Nuclear Export of Drosophila PERIOD contributes to temperature compensation of the circadian clock

Circadian clocks are self-sustained molecular oscillators controlling daily changes of behavioral activity and physiology. For functional reliability and precision the frequency of these molecular oscillations must be stable at different environmental temperatures, known as ‘temperature compensation’. Despite being an intrinsic property of all circadian clocks, this phenomenon is not well understood at the molecular level. Here we use behavioral and molecular approaches to characterize a novel mutation in the period (per) clock gene of Drosophila melanogaster, which alters a predicted nuclear export sequence (NES) of the PER protein. We show that this new perI530A allele leads to progressively longer behavioral periods and clock oscillations with increasing temperature in both clock neurons and peripheral clock cells. While the mutant PERI530A protein shows normal circadian fluctuations and post-translational modifications at cool temperatures, increasing temperatures lead to both, severe amplitude dampening and hypophosphorylation of PERI530A. We further show that PERI530A displays reduced repressor activity at warmer temperatures, presumably because it cannot inactivate the transcription factor CLOCK (CLK). With increasing temperatures nuclear accumulation of PERI530A within clock neurons is increased, suggesting that PER is normally exported out of the nucleus at warm temperatures. Consequently, downregulating the nuclear export factor CRM1 also leads to temperature-dependent changes of behavioral rhythms. In summary, our results suggest that the PER NES and the nuclear export of clock proteins play an important role in temperature compensation of the Drosophila circadian clock.


Introduction:
Circadian clocks allow organisms to anticipate the daily changes of their environment, such as daily fluctuations of light intensity and temperature. They enable animals to restrict their behavioral and physiological activities to species-specific advantageous times of day and night, increasing their overall fitness. Circadian clocks are composed of negative molecular feedback loops, within which clock gene products oscillate in abundance and subcellular localization with a 24-h period (1). In Drosophila melanogaster, expression of the clock genes per and timeless (tim) is mediated by the transcription factors CLK and CYCLE (CYC). PER and TIM proteins accumulate in the cytoplasm before they translocate to the nucleus, bind to the CLK/CYC dimer and repress their own transcription. Eventually, PER and TIM proteins are degraded, allowing CLK and CYC to start a new round of the cycle (1). Posttranslational modifications regulate this process to maintain the period length of the molecular oscillations at ~ 24 h. For example, cytoplasmic PER and TIM phosphorylation by the kinases DBT, CK2 and SGG determines timed nuclear translocation (2). In addition, hyperphosphorylated nuclear PER coincides with hyperphosphorylated, inactive CLK, and the DOUBLETIME (DBT) kinase mediates phosphorylation of both proteins within the same complex leading to transcriptional repression (3,4). In the so called PER 'phosphotimer' the NLK kinase NEMO initially phosphorylates PER S596, which stimulates phosphorylation of S589, S585 and T583 by the CK1ε kinase DOUBLETIME (DBT). All these residues belong to the per Short phosphocluster, and their phospho-occupancy somehow delays DBT-dependent PER phosphorylation at other residues, most importantly PER S47 (5). Since phosphorylation at PER S47 enables binding of the F-box protein SLIMB and subsequent degradation of PER (6), these consecutive phosphorylation events regulate temporal PER stability and thereby period length of the circadian clock (5). Based on the involvement kinases and phosphatases and other post-translational modifications in clock regulation it could be expected that increasing temperature leads to faster enzyme kinetics and therefore shorter periods of molecular oscillations and vice versa. However, circadian clocks are temperature compensated, meaning periods of circadian clock oscillations do not change across a wide range of temperatures (e.g., between ~15°C and 29°C in fruit flies) (7). Although temperature compensation is a hallmark of all circadian clocks (8,9), its molecular etiology is still not well understood. Circadian clocks can be synchronized by temperature cycles, which suggests that they are temperatureresponsive with compensatory mechanisms (10). This compensation is traditionally explained with the assumption that temperature-mediated changes of different rates across the circadian clock cancel each other out (11). This model is less likely, given that any mutation in circadian genes would yield a temperature-sensitive circadian clock.
Temperature compensation could also be achieved by considering the circadian clock as two coupled oscillators with complementary temperature dependence of period length (10), or as one oscillator composed of biochemical reactions with opposite temperature coefficients (12).
This would require different clocks within one organism to have distinct regulatory mechanisms (13). The circadian clock of Drosophila consists of multiple coupled oscillators that vary in their intrinsic period length (e.g. (14)), but it is not known how they behave at different temperatures. Subsets of these oscillators, such as the so-called small lateral neurons (s-LNv), sustain free-running behavioral rhythms in constants darkness (DD) (15), while others such as the Dorsal Neurons 1 -3 (DN1, DN2, and DN3) are unable, despite sustaining molecular rhythms in DD (16,17). It is therefore conceivable that the different clock neuronal subsets also differ in their ability to maintain temperature-compensated molecular oscillations.
Another possible mechanism for temperature compensation, is that each step in the clock is individually temperature compensated (18). For example, timed nuclear accumulation and timed nuclear degradation of PER/TIM would each be temperature compensated. Some alleles of per and tim that are defective in nuclear accululation of the PER/TIM complex also affect temperature compensation. per L flies (19) exhibit an increase of behavioral period by four hours from 18°C (27 h) to 29°C (31 h), which correlates well with the ~4-hour delay of PER L nuclear entry (20,21). This delay can be explained by the weaker interaction of PER L with TIM with increasing temperatures, because TIM stabilizes PER and PER L therefore accumulates even slower at higher temperatures (22).
Similarly, the tim blind allele increases behavioral period by four hours in a temperaturedependent manner from 18°C (24 h) to 29°C (28 h) (23). This, and several other new tim alleles map to nuclear export signals (NESs) throughout TIM and also show similar temperature compensation defects (23). Indeed, TIM and PER are subject to active Cargo Protein and Chromosome Maintenance 1 (CRM1)-mediated nuclear export in Drosophila larvae (24). With TIM export, this mechanism relies on CKII-dependent phosphorylation of residue S1404 near a putative NES, which reduces the interaction between TIM and CRM1, and retains normal nuclear levels of TIM and regulation of CLK activity (3). In mammals, mPER1, mPER2, and mPER3 proteins also contain canonical leucine-rich NES motifs, which in cell culture shuttle PER proteins from the nucleus to the cytoplasm in order to target them for degradation (25,26), but their roles in temperature compensation have not been examined.
While it seems clear from these studies that nuclear export of clock proteins plays a role in circadian clock function in both mammals and fruit flies, its role in temperature compensation has not been examined. Here, we generated a point mutation (per I530A ) in the single leucinerich putative PER NES, which is highly conserved to the mPER NES ( Figure 1A) (25,27,28) to systematically study the role of the PER NES in the context of temperature compensation.
per I530A mutant flies show substantial temperature-dependent increase of their behavioral period, which is correlates well with longer molecular oscillations in several subsets of clock neurons and in peripheral clock cells. PER I530A shows temperature-dependent alterations of posttranscriptional modifications, accompanied by reduced ability to function as a transcriptional repressor of CLK. Finally, enhanced nuclear PER I530A accumulation with increasing temperatures, as well as temperature-dependent period lengthening of flies with reduced CRM1 expression strongly implicate nuclear export of PER as an important mechanism for temperature compensation in Drosophila. The knockdown of CRM1 weakens the observed temperature-compensation defect of per I530A , suggesting that nuclear export of other clock proteins may also be affected, and implicating nuclear export as a key mechanism in establishing temperature compensation.

Results:
The per I530A mutant causes circadian period lengthening with increasing temperatures To analyze the function of the nuclear export signal (NES) in the Drosophila Period (PER) protein, we introduced a single amino acid change into the NES by replacing the conserved isoleucine at position 530 with alanine (PER I530A ). To do this, the endogenous per gene was replaced with the mutated per I530A allele by CRISPR/Cas9 mediated homologous recombination ( Figure 1A and Methods). To aid detection of wild type and mutated PER I530A we also generated flies where both proteins are endogenously tagged with the V5 epitope (per V5 and per I530A-V5 respectively, Figure 1A and Methods). To investigate if the per I530A allele influences circadian clock function and behavioral rhythms, we analyzed locomotor activity in constant darkness at 25°C (DD25). While y w control flies showed the typical robust activity rhythms with period values (τ) of 23.5 hr, τ values of the per I530A flies were > 2hr longer (25.9 hr) ( Figure 1B Figure 1B, C, Table 1). The results unequivocally show that specific mutation of the PER NES leads to a temperature-dependent lengthening of τ, pointing to temperature over-compensation of the circadian clock.
The per I530A mutation leads to temperature dependent molecular period lengthening within central clock neurons and peripheral clock cells Behavioral rhythms in constant darkness are driven by molecular oscillations of clock gene expression within circadian pacemaker neurons in the fly brain. It is therefore expected that temperature-dependent period-lengthening also occurs in clock neurons of per I530A mutant flies. To test this, and to see if temperature compensation is affected equally in all clock neurons, we applied a novel, cell-type restricted real-time luciferase assay (Locally Activatable Bioluminescence or LABL). Central to LABL is a period-luciferase transgene, in which luciferase expression is compromised by a mCherry reporter gene followed by stop codons. This cassette is inserted between the period promoter and the luciferase coding sequences, and flanked by FRT recombination sites ( Figure S1A). period-driven luciferase expression can therefore be induced and restricted to subsets of per expressing cells using Gal4 driven expression of UAS-FLP ( Figure S1A). Details of this technique are described in an accompanying paper (Johnstone et al 2021, this issue). Using LABL we first compared period-luciferase expression in different sets of clock neurons using the Pdf-gal4 (PDF positive s-LNv and l-LNv), DvPdf-gal4 (PDF positive s-LNv and l-LNv, 5 th s-LNv, 4 LNd) and R18H11-gal4 (subset of the DN1p) at three different temperatures in wild type flies. Bioluminescence of 15 male flies per genotype housed in a customized arena was measured using the LumiCycle 32 Luminometer (Actymterics) for 9 days in constant darkness at either 18°C, 25°C, or 29°C (see Materiala and Methods and Johnstone et al 2021, this issue, for details). In wild type flies, all driver lines generated robust bioluminescence oscillations with stable period lengths close to 24 hr at all temperatures ( Figure 2A, C, S1B), indicating that all subsets of clock neurons are temperature compensated (cf. (29)). In contrast, the same analysis in the background of the per I530A mutation revealed a significant linear period-lengthening from 18°C to 29°C for Pdf and R18H11-gal4 driver lines and between 18°C and 25°C for DvPdf-gal4 (Figure 2A, C, S1B) (for unknown reasons the genotype involving per I530A and DvPdf-gal4 died at 29°C). At the lower temperature, bioluminescence oscillations showed a period close to 24 hr, while at 25°C it was lengthened from 26-28 hr, and further to 30-32 hr at 29°C, depending on the Gal4 driver used to activate LABL ( Figure 2D, S1B). Therefore, the period length of the molecular oscillations and behavioral rhythms are equally affected by the per I530A mutation, showing normal values at 18°C and temperature-dependent period lengthening. As a control, we also analyzed period-luciferase expression in per L flies, which have a period of 27.1 hr at 18°C, increasing to 29.6 hr at 25°C and 31.4 hr at 29°C (20). Again, using Pdf-gal4 to drive LABL, the molecular oscillations closely matched the behavioral ones, with periods of 27-28 hr at 18°C lengthening to 29-30 hr at 25 °C, and 32 hr at 29°C ( Figure S1C). We conclude that LABL is suitable for reliable period estimation within clock neurons at different temperatures (see also accompanying paper by Johnstone et al).
To determine whether temperature-dependent defects of the per I530A allele also extend to peripheral clocks, we next used LABL to measure period-luciferase expression in all clock cells.
For this, we applied the tim-UAS-Gal4 driver expressed in all clock cells (30)  as previously shown for period-luciferase transgenes expressed in all clock cells (31). As expected, the bioluminescence levels emanating from tim-UAS-Gal4 LABL flies were drastically (40-50 fold) increased compared to the three clock neuronal drivers, reflecting expression in peripheral clock cells ( Figure 2B). Similar to the results obtained for the clock neurons, oscillations in peripheral clock cells in wild type flies were temperaturecompensated, with stable periods close to 24 hr at all three temperatures ( Figure 2B, C, S1B).
Moreover, when analyzed in a per I530A and per L mutant background, tim-UAS-Gal4 driven period-luciferase oscillations showed the same temperature-dependent period-lengthening as observed with the neuronal drivers ( Figure 2B, C, S1B, C). Therefore, the per I530A mutation has a similar effect on both central and peripheral clocks, indicating a common mechanism of temperature compensation.
The per I530A mutation affects PER and TIM stability and post-translational modification at higher temperatures during LD The temperature-dependent period-lengthening of per I530A flies could be due to decreased PER I530A stability at higher temperatures. To test this, we performed Western blots of total head protein extracts of control (per V5 ) and per I530A-V5 flies raised at different temperatures and collected every six hours during one day in LD (ZT2, ZT8, ZT14, ZT20). To see how wild type PER levels behave at different temperatures throughout the day we first analyzed per V5 flies at 18°C, 25°C and 29°C ( Figure 3). As expected, we observed robust PER oscillations at all three temperatures, with trough levels at ZT8 and peak expression from ZT20 to ZT2 ( Figure 3A) (32). At 18°C PER levels appear somewhat reduced from ZT2 to ZT14 compared to the higher temperatures, but the differences were not significant ( Figure 3A). We also did not detect obvious differences in the migration speed of PER, which is indicative of its phosphorylation status (33), except that at ZT2 during 29°C PER migrates faster compared to the cooler temperatures and therefore appears hypophosphorylated. These results show that wild type PER levels as well as rhythmic changes in abundance and phosphorylation are well compensated against changes of ambient temperature.
Next we compared PER and PER I530A expression at different temperatures. At 18°C no obvious differences in protein amounts or mobility could be observed during the four time points we analyzed ( Figure 3B, top panel). In contrast, at 25°C an amplitude reduction of PER I530A cycling due to increased trough levels at ZT8 was apparent ( Figure 3B, middle panel). In addition, there was a striking effect on mobility at ZT2 and ZT8. At these time point wild type PER appears mainly in its slow migrating form, which is not observable for PER I530A . Instead the faster migrating form is still present at ZT8, indicating higher stability of PER I530A due to incomplete post-translational modifications and degradation. This could ultimately result in the lower-amplitude protein cycling we observed ( Figure 3B, middle panel). Strikingly, at 29°C PER I530A oscillations were completely abolished and presumably hypophosphorylated protein was present at peak levels at all time points analyzed ( Figure 3B, lower panel). To summarize, in parallel to the behavioral and luciferase observations, PER I530A protein behaves normal at 18°C, but both, protein cycling and the phosphorylation pattern become increasingly aberrant with rising temperature. Our results suggest that a temperature-dependent phosphorylation defect underlies the increased stability and reduced oscillation of the PER I530A protein at higher temperatures.
Next, we examined if the per I530A mutation had similar effects on the clock protein TIM. First, we again checked if TIM expression varied in wild type in a temperature-dependent manner.
We did not detect any differences in levels or migration properties, except for the expected detection of the shorter, cold-induced TIM-SC protein at 18°C ( Figure 3A, lower panel) (34,35).
As expected from the results with PER, wild type TIM levels and phosphorylation patterns are therefore well compensated against temperature changes. Comparison of TIM between wild type and per I530A mutants revealed only mild differences at the warmer temperatures, while at 18°C they were essentially identical at all time points ( Figure S2). At 25°C TIM levels in per I530A were slightly increased at ZT2 and ZT8, and this effect was further pronounced at 29°C (Extended Data Figure 3). Nevertheless, even though the amplitude is reduced, TIM levels still oscillated. We were not able to detect any differences in TIM migration between wild type and per I530A , but phosphorylation-dependent migration differences are generally not as pronounced for TIM compared to PER (compare Figure 2A, B with S2).
The per I530A mutation affects PER and TIM stability and post-translational modification at higher temperatures during DD Because the temperature-dependent effects on behavioral period length are necessarily determined in constant darkness, we wanted to analyze the effects of the per I530A allele on PER and TIM expression also in DD. PER and TIM isolated from total fly heads mainly reflect expression in the photoreceptor cells, which contain peripheral clocks that dampen out rapidly in DD (16). We therefore analyzed PER and TIM levels during the first day in DD after LD entrainment from both wild type and per I530A flies at different temperatures. At 18°C, both PER and TIM cycled with similar amplitude in mutants and controls, although both proteins appeared more stable in per I530A ( Figure S3). At 25°C and 29°C weak PER and TIM oscillations were observed in wild-type, whereas in the per I530A mutants both proteins were at high constitutive levels ( Figure S3). Similar as in LD, PER I530A protein showed normal daily migration changes at 18°C, while at 25°C and 29°C the slow migrating, hyperphosphorylated forms that normally occur in the subjective morning (CT2 and CT8) were absent and instead high levels of faster migrating hypophosphorylated PER I530A were visible ( Figure S3).
In order to track the effects of per I530A on PER oscillations with high temporal resolution, we turned to the BG-luc reporter, a period-luciferase transgene encoding for ~2/3rds of the PER protein. This per transgene does not rescue behavioral and molecular arrhythmicity induced by the per 01 allele, but it accurately reports PER protein cycling in a wild-type background (31), and is therefore suitable to determine the effects of per alleles on PER protein expression.
Bioluminescence rhythms emanating from individual adult male BG-luc flies were measured for four days in LD followed by three days in DD at the three different temperatures. Strikingly, at 18°C BG-luc oscillations in control and per I530A were of similar amplitude in both LD and DD conditions, with somewhat increased overall expression levels in the mutant ( Figure 4A). At 25°C BG-luc oscillations were strongly reduced in the per I530A mutant background compared to the controls, and this effect was further enhanced at 29°C ( Figure 4A). Oscillations in a wild type background were similarly robust at 18°C and 25°C, but amplitude was reduced at 29°C.
Next we analyzed the expression of a tim-luciferase (ptim-TIM-LUC) reporter gene that accurately reflects temporal TIM expression in peripheral clock cells (29,36). In a wild type background, ptim-TIM-LUC oscillations were indistinguishable at 18°C and 25°C, while oscillations were reduced in amplitude at 29°C ( Figure S4A), similar to what we observed for BG-luc ( Figure 4A). This was particularly the case in DD at 29°C, presumably because oscillations of this peripheral clock reporter dampen rapidly in DD even at lower temperatures ( Figure S4A) (36). In the per I530A mutant background, ptim-TIM-LUC oscillations were unaffected at 18°C, except for a reduced amplitude in DD ( Figure S4A). At 25°C already in LD a strong reduction of TIM amplitude caused by increased trough levels was observed ( Figure   S4A). In fact, the remaining oscillations are comparable in amplitude to those observed in clock less per 01 ; ptim-TIM-LUC flies at 25°C, indicating that they are mainly driven by lightdependent TIM degradation during the day ( Figure S4B) (37). At 29°C, the mutants' effects on ptim-TIM-LUC expression are further enhanced, leading to largely arrhythmic expression during LD and DD ( Figure S4A). In summary, the luciferase reporter experiments confirm that the per I530A mutant causes a temperature-dependent impairment of clock protein expression, ranging from normal function at 18°C to severe impairments at 29°C.

Mutations in the per Short phosphocluster do not cause temperature-dependent period lengthening.
Our Western blot results indicate a temperature-dependent phosphorylation defect caused by per I530A . Because the isoleucine at position 530 is not a target for phosphorylation by a known PER kinase, we investigated if per I530A may affect PER phosphorylation by some of the known PER kinases. To this end, we analyzed mutations affecting the PER 'phosphotimer' known to set clock speed in Drosophila (5). As expected, mutation of the S596 NEMO site (per S596A ) resulted in a drastic period shortening at 25°C (16.5 hr, Table 2), but periods were similarly short at both 18°C (16.9 hr) and 29°C (15.8 hr). Similarly, mutation of a single DBT target site in the per Short cluster (per S585A ) resulted at short periods at all three temperatures, with an overall 1-hr period shortening at 29°C compared to 18°C ( Table 2). Mutating all 4 phosphorylation sites within the per Short cluster (per TS583-596 ) also shortened period at all temperatures with a 1 hr increase in clock speed at 29°C compared to 18°C (Table 2). Finally, mutation of S47A led to the expected period lengthening at 25°C (29.8 hr, Chiu et al 2008), but also at 18°C (29.3 hr) and 29°C (30.3 hr) ( Table 2). While these results do not rule out a contribution of the PER phosphotimer in temperature compensation (i.e., mutations within the per Short cluster show a tendency to increase clock speed with rising temperature), it seems clear that disruption of this timer does not underlie the drastic overcompensation phenotype of per I530A .

period and timeless transcription is enhanced in per I530A mutants
Hyperphosphorylated PER is required for the repression of CLK/CYC mediated per and tim transcription (4,38,39). To test if the lack of fully hyperphosporylated PER in per I530A flies at 25°C and 29°C is associated with reduced repressor function we analyzed per and tim transcription using appropriate per-luc and tim-luc reporters (40,41). Bioluminescence emanating from whole flies was recorded as described above in LD and DD conditions at three different temperatures. At 18°C, both control and per I530A flies showed robust oscillations of per-luc expression in LD conditions, which rapidly dampen out in DD ( Figure 4B). However, expression levels in the per I530A mutants were about 2.5 times higher compared to the controls, indicating reduced repressor or enhanced transcriptional activation function of the mutant protein at this temperature ( Figure 4B, C). On average, levels of per transcription in the per I530A mutant were somewhat reduced compared to those in a per 01 background ( Figure   4B, C). At 25°C the amplitude of per-luc oscillations in the per I530A mutants was reduced compared to the controls ( Figure 4B). Moreover, expression levels were now higher compared to those in the complete absence of PER, in agreement with an enhanced impairment of PER I530A function at 25°C compared to 18°C. The effect on per-luc expression was further enhanced at 29°C, indicated by significantly higher expression levels compared to those in per 01 ( Figure 4B, C). Furthermore and like in per 01 , expression at 29°C was increased during the light phase of the LD cycle in contrast to the clock-dependent per-luc peak occurring during the dark phase ( Figure 4B) (40). The most likely explanation for this light-dependent peak in  Figure 5A). Interestingly, at ZT22 this trend was reversed, with higher levels of PER I530A :TIM dimers at 25°C and 29°C compared to PER:TIM heterodimers at these temperatures and also compared to PER I530A :TIM dimers at 18°C ( Figure 5A). The results clearly show that the PER I530A protein is able to form heterodimers with TIM at all temperatures tested, both at ZT16 and ZT22. Faulty PER:TIM heterodimerization is therefore unlikely to explain to observed behavioral and molecular phenotypes of per I530A flies.

PER:PER homodimerization is not grossly affected in per I530A flies
In addition to PER:TIM heterodimerization, PER:PER homodimerization is important for PER nuclear entry and PER repressor function (43). We therefore tested if PER I530A is able to form homodimers at different temperatures. For this, we added HA or MYC antigen tags to a full length per transgene containing the per I530A mutation and crossed these transgenes into a per 01 background. Wild type versions of these constructs (per c-myc and per HA ) restore robust behavioral rhythms in per 01 flies at 25°C these rhythms are also temperature compensated (43) (Table S1). As expected per 01 flies expressing the tagged PER I530A-MYC and PER I530A-HA constructs lengthen their free-running period with increasing temperature, similar to the endogenous per I530A CRISPR mutants (Table S1). Using Co-IP experiments we have previously shown the abundant existence of PER HA :PER MYC dimers in fly heads at 25°C (43). Here we show that PER homodimers also form at 18°C and 29°C ( Figure 5B). Next, we compared PER:PER and PER I530A :PER I530A dimer formation at the different temperatures in the tagged flies at ZT22 ( Figure 5B). The results were very similar to those obtained for PER:TIM heterodimerization: While PER I530A :PER I530A homodimers do form at all three temperatures, they appear more abundant at 25°C and 29°C compared to PER:PER homodimers, and this trend is again reversed at 18°C ( Figure 5B). Overall, the Co-IP experiments suggest that at ZT22 levels of both PER I530A :PER I530A homodimers, as well as PER I530A :TIM heterodimers are increased compared to the wild type dimers at 25°C and 29°C. Increased levels of nuclear PER I530A caused by faulty nuclear export offers a potential explanation for this observation.

Phosphorylation of the transcription factor CLOCK is altered in per I530A mutants.
An important function of PER in the nucleus is to bind to the transcription factor CLK, thereby recruiting kinases that phosphorylate and thereby inactivate CLK, resulting in repression of per and tim transcription. To see, if the increase per and tim transcription levels in per I530A flies are caused by inefficient inactivation of CLK, we compared CLK levels and phosphorylation status at ZT2, a time when CLK is maximally phosphorylated and repressed in wild type flies (reference). Strikingly, in per I530A flies, CLK phosphorylation was severely affected and mainly hypophosphorylated forms were visible at 25°C and CLK was barely detectable at 29° C ( Figure   5C, compare blue and orange arrows in middle panel). Indeed, at ZT2 and 25°C PER I530A migrated indistinguishably from wild type PER at ZT14, a time when CLK is transcriptionally active and hypophosporylated ( Figure 5C, orange arrow) (4,38). In contrast, although CLK levels were reduced in per I530A mutants at 18°C, CLK migrated similar in wild type and per I530A mutants, indicating normal phosphorylation and repression of CLK ( Figure 5C, black bar).
Although PER I530A can form PER:TIM and PER:PER dimers, it appears that its ability to repress CLK (or even binding to CLK) seems to be affected in a temperature dependent manner.

Subcellular localization of PER I530A is altered at warmer temperatures
If the per I530A mutation alters nuclear export in a temperature-dependent way, it should be possible to detect differences in PER I530A subcellular localization at different temperatures. To see if this the case, we first looked at wild type PER expression in the small ventral lateral neurons (sLNv), which are pacemaker neurons important for the maintenance of behavioral rhythms and for determining the period length in DD (15). Brains of wild type flies were dissected at two time points: ZT16, when PER is largely hypophosphorylated and cytoplasmic, and ZT22, when PER is largely hyperphosphorylated and nuclear. We found that the levels and subcellular localization (ratio of nuclear versus cytoplasmic signals) of PER were similar at 18°C, and 29°C with expected cytoplasmic and nuclear redistribution between ZT16 and ZT22 ( Figure 6A, C, S5). This is in-line with the Western blot results ( Figure 3A) and the temperature compensated behavioral period length ( Figure 1, Table 1). In contrast, PER I530A showed temperature-dependent variations in nuclear/cytoplasmic distribution. Compared to wild type, nuclear PER I530A levels were drastically reduced at ZT22 and 29°C ( Figure S5). Strikingly, at ZT16 and the same high temperature, PER I530A was predominantly nuclear, despite overall lower PER levels ( Figure 6, S5). These results indicate that at 29°C, wild type PER enters the nucleus earlier (≤ ZT16) compared to cooler temperatures (~ZT18) and is then exported again to mitigate premature nuclear accumulation and period shortening at warmer temperatures.
We have no explanation for why we observed high PER I530A at 29°C in head extracts, but low levels in the sLNv. One possibility is that in the sLNv premature, nuclear-trapped PER I530A , is rapidly degraded, presumably due to aberrant phosphorylation. In contrast, in the photoreceptor cells, which do not contain a self-sustained molecular clock, clearance of premature-nuclear PER I530A may not occur, because it may not be required in peripheral clock cells.

The nuclear export factor CRM1/Embargoed contributes to temperature compensation
If PER is transported by the general nuclear export machinery mediated by CRM1/Embargoed, RNAi constructs (44). While control flies containing only the gal4 or RNAi construct showed stable locomotor activity periods at 18°C, 25°C, and 29°C, this was not the case for lines expressing emb RNAi constructs in clock cells and clock neurons ( Figure 7A, B, Table 3).
Strikingly, and similar to per I530A mutants, knock down of emb led to significantly longer periods at 25°C compared to 18°C. Depending on the driver and the RNAi line the period lengthening varied between 0.7 h and 5 h (average 2.8 h), indicating a massive disturbance of temperature compensation ( Figure 7A, B, Table 3). Interestingly, further increasing the temperature to 29°C reversed the effect on clock speed, with periods shortening between 0.5 h and 2.6 hr (average 1.5 h), but not shortening back to the values observed at 18°C (Figure 7 A, B, Table 3). While this effect on temperature compensation is different to the linear increase of period length in per I530A mutants with temperature, it nevertheless confirms the involvement of nuclear export in temperature compensation. Moreover, it rules out that the observed period lengthening at 25°C is simply caused by increased knock down efficiency due to the temperature dependency of the Gal4 system. The period shortening observed at 29°C most likely indicates that other clock proteins are also subject to temperature dependent nuclear export. To directly test the interaction between specific knockdown of nuclear export and the per I530A mutation, we analyzed the behavior of flies expressing emb-RNAi in the per I530A background ( Figure 7C, Table 4). Because we could not co-express UAS-dicer as in the RNAi experiments described above (both UAS-dicer and per I530A are located on the Xchromosome), we compared the period length of the double mutants to those of per I530A single mutants carrying the gal4 only. Using Pdf-gal4 to drive two different emb-RNAi constructs in per I530A mutant background resulted in a significant blunting of the temperaturedependent period increase observed in per I530A single mutants ( Figure 7C, Table 4). Moreover, at 25°C expression of emb-RNAi BL34021 resulted on 100 % arrhythmic flies, which is not observed in per I530A single mutants or Pdf > emb-RNAi flies in a per + background (Tables 1, 3,   4). Using tim-gal4 to drive emb-RNAi BL31353 in a per I530A mutant background also blunted temperature-dependent period increase compared to per I530A , tim-gal4 flies ( Figure 7C, Table   4). Taken together, these results indicate an interaction between nuclear export and the per I530A mutant and suggest that in addition to PER, nuclear export of other clock proteins is regulated in a temperature-dependent manner (see Discussion).

Discussion
Circadian clocks are able to compensate for the acceleration of biological reactions associated with an increase in temperature. An impairment of temperature compensation should therefore result in a period shortening with increasing temperature. The per I530A mutation has the opposite effect, indicating that this mutation leads to an overcompensation phenotype.
Nevertheless, the molecular defects associated with per I530A could shed light on the mechanism underlying temperature compensation, and indeed our findings indicate that nuclear export plays an important role in this process. First, nuclear levels of both PER I530A :TIM heterodimers and PER I530A :PER I530A homodimer levels appear increased in per I530A mutants compared to controls, which could be explained by reduced nuclear export efficiency of the mutant protein ( Figure 5A, B). Second, subcellular localization of PER I530A revealed a temperature-dependent atypical nuclear enrichment at ZT16 at 29°C, in agreement with the idea that wild type PER is exported from the nucleus at this temperature to mitigate early nuclear accumulation. Finally, knockdown of the nuclear export factor CRM1 effected temperature compensation in opposite ways. In the range from 18°C to 25°C emb-RNAi expression resulted in a significant lengthening of period with the same magnitude (about 3-4 h) as observed in per I530A flies ( Figure 7B, Table 3). In contrast, further increase of temperature to 29°C lead to a significant period-shortening by ~1.5 h. These results unequivocally show that nuclear export is involved in the temperature compensation mechanism. Moreover, they indicate that in addition to PER, other clock proteins are subject to nuclear export. At low temperatures (18°C) nuclear export does not seem to play a role in adjusting period length, as both per I530A and emb-RNAi flies show periods in the normal 24 h range. At 25°C, both per I530A and emb-RNAi flies show essentially the same period-lengthening phenotype, indicating that in wild type flies, PER needs to be exported out of the nucleus in order to maintain a normal period length. At 29°C per I530A flies further increase their period length, indicating the necessity for continued temperature compensation of PER nuclear export above 25°C. In contrast, emb-RNAi flies shorten their period length from 25°C to 29°C by more than one hour. This suggests that at least one other clock protein needs to be exported at temperatures above 25°C to avoid speeding up the clock at temperatures above 25°C. A good candidate would be the unknown CLK-kinase, required to shut down CLK transcriptional activity (4,39). Increased kinase activity with rising temperature would be counteracted by increased nuclear export of the kinase. This idea is also consistent with the phenotype of flies with knocked down emb expression in the per I530A mutant background: at 29°C this resulted in reduction of the per I530A -induced period lengthening by 1-2 h, while there was little effect at 25°C ( Figure 7C, Table 4). Since mutations affecting TIM NES function also show temperature dependent period lengthening from 18°C to 29°C, it is unlikely that blocking TIM nuclear export is responsible for the period shortening observed in emb-RNAi flies above 25°C. We therefore propose that PER and TIM (presumably as PER:TIM heterodimer and PER:PER homodimer) are both subject to increased nuclear export with increasing temperature.

Molecular consequences of nuclear PER I530A retention at higher temperatures.
The temperature-dependent phosphorylation defect of PER I530A ( Figure 3B) combined with its reduced repressor activity ( Figure 4B, C) point to an impairment of this protein to recruit kinases to the CLK transcription factor, which are important for inactivating this transcription factor (4,39). Based on the striking phosphorylation defect of PER I530A at 25°C and 29°C ( Figure   3B), we propose that the mutant protein is retained in the nucleus and 'protected' from kinases that would normally phosphorylate PER in the cytoplasm to promote nuclear entry (45)(46)(47)(48). At 18°C, where nuclear export is less important, PER I530A would be normally phosphorylated and imported into to the nucleus. If true, this implies that wild type PER proteins are shuttled in and out of the nucleus at warmer temperatures to prevent early nuclear accumulation and speeding up of the molecular feedback loop with increasing temperature. Overall a relatively simple model emerges in which temperature-dependent enzymatic events (e.g. kinase activities) leading to earlier accumulation of PER and TIM in the nucleus and presumably to premature inactivation of CLK at warm temperatures (> 25°C), are counteracted by increased nuclear export. Although possibly too simplistic, this model is attractive, because it concedes that all enzymes and other factors involved naturally increase their reaction rate and activity with increasing temperature.

Temperature compensation: Network versus cell autonomous property
The Drosophila circadian clock network driving behavioral activity rhythms consists of several coupled neuronal groups (14). Following Pittendrigh's proposition, it is therefore conceivable that in order to compensate temperature changes some of these groups speed up with temperature, while others slowdown (10). While such a scenario is possible, available evidence points to temperature compensation being a cell autonomous property, affecting distinct steps in clock regulation as suggested by others (18). Using the novel LABL technique, we were able to monitor period promoter activity within specific subsets of the clock network and found that all groups are properly temperature compensated. Importantly, this included the DN1 neurons, which on their own are not able to sustain behavioral rhythmicity in DD, as well as peripheral clock cells (Figure 2A, B). Although it can be argued that signals were recorded from specific neuronal subsets, these subsets within the context of the entire network, which could of course influence the oscillations within one group. Nevertheless, even in the background of a Pdf 01 mutation, which largely prevents network coupling due to the lack of the PDF neuropeptide (17), temperature compensated clock gene expression was observed in dorsal subsets of the clock neurons (29). However, the same study revealed that circadian clocks in isolated fly tissues (in halteres and antennae) are over compensated, pointing to differences between neuronal and non-neuronal circadian clocks (29). Taken together, the currently available data support a model of cell autonomous temperature compensation, at least in central circadian clock neurons.

Conclusions:
The per I530A NES mutation equally effects temperature compensation on a behavioral and molecular level. Combined with the temperature compensation defects observed after knockdown of the nuclear export factor CRM1, this strongly implicates nuclear export of PER and other clock proteins as important part of the underlying molecular mechanism. Nuclear retention of PER I530A leads to increased CLK-mediated transcriptional activity, most likely caused by a temperature-dependent PER I530A phosophorylation defect. Our results support a model in which premature nuclear accumulation and CLK-inactivation at warmer temperatures is counteracted by increased nuclear export of clock factors, resulting a temperature compensated circadian clock. Future work will reveal, which other clock factors show temperature dependent control of subcellular localization.

Acknowledgements:
We thank Patrick Emery for discussions and for sharing unpublished results. We thank Isaac homology arm and 3' homology arm) were assembled into plasmid pBS-KS-attB1-2-PT-SA-SD-0-2xTY1-V5 (Addgene) that was linearized with XbaI and HindIII using In Fusion cloning. In a second round of cloning the homology arms were amplified again using the pBS donor plasmid from the previous round as a template. Outside primers were as described above while the inside primers introduced either a silent HindIII site that can be used to screen for transformants or the V5 tag. In fusion cloning was used to assemble the fragments as described above. The resulting plasmid was then used in a final round of PCR to introduce the per I530A mutation. See table_ for a detailed list of all primers.

Embryo injection and screening for transformants:
Donor plasmids containing the desired mutation along with gRNA plasmids were verified by sequence analysis and scaled up for injections using Qiagen plasmid midiprep. 6 µg of each plasmid were precipitated and eluted in injection buffer. gRNA construct and donor plasmids were mixed prior to injection and the mix was injected into freshly laid embryos of nos-Cas9 flies that were crossed to y w flies (55). Surviving adults were backcrossed in batch crosses to

Behavior:
Two to four days old males were loaded into glass tubes containing 5% sucrose in 2% agar and loaded into the DAM2 TriKinetics system (Waltham, MA). Flies were exposed to LD for 3 days,  (56). For statistical analysis shown in Figures 1C and 7B, estimation statistics has been used. This approach gives a more informative way to analyze and interpret results (57). It focuses on the effect size, as opposed to significance testing. While significance testing (p-values) focus on the acceptance or rejection of the null hypothesis, estimation stats focus on the magnitude of the effect size (i.e. mean difference) and its precision (57). Data were analyzed using DABEST (57), using the website available under https://www.estimationstats.com/#/ as described (29). Protein concentration was measured using Coomassie Plus Protein Assay Reagent (Thermo scientific) and equal amounts of protein were loaded onto 6% SDS gels for anti-V5 and anti-TIM blots or 8% SDS gels for anti-CLK blots respectively. For anti-V5 and anti-TIM blots proteins were transferred onto PVDF membranes while for anti-CLK blots proteins were transferred onto Nitrocellulose membranes using a Semi-dry electro blotting unit at 240 mA for 40 min.
After blotting, membranes were blocked with 5% milk in TBS-T at room temperature for 1 h.

Immunoprecipitations
CoIPs were performed as previously described (43). Briefly, adult flies from V5-tagged strains Average values were then calculated from three individual blots and were blotted as bar graphs (± SEM).

Immunostainings:
The protocol used was the same as previously described (60). Flies were entrained in LD for 6 days. Flies were collected in the dark and stayed in the dark until dissections. Brains were dissected in PBST 0.1% and fixed for 20 min at room temperature in PFA 4%. After 3 washes brains were blocked for one hour at room temperature in PBST 0.1% + 5% goat serum. Primary antibodies were incubated for 48 h (in PBST 0.1% + 5% goat serum) at 4°C, while secondary incubation was done overnight at 4°C. Brains were mounted using Vectashield. Monoclonal anti-PDF (DSHB) was used at 1/1000, and pre-absorbed Rabbit anti-PER (61) Table 4.