be circadian gated. Only transcripts with average CPM > 1 under control temperature conditions or cold temperature conditions, across the time-series, were included in the analysis to eliminate low-expressed genes with large relative variation between replicates due to the low counts. The difference between mean normalised CPM values (∆mean-normCPM) was used to assign transcripts to three directional response groups: circadian-gated inductions by cold (∆mean-normCPM > 0.5), cold induced displacement of transcript cycling (-0.5 < ∆mean-normCPM < 0.5) and circadian gated reductions by cold (∆mean-normCPM > -0.5).

The estimated time of maximum cold responsiveness (ETMR) refers to the time of the greatest ∆CPM. This was estimated from the phase measured using meta2d analysis, which provides the peak phase of the time-series data. For the ETMR for circadian gated cold reductions, the ETMR was determined using meta2d phase estimates for the inverse ∆CPM values, because they represent the phase of the trough. This allowed a total of 18 directional ETMR groups to be defined (Fig 3C). Extra filters were applied to ETMR groups < 4 h and > 20 h for inductions and displaced transcripts, to remove 135 trend anomalies that had greater ∆CPM at the opposite timepoint (+ 12 h) to the meta2d phase estimate.

Gene ontology-term analysis

GO-term enrichment analysis for biological processes was performed using TopGo [97] for each cold-induction ETMR group. The group sizes were sufficiently large for this approach (n>20). GO-terms were retrieved from the IWGSC RefSeqv1.0 annotation and transferred to annotation v1.1 using the method described by [98]. Enrichment analysis was performed on gene lists through comparison against a gene universe of 75251 genes, which comprised all genes expressed in our dataset that had high quality annotation and could be associated with a GO term (nodeSize = 10, algorithm = weight01) [97]. WeightedFisher values were adjusted for multiple testing using the Benjamini-Hochberg correction method to generate adjusted p-values, filtered stringently for adjusted p < 0.001. The same pipeline was used for genes with rhythmic control and cold CPM profiles, and rhythmic ∆CPM that had a displacement of abundance for the duration of the cold treatment (displacement ratio <0.5).

Investigation of the contribution of wheat subgenomes to triad responsiveness

Subgenome occupancy was determined using information within the gene identifier (e.g. TraesCS1A02G123456). Genes were organised into triads with a 1:1:1 correspondence using the triad numbers in the High Confidence Triad table from [49] (https://github.com/Uauy-Lab/WheatHomoeologExpression/Data/data/TablesForExploration/HCTriads.csv). Triads were considered to be part of a circadian gated induction ETMR group if any homoeolog from the triad was categorised into the group. Individual homoeologs of triads were sometimes within different ETMR groups (Fig 5F). An average ∆CPM value for the ETMR for each gene was determined as: where ZTlower is the ZT timepoint corresponding to the lower time limit of the ETMR group and ZTupper is the ZT timepoint corresponding to the upper time limit of the ETMR group, e.g for ETMR group 4–8 h, ZTlower = ZT 28 and ZTupper = ZT 32. The relative contribution of each homoeolog to the triad responsiveness for the ETMR was determined as: where x is the subgenome of the homoeolog being investigated, and y and z are the remaining two subgenomes. This used a method similar to Ramirez-Gonzalez et al. [49]. The relative contributions were determined and visualised in R using ggtern [99].

Identification of transcripts where cycling is stabilized during an acute cold treatment

We wished to identify the entire set of transcripts that had a stabilization of abundance during the cold treatments. To achieve this, transcripts with rhythmicity at control temperatures and in response to 3 h cold treatments were identified using the ARSER algorithm in MetaCycle (using meta3d) [95,100]. Genes previously defined as having a circadian gated upregulation or downregulation in response to cold were removed from this group to prevent double-counting of individual genes. Adjusted ∆CPM (Adj.∆CPM; S6 Fig) values were determined for timepoints ZT28-44 for each gene as:

Where ZTx is the is sampling timepoint. A time course mean ∆CPM for each gene was determined as:

And a mean for adjusted ∆CPM for each gene was determined as:

A displacement ratio value was determined for each gene in this group using mean ∆CPM and mean adjusted ∆CPM values as:

Displacement ratios less than 1 indicate a reduced mean ∆CPM following adjustment. Gene identifiers for putative wheat circadian clock genes were obtained from [43], and displacement ratios determined for each.

cDNA synthesis and RT-qPCR analysis

For RT-qPCR experiments, the amount of RNA was standardised across samples to to 0.1 μg/μl and cDNA was synthesised using a High-Capacity cDNA Reverse Transcription Kit and random primers (Applied Biosystems). Transcript abundance in cDNA dilutions of 1/100 was measured in 25 μl reactions using qPCRBIO SyGreen LO-ROX RT-qPCR reagent (PCR Biosystems) with relevant primers on a CFX96 qPCR machine (Bio-Rad, California, US). Transcript abundance was normalized to TaHK4 (TraesCS5A02G015600) [101] according to the ΔΔCT method. Primers (5’ -> 3’) were ATCAATAATGGTACTTCTCCGGG (forward) and TGATTTTGGAACTTCTCTGGTG (reverse) for TaLHY, and TCTAAATGTCCAGGAAGCTGTTA (forward) and CCTGTGGTGCCCAACTATT (reverse) for TaHK4.