Natural Variation in Drosophila melanogaster Clock PolyQ Length: Geographic Gradients and Functional Properties

ABSTRACT

Natural variation in circadian clock genes provides a powerful framework for understanding how organisms respond to environmental heterogeneity. The Clock (Clk) gene encodes a core transcriptional regulator of circadian rhythms and contains a polymorphic polyglutamine (polyQ) tract whose evolutionary significance remains unclear. Here, we integrate population genomic, behavioural, and molecular analyses to investigate the functional and geographic patterns of Clk polyQ variation in Drosophila melanogaster . Using data from 127 European populations, we identify 11 Clk polyQ alleles and find that their frequencies are geographically structured, most consistently along an east–west gradient: the Q25 and Q27 alleles show robust clines in longitude and in a bioclimatic axis of continentality, whereas latitudinal and altitudinal trends are weaker. Behavioural assays of near‐isogenic lines revealed that polyQ length modulates circadian function under thermal challenge: most alleles maintained stable free‐running periods across temperatures, whereas the intermediate‐length Q25 allele showed the strongest, though modest, temperature sensitivity. Circadian phase showed pronounced allele‐specific sensitivity to elevated temperature in laboratory assays, although phase variation did not display a consistent relationship with geographic variables. At the molecular level, luciferase reporter assays showed that longer polyQ alleles exhibited higher transcriptional activity, linking polyQ length to CLK‐mediated gene expression. Together, these results demonstrate that natural variation in Clk polyQ length has measurable functional consequences for circadian regulation and is geographically structured in patterns consistent with underlying climatic variation, highlighting the potential for low‐complexity regions to modulate clock function in a context‐dependent manner across environmental gradients.

1. Introduction

Circadian clocks are endogenous timekeeping systems that synchronize physiology and behaviour with the 24‐h environmental cycle. They confer a critical adaptive advantage by aligning internal biological processes such as metabolism, reproduction, and locomotor activity with predictable daily changes in light and temperature. Across taxa, disruption of circadian organization reduces fitness through impaired survival, fecundity, and competitive performance (reviewed by Jabbur et al. ²⁰²⁴). The circadian system therefore serves as a key interface between genetic regulation and environmental periodicity, allowing organisms to anticipate rather than simply react to environmental fluctuations.

At the core of the circadian system lies Clock (Clk), a transcription factor that functions as a master regulator of rhythmic gene expression. In Drosophila melanogaster , CLK forms a heterodimer with CYCLE (CYC) to activate downstream clock genes, including period (per) and timeless (tim), through E‐box enhancer elements (Darlington et al. 1998; Patke et al. ²⁰²⁰). The resulting feedback loops involving PER and TIM proteins generate near 24‐h oscillations that persist under constant conditions (Vosshall et al. 1994; Price et al. ¹⁹⁹⁸). CLK activity is further modulated by additional feedback components such as vrille (vri), pdp1ɛ and clockwork orange (cwo), which fine‐tune transcriptional amplitude and phase (Damulewicz and Mazzotta 2025). Because CLK sits at the apex of this transcriptional hierarchy, variation in its structure or activity has the potential to influence the entire circadian network.

Circadian rhythms are not rigidly fixed but can evolve in response to environmental pressures. Two key properties, entrainment to external cues and temperature compensation, enable organisms to maintain synchrony with local conditions. In natural populations, clock genes frequently exhibit molecular polymorphisms that affect these traits, suggesting that variation in clock components contributes to local adaptation (Bell‐Pedersen et al. ²⁰⁰⁵; Khatib et al. ²⁰²³). For example, Drosophila populations across Europe show subtle but significant geographic structure in allele frequencies of several circadian genes, with some variants linked to differences in free‐running period, activity phase and seasonal traits such as diapause and cold tolerance (Khatib et al. 2023). These patterns indicate that natural selection can act on circadian regulatory variation.

Among the core clock components, Clk stands out as a particularly promising locus for adaptive evolution. The gene contains a repetitive polyglutamine (polyQ) region in its C‐terminal domain, where variation in repeat length has been observed across multiple species. This region influences protein–protein interactions and transcriptional activity, suggesting that natural variation in Clk polyQ length could modulate circadian properties such as temperature compensation.

Natural variation in circadian clock genes has been widely documented and often exhibits geographic patterns consistent with local adaptation. A well‐studied example involves a repetitive region in the per gene that encodes pairs of Threonine–Glycine (Thr‐Gly) residues. The most common alleles (Thr‐Gly)₁₇ and (Thr‐Gly)₂₀, show a clear latitudinal cline in European populations of Drosophila melanogaster , with the longer repeat more frequent at higher latitudes and a similar, although weaker, trend observed in Australia (Costa et al. 1992; Sawyer et al. ²⁰⁰⁶). Behavioural studies revealed that flies carrying the (Thr‐Gly)₂₀ allele exhibit more effective temperature compensation, whereas the (Thr‐Gly)₁₇ variant shortens the free‐running period at low temperatures (Sawyer et al. 1997). These differences suggest that thermal selection maintains allele frequency gradients, favouring variants that optimize circadian performance under local climatic conditions (Hut et al. 2013; Kyriacou et al. ²⁰⁰⁸).

Similar geographic trends have been identified in Clk, highlighting its potential role in adaptive modulation of circadian timing. The Clk protein contains a repetitive polyglutamine (polyQ) tract located in its C‐terminal glutamine‐rich domain, encoded by a tandem array of CAG/CAA codons. Variation in polyQ length has been observed in several taxa and is often associated with environmental gradients. In Chinook salmon ( Oncorhynchus tshawytscha ), for instance, eight distinct Clk polyQ alleles were identified, with shorter alleles predominating in southern, low‐latitude populations and longer alleles more common at higher latitudes (O'Malley and Banks ²⁰⁰⁸). This variation correlates with migration timing, as individuals carrying longer alleles tend to migrate later in the season, suggesting that polyQ length diversity contributes to photoperiodic adaptation.

A similarly patterned relationship has recently been reported across a much broader taxonomic sample of coral reef fish. A comparative analysis of Clk polyQ and Q‐rich repeat length across 20 damselfish species (Pomacentridae) found a significant positive correlation between combined repeat length and pelagic larval duration, a key life‐history trait governing the developmental window before reef settlement. This relationship held at the interspecific level but not within a single species once local habitat availability was accounted for, suggesting that polyQ length may track developmental timing across lineages while additional ecological factors modulate its expression within populations (Schalm et al. 2025).

Comparable Clk polymorphisms have also been described in avian species. In the bluethroat ( Luscinia svecica ) and the blue tit ( Cyanistes caeruleus ), seven and nine polyQ alleles were identified, respectively. In blue tits, allele length showed a positive correlation with breeding latitude, whereas no such relationship was observed in the migratory bluethroat, possibly due to differences in reliance on local photoperiodic cues (Johnsen et al. 2007). These studies together support the idea that Clk polyQ length variation may provide a molecular basis for latitude‐dependent adjustments in circadian or seasonal timing.

The functional roles of polyQ tracts further reinforce this interpretation. PolyQ regions are common in transcriptional regulators and can modulate protein–protein interactions, structural flexibility, and transcriptional activity (Gemayel et al. 2015; Mier and Andrade‐Navarro ²⁰²¹). Moderate variation in polyQ length can tune transcriptional output, while extreme expansions beyond a critical threshold can lead to pathological aggregation (Lieberman et al. 2019). Experimental evidence from plants such as Arabidopsis thaliana and Populus shows that polyQ variation influences thermal responsiveness and protein localization (Sureshkumar et al. 2025). These findings suggest that naturally occurring variation in Clk polyQ length could serve as a mechanism for fine‐tuning transcriptional regulation and temperature compensation within the circadian system.

Despite extensive knowledge of the circadian molecular network, the contribution of Clk polymorphism to adaptive regulation of clock function remains unclear. The presence of a variable polyQ tract in Clk, coupled with its central regulatory role, suggests a plausible mechanism through which natural selection could fine‐tune circadian timing and temperature compensation. This study aims to test whether variation in Clk polyQ length influences circadian performance under different thermal conditions and to evaluate its potential role in local adaptation.

2. Methods and Materials

2.1. Analysis of Clk polyQ Allele Frequency Across Europe

A large genomic dataset of wild Drosophila melanogaster populations across Europe was obtained from the DrosEU consortium (Kapun et al. 2020). DNA sequences from 169 wild populations, each sequenced using Pool‐seq, were assembled with the SAMtools package (Li et al. 2009). Reads spanning the Clk polyQ region (Chr3L:7,764,779–7,764,908) were extracted together with flanking indel polymorphisms. Known indel variants corresponding to distinct polyQ alleles were manually counted by visual inspection of read alignments in the Integrative Genomics Viewer (IGV; Robinson et al. ²⁰¹¹). Allele frequencies were calculated by dividing the number of reads supporting each indel by the total read depth at that site.

To assess environmental associations, a principal component analysis (PCA) was conducted using 19 bioclimatic variables obtained from the WorldClim database (https://worldclim.org; Fick and Hijmans ²⁰¹⁷). The variables included annual means (e.g., temperature and precipitation), measures of seasonality (e.g., annual ranges in temperature and precipitation), and extreme environmental values (e.g., temperature of the coldest and warmest months and precipitation of the wettest and driest quarters).

To test for associations between Clk polyQ allele frequencies and geographic or climatic variables, we used generalized linear models (GLMs) implemented in R (RStudio version 2025.05.1). Allele frequencies (proportion of reads supporting a given allele) were modelled with a quasibinomial error and a logit link, which accommodates the overdispersion of Pool‐seq allele counts relative to the binomial expectation. For each allele, one model included latitude, altitude, and longitude as joint predictors, and a separate model included the first two climatic principal components (PC1 and PC2) derived from the bioclimatic PCA as explanatory variables. To avoid conflating spatial with seasonal variation, the geographic and climatic association analyses were restricted to spring‐collected samples (101 of the 127 populations). Because DrosEU populations were sampled in different seasons and allele frequencies at clock loci can shift seasonally (Bergland et al. 2014), analysing a single season isolates the spatial (clinal) signal from seasonal change. Effect sizes were expressed as odds ratios (OR), which describe how the odds of observing a given allele change with increasing values of the predictor variable (OR > 1 indicates a positive association, OR < 1 a negative association). Model significance was assessed using Wald tests, and p‐values were adjusted for multiple comparisons where appropriate.

2.2. Generation of Clk polyQ Allele Series

To establish populations homozygous for specific Clk polyQ alleles, virgin females were collected from several DrosEU isofemale lines. Each virgin female was paired with a male from the same isofemale line. DNA from the parental flies was extracted using a single‐fly squashing protocol and sequenced to determine their Clk polyQ genotype. Only F1 progeny from parental pairs that were homozygous for the same allele were retained. This procedure produced six homozygous populations carrying polyQ alleles Q21, Q25, Q29, Q30, Q31 and Q33. Sequence alignment confirmed that these alleles differed only in the number of consecutive glutamine residues within the uninterrupted polyQ stretch (Figure 1).

FIGURE 1.

Alignment of six Clk polyQ variants used in this study. Each sequence is labelled with ‘Q’ followed by the number of consecutive glutamine residues within the uninterrupted polyQ stretch.

2.3. Generation of Near‐Isogenic Lines

Near‐isogenic lines (NILs) are strains that are genetically identical except for one or a few loci of interest, which minimizes the effects of background genetic variation on phenotypes (Yuan et al. 2017). These lines were produced through a series of backcrosses in which the Clk locus from a donor parent was introgressed into the background of a deficiency line, with selection for the desired genotype at each generation (Kooke et al. 2012).

To generate Clk polyQ NILs, virgin females from each homozygous Clk polyQ line (Q21, Q25, Q29, Q30, Q31 and Q33) were crossed with males carrying a deficiency spanning the Clk locus (Df(3L)pbl‐X1/TM6C, Sb[1]; BDSC #1420). Wild‐type bristled hemizygous females from the F1 offspring were selected and backcrossed with males from the same deficiency stock. This process was repeated for four backcross generations (BC1–BC4). By the fourth generation, progeny were expected to share approximately 96% of their genome with the deficiency background, differing primarily at the Clk locus. Hemizygous BC4 males were collected for behavioural testing, while females were used for continued backcrossing until at least 20–30 flies per allele were obtained for behavioural assays. The crossing design used to generate the NILs is illustrated in Figure S1.

2.4. Analysis of Circadian Behaviour

Circadian activity was monitored using the Drosophila Activity Monitoring (DAM) system (Trikinetics, Waltham, MA). Flies were first entrained under a 12:12 h light–dark (LD) cycle for five days, followed by nine days in constant darkness (DD). Temperature compensation was examined by testing flies at 24°C and at 18°C. Recordings at 29°C were also attempted, but flies of these near‐isogenic lines showed high mortality and loss of measurable rhythmic activity at this temperature and did not yield usable data.

Activity data were analysed using the BioDare2 webserver (Zielinski et al. 2014) with the Maximum Entropy Spectral Analysis (MESA) method to determine each fly's free‐running period (FRP) that is the intrinsic circadian period measured under constant conditions without light entrainment. Flies were excluded from analysis if they (i) died before completing at least 10 days (5 LD + 5 DD), (ii) showed arrhythmic behaviour, or (iii) displayed FRP values outside the range of 22–28 h. These estimates were flagged by BioDare2 as requiring attention and correspond to weakly rhythmic or otherwise unreliable fits rather than to reliably estimated circadian periods. The criterion was applied uniformly to all genotypes and both temperatures without reference to allele identity (see per‐allele counts in Table S4).

Circadian phase was quantified from the entrained 12:12 LD recordings using a single deterministic pipeline written in base R (v4.3.3). Activity was recorded in 30‐min bins (48 bins per 24 h). Because the DAM monitors were started at lights‐off (ZT12, ~20:00), each folded profile was rotated so that bin 1 corresponded to ZT0 (lights‐on) before any phase reference was computed. For each fly, the LD recording days were folded into an average 24‐h activity profile (NA‐aware mean per bin). Flies were retained if they showed a mean activity of at least 2 counts per bin over the folding window and non‐zero activity on the final day (i.e., alive at the end of recording); only the six established NIL alleles (Q21, Q25, Q29, Q30, Q31 and Q33) were included. Because this criterion selects for a scorable LD activity profile rather than for a clean free‐running rhythm in DD, the phase dataset (n = 352 at 24°C; n = 396 at 18°C) is larger than, and analysed separately from, the FRP dataset described above.

Seven phase references were pre‐specified per fly. Within a morning window (ZT21–06) and an evening window (ZT07–16), we extracted the onset, peak, and offset of each activity component, together with the morning–evening (M–E) phase angle, defined as the interval between the morning and evening peaks. Onset and offset were taken as the half‐maximum up‐ and down‐crossings of the component (linearly interpolated between bins); the peak was estimated as the centroid (centre of mass above baseline) rather than the single highest bin, giving a more robust estimate for broad or bimodal peaks. Morning references were expressed relative to ZT0 (values after midnight rendered as negative) and the M–E angle as E_peak − M_peak (mod 24 h). The evening references fall predominantly in the anticipatory phase preceding lights‐off, limiting the contribution of the light‐off startle response.

Genotype effects were tested within each temperature by one‐way ANOVA and Kruskal–Wallis on each reference, with Benjamini–Hochberg (BH) correction applied across exactly the seven pre‐specified references; Tukey HSD was used for post hoc pairwise comparisons on references significant after correction. Genotype × temperature effects were tested on the combined dataset by two‐way ANOVA, with η ² reported as the effect size for each term and the interaction BH‐corrected across references and additionally verified by a distribution‐free Freedman–Lane permutation test (2000 permutations). For each reference we report the mean resultant length (R̄) as a diagnostic of phase concentration; because all references were strongly concentrated (R̄ ≥ 0.84, and ≥ 0.98 for the evening peak) and lie away from the ZT0 boundary, phases were summarized with linear statistics. A circular Watson–Williams test gave results concordant with the parametric analysis for all references. All statistics were computed in base R.

2.5. Luciferase Assay Constructs and Transformation

Two reporter plasmids were used for the luciferase assay: a Renilla reniformis luciferase reporter driven by the copia promoter (Renilla luciferase) for normalization, and a Drosophila per E‐box firefly luciferase reporter (pGL3‐4E‐hsp‐luc) containing four tandem per E‐box sequences (CACGTG) with 18 bp of flanking sequence upstream of a minimal hsp70 promoter. These constructs were generously provided by Sebastian Kadener and Lin Zhang.

The Clk expression vector pAC‐dCLK was generated by cloning Clk cDNA into the pBS‐SK(−) vector, followed by subcloning into the NotI site of the pAc5.1‐V5/His vector (Invitrogen) for expression in S2 cells (McDonald et al. ²⁰⁰¹). The empty pAc5.1‐V5/His vector was used as a negative control. All plasmids contained an ampicillin resistance gene for selection.

Escherichia coli DH5α cells were transformed using the standard heat‐shock protocol. Plasmid DNA was purified with the NucleoSpin Plasmid EasyPure Mini Kit (MACHEREY‐NAGEL). The Clk polyQ allele in the pAC‐dCLK plasmid was verified by Sanger sequencing using primers flanking the polyQ region (pAc‐clkL5: AATGGCTGGAAATGCGTGTC; pAc‐clkL3: GCTCCGAGTTCAGTGCAAAT; pAc‐clkR5: CAGCAACATCAGAGCCACTC; pAc‐clkR3: TGCTGCAGGATTTGTTGTTGT). Sequence assembly and analysis were performed in BioEdit, confirming that the original plasmid carried the polyQ33 allele (designated pAc‐polyQ33).

Allele sequences from DrosEU isofemale lines were used to generate additional Clk expression plasmids carrying polyQ31, polyQ28 and polyQ21 variants. Five μg of pAC‐dCLK plasmid DNA and the sequence information for each variant were submitted to GenScript for plasmid construction. The polyQ33 and polyQ31 alleles were selected for analysis due to their high frequencies in the DrosEU dataset, while polyQ28 and polyQ21 were included for their distinct clustering patterns in PCA and unique sequence features, including an internal histidine residue in the Q21 variant. All constructs were verified by Sanger sequencing using the primer sets described above.

2.6. S2 Cell Culture, Transfection and Luciferase Assay

Drosophila Schneider 2 (S2) cells, which lack expression of most endogenous clock genes except cyc, were used to evaluate the transcriptional activity of Clk polyQ variants. Cells were co‐transfected with pAc‐Clk constructs (polyQ33, polyQ31, polyQ28, or polyQ21) together with the pGL3‐4E‐hsp‐luc reporter plasmid and the Renilla luciferase plasmid for normalization. Cells transfected with the empty pAc5.1‐V5/His vector served as negative controls. Luciferase activity was measured 48 h after transfection.

Luciferase activity was measured to assess the transcriptional regulation of the Drosophila per E‐box by CLK variants containing different polyQ alleles. Measurements were performed using the Dual‐Luciferase Reporter Assay System (Promega) on a BioTek Synergy HTX plate reader with Gen5 software, following the manufacturer's protocol. Firefly luminescence driven by the pGL3‐4E‐hsp‐luc reporter was measured first, then quenched before measuring Renilla luminescence from the copia‐driven reporter (pCopia‐Renilla). Relative luciferase activity for each replicate was calculated as the ratio of firefly to Renilla luminescence.

3. Results

3.1. Spatial Analysis of Clk polyQ Allele Frequency Across Europe

Genomic data from 169 Drosophila melanogaster populations obtained from the DrosEU dataset (Kapun et al. 2020) were analysed. After excluding populations with low read coverage and those from extreme latitudes, longitudes, or altitudes, 127 populations remained for analysis. Retained populations spanned latitude 35.1° to 62.6° N, longitude −8.4° to 38.7° E and altitude 11 to 870 m. Within the Clk polyQ region, 11 distinct length variants were identified, ranging from Q21 to Q33. To confirm these variants at the sequence level, we extracted the polyQ tract directly from reads spanning the Clk repeat across all populations and reconstructed a consensus sequence for each length class (Figure S2). The eleven alleles form a nested length series differing only in the number of glutamine codons within the tract, with the longest allele (Q33) recovered from nearly all populations. Within the tract, glutamine is encoded by an interspersed mixture of CAG and CAA codons, and a histidine codon is present near the tract margin in most alleles. The two rarest variants, Q21 and Q32, were each supported by only a small number of reads.

The most common alleles were Q33 and Q27, which occurred in nearly all populations (127 and 122 populations, respectively). Alleles Q25 and Q31 were also frequent, appearing in over 100 populations each. The Q28 allele was moderately common (70 populations), while Q29 and the remaining shorter or longer variants (Q21, Q24, Q26, Q30 and Q32) were relatively rare, detected in fewer than 30 populations each.

PolyQ length for each read was determined by calculating the number of base pairs deleted relative to the longest allele, Q33, which consists of 99 base pairs and contains no indels. Structural inspection revealed variation in the positions of insertions and deletions within the polyQ region, indicating that indel events occurred independently across populations. In some alleles, deletions formed a single continuous gap, while in others they appeared as multiple smaller deletions with equivalent total lengths. The Q27 allele showed the greatest diversity of indel configurations, with ten distinct patterns identified. Because the goal of this study was to examine the effect of polyQ length variation rather than indel structure, subsequent analyses focused only on allele length.

To summarize variation in Clk polyQ allele composition across Europe, we first performed a principal component analysis (PCA) of allele frequencies for the five most common variants (Q33, Q31, Q28, Q27 and Q25). The first two components explained 58.1% of the total variance (Figure 2; Table S1). Populations were distributed continuously along the first two axes, indicating gradual changes in allele composition rather than discrete clusters. PC2 was driven mainly by variation in the frequency of Q28, suggesting that this allele contributes disproportionately to among‐population differentiation. These broad patterns motivated subsequent analyses testing the associations between allele frequencies and geographic and climatic variables.

FIGURE 2.

Principal component analysis (PCA) of Clk polyQ allele frequencies across European populations. Each black point represents a Drosophila melanogaster population, and blue arrows indicate the loadings of individual polyQ alleles. PC1 and PC2 together explain 58.1% of the total genetic variation in allele frequencies among populations. For clarity, one population with an extreme PC2 score (PC2 ≈ −7.8) falls below the plotted range; it was retained in all analyses.

3.2. Geographic and Environmental Correlates of Clk polyQ Allele Frequencies

To examine spatial patterns in Clk polyQ allele frequencies, we tested associations between allele frequencies and geographic variables (latitude, altitude, and longitude). To separate spatial from seasonal variation, this analysis used only the 101 spring‐collected populations. Because allele frequencies estimated from Pool‐seq data are overdispersed relative to a binomial expectation (residual deviance approximately 2.0–4.6 times the residual degrees of freedom), the models were fitted with a quasibinomial error, which estimates the dispersion from the data and scales the standard errors accordingly. Among the five most common alleles, the longitudinal associations were the most robust: Q25 and Q27 were significantly associated with longitude (quasibinomial p = 0.0013 and p = 0.0044, respectively), Q25 becoming more frequent and Q27 less frequent towards the east (Figure 3). Q28 showed a weaker latitudinal association (p = 0.036), decreasing towards higher latitudes (Figure S3). The remaining geographic and altitudinal effects, including those of Q31 and Q33, were not significant after accounting for overdispersion (all p > 0.12; Figure S4) and are not interpreted further. These findings indicate spatially structured genetic variation across Europe along a predominantly east–west axis, consistent with potential clinal differentiation.

FIGURE 3.

Correlations between polyQ alleles frequency with longitude across spring‐collected populations (n = 101). Points are per‐population allele frequencies; red lines are linear fits and p‐values are from quasibinomial GLMs accounting for overdispersion. Q25 (p = 0.0013) and Q27 (p = 0.0044) are significant, defining the east–west cline.

To interpret whether these spatial trends reflect underlying climatic differences, we summarized the main environmental gradients across Europe using principal component analysis of the bioclimatic dataset. The first two components (PC1 and PC2) together explained 73.3% of the total climatic variation (Figure S5). PC1 represented a temperature–precipitation gradient, with high positive loadings for temperature variables and negative loadings for precipitation, such that higher PC1 values correspond to warmer and drier environments. PC2 captured a gradient of continentality, with strong positive loadings for minimum temperature of the coldest month, annual precipitation, and a strong negative loading for temperature seasonality. High PC2 values describe regions with mild, wet winters and low seasonal temperature variation, while low PC2 values correspond to colder, more seasonal climates (Table S2).

Geographic variables were closely correlated with these climatic axes (Figure S6). As expected, PC1 decreased with latitude (t ₁₂₃ = −11.90, p < 2 × 10⁻¹⁶), confirming that southern populations experience warmer and drier conditions. More importantly, PC2 showed a pronounced negative association with longitude (t ₁₂₃ = −10.50, p < 2 × 10⁻¹⁶), revealing an east–west gradient in continentality. This longitudinal climatic gradient, which we previously highlighted as an underappreciated axis of environmental variation across Europe (Khatib et al. 2023; Kiat et al. ²⁰²¹), provides an important spatial context for the variation observed in Clk allele frequencies.

Applying the same quasibinomial correction to the climatic models (Figure 4; Table S3), the associations with PC2, the continentality axis, were the most robust. Q27 was positively associated with PC2 (OR = 1.228, quasibinomial p = 0.0004), indicating higher frequencies in milder, western environments, whereas Q25 was negatively associated with PC2 (OR = 0.787, p = 0.0040), indicating higher frequencies in colder, more seasonal eastern regions. The PC2 association of Q33 was not significant after correcting for overdispersion (OR = 0.916, p = 0.066) and is not emphasized.

FIGURE 4.

Spatial climatic gradients shape variation in C lk polyQ allele frequencies. PC1 captures a north–south gradient across Europe, while PC2 reflects an east–west gradient. Allele Q31 shows marginally significant associations with PC1, whereas Q25 and Q27 are significantly associated with PC2.

The PC1 associations of Q31 and Q28 (OR = 0.862 and OR = 1.239 in the uncorrected binomial fit) did not remain significant after correcting for overdispersion (quasibinomial p = 0.050 and p = 0.080, respectively) and are therefore not interpreted as robust climatic clines.

Together, these analyses indicate that Clk polyQ allele frequencies exhibit spatial structuring across Europe, reflecting geographic gradients and the continentality axis summarized by PC2. Although the effect sizes were modest, the consistent patterns across alleles suggest that temperature and seasonality may contribute to maintaining polymorphism at the Clk locus through spatially varying selection.

3.3. Analysis of Circadian Locomotor Activity

To evaluate whether variation in Clk polyQ length affects circadian temperature compensation, we measured locomotor activity rhythms in flies carrying different Clk polyQ alleles on a nearly isogenic background at 24°C and 18°C. After excluding individuals that died early, displayed arrhythmic activity, or showed extreme period values, 264 flies were included in the final analysis (12–36 individuals per allele per temperature; Figure 5). Detailed mean ± SD values and sample sizes for each allele and temperature are provided in Table S4. At 24°C, flies carrying the Q25, Q31 and Q30 alleles exhibited shorter mean free‐running periods (FRPs) than at 18°C, whereas Q21 showed little change and Q29 and Q33 slightly longer periods.

FIGURE 5.

Effects of Clk polyQ variation and temperature on circadian free‐running period (FRP). Boxplots show the distribution of FRP (hours) for flies carrying six Clk polyQ alleles (Q21, Q25, Q29, Q30, Q31, Q33) measured at 18°C (blue) and 24°C (red) under constant darkness. Each point represents an individual fly. Boxes indicate the interquartile range, horizontal lines mark medians, and whiskers extend to 1.5 × the interquartile range. The plot illustrates variation in circadian period across genotypes and temperatures. Detailed sample sizes and descriptive statistics are provided in Table S4.

A two‐way ANOVA was performed with temperature, allele, and their interaction as fixed factors. There was no significant main effect of temperature on FRP (F ₁,₂₅₂ = 2.42, p = 0.121), although a weak trend towards shorter FRP at higher temperature was observed (Figure 5). Neither the main effect of allele (F ₅,₂₅₂ = 0.50, p = 0.779) nor the interaction between allele and temperature (F _5,252 = 1.40, p = 0.224) was significant, indicating that FRP did not differ among polyQ variants and that temperature effects, where present, were consistent across alleles.

To assess temperature compensation at the level of individual genotypes, we next examined within allele changes in FRP using a Bayesian model (Table 1). For each allele, the magnitude of temperature dependent change in FRP (ΔFRP, 18°C–24°C) was summarized by the posterior mean and 95% credible interval (CI). Q25 showed a credible shortening of FRP at higher temperature (posterior mean ΔFRP = −0.68 h, 95% CI = −1.27 to −0.09, P(Δ < 0) = 0.99). Q31 exhibited a moderate decrease in FRP, although with greater uncertainty (ΔFRP = −0.52 h, 95% CI = −1.16 to 0.11, P(Δ < 0) = 0.95), with a similar but weaker pattern for Q30 (ΔFRP = −0.42 h, 95% CI = −1.18 to 0.34, P(Δ < 0) = 0.86). For the remaining alleles, credible intervals overlapped zero, indicating no detectable temperature dependent change in FRP. Pairwise comparisons supported a stronger temperature response in Q25 relative to most other alleles, although overlap among posterior distributions indicated that differences were modest overall (Table S5).

TABLE 1.

Bayesian model estimates for ΔFRP (18°C → 24°C) across Clk polyQ alleles. Posterior mean ΔFRP, 95% credible interval (CI), and posterior probability P(Δ < 0).

Allele	ΔFRP (h)	95% CI (h)	P(Δ < 0)
Q21	−0.14	[−0.71, 0.43]	0.68
Q25	−0.68	[−1.27, −0.09]	0.99
Q29	0.31	[−0.25, 0.87]	0.14
Q30	−0.42	[−1.18, 0.34]	0.86
Q31	−0.52	[−1.16, 0.11]	0.95
Q33	0.23	[−0.46, 0.92]	0.25

Overall, these results indicate that FRP was only weakly influenced by temperature across alleles. Among the variants examined, Q25 showed the strongest evidence for temperature sensitivity, consistent with a modest reduction in temperature compensation relative to the other alleles.

We next examined whether polyQ length affects the timing of locomotor activity under LD. The day‐folded average activity profiles are presented in Figure S7. Although phase differences are modest in magnitude, statistical analysis using seven pre‐specified phase references revealed significant differences (Table S6). Temperature had a strong and consistent effect on every phase reference (two‐way ANOVA, main effect of temperature p < 10⁻⁶ for all seven references; e.g., evening peak F _1,736 = 166.6, p = 1.6 × 10⁻³⁴). Cooling from 24°C to 18°C advanced the evening peak (from a genotype range of ZT 11.3 to 12.0 down to ZT 10.9 to 11.1) and delayed the morning peak (from ZT 0.4 to 1.1 up to ZT 1.2 to 1.5), so that the two activity components moved closer together and the morning‐to‐evening interval contracted (from ZT 10.3 to 11.6 at 24°C to ZT 9.5 to 9.8 at 18°C). These shifts were common to all genotypes.

Against this background, polyQ genotype affected activity phase specifically at the warmer temperature. At 24°C the genotypes differed significantly in all three evening references and in the morning‐to‐evening interval (one‐way ANOVA: evening onset F _5,346 = 7.25, p = 1.8 × 10⁻⁶; evening peak F _5,346 = 9.48, p = 1.7 × 10⁻⁸; evening offset F _5,346 = 4.33, p = 7.7 × 10⁻⁴; morning‐to‐evening interval F _5,346 = 8.87, p = 6.1 × 10⁻⁸; morning peak F _5,346 = 5.98, p = 2.5 × 10⁻⁵; all significant after Benjamini‐Hochberg correction across the seven references; Table S6). Post hoc comparisons (Tukey HSD) resolved two consistent groups: Q29 and Q33, together with Q25, showed the latest evening peaks and the longest morning‐to‐evening intervals, whereas Q21 and Q30 showed the earliest evening peaks and the shortest intervals, with Q31 intermediate (Figure 6; Table S6). The effects were modest in magnitude, with genotype explaining approximately 10 to 12% of the among‐fly variance in the strongest references (evening peak and interval) and with substantial overlap among individuals; consequently, these differences were not apparent in the population‐average waveforms and emerged only from the per‐fly analysis. At 18°C, by contrast, no phase reference differed among genotypes after correction, and the genotype means converged onto a common phenotype (Figure 6; Table S6).

FIGURE 6.

Clk polyQ genotype shapes evening phase timing at 24°C but not at 18°C. Evening onset (A), evening peak (B), and the morning to evening interval (C), expressed in Zeitgeber time (ZT, hours), for six Clk polyQ genotypes (Q21, Q25, Q29, Q30, Q31, Q33) recorded under 12:12 LD at 18°C (blue) and 24°C (orange). Faint points are individual flies; filled symbols show genotype means with 95% confidence intervals (n = 352 flies at 24°C, 396 at 18°C). At 24°C, genotype means sharing a letter do not differ (Tukey HSD, p < 0.05); at 18°C the genotypes are statistically indistinguishable (n.s.). The divergence of genotypes at 24°C together with their convergence at 18°C reflects a significant genotype by temperature interaction for all three references (Freedman‐Lane permutation, p < 0.05).

This temperature dependence was confirmed by a significant genotype by temperature interaction, tested with a distribution‐free Freedman‐Lane permutation procedure. The interaction was significant for the evening references and the morning‐to‐evening interval (evening peak and interval p = 0.002, evening onset p = 0.006, evening offset p = 0.037, Benjamini‐Hochberg corrected) and marginal for the morning peak (p = 0.050), but absent for the morning onset and offset (Table S6). Variation in Clk polyQ length therefore has little effect on the phase of activity under cooler conditions but produces reproducible differences in the timing of the evening component under warmer conditions, primarily by advancing or delaying the evening peak and thereby narrowing or widening the interval between the morning and evening activity bouts.

Overall, these analyses indicate that variation in Clk polyQ length does not strongly alter temperature compensation of circadian period, as FRP remained largely stable across alleles. In contrast, circadian phase exhibited greater sensitivity to temperature, with pronounced allele‐specific differences emerging at elevated temperature. Among the variants examined, Q25 showed the clearest reduction in temperature compensation of period, whereas the temperature‐dependent divergence in phase was greatest for Q29 and Q33, which displayed the latest evening activity peaks at 24°C.

3.4. Effects of Clk polyQ Variation on Transcriptional Activity

Luciferase reporter assays were used to assess how variation in Clk polyQ length influences CLK transcriptional activity. Control transfections containing the empty expression vector showed very low relative luciferase activity, approximately 95‐fold lower than that observed for the least active CLK variant. The analysis revealed a significant overall effect of polyQ length on reporter expression (Kruskal–Wallis test: χ² = 14.58, df = 3, p = 0.002; Figure 7). Pairwise comparisons indicated that the longer alleles, polyQ31 and polyQ33, exhibited significantly higher transcriptional activity than polyQ21 (p = 0.015 for both) and polyQ28 (p = 0.015 and p = 0.023, respectively). No significant differences were found between polyQ31 and polyQ33 or between polyQ21 and polyQ28. These results show that CLK variants with longer polyQ tracts promote stronger activation of the per E‐box reporter, suggesting that natural variation in polyQ length modulates CLK's transcriptional potency.

FIGURE 7.

Variation in CLK polyQ length influences transcriptional activity in a luciferase reporter assay. Luciferase activity was measured for different CLK polyQ alleles (Q21, Q28, Q31, Q33) and the pAc control. Each point represents an independent biological replicate (eight for each allele, three for the control). PolyQ31 and PolyQ33 showed significantly higher activity than PolyQ21 and PolyQ28 (p < 0.05, Wilcoxon test with Benjamini–Hochberg correction), while no significant differences were detected between PolyQ31 and PolyQ33 or between PolyQ21 and PolyQ28.

Overall, these results indicate that CLK variants with longer polyQ tracts exhibit enhanced transcriptional activation of the per E‐box reporter, consistent with length‐dependent modulation of CLK activity.

4. Discussion

Natural variation in circadian clock genes provides a valuable framework for understanding how organisms adapt to environmental heterogeneity. In Drosophila melanogaster , polymorphisms in the Clock (Clk) gene are of particular interest since CLK plays a central role in the transcriptional feedback network that sustains circadian oscillations. This study reveals that variation in the length of the polyQ tract within Clk contributes to both the behavioural and molecular tuning of the circadian system. Using European population‐genomic data, behavioural assays in near‐isogenic lines, and luciferase reporter analyses, we found that Clk polyQ alleles differ in frequency across climatic gradients and exhibit functional variation under laboratory conditions. These findings reveal potential mechanisms through which environmental variation could influence allele maintenance in natural populations.

Patterns of Clk allele distribution across Europe were structured along geographic axes. The most robust clines were longitudinal: Q25 was more frequent in eastern, more seasonal regions and Q27 in milder, western regions, whereas the weaker latitudinal and altitudinal trends of the longer alleles Q31 and Q33 did not remain significant after accounting for overdispersion. Although longitude is not itself a selective agent, in Europe it corresponds to a gradient of continentality, from the mild oceanic west to the colder, more seasonal east, captured here by PC2 (Table S2). We have previously documented longitudinal clines in circadian clock genes along this frequently overlooked axis (Khatib et al. 2023), so a longitudinal cline in Clk polyQ frequency is consistent with a real environmental gradient, though it could also partly reflect colonization history. The longitudinal cline in Q25 and Q27 aligns with this environmental axis; however, such spatial patterns can arise from colonization history, demographic processes, or selection, and our data do not permit distinguishing among these mechanisms. More broadly, clinal variation in clock genes has been reported across taxa, including the per Thr‐Gly polymorphism (Costa et al. 1992) and Clk polyQ variation in salmon (O'Malley and Banks ²⁰⁰⁸) and birds (Johnsen et al. 2007). Although the associations between allele frequencies and climatic or geographic gradients were generally weak to moderate, their consistency across datasets and correspondence with environmental axes warrant further investigation into potential selective mechanisms, possibly linked to the thermal and photoperiodic stability of local environments.

The polyQ polymorphism in Drosophila melanogaster was previously examined in Australian populations, where Weeks et al. (2006) reported no evidence of a cline in Clk repeat length along the east coast. In the same study, they also failed to detect a latitudinal pattern in the period (per) Thr‐Gly repeat, a result that contrasted with the strong European cline originally described by Costa et al. (1992). This absence of a pattern in Australia prompted debate about whether clinal variation in clock genes truly reflects adaptive differentiation or arises from sampling and methodological artefacts. However, subsequent analyses by Kyriacou et al. (2007) and Sawyer et al. (2006) re‐examined Australian and African data and identified a significant, though weaker, cline in per that was consistent in direction with the European gradient, supporting the thermal‐adaptation hypothesis. These authors argued that the apparent lack of a cline in Weeks et al. (2006) likely resulted from limited allele resolution, potential species misidentification, and the relatively recent colonization history of D. melanogaster in Australia, rather than the absence of selection. This historical debate over per underscores how demographic history and sampling scale can obscure adaptive spatial patterns, a consideration directly relevant to interpreting the geographic structuring of Clk polyQ variation in our European dataset.

Behavioural analyses of near‐isogenic lines showed that most Clk polyQ variants maintained a largely stable free‐running period (FRP) across 18°C and 24°C, indicating effective temperature compensation. However, flies carrying the Q25 allele displayed a consistent shortening of FRP at higher temperature, suggesting reduced compensation. This pattern is analogous to that observed for per Thr‐Gly polymorphisms, where specific repeat lengths altered thermal robustness (Sawyer et al. 1997). Interestingly, both the shortest (Q21) and longest (Q33) alleles exhibited strong compensation, while intermediate alleles, particularly Q25, were more temperature‐sensitive, revealing a non‐linear relationship between polyQ length and circadian stability. In contrast to these allele‐specific effects on circadian traits under laboratory conditions, circadian phase did not show a consistent relationship with geographic variables, suggesting that selection on Clk polyQ length is unlikely to act directly on phase timing alone.

At the molecular level, luciferase reporter assays demonstrated that CLK transcriptional activity also depends on polyQ length. Longer variants (Q31, Q33) exhibited higher activation of a per E‐box reporter compared with shorter alleles (Q21, Q28), suggesting that polyQ extension enhances CLK's ability to promote target gene expression. Similar effects of polyQ length on transcriptional activation have been reported in other regulatory proteins (Atanesyan et al. 2012; Gemayel et al. ²⁰¹⁵). The correspondence between transcriptional strength and aspects of behavioural robustness suggests that differences in CLK activity, specifically the higher transactivation of longer polyQ variants, could contribute to circadian function. Testing CLK transactivation under varying thermal conditions would be necessary to directly assess whether transcriptional differences mediate temperature compensation. The transcriptional constructs (Q21, Q28, Q31 and Q33) were selected a priori from allele frequency and PCA clustering, before the behavioural assays identified Q25 as the temperature‐sensitive allele; a Q25 construct is therefore a priority for follow‐up work. Notably, although the Q21 variant carries an internal histidine that interrupts its glutamine tract, its transcriptional activity was comparable to that of the other short allele, Q28, consistent with the length of the uninterrupted glutamine stretch, rather than the total repeat region, being the primary determinant of transcriptional activity.

With respect to circadian phase specifically, the transcriptional and behavioural phenotypes are best viewed as complementary rather than directly predictive. Among the alleles common to both assays, reporter activity and evening phase did not co‐vary monotonically, indicating that steady state transactivation alone does not set phase. A plausible link is that differences in CLK transactivation alter the kinetics of the PER and TIM feedback loop, and thus the timing of the evening peak, so that reporter magnitude and phase shift need not be tightly coupled; time resolved assays capturing oscillation phase will be needed to test this.

The physical basis of these functional effects likely involves the intrinsic structural properties of polyQ domains. PolyQ sequences facilitate multivalent, weak interactions that promote the formation of dynamic, membraneless condensates through liquid–liquid phase separation (Banani et al. 2017; Peng et al. ²⁰²⁰; Kawasaki and Fukaya ²⁰²³). In the Arabidopsis clock protein ELF3, phase separation of the prion‐like domain is driven by rising temperature, so that warming, rather than cooling, promotes condensate formation, and lengthening the polyQ tract broadens the range of temperature and pH over which the condensed phase forms and persists while having a comparatively small effect on the transition temperature itself (Hutin et al. 2023). By analogy, longer Clk polyQ alleles may promote and stabilize CLK condensates across a broader range of thermal conditions and thereby support stronger transactivation, consistent with the higher reporter activity we observed for Q31 and Q33, whereas shorter tracts may sustain condensates over a narrower range. This interpretation is consistent with biophysical studies showing that shorter polyQ tracts undergo greater temperature‐induced conformational changes (Cui et al. 2014).

This mechanism appears to be a broadly conserved feature of circadian clock architecture: a recent survey of intrinsically disordered regions across fungal, insect, and mammalian core clock proteins highlights how such disordered, low‐complexity domains, including polyQ and prion‐like tracts, function as tunable, temperature‐sensitive interaction modules that support condensate assembly and modulate clock protein activity (Usher and Pelham 2026).

The unexpected abundance of the temperature‐sensitive Q25 allele in eastern populations may reflect additional or indirect selective pressures. One possibility is that greater structural flexibility of Q25 confers advantages for other temperature‐linked traits such as diapause or cold response, both regulated by circadian pathways (Hidalgo and Chiu 2024; Goto et al. ²⁰⁰⁶; Cai et al. ²⁰²⁴). Alternatively, heterozygosity for Q25 and longer alleles like Q33 could produce composite phenotypes with broader environmental tolerance, maintaining diversity through balancing selection. Similar maintenance of Clk polyQ diversity has been reported in migratory birds exposed to variable photoperiods (Justen et al. 2022).

Beyond repeat length, additional molecular features likely shape CLK function. The polyQ region can be encoded by mixed CAG and CAA codons, whose differing translation speeds influence protein folding and stability (Fu et al. 2016; Moreno‐Rodríguez et al. ²⁰²⁵). Moreover, the structure of neighbouring domains modulates the conformational flexibility and accessibility of the polyQ tract (Mier and Andrade‐Navarro ²⁰²¹). Such interactions may explain why alleles at opposite ends of the length range, such as Q21 and Q33, exhibit comparable temperature robustness, whereas Q25 shows distinct behaviour. Detailed analyses of codon composition, adjacent sequence variation, and post‐translational modifications will be essential to fully understand how natural Clk variants influence clock performance.

In summary, our results demonstrate that Clk polyQ length polymorphism affects both the molecular activity of CLK and the thermal stability of circadian behaviour in Drosophila melanogaster . The geographic distribution of alleles, particularly the pronounced longitudinal cline in Q25 and Q27, is structured in patterns consistent with underlying environmental variation, though demographic and selective mechanisms cannot be distinguished with current data. The non‐linear relationship between polyQ length and function highlights the adaptive potential of low‐complexity regions, which can modulate protein dynamics and environmental responsiveness without major changes in overall structure. These findings contribute to a broader understanding of how subtle molecular variation in key regulatory proteins underpins the evolution of environmental adaptation in circadian systems.

Author Contributions

M.Y. conducted the experiments, analysed the data and wrote the first draft of the manuscript. B.F. assisted with the experiments and supervised the experimental work. M.K. assembled and processed the sequencing data. E.T. designed the research and contributed to data analysis. All authors revised the manuscript and approved the final version.

Funding

M.Y. was supported by a graduate scholarship from the University of Haifa.

Disclosure

The authors have nothing to report.

Ethics Statement

All experiments in this study were performed on the invertebrate Drosophila melanogaster , an organism that is not subject to the ethical regulations governing vertebrate animal research. No ethical approval or licence was therefore required for this work.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Figure S1: Crossing scheme for generating the polyQ NILs. The 3rd chromosome consists of WT 3 L (grey), TM6C balancer (red), and the chromosomes with the deficiency (Df(3L)pbl‐X1; black). An initial cross between deficiency‐ carrying males, and WT females from each polyQ line was conducted. Hemizygote virgin females with WT bristles (no balancer) were collected from F1 offspring and backcrossed with w [1118]; Df(3L)pbl‐X1/TM6C males to generate BC1. The following BC2‐BC4 were generated by backcrossing white‐eyed hemizygote virgin females offsprings of the previous backcross with the deficiency‐carrying males three more times. In each step, the contribution of the donor parent genome is reduced by half, so that BC4 flies have 96% genomic similarity. From each BC4 population, hemizygote white‐eyed and WT bristled females and males were collected. Hemizygous males were tested for their circadian activity while the females were crossed to generate the next BC.

Figure S2: Sequence variation among the eleven Clk polyQ length alleles in European Drosophila melanogaster . Coding‐strand codon alignment (N to C terminus) of the eleven length variants (Q21 to Q33) recovered from DrosEU Pool‐seq data. Each cell represents one codon, coloured by identity: CAG (teal) and CAA (amber) both encode glutamine, showing the synonymous codon interspersion within the tract; CAT (red) encodes a histidine near the tract margin, present in nine alleles and replaced by glutamine in Q24 and Q27; flanking codons are shown in grey. Length differences relative to the longest allele (Q33) are drawn as a single nested deletion block; because deletions cannot be localized within a tandem repeat, gap placement within the tract is arbitrary. The number of spanning reads supporting each allele is given at left; Q21 and Q32 (n = 4) are the least supported.

Figure S3: Correlations between Clk polyQ allele frequencies and latitude across spring‐collected populations (n = 101). Points are per‐population allele frequencies; red lines are linear fits and p‐values are from quasibinomial GLMs accounting for overdispersion. Only Q28 is significant (p = 0.036).

Figure S4: Correlations between polyQ alleles frequency with altitude across spring‐collected populations (n = 101). Points are per‐population allele frequencies; red lines are linear fits and p‐values are from quasibinomial GLMs accounting for overdispersion. No allele is significant after correction.

Figure S5: PCA analysis of 19 Bioclimatic variables measured across Europe, corresponds to locations of DrosEU drosophila populations. The loading of each of the variables is shown by black arrow. PC1 and PC2 explain 73.3% of the variation across the population.

Figure S6: Principal components (PC1 and PC2) of 19 bioclimatic variables and their correlations with latitude, longitude and altitude. PC1, strongly correlated with latitude, represents a north–south gradient in temperature and precipitation, with high values (south) indicating warmer, drier climates and low values (north) cooler, wetter ones. PC2, correlated with longitude, reflects an east–west gradient from mild, humid western regions to colder, more seasonal eastern environments.

Figure S7: Day‐folded average locomotor activity profiles of Clk polyQ alleles at 18°C and 24°C. Each panel shows the mean daily activity profile for flies carrying one of six Clk polyQ alleles (Q21, Q25, Q29, Q30, Q31, Q33) recorded under a 12:12 light–dark cycle at 24°C (red) and 18°C (blue). Separate cohorts of flies were assayed at each temperature. Activity (beam crossings per 30‐min bin) was averaged across recording days within each fly and then across flies within each allele and temperature; shaded envelopes show the standard error of the mean across flies. The x‐axis is Zeitgeber time (ZT), with ZT0 corresponding to lights‐on and the grey region marking the dark phase (ZT12 to ZT24). Profiles were folded over days 1 to 4 at 24°C and days 1 to 5 at 18°C; the number of flies contributing to each profile is given in the panel headers.

Table S1: Loadings of polyQ alleles on the first five principal components (PC1–PC5) derived from allele frequency data.

Table S2: Loadings of 19 Bioclimatic variables measured across Europe on the first two principal components derived from WorldClim.

Table S3: Quasibinomial GLM analyses (dispersion‐corrected) testing associations between allele frequencies and climatic principal components. Estimates, odds ratios, standard errors and z‐values are from the binomial fit; quasibinomial p ‐values are scaled by the estimated dispersion (phi). Effect direction is given only for associations significant under the quasibinomial model (p < 0.05).

Table S4: Sample size (n), mean FRP ± SD (h) for each allele group tested at 18°C and 24°C.

Table S5: Pairwise comparisons of ΔFRP between alleles results.

Table S6: Effect of Clk polyQ genotype on circadian activity phase, and its interaction with temperature.

Data Availability Statement

All data, analysis scripts (R and RMarkdown), and intermediate datasets are archived at Zenodo (https://doi.org/10.5281/zenodo.22030098).