Compensatory Mutations in GI and ZTL May Modulate Temperature Compensation in the Circadian Clock^,

The physical interaction between naturally occurring mutant variants of Arabidopsis circadian clock components contributes to temperature compensation in the Cape Verde Islands.

Abstract

Circadian systems share the three properties of entrainment, free-running period, and temperature compensation (TC). TC ensures nearly the same period over a broad range of physiologically relevant temperatures; however, the mechanisms behind TC remain poorly understood. Here, we identify single point mutations in two key elements of the Arabidopsis circadian clock, GIGANTEA (GI) and ZEITLUPE (ZTL), which likely act as compensatory substitutions to establish a remarkably constant free-running period over a wide range of temperatures. Using near-isogenic lines generated from the introgression of the Cape Verde Islands (Cvi) alleles of GI and ZTL into the Landsberg erecta (Ler) background, we show how longer periods in the Cvi background at higher temperatures correlate with a difference in strength of the GI/ZTL interaction. Pairwise interaction testing of all GI/ZTL allelic combinations shows similar affinities for isogenic alleles at 22°C, but very poor interaction between GI (Cvi) and ZTL (Cvi) at higher temperature. In vivo, this would result in lower ZTL levels at high temperatures leading to longer periods in the Cvi background. Mismatched allelic combinations result in extremely strong or weak GI/ZTL interactions, indicating how the corresponding natural variants likely became fixed through epistatic selection. Additionally, molecular characterization of GI (Cvi) reveals a novel functional motif that can modulate the GI/ZTL interaction as well as nucleocytoplasmic partitioning. Taken together, these results identify a plausible temperature-dependent molecular mechanism, which contributes to the robustness of TC through natural variation in GI and ZTL alleles found on the Cape Verde Islands.

Circadian oscillators in eukaryotes are comprised of multiply interlocked transcriptional-translational circuits that create and sustain rhythms of ∼24 h (Hurley et al., 2016). The Arabidopsis circadian system complex incorporates at least 20 genes in the maintenance of a robust circadian period (Hsu and Harmer, 2014; Nohales and Kay, 2016). Most current models are largely based on transcriptional relationships among clock components, but post-translational processes, such as regulated proteolysis and splicing, are increasingly being viewed as equally important (Fujiwara et al., 2008; Kim et al., 2007, 2011; Seo and Mas, 2014).

Temperature compensation (TC) is a defining feature of all known circadian systems (Edwards et al., 2005; Narasimamurthy and Virshup, 2017). This mechanism maintains the pace, or period, of the clock stably over a broad range of physiologically relevant temperatures, establishing an accurate timing mechanism independent of ambient temperature. The molecular basis of TC is unclear. Numerous studies in a variety of circadian systems have implicated different genes and mechanisms as central to its maintenance, suggesting that more than one approach to solving this problem has evolved. The strongest common element across species is the role of protein phosphorylation. In Neurospora crassa, phosphorylation of the frequency protein by casein kinase2 (CK2) alters protein turnover, especially at higher temperatures, contributing to normal TC (Mehra et al., 2009). In mammals, CK1 phosphorylates PER2 and contributes to the action of a two-site phosphoswitch that controls PER2 stability in a temperature-dependent way, contributing to TC (Isojima et al., 2009; Narasimamurthy and Virshup, 2017).

In Arabidopsis, CK2 phosphorylates CCA1, reducing its binding to oscillator genes (Portolés and Más, 2010). Altering CK2 abundance disrupts TC, in part by altering CCA1 chromatin presence (Portolés and Más, 2010). Additional elements implicated in TC in Arabidopsis include the importance of the expression levels of certain clock proteins (e.g. LHY, CCA1, PRR7, and PRR9; Gould et al., 2006; Salomé et al., 2010), factors involved in alternative splicing (James et al., 2012; Wang et al., 2012; Schlaen et al., 2015; Marshall et al., 2016), and the photoreceptor cryptochrome (Gould et al., 2013).

Recent reports on the role of GIGANTEA (GI) have clarified its central role as a post-translational stabilizer of the F-box proteins ZEITLUPE (ZTL) and FKF1, in the control of circadian period and flowering time, respectively (Kim et al., 2007). In the absence of GI or in the presence of mutations that disrupt the GI–ZTL interaction, peak ZTL levels are reduced 4- to 5-fold (Kim et al., 2007). GI acts as a cochaperone of the ubiquitous protein chaperone, HSP90, to specify the maturation of ZTL into a functional component of the SCF^ZTL E3 ligase that targets the transcriptional repressors TOC1 and PRR5 for ubiquitylation and degradation (Cha et al., 2017). GI protein oscillates with a late evening phase, and a blue light-enhanced GI–ZTL interaction helps create and sustain an evening-phased post-translational rhythm of ZTL abundance (Kim et al., 2007). This contributes to the maintenance of high-amplitude oscillations of TOC1 and PRR5, which is essential for normal circadian period in Arabidopsis (Kim et al., 2007; Somers and Fujiwara, 2009).

TC in Arabidopsis has been previously examined by exploiting the natural variation between two accessions of Arabidopsis adapted to different thermal environments. One population was obtained from a warm climate, the Cape Verde Islands (Cvi; 15° N to 17° N), and the second from a temperate European location (Landsberg erecta, Ler; 53° N; Edwards et al., 2005). Recombinant inbred lines (Alonso-Blanco et al., 1998) from the two accessions were used to identify several QTLs for TC. Two strong candidates identified were the core clock components GI and ZTL, which were further resolved using near-isogenic lines (NILs; Edwards et al., 2005; Keurentjes et al., 2007). ZTL^Cvi was implicated to promote period lengthening at high temperatures whereas GI^Cvi was associated with shortened period, at least at 22°C (Edwards et al., 2005). These results suggested opposing effects of these alleles on period at different temperatures, though no mechanism was suggested.

Here, we identify a likely molecular mechanism for TC that arises through the cooperative action of GI^Cvi and ZTL^Cvi alleles. We propose that each variant alone drives opposing period effects but ZTL^Cvi compensates for the overall period shortening of GI^Cvi at higher temperatures, resulting in strong TC. This period compensation effect is mediated by changes in ZTL abundance through a thermo-dependent GI-ZTL interaction. Additionally, the GI^Cvi allele resides in a novel functional motif that modulates the GI/ZTL interaction as well as GI nuclear trafficking. Taken together, these results identify a probable molecular mechanism that increases the robustness of Arabidopsis TC capability through the natural variation of GI and ZTL at the high thermal environments of Cape Verde Island.

RESULTS

Natural Variants of GI and ZTL Mediate Temperature-Dependent Period Variations

A previous report demonstrated temperature-specific period effects associated with the Cvi alleles of GI and ZTL (GI^Cvi and ZTL^Cvi) through leaf movement analysis using NILs, suggesting GI and ZTL are candidate loci underlying the temperature effects (Edwards et al., 2005). We obtained a more direct and molecular assessment of clock activity by crossing the appropriate NILs to a CCA1:LUC (Ler) reporter line and selecting for NILs expressing CCA1:LUC and harboring either GI^Cvi only (G^CZ^L), ZTL^Cvi only (G^LZ^C), or both GI^Cvi and ZTL^Cvi (G^CZ^C), which were compared to the Ler (G^LZ^L) control (Supplemental Fig. S1). We then characterized the temperature effect of these genotypes on circadian period.

We first observed circadian rhythmicity for the NILs and wild-type Ler (G^LZ^L) under continuous light at 15°C, 22°C, and 27°C (Fig. 1). The periods of each genotype were not significantly different within a genotype at 15°C and 22°C (Fig. 1A). However, for most genotypes periods shortened significantly at 27°C (Fig. 1A), although the magnitudes of the period changes were different among the four genotypes tested (Fig. 1, A and C; Supplemental Fig. S2, A and B). Most notably, G^CZ^C displayed the least shortened period among the four lines, showing strong TC over the full 12°C temperature range. The two other NILs and Ler (G^LZ^L) were extremely similar, with a strong tendency toward period shortening at higher ambient temperatures (Fig. 1C).

Figure 1.

Circadian phenotypes mediated by interactions between GI and ZTL natural alleles. A, Mean circadian period (CCA1pro:LUC) in Ler (G^LZ^L) and three NILs harboring GI^Cvi alone (G^CZ^L), ZTL^Cvi alone (G^LZ^C), or both GI^Cvi and ZTL^Cvi (G^CZ^C) at 15°C, 22°C, and 27°C under constant red light (35 μmol m⁻² s⁻¹) after entrainment in 12-h light/12-h dark cycles (LD). Error bars indicates ±se (n = 2 for 15°C, n = 3 for 22°C and 27°C; see Supplemental Fig. S2 for details). Different letters indicate significant differences at a given temperature within each line according to Tukey’s multiple comparisons. B, Normalized circadian traces. C, TC profiles determined by the mean period values of Ler and the NILs.

In addition, lines with GI^Cvi (G^CZ^C and G^CZ^L) showed a significantly shorter period at all temperatures regardless of the ZTL allele (Fig. 1, A and C; Supplemental Fig. S2B). The periods of G^CZ^C and G^CZ^L were significantly shortened by ∼1.1 and 1.5 h at 22°C relative to G^LZ^L, respectively (Supplemental Fig. S2B). The period of G^CZ^L was similar to G^CZ^C at 15°C and 22°C but was further shortened at 27°C. This indicates that the shortening effect of GI^Cvi is most acute at high temperatures.

In contrast, ZTL^Cvi modulates the temperature-dependent period shortening effect of GI^Cvi (Fig. 1). The two G^Ler lines (G^LZ^C and G^LZ^L) were similar in period at 15°C but at higher temperature the period of G^LZ^C remained longer, relative to the Ler control (Supplemental Fig. S2B). A high temperature-dependent period lengthening effect of ZTL^Cvi is consistent with this difference, and when paired together with GI^Cvi (G^CZ^C) at 27°C, it can explain the strong TC seen in this line (Fig. 1, A and C). Overall, the NIL data support the notion that at higher temperatures GI^Cvi has a period shortening effect that is counterbalanced by ZTL^Cvi, supporting a nearly unchanging period over this range of temperatures when the two alleles are together (G^CZ^C; Fig. 1C). The findings are similar under white light, with the trend of G^CZ^C less acute than for the other two NILs and the Ler parent line (Supplemental Fig. S3).

These conclusions are bolstered by statistical modeling analysis (Supplemental Methods). We analyzed the red and white-light data sets using a linear model with the period as an outcome and the loci GI, ZTL, and temperature (15°C, 22°C, and 27°C) as categorical covariates. This approach allowed consideration of which of these three factors, and their potential interactions, contributed to period variation. Based on this analysis, we found that GI^Cvi has a significant period-shortening effect under red light, compared to GI^Ler. In contrast, ZTL^Cvi has, to a lesser degree, significant period-lengthening effect, compared to ZTL^Ler (see Supplemental Methods; Supplemental Table S1). High temperature (27°C) also contributes to period shortening, as visually apparent in Figure 1. Under white light, the independent effects of the alleles are the same, but additionally there is a strong period lengthening effect contributed by the joint effect of GI^Cvi and ZTL^Cvi at high temperature (see Supplemental Methods; Supplemental Table S2).

These period effects mediated by GI^Cvi and ZTL^Cvi are unlikely to be due to background differences at other loci because independent segregants derived from the original NILs x Ler (CCA1-LUC) crosses (Supplemental Fig. S1) showed similar period effects among them (Supplemental Fig. S4), and additional segregants derived from the G^CZ^C x Ler (CCA1-LUC) crosses (Supplemental Figs. S5 and S6) exhibited the expected period phenotypes.

GI-ZTL Allelic Interactions Can Account for High Temperature-Enhanced ZTL Accumulation

Circadian period and ZTL protein level are inversely correlated in Arabidopsis (Somers et al., 2004) and GI positively regulates ZTL protein levels through direct protein interactions (Kim et al., 2007). Therefore, we next tested whether the opposing effects GI^Cvi and ZTL^Cvi have on period might arise from altered ZTL levels. We monitored ZTL levels at Zeitgeber time (ZT) 1 and 13 in seedlings of the NILs (G^CZ^C, G^LZ^C, and G^CZ^L) and Ler (G^LZ^L) grown at 22°C and 27°C under light/dark (LD) cycles (Fig. 2). We observed differences in steady-state levels of ZTL among the lines and at the different temperatures.

Figure 2.

Temperature effects on endogenous ZTL abundance. A, Endogenous ZTL levels in NILs and the recurrent parent (RP). Seedlings were grown for 7 d in 12-h light/12-h dark cycles (LD; 22°C), then continued at 22°C or 27°C for five additional days under LD cycles and sampled at ZT1 and ZT13. Immunoblot representative of at least three trials. B, Quantitation of ZTL levels. Error bars represent ±se (n = 3). Different letters indicate significant differences at a given temperature according to Tukey’s multiple comparisons, followed by ANOVA. ADK was used as loading control. Quantitation relative to ADK band and normalized to highest protein level at a given temperature.

Figure 2 shows that the shorter periods consistently associated with GI^Cvi are not due to marked differences in ZTL levels. All genotypes have nearly identical ZTL levels at ZT1 at 22°C. At ZT13 (22°C), ZTL levels in both G^LZ^C and G^LZ^L are very similar to those in the shorter period G^CZ^C line (Fig. 2B). Similarly, at 27°C, both GI^Ler lines show longer periods than GI^Cvi lines (Fig. 1, A and C), despite similar ZT13 ZTL levels in, for example, G^CZ^L and G^LZ^L at both time points (Fig. 2). This indicates that the consistently shorter periods of GI^Cvi genotypes arise from effects of GI on other aspects of the circadian system.

Within a GI genotype, however, the period reflects the expected effects of ZTL and GI interactions inferred from the genetics, particularly at 27°C. The longer period of G^LZ^C relative to G^LZ^L at 27°C (Fig. 1) correlates well with the higher ZTL levels in G^LZ^L (at ZT13) and its shorter period (Fig. 2). ZTL^Cvi is less abundant in the presence of GI^Ler, especially at 27°C (Fig. 2), correlating with the longer period (Fig. 1).

For GI^Cvi, ZTL protein abundance also correlates strongly with period in a ZTL allele-specific way. At 22°C, the ZTL level of G^CZ^L is significantly higher relative to G^CZ^C at ZT13, supporting a positive role of GI^Cvi on ZTL^Ler stability and a shorter period (Fig. 2; Fig. Supplemental S2B). This effect is further accentuated at 27°C, where ZTL levels in G^CZ^L are nearly 3-fold higher than those found in G^CZ^C at ZT13 (Fig. 2B), correlating with the much shorter period of G^CZ^L (Fig. 1; Supplemental Fig. S2B). Taken together, these results support the notion that GI controls ZTL accumulation depending on the particular combination of alleles. Hence, to account for the better TC of G^CZ^C, we hypothesize that the longer period at 27°C arises from an inherently strong interaction of GI^Cvi with ZTL being counterbalanced by an inherently weaker ZTL^Cvi interaction. This would result in lower accumulation of ZTL and the longer period.

Examination of the mRNA expression patterns of the GI and ZTL loci under the same temperatures did not show any significant differences among them, indicating these effects occur post-transcriptionally (Supplemental Fig. S7).

Temperature-Dependent ZTL Accumulation Results from Temperature Effects on GI-ZTL Interactions

We tested the above hypothesis by addressing the molecular basis of the different GI and ZTL allelic combinations on ZTL levels. Amino acid sequence alignments of the GI and ZTL alleles of Col, Ler, and Cvi accessions revealed that GI^Cvi and ZTL^Cvi differ from GI^Ler and ZTL^Ler by single amino acid substitutions of a Leu-to-Phe (L718F) and a Pro-to-Thr (P35T), respectively (Supplemental Fig. S8). Additionally, GI^Col is nearly identical to GI^Ler except for an I-to-V substitution at residue 113 (Supplemental Fig. S8).

To determine whether these differences alter the GI-ZTL interaction, we performed quantitative coimmunoprecipitation (co-IP) assays analyzing all combinations of GI and ZTL natural alleles through transient expression in Nicotiana benthamiana (Fig. 3). In these interaction experiments, we compared the more commonly used Columbia (Col-0) alleles to Cvi alleles, so we first established that the interactibility of GI^Col and GI^Ler with ZTL^Col/Ler are not significantly different (P = 0.210, two-sample Student’s t test). In vivo, ZTL protein levels in the Col and Ler accessions are very similar, suggesting that ZTL interacts equally well with both GI alleles (Supplemental Fig. S9).

Figure 3.

GI and ZTL temperature-dependent interactions. A and B, N. benthamiana plants were grown at 22°C under 16-h light/8-h dark cycles (LD) and protein extracts from leaves transiently expressing GI^Col-GFP and GI^Cvi –GFP (GI^Cvi) coexpressed with ZTL^Col or ZTL^Cvi were harvested at ZT12. Anti-GFP immunoprecipitates (IP) were probed with anti-ZTL polyclonal antibody (co-IP). Quantitation based on the ratio of co-IP/IP signals, normalized to strongest interaction. Image is representative of at least three trials; error bars represent ±se (n = 3). C and D, As in (A) and (B), but for plants maintained in 22°C or 27°C before extraction and processing. Statistical difference between pairs (Student’s t test) is shown as ^*P < 0.05, ^**P < 0.01. NS, not significant.

ZTL^Col or ZTL^Cvi were coinfiltrated with tagged GI^Col-GFP or GI^Cvi-GFP into N. benthamiana and the association of ZTL with GI-GFP in the immunoprecipitates was measured at 22°C (Fig. 3). When a ZTL allele was paired with the complementary GI variant naturally found in that accession, the extent of the interaction was very similar: GI^Col/ZTL^Col and GI^Cvi/ZTL^Cvi interactions were nearly identical (Fig. 3B). In contrast, the test of the GI^Cvi /ZTL^Col interaction resulted in significantly larger amounts of coimmunoprecipitated ZTL, whereas the GI^Col /ZTL^Cvi combination showed the weakest interactions (Fig. 3B). These results strongly suggest an epistatic selection for similar ZTL-GI interactiveness within an accession under 22°C.

We extended these tests in N. benthamiana that were maintained at 27°C for 3 d after infiltration. Here, we found that the paired Col-0 alleles of GI and ZTL interacted >2-fold better at 27°C relative to that seen at 22°C (Fig. 3, C and D). Conversely, ZTL-GI Cvi interactions at 27°C were reduced by 2-fold (Fig. 3). These results are consistent with the long period (Fig. 1, A and C) and lower amounts (Fig. 2) of ZTL^Cvi observed in the G^CZ^C NIL at 27°C.

We conducted a more extensive time series and compared endogenous ZTL protein accumulation within two Cvi accessions and Col at 22°C and 27°C. ZTL levels in Col were consistently higher at all-time points at 27°C, consistent with a shorter period at that temperature (Fig. 4, A and B). In contrast, peak ZTL^Cvi levels were lower at 27°C over the time series (Fig. 4, A and C). When samples were compared on the same immunoblot with the same exposure, similar trends were observed (Supplemental Fig. S10). ZTL transcript levels are not markedly different under these conditions, indicating that the temperature-dependent effects on ZTL protein levels are determined post-translationally (Fig. 4D). Taken together, results from both the transient coexpression data and in vivo ZTL levels demonstrate that a temperature-dependent interaction between GI and ZTL results in changes in the steady state of ZTL over a diurnal time course.

Figure 4.

Temperature-dependent ZTL accumulation depends on genotype. A to C, Seedlings (Col, Cvi) were grown for 7 d in 12-h light/12-h dark cycles (LD) at 22°C, continued at 22°C or 27°C for an additional 5 d under LD cycles, then sampled at the indicated ZT. Endogenous ZTL protein was probed with anti-ZTL polyclonal antibody. ZTL abundance was quantitated relative to a nonspecific band above the target protein and normalized to the most abundant level within each genotype for both temperatures. D, ZTL transcript abundance determined by reverse transcription quantitative PCR and normalized to ACTIN2. Cvi data are from four independent trials; two with Cvi-0 and two with Cvi-1 together with the corresponding Col accession. Statistical difference between pairs (Student’s t test) is shown as ^*P < 0.05, ^**P < 0.01. Error bars (B–D) represent ±se (n = 4).

GI^Cvi Enhances GI Cytosolic Localization and Strengthens the GI-ZTL Interaction

GI is present in both the nucleus and cytosol and the relative partitioning of GI can affect ZTL levels (Kim et al., 2013a). Previous work indicated residues 543–783 are sufficient for nuclear targeting, within which the Cvi/Ler polymorphisms (GI⁷¹⁸) reside (Huq et al., 2000). Therefore, we next tested whether the differences in ZTL^Cvi and ZTL^Col/Ler accumulation might also arise, in part, from differences in the nucleocytoplasmic partitioning of the different GI alleles. We tested the effect of GI^Cvi on nucleocytoplasmic partitioning through transient expression. GI^Cvi displayed 2-fold enriched presence in the cytosol relative to GI^Col (P < 0.001, one-way ANOVA; Fig. 5, A and B).

Figure 5.

GI^Cvi enhances cytosolic partitioning and ZTL interaction. A and B, GI^Col-GFP and GI^Cvi-GFP transiently expressed in N. benthamiana and separated into cytosolic and nuclear fractions. Immunoblot is representative of three trials. Levels of cytosolic and nuclear GI protein were normalized to ADK and H3. Error bars represent ±se of three independent trials. C and D, N. benthamiana leaves transiently and ectopically expressing GI^Col –GFP or GI^Cvi –GFP coexpressed with ZTL^Col were harvested at ZT12. Anti-GFP immunoprecipitates (IP) were probed with anti-ZTL polyclonal antibody (CoIP). Quantitation based on the ratio of co-IP/IP signals and normalized to the GI^Cvi-ZTL total interaction. Image representative of at least three trials; error bars represent ±se (n = 3). S.E, short exposure; L.E, long exposure. Statistical differences between pairs (Student’s t test) is shown as ^*P < 0.05, ^**P < 0.01.

As ZTL is localized in the cytosol, we tested whether the allele-specific differential interactions that we observed among the different GI and ZTL alleles were due to the higher affinity of GI^L718F for ZTL or its predominant cytosolic localization. We compared the interaction of GI^Col and GI^L718F with ZTL and found that in both total and cytosolic extracts GI^L718F interacts 3-fold more strongly with ZTL relative to GI^Col, indicating that the enhanced protein accessibility of GI^L718F to ZTL cannot explain the strong interaction (Fig. 5, C and D). We confirmed in vitro that the GI^Cvi allele has intrinsically higher affinity for ZTL using GST-tagged ZTL (P < 0.05, two-sample Student’s t test; Supplemental Fig. S11).

GI^718L Is within a Functional Unit that Modulates GI/ZTL Interaction and GI Nucleocytoplasmic Localization

Closer inspection of the sequence shows that the L718F substitution occurs within a highly conserved Leu-rich region at residues 712–721 (LXXLXXLXXL; Supplemental Fig. S12) that is found across many taxa. Hence, it is plausible that the L-to-F difference in GI^Cvi at GI(718) interrupts nuclear targeting either by modifying this motif or a neighboring nuclear localization sequence.

To determine whether this motif is required for either the GI/ZTL interaction or nuclear import, we created serial L-to-A mutations in the LXXLXXLXXL region (Fig. 6). We found that single L-to-A mutations (L712A, L715A, and L718A) within this region diminished the relative nuclear abundance, with substitutions closer to the N terminus being the most effective. Moreover, the triple mutant (L715A, L718A, and L721A) markedly suppressed the nuclear presence of GI relative to that of the single mutants (Fig. 6, A and B). Notably, the nuclear accumulation of GI^L721A is comparable to the wild type (GI^Col), demonstrating the necessity of only the L⁷¹²XXL⁷¹⁵XXL⁷¹⁸ motif for nuclear accumulation (Fig. 6, A and B).

Figure 6.

The L⁷¹²XXL⁷¹⁵XXL⁷¹⁸ motif determines GI nucleocytoplasmic partitioning and ZTL interaction. A and B, N. benthamiana leaves transiently and ectopically expressing GI^Col-GFP, GI^L712A-GFP, GI^L715A-GFP, GI^L718A-GFP, GI^L721A-GFP or GI^LAAA-GFP were separated into cytosolic and nuclear protein fractions and detected with anti-GFP antibody. Immunoblots quantitated as the ratios of cytosolic to nuclear GI. Levels of cytosolic and nuclear GI protein were normalized to ADK and H3, respectively. Error bars represent ±se of three independent trials. C and D, N. benthamiana leaves transiently and ectopically expressing GI alleles in (A) and (B) coexpressed with ZTL^Col were harvested at ZT12. Anti-GFP immunoprecipitates (IP) were probed with anti-ZTL polyclonal antibody (Co-IP). Interactions were quantitated by the ratio of co-IP/IP signals, normalized to the strongest interaction. Gel image is representative of at least three trials; error bars represent ±se (n = 3). Statistical difference between pairs (Student’s t test) is shown as ^*P < 0.05 and ^**P < 0.01. WT, wild type.

We next determined if this same motif affects the GI-ZTL interaction. Quantitative co-IPs with these GI mutant alleles and ZTL showed that mutations in the motif weaken the GI-ZTL interaction (Fig. 6, C and D). GI^L718A diminished the interaction slightly (<2-fold) relative to the wild type, but GI ^L712A and GI^L715A strongly decreased the interaction further. Furthermore, the triple GI mutant allele completely abrogated the interaction (Fig. 6, C and D). Together, these data indicate that the LXXLXXL motif is a necessary determinant for both the GI-ZTL protein interaction as well as nucleocytoplasmic localization. Both functions are most strongly affected by loss of the more N-terminal Leu residues (Fig. 6, B and D).

Natural Variants of GI^L718F and ZTL^P35T Are Unique

To investigate how frequently the GI^L718F and ZTL^P35T alleles might emerge in nature, we used Polymorph 1001 to examine the sequences of the GI and ZTL loci in 1135 Arabidopsis accessions (https://tools.1001genomes.org/polymorph/), in addition to ecotypes adapted to hot areas, including Tanzania (Tanz-1 and -2) and Marrakesh, Morocco (Aitba-2 and Toufl-1; Supplemental Fig. S13). With the exception of Cvi, none of these Arabidopsis accessions, including hot-adapted ecotypes, harbored either GI^L718F or ZTL^P35T. Considering the functional impacts of the natural alleles for TC, the co-occurrence of these alleles in Cvi may be an evolutionary outcome unique to Cvi, where the average daily temperature is much higher than in many other Arabidopsis habitats.

DISCUSSION

Temperature-Dependent GI-ZTL Interactions Alter Period through Effects on ZTL Abundance

In this study, we have exploited the natural genetic variation between two accessions of Arabidopsis to identify a role for ZTL and GI in the TC of the circadian clock. We propose that the mechanism behind this process hinges on a differential, temperature-dependent interaction between the two proteins based on two single amino acid changes within each of them, specifically L718F in GI and P35T in ZTL, which are present in the Cvi accession and absent in the Col and Ler alleles. We show that pairings between mismatched alleles result in a much reduced interaction (GI^Col/ZTL^Cvi) or an enhanced interaction (GI^Cvi/ZTL^Col) relative to like allelic pairings (GI^Col/ZTL^Col and (GI^Cvi/ZTL^Cvi; Fig. 3) at 22°C, which are balanced to result in very similar interactivity. However, at higher temperatures (27°C) this balance is lost, with a much reduced interaction seen between GI^Cvi/ZTL^Cvi and more enhanced interaction apparent between GI^Col/ZTL^Col (Fig. 3D). We propose that these interaction differences are sufficient to account for difference in the net accumulation of ZTL in the two backgrounds: a much higher level of ZTL in Col-0 at 27°C versus 22°C compared to in the Cvi accession under the same conditions (Fig. 4). This finding is consistent with the role of GI in the maturation of functional ZTL (Kim et al., 2007; Cha et al., 2017).

In turn, the lower ZTL levels correlate with the longer period in the Cvi accession at higher temperatures (Fig. 4; Somers et al., 2004). The period-shortening effect of the GI^Cvi allele is more acute at higher temperatures and it is the weaker interaction between GI^Cvi and ZTL^Cvi that results in the longer free-running period. Thus, the evidence suggests that temperature-dependent interactions between different ZTL and GI alleles underlie TC in this environmental setting.

Interestingly, an analogous relationship exists in the mouse (Mus musculus) clock, also as a mechanism of TC. Here, a balance between temperature-sensitive substrate binding and temperature-sensitive product binding underlies how CKIδ-dependent phosphorylation of its substrate (PER) is subject to TC (Shinohara et al., 2017). A single amino acid change at the enzyme′phosphopeptide binding site of CKIδ markedly alters the temperature dependency of CKIδ-dependent phosphorylation of PER by impairing ADP-dependent product binding. This mutant shows much greater degradation activity of PER2 protein at a higher temperature compared to wild-type CKIδ, resulting in shorter periods. Thus, a temperature dependency in substrate binding and product release is inherent in CKIδ structure, akin to how the temperature-dependent interaction between GI and ZTL turns on two specific, complementary residues within the two proteins.

GI^Cvi Shortens Period Independent of the ZTL Allele

The single L718F transition of GI^Cvi results in a consistently shorter period than the GI^Ler allele regardless of the ZTL allele with which it is paired and under all the growth temperatures tested (Fig. 1; Supplemental Fig. S2). This could indicate that GI^Cvi is expressed at higher levels than GI^Ler, hence raising ZTL levels, resulting in shorter periods. However, G^CZ^L and G^LZ^L have similar ZTL levels at 27°C (Fig. 2B), yet differ in the period by >2 h (Fig. 1; Supplemental Fig. S2). Similarly, at 22°C ZTL levels in G^CZ^C and G^LZ^C are about the same (Fig. 2B), yet G^CZ^C period is >1 h shorter than G^LZ^C (Supplemental Fig. S2). Thus, it appears that GI can also control circadian period through other mechanisms, separate from its effects on ZTL accumulation. This notion is supported by the fact that many GI mutants display both short periods and reduced ZTL levels (Kim et al., 2007; Somers et al., 2007). Interestingly, this strong period shortening effect of GI^Cvi is diminished under white light (Supplemental Fig. S3), although the tendencies seen under red light for each NIL are still present. This finding suggests that the added presence of blue light may differentially affect one variant of GI over the other.

Two known GI interactors, ELF3 and ELF4, are core components of the tripartite Evening Complex (EC; ELF3-ELF4-LUX), a key element of the circadian clock that also helps integrate environmental signals with endogenous processes (Nusinow et al., 2011; Herrero et al., 2012; Ezer et al., 2017). ELF4 interacts with and sequesters GI in the nucleus, restricting its chromatin access (Kim et al., 2013b). Further, ELF3, together with COP1, controls GI turnover in the dark (Yu et al., 2008). Hence, allelic differences among the Cvi and Ler/Col accessions at one or more of these loci could be partially responsible for the ZTL-independent period differences observed in the NILs that are linked to GI^Cvi. For example, natural variation in ELF3 nuclear abundance has been associated with shortened free-running period, though the molecular basis of the difference in its nucleocytoplasmic partitioning is unknown (Anwer et al., 2014). Similarly, known variation in the poly-Gln (polyQ) tract length in ELF3 (7–26 polyQs) can correlate with ELF3-dependent phenotypes (Undurraga et al., 2012; Press and Queitsch, 2017). ELF3^Cvi (nine polyQs) and ELF3^Ler (17 polyQs) may interact differently with GI^Cvi, leading to period differences.

Although nuclear levels of GI have only been linked to effects on flowering time, through interaction with FKF1, natural variation in ELF4 nuclear sequestration of GI may indirectly affect period through GI effects on clock gene expression. Similarly, GI expression levels are affected by temperature, which is mediated through direct binding of the EC to the GI promoter (Mizuno et al., 2014). Natural variation in the relative affinity of the EC components for each other, and its chromatin residence, may subsequently alter GI levels in a temperature-specific manner. Further testing of the expression levels and interaction features of the Cvi and Ler/Col alleles of GI, ELF3, and ELF4 may resolve the ZTL-independent effects of GI on period.

The GI Leu-Rich Motif (LXXLXXL) Controls the GI-ZTL Interaction and GI Partitioning

We found that the L718F substitution in GI increases its affinity to ZTL but L718A strongly diminishes the interaction. Furthermore, serial changes at L712, 715, and 718 to A but not at L721 additively compromise both the GI-ZTL interaction and GI nuclear presence. Taken together, these results clearly support the notion that the L718 residue of GI is a part of Leu-rich unit motif (L712xxL715xxL718) that facilitates both the GI/ZTL interaction and nucleocytoplasmic partitioning. This sequence is superficially similar to the Leu-rich repeat sequence motifs (LxxLxLxxNxL or LxxLxLxxCxxL) that form the core of this well-characterized protein interaction domain (Matsushima and Miyashita, 2012). However, this stand-alone motif in GI is not part of a larger, repeating sequence (typically 4–30 repeats) and is likely unrelated to the structural and signaling functions found in plant receptor kinases (Leu-rich repeat receptor kinases).

Using the nuclear export sequence (NES) predictor, LocNES, the L712xxL715xxL718 motif was shown to reside within the second most likely NES of GI (Xu et al., 2015). Replacing L718 with either A or F greatly diminishes the predicted likelihood of this motif as a functional NES. If these changes disrupted an NES, we would predict a greater nuclear accumulation of GI. However, we observe a stronger cytosolic GI accumulation with both mutations (Figs. 5 and 6), suggesting that this region affects GI nucleocytoplasmic partitioning through a different mechanism. The Leu-rich motif also lies within a region (residues 543–783) previously identified as controlling nucleocytoplasmic partitioning (Huq et al., 2000). Strongly predicted nuclear localization sequence (SGSKRPRSEY) and NES (CAGVELASRLLFVV) motifs also reside within this stretch. Hence, our findings corroborate this previous report as defining an essential nucleocytoplasmic partitioning region of GI.

Our previous mapping of the GI-ZTL interaction domains identified the ZTL N terminus as essential for GI interaction but did not resolve the necessary GI elements (Kim et al., 2007). The Leu-rich motif now identifies a unique region within GI that is clearly essential for ZTL binding. The very large size of GI (1,173 residues) and many binding partners (Mishra and Panigrahi, 2015; Cha et al., 2017) suggests that it can fold to present multiple interaction surfaces. GI, ZTL, and HSP90 form a tripartite complex in which the GI N terminus (amino acids 1–391) interacts with the middle domain of HSP90 (Cha et al., 2017), and now a central region (amino acids 712–718) is essential for the GI-ZTL interaction. Taken together, these findings begin to establish functional domains of GI essential to its role in the circadian clock.

Potential Epistatic Selection of GI^Cvi and ZTL^Cvi

Epistatic interactions have been well-documented as underlying autoimmune responses and pathogen resistance in Arabidopsis (Bomblies et al., 2007; Alcázar et al., 2009; Chae et al., 2014; Zhu et al., 2018). Genetic analyses in these studies have implicated certain gene clusters or individual loci as important, but the biochemical or physical mechanism underlying the resultant phenotypes (e.g. poor growth, necrosis) has rarely been identified. Fixation of allelic pairs through the simultaneous selection of combinations of genotypes at two loci may arise through epistatic selection (Schlosser and Wagner, 2008; Nasrallah, 2013). Compensatory mutations, or amino acid substitutions, in interacting proteins that, together, offset their individual effects, are one type of epistatic interaction.

Compensatory intramolecular substitutions that affect RNA secondary structure have been described for HIV and the pre-mRNA of alcohol dehydrogenase (Kirby et al., 1995; Assis, 2014). In these examples, mutations that disrupt base pairing in an element of RNA secondary structure are considered deleterious if they destabilize a functionally important structure. If a second compensatory mutation arises, the original structure may be restabilized and fitness restored (Kirby et al., 1995). Similarly, intraprotein compensatory mutations can be recovered, for example, as second-site suppressors of temperature-sensitive folding mutations, which can restore protein activity or overall fitness (Mitraki et al., 1991, 1993; DePristo et al., 2005).

This concept can be expanded to consider the interaction between two or more proteins (Schlosser and Wagner, 2008). For example, the strong interactions between two key sex determination proteins, FEM-3 and TRA-2, which occur within three different species of Caenorhabditis, fail when tested across the species (Haag et al., 2002). These results strongly suggest a compensatory coevolution of binding partners. For GI and ZTL, we find that cross-accession alleles still interact, but with altered affinities; the divergence of interaction has not progressed to the extent found in Caenorhabditis species. Further interaction testing of GI and ZTL alleles from other Arabidopsis species and relatives could shed insight into how important their affinities are for period maintenance in general.

Compensatory substitutions may arise in different ways, depending on the strength of natural selection. If selection is very weak and the population size is small, a deleterious mutation may become fixed in the population before a second (compensatory) mutation arises, and then subsequently the pair becomes fixed. Under stronger selection, the first mutation may appear deleterious and compensatory substitution can only occur if the second mutation arises before the first is selected against (Nasrallah, 2013). However, without knowing more about the evolutionary history of the origins of GI^Cvi and ZTL^Cvi or the conditions under which their interaction was selected, it is not possible to know how they became fixed in the Cvi population.

The pairing of these specific GI and ZTL alleles in Cvi is unique among known Arabidopsis accessions. Our data supports the notion that both alleles were selected to help establish strong TC at high temperatures. Whereas it might be expected that other populations indigenous to high-temperature environments would show the same or similar compensatory mutations in GI and ZTL, no common or universal molecular mechanism of TC has been identified in either plants or animals. Hence, it is likely different solutions have independently evolved to attain period stability at high ambient temperatures. In this case, field tests could, in principle, be performed in which all four combinations of the Cvi and Ler/Col-0 GI and ZTL alleles were created (possibly via CRISPR-Cas mediated genome editing) in the Cvi background and tested in a high-temperature field environment for reproductive success. Reduced seed set or reduced survival of populations with Col/Ler alleles in place of the Cvi alleles would support the notion of a selective advantage conferred by the endogenous Cvi alleles. This outcome would be strong support for our hypothesis that the GI and ZTL Cvi alleles were selected for high temperature stabilization of circadian period.

Recent genomic analysis of Arabidopsis accessions sampled from within Africa and Asia suggests a history of the species that includes a more widespread historical distribution across Africa than previously considered (Durvasula et al., 2017; Zou et al., 2017; Fulgione and Hancock, 2018). Earlier assessments, based on 1,135 Arabidopsis accessions, concluded that Arabidopsis is comprised largely of a single majority group (1,109 accessions) out of six diverged groups, all survivors from multiple glacial refugia. From this analysis, Cvi-0 is in an outlier group, with recent evidence indicating an African origin based on S-haplotype analysis (Nordborg et al., 2005; The 1001 Genomes Consortium, 2016; Durvasula et al., 2017). The average daily temperature of the Cape Verde Islands tropical bioclimatic zone is higher over the year, with only mild variation, ranging from 25°C (77°F) in January to 29°C (84.2°F) in September (Abarca et al., 2001). Thus, the colonizing Arabidopsis populations on the Cape Verde Islands were likely exposed to thermal selective pressures very different from continental accessions, particularly at the high end of the temperature range.

The Cape Verde Islands environment is geographically remote and genetically isolated, yet GI and ZTL are remarkably conserved in their peptide similarity compared to alleles found elsewhere, except for these two substitutions. The evidence we provide here strongly suggests these are compensatory mutations, possibly arising through epistatic selection, which create an appropriately balanced GI/ZTL interaction at high temperatures to attain an adaptively temperature-compensated circadian clock.

MATERIALS AND METHODS

Plant Material

To obtain NILs covering GI^Cvi and ZTL^Cvi, we screened candidate NILs through SNPs in GI (C to T at nucleotide 3,061, AT1G22770) and ZTL (C to A at 103, AT5G57360) from NILs in which the Cvi accession had been introgressed into the Landsberg erecta (Ler) background (Keurentjes et al., 2007). CCA1: luciferase (Ler; Wang et al., 2013) were crossed to the relevant NIL lines and luminescent progeny homozygous for GI^Cvi or ZTL^Cvi or both were selected for subsequent rhythmic assay and protein analysis.

CCA1:luciferase was transformed into the Cvi-0 background via standard techniques. Two independent transformants were selected for subsequent study.

The NILs (CS717143–CS717147), Arabidopsis (Arabidopsis thaliana) ecotypes Cvi-0 (CS28198), Cvi-1 (CS28199), Tanz-1 (CS75719), Tanz-2 (CS75720), Aitba-2 (CS76347), and Toufl-1 (CS76348)) were obtained from the Arabidopsis Biological Resource Center (The Ohio State University).

Growth Condition and Temperature Treatments

All plant materials were grown under 12-h white fluorescent light (50–60 μmol m⁻² s⁻¹)/12-h dark cycles (LD) for 8–12 d on Murashige and Skoog (MS; GIBCO BRL) medium with 3% (w/v) Suc and 1% (w/v) agar as described in Somers et al. (2004). For all period analyses, seedlings were grown on MS agar under LD cycles for 5–7 d at 22°C and then transferred to constant red light (35 µmol m⁻² s⁻¹) at 15°C, 22°C, or 27°C. For protein expression experiments, 7-d–old LD-grown (12-h light/12-h dark) seedlings (MS agar media) at 22°C were then incubated at 15°C, 22°C, or 27°C for five additional days under the LD cycles. Seedlings were harvested at the indicated ZT times.

Rhythmic Assay

Seedlings for luminescence rhythm analysis were grown and period estimates determined as described in Somers et al. (2004). Constant white light (5,700 K) during circadian imaging experiments was provided by LED RX30 lamps (Heliospectra). Red light was provided by a 660-nm LED array.

Vector Construction

The 35S::GI-GFP, GI complementary DNA (cDNA) clone was PCR-amplified using the full-length GI cDNA derived from the Columbia ecotype (Col-0; Kim et al., 2007) using the primers 5′-GAA TTC ATG GCT AGT TCA TCT TCA T-3′ and 5′-CGC GGA TCC ATG GCT AGT TCA TCT TCA TCT GAG-3′. The amplified GI cDNA digested with EcoRI and BamHI was ligated into pENTR2B and transferred to pMDC85 via Gateway cloning. The 35S::ZTL was prepared by recombination of an entry clone that contained the ZTL cDNA (Kim et al., 2013a) to pMDC32. GI alleles, including GI^L718F, GI^L712A, GI^L715A, GI^L718A, and GI^L721A, and the ZTL allele ZTL^P35T were created through the site-directed mutagenesis with a QuikChange system (Stratagene) using GI and ZTL cDNA sequences in the entry clone.

Protein Analysis

Detection of ZTL and GI-GFP, adenosine kinase (ADK), and Histone H3 from Arabidopsis protein extracts was performed as described in Kim et al. (2007). For immunoprecipitation of GI-GFP and Co-IP of ZTL, anti-GFP polyclonal antibody was preincubated with protein A agarose (Invitrogen) at 4°C. After incubation for 1 h with gentle rotation, immune complexes were washed three times, then resuspended in SDS-PAGE sample buffer, briefly heated, and subjected to SDS-PAGE followed by immunoblotting. To quantify the protein interactions, the signal intensity of coimmunoprecipitated protein was normalized to that of primary-immunoprecipitate. For ZTL immunoblotting, protein was transferred to a PVDF membrane (Bio-Rad) and incubated with affinity-purified anti-ZTL antibody (1:500) for 4 h at room temperature or overnight at 4°C. The signal was visualized using peroxidase-conjugated secondary antibody (1:3,000–5,000; GE or Sigma-Aldrich) by enhanced chemiluminescence (ECL SuperSignal West Pico Chemiluminescent substrate; Pierce Biotechnology).

Isolation of nuclear and cytosolic proteins from tobacco tissue was performed with a CelLytic PN Isolation/Extraction Kit (Sigma-Aldrich) as described in Wang et al. (2010). Purification of each compartment was validated by immunodetection using cytosolic (ADK) and nuclear (Histone H3) marker proteins.

For the GST binding assay, GST and GST-ZTL were expressed in Escherichia coli BL21(DE3) and purified with glutathione-agarose (Sigma-Aldrich). Concentration of each protein bound to the glutathione-agarose was determined by Coomassie staining. Equal amounts of glutathione beads of GST and GST-ZTL were used for the in vitro binding assay. Protein extracts from Nicotiana benthamiana leaves transiently expressing GI-GFP were incubated with ZTL-GST resin. ZTL-bound GI alleles-GFP was detected with an anti-GFP antibody (cat. no. ab6556; Abcam).

Interactions between GI and ZTL were performed with total and cytosolic proteins from transient expression in N. benthamiana of 35S::GI-GFP and 35::ZTL and immunoprecipitations were performed as reported in Wang et al. (2010).

Reverse Transcription Quantitative PCR

Reverse transcription quantitative PCR was performed using conditions and primers as described in Kim et al. (2013a).

Polymorphism Analysis

The tool Polymorph 1001 (https://tools.1001genomes.org/polymorph/) was used to scan 1,135 accessions for amino acid polymorphisms at ZTL and GI, with the settings: Variant type, SNP; Impact, moderate; and effect type, missense. There were 96 variants found for ZTL (relative to Col) and only Cvi (strain 6911) diverged from Col with a missense that resulted in a P35T change. Similarly, using the same criteria for GI, there were 783 variants and only Cvi diverged from Col with a missense that resulted in an L718F change.

Accession Numbers

Sequence data from this article can be found in the European Molecular Biology Laboratory databases under accession numbers: At1g22770 (GI;https://www.ncbi.nlm.nih.gov/gene/838883) and At5g57360 (ZTL;https://www.ncbi.nlm.nih.gov/gene/835842).

Supplemental Data

The following supplemental materials are available:

• Supplemental Figure S1. Creation and identification of CCA1:LUC expressing NILs.

• Supplemental Figure S2. Tabular and graphic summary of the temperature-dependent period effects of Ler and the NILs.

• Supplemental Figure S3. Circadian phenotypes under white light mediated by interactions between GI and ZTL natural alleles.

• Supplemental Figure S4. Allele-dependent period effects of the NILs supported by additional segregants.

• Supplemental Figure S5. Period estimates of independent G^CZ^L lines derived from G^CZ^C.

• Supplemental Figure S6. Period estimates of independent G^LZ^C lines derived from G^CZ^C.

• Supplemental Figure S7. The period phenotypes of NILs are not based on transcript abundance differences.

• Supplemental Figure S8. Amino acid sequence profile of GI and ZTL alleles in Col, Ler, and Cvi accessions.

• Supplemental Figure S9. GI^Ler and GI^Col interact similarly with ZTL.

• Supplemental Figure S10. Endogenous ZTL levels in Col-0 and Cvi.

• Supplemental Figure S11. GI ^Cvi strongly interacts with ZTL in vitro.

• Supplemental Figure S12. L718F occurs at a highly conserved LxxLxxLxxL region.

• Supplemental Figure S13. GI^L718F and ZTL^P35T alleles are unique.

• Supplemental Table S1. Table of regression coefficients from a linear model (lm in R) with the periods as outcomes and the genetic loci GI and ZTL and temperature (15°C, 22°C, and 27°C) as categorical covariates under red light.

• Supplemental Table S2. Table of regression coefficients from a linear model with the periods as outcomes and the genetic loci GI and ZTL and temperature (15°C, 22°C, and 27°C) as categorical covariates under white light.

• Supplemental Methods. Linear model analysis.