Transcriptional interference by antisense RNA is required for circadian clock function.

Eukaryotic circadian oscillators consist of negative feedback loops that generate endogenous rhythmicities. Natural antisense RNAs are found in a wide range of eukaryotic organisms. Nevertheless, the physiological importance and mode of action of most antisense RNAs are not clear. frequency (frq) encodes a component of the Neurospora core circadian negative feedback loop, which was thought to generate sustained rhythmicity. Transcription of qrf, the long non-coding frq antisense RNA, is induced by light, and its level oscillates in antiphase to frq sense RNA. Here we show that qrf transcription is regulated by both light-dependent and light-independent mechanisms. Light-dependent qrf transcription represses frq expression and regulates clock resetting. Light-independent qrf expression, on the other hand, is required for circadian rhythmicity. frq transcription also inhibits qrf expression and drives the antiphasic rhythm of qrf transcripts. The mutual inhibition of frq and qrf transcription thus forms a double negative feedback loop that is interlocked with the core feedback loop. Genetic and mathematical modelling analyses indicate that such an arrangement is required for robust and sustained circadian rhythmicity. Moreover, our results suggest that antisense transcription inhibits sense expression by mediating chromatin modifications and premature termination of transcription. Taken together, our results establish antisense transcription as an essential feature in a circadian system and shed light on the importance and mechanism of antisense action.

The transcription factors WHITE COLLAR (WC) -1 and -2 form a complex that activates frq transcription in the dark (DD) and mediates light-induced frq transcription for light-resetting of the clock by binding to light-responsive elements (LREs) on the frq promoter[11-13]. 3′ RACE and RNA sequencing showed that frq and qrf transcripts almost completely overlap (Figure 1a and Extended Figure 1a). In the wc mutants, frq expression was nearly abolished but qrf transcript was observed at ∼25% of wild-type levels (Figure 1b), indicating that both WC-dependent and -independent mechanisms mediate qrf transcription. WC complex binds to the qrf promoter region[14]. frq constructs with point mutations in each of the five putative binding sites in the qrf promoter were individually introduced into a frq (frq and qrf null) strain[15]. Mutation of only one site, qLRE, dramatically reduced qrf level (Figure 1c and Extended Data Figure 1b-d). In the frqqLREmut strain, the qrf level was comparable to that of the wc mutant. In a qLRE knock-in strain (frqKI(qLREmut), Extended Data Figure 1e), qrf levels were also much lower than in the control knock-in strain and WC binding at the qrf promoter was completely abolished (Figure 1c-d and Extended Data Figure 1f), indicating that qLRE is the only WC binding site in the qrf promoter.

Figure 1

WC-mediated light-induction of qrf represses frq to regulate light resetting of the clock. (a) A diagram showing the frq locus. E: EcoR V site. C: Cla I site. (b) Strand-specific RT-qPCR analyses showing the levels of frq and qrf in LL. frq-RT and qrf-RT are control reactions without reverse transcriptase. Error bars are SD (n=3). (c) Northern blot analyses showing the expression of frq and qrf in LL. (d) WC-2 ChIP assays showing the WC binding levels at the qLRE of qrf promoter in LL. (e-f) Strand-specific qRT-PCR analyses showing the levels of frq and qrf after a 2 min of light induction at DD24. (g) Phase response curves of conidiation rhythms after 2 min of light pulse at different circadian time points. *P<0.05, **P<0.01, ***P<0.001.

Extended Data Figure 1

Light-induced qrf expression represses frq transcription and regulates light resetting of the clock. (a) Strand-specific RNA-Seq result of the frq locus showing the overlapping frq and qrf transcripts. (b) Sequence alignment of known LRE elements. The qLRE and the mutated regions in the qrf promoter are shown. (c) Diagrams showing the chromosomal modifications in the indicated loci in the frq;frq and frq;frq strains. C, Cla I; B, Bgl II; E, EcoR V. (d) Strand-specific RT-qPCR results showing the expression levels of frq and qrf in indicated strains in LL. Error bars are standard deviations (n=3). Asterisks indicate P value < 0.01. n.s. indicates that the difference is not statistically significant. frq-RT and qrf –RT represent the non-RT reaction control. (e) Diagram describes the strategy used to obtain the knock-in strains by homologous recombination. (f) Strand-specific RT-qPCR results showing the expression levels of frq and qrf in the indicated knock-in strains in LL. (g) Western blot results showing the FRQ expression levels in the indicated strains in LL. The densitometric analysis of western blot results from three independent experiments is shown at right. Error bars are standard deviations. Asterisks indicate P value < 0.01. (h) Phase response curves of circadian conidiation rhythms of the indicated knock-in strains after 2 min of light pulse at different circadian time (CT) points.

Coinciding with low levels of qrf in the qLRE mutant, higher than wild-type levels of frq mRNA and FRQ protein were observed (Figure 1c and Extended Data Figure 1d, 1f, 1g). frq and qrf are rapidly light-induced in the wild-type (Figure 1e) but in the frq,frqqLREmut strain, whereas qrf induction was completely abolished (Figure 1f), light-induction of frq was significantly elevated when compared to the wild-type, indicating that light induction of qrf represses light-induced frq. Circadian conidiation rhythms of qLRE mutant strains in DD were near normal, however, a light pulse resulted in significantly greater phase-shifts in the qLRE mutants than the controls (Figure 1g and Extended Data Figure 1h). These results are consistent with a previous report[3] and indicate that qrf regulates light resetting of the clock by repressing light-induced frq.

Similar levels of qrf transcripts seen in the frq) and frq) strains at DD24 (Extended Data Figure 2a) indicate that qLRE does not regulate qrf expression in DD. In a strain (frq;frq.aq) in which the promoter of qrf was replaced with the quinic acid (QA)-inducible qa-2 promoter, qrf expression was abolished in the absence of QA (Figure 2a). On addition of QA, qrf was induced but frq levels were significantly reduced, further indicating repression of frq by qrf.

Extended Data Figure 2

qrf expression is required for circadian rhythmicities. (a) Strand-specific RT-qPCR results showing the expression levels of qrf in the indicated knock-in strains at DD24. (b) Race tube analyses of the frq;frqWT and frq;frq.aq strains in medium containing 0% glucose, 0.17% arginine with the indicated concentrations of QA in DD. The lack of glucose in medium is known to allow more efficient expression from the qa-2 promoter. The black lines on race tubes indicate the daily growth fronts. (c) The unnormalized luciferase activity of the experiments in Figure 2c. (d) The normalized result of Figure 2c in 10× scale showing fluctuation of the luciferase activity in the frq;frq.aq strain is random and not rhythmic in the frq;frq.aq strain. (e) A table showing the phases of the first conidiation band in DD of the race tube results shown in Figure 2b. (f) Western blot analysis showing the FRQ expression profiles in the frq;frq.aq strain with/without QA in DD at the indicated time points. (g) Northern blot analysis showing frq expression profiles in the frq;frq.aq strain in DD at the indicated time points. The densitometric analysis is shown below.

Figure 2

qrf expression is required for circadian rhythmicities. (a) Northern blot analysis showing the levels of frq and qrf with/without 1×10-3 M QA (0.1% glucose) in LL. Right panel shows the densitometric analyses of the results. Error bars are SD. (b) Race tube analyses of the indicated strains (0.1% glucose with different concentrations of QA) in DD. (c) Luciferase reporter assay showing the normalized frq promoter activity after two days in DD.

A number of observations indicate that qrf expression must be within a certain range to permit a functional clock. Without QA, the frq;frq.aq strain showed arrhythmic conidiation after the first day (Figure 2b and Extended Data Figure 2b). Moreover, a luciferase reporter (Pfrq-luc)[16] in the frq;frq.aq strain showed that the circadian luciferase activity seen in the control strain was abolished, indicating that qrf expression is required for clock function (Figure 2c and Extended Data Figure 2c, 2d). As QA concentration increased, circadian conidiation rhythms were gradually restored in the frq;frq.aq strain and at 10-5 -10-4 M QA, the rhythms were similar to the control strain (Figure 2b and Extended Data Figure 2b). At higher QA concentrations, however, the amplitudes of the rhythms reduced (Figure 2b) or became arrhythmic (Extended Data Figure 2b). In addition, the phase of the rhythms was significantly delayed without QA (Extended Data Figure 2e), a defect that was also rescued by qrf induction. Restoration of circadian rhythms of FRQ expression, FRQ phosphorylation profiles and frq mRNA oscillation by QA was also seen in the frq;frq.aq strain (Extended Data Figure 2f-g).

qrf RNA oscillates in DD in the wild-type in antiphase to frq (Figure 3a)[3] but WC-2 does not bind to the qrf promoter in DD (Extended Data Figure 3a). Moreover, the qLRE mutation did not affect either frq or qrf levels in DD (Extended Data Figure 3b), indicating that the WC complex does not regulate qrf transcription in DD. Crucially, a luciferase reporter (Pqrf-luc) driven by the qrf promoter showed that the qrf promoter activity is not rhythmic in a wild-type strain (Figure 3b and Extended Data Figure 3c-d).

Figure 3

Mutual inhibition of frq and qrf transcription forms a double negative feedback loop that is required for clock function. (a) Strand-specific RT-qPCR results showing the oscillations of frq and qrf in DD. Error bars are SD (n=3). (b) Luciferase reporter assay showing the normalized frq or qrf promoter activity after one day in DD in wild-type strains with the Pfrq:luc or Pqrf:luc construct. (c) Strand-specific RT-qPCR results showing the levels of frq and qrf in the frq;frqqLRE mut strain at DD24 or 60 min after the dark to light transfer. (d) Levels of frq and qrf in the wild-type and wcc strains measured by strand-specific RT-qPCR in DD. (e) Luciferase reporter assays showing the normalized luminescence levels in a wild-type strain that carries the Pfrq.luc (black) or Pmini.luc.Pfrq (red) construct. (f) A model of the Neurospora oscillator. (g) Mathematical simulation of relative frq levels in DD without qrf (k19 =0) (left), with qrf (k16=0.5, k17=0.35, k19=0.1) (middle), and with qrf over-expression (k16 =0.91) (right).

Extended Data Figure 3

(a) WC complex does not bind to the qrf promoter in DD. WC-2 ChIP assays showing the relative levels of WC binding at the frq (dLRE) and qrf (qLRE) promoters at the indicated time points in DD. (b) Strand-specific RT-qPCR results showing the levels of frq and qrf at DD24 in the indicated strains. (c) The unnormalized luciferase activity of the experiments in Figure 3b. (d) The normalized luciferase activity of the Pqrf:luc construct in Figure 3b shown in 10× scale (e) Densitometric analyses of three independent northern blot results indicate that levels of qrf transcripts are reduced in the frq9 strains as a result of increased frq expression.

Several results indicate that frq and qrf mutually inhibit each other. 1) In a frq9 mutant, only truncated FRQ protein is made, resulting in high frq levels [15] but reduced qrf levels (Extended Data Figure 3e). 2) When the frq,frqqLREmut strain was exposed to light, induction of frq resulted in decreased qrf (Figure 3c). 3) Low frq mRNA levels in the wcc mutant led to elevated qrf levels in DD (Figure 3d).

To further investigate the regulation bewteen frq and qrf, we created a luciferase reporter construct (Pmin-luc-Pfrq), in which the luciferase sense mRNA is driven by a constitutive promoter[17] and antisense luciferase mRNA is driven by the frq promoter (Figure 3e). Wild-type strains containing the Pfrq-luc or Pmin-luc (lacking antisense luciferase RNA) construct were used as controls. Luminescence in the Pmin-luc strain was arrhythmic but the Pmin-luc-Pfrq strain exhibited a robust circadian luminescence rhythm antiphase to that of the Pfrq-luc rhythm (Figure 3e and Extended Data Figure 4a-b). These results indicate that the antiphasic rhythm of qrf expression is driven by rhythmic frq transcription independent of RNA sequence. Therefore, frq and qrf transcription forms a double negative feedback loop that results in antiphasic rhythms of frq and qrf (Figure 3f).

Extended Data Figure 4

(a) The unnormalized luciferase activity of the experiments in Figure 3e. (b) The unnormalized luciferase activity of the wild-type strain carrying the Pmin-luc construct. Results of two independent transformants were shown. (c) Mathematical modeling of the Neurospora circadian oscillator with the double negative feedback loop. The differential equations used in the model are shown. The model is identical to a previously developed model[23] with the exception of equation 1, which in this case includes the inhibition of frq transcription by qrf, and the equation 8, which includes the inhibition of qrf transcription by frq. The rate constants used in the simulations were listed below.

Mathematical modeling (Extended Data Figure 4c)[18] demonstrated that without qrf the Neurospora circadian oscillator can only generate a low amplitude frq oscillation that damps out quickly (Figure 3g). When the double negative feedback loops were introduced into the model, both frq and qrf levels oscillated robustly with antiphasic rhythms. When qrf was over-expressed, the frq oscillation could not sustain. These results suggest that the previously known circadian feedback loops, although not sufficient to sustain a persisting rhythm, are the source of the rhythmicity that is amplified and sustained by the resonance of mutually inhibitory and antiphasic expression of frq and qrf.

How do qrf and frq inhibit each other? Light-induced frq transcription, frq and qrf levels and circadian rhythms were normal in RNAi mutants (Extended Data Figure 5a-d)[19]. Convergent transcription is also known to induce DNA methylation in the frq region[20,21] but deletion of genes required for DNA methylation also did not affect frq expression (Extended Data Figure 5e-f).

Extended Data Figure 5

Neither RNAi nor DNA methylation pathways play a significant role in the clock. (a-b) Strand-specific RT-qPCR results showing the induction of frq after a light pulse are similar in the indicated trains at DD24. (c) Strand-specific RT-qPCR results showing similar expression levels of frq and qrf in the indicated strains in LL. (d) A table showing the period lengths of the conidiation rhythms in the wild-type and different RNAi mutants. (e-f) Strand-specific RT-qPCR results showing the induction of frq after a light pulse is similar in the indicated trains at DD24. Error bars are standard deviations (n=3).

We introduced a frq construct (qrf) with the frq promoter deleted into the frq and frq;frqqLRE mut strains at the csr-1 locus (Extended Data Figure 6a). This transgene can express normal levels of qrf without detectable frq expression (Extended Data Figure 6b). In the frq;frqqLRE mut;qrf strains, even though qrf expression was restored to normal levels, frq levels were not rescued (Extended Data Figure 6c-d), indicating that qrf regulates frq in cis.

Extended Data Figure 6

(a) A diagram showing the chromosomal modifications in the frq;qrf strains that allow the expression of qrf in trans. The frq construct is at the his-3 locus, and qrf is expressed only from the csr-1 locus. The red dashed line indicates that the frq promoter region is deleted in the qrf construct to abolish frq expression. (b) Strand-specific RT-qPCR results showing that only qrf is expressed from the qrf construct in the frq; qrf strain. (c) qrf expression does not repress frq transcription in trans. Strand-specific RT-qPCR results showing the levels of frq and qrf transcripts in the indicated strains in LL. Error bars are standard deviations. Asterisk indicates P value < 0.05 (n=3). (d) Northern blot results showing that expression of qrf in trans in the frq;frq;qrf strains does not repress frq expression. Densitometric analysis of the northern blot results is shown at right. (e-f) WC ChIP assays showing the relative WC binding levels at the frq promoter in LL in the indicated strains. The wc double mutant (wcc) was used as a negative control for ChIP. n.s. indicates the lack of statistically significance (n=3). (g) Strand-specific RT-qPCR results showing that the mutation of the qLRE element in the qrf promoter results in significant increases in light-induced frq pre-mRNA expression. (h) Northern blot results showing that the stability of frq mRNA is not affected by the transcription of qrf. The frq;frq.aq strain that can induce qrf expression in the presence of QA was used. Thiolutin, a transcription inhibitor[30], was added in the culture to block frq transcription so that frq mRNA stability could be determined. Cultures were harvested at the indicated time points after the addition of thiolutin.

WC binding to the frq promoter initiates WC-dependent frq transcription, but the qLRE mutation did not affect WC binding at the frq promoter (Extended Data Figure 6e-f). However, However, levels of frq pre-mRNA were significantly elevated in the qLRE mutants (Figure 4a and Extended Data Figure 6g). Moreover, qrf expression in the frq;frq.aq strain did not affect frq RNA stability (Extended Data Figure 6h). These results suggest that qrf regulates frq after transcriptional initiation.

Figure 4

qrf transcription results in pol II stalling, premature transcription termination and chromatin modifications. (a) Strand-specific RT-qPCR results showing the levels of intron-containing frq after dark to light transfer. (b) ChIP assays showing the relative enrichment of pol II Ser 5, Ser 2, and H3K36me3 in the frq locus in LL. IgG was used as the mock control for IP. (c-d) Northern blot analyses showing the levels of frq and qrf in LL. (e-f) Northern blot analysis in LL. frq-N term and frq-C term are specific for the 5′ or 3′ half of the frq transcripts, respectively (shown in Figure 4b). The addition of QA results in rrp44 silencing. The ratios between truncated and full-length frq transcripts from three independent experiments in (f) are shown. Error bars are SD. *P<0.05, **P<0.01.

After transcriptional initiation, RNA polymerase II CTD is phosphorylated at serines 2 and 5 with Ser 5 and Ser 2 phosphorylation enriched near the 5′ and 3′ ends of transcribed regions, respectively[22]. Similar Ser 5 and Ser 2 phosphorylation profiles were seen at a Neurospora locus without antisense transcripts (Extended Data Figure 7a-b). In contrast, both modifications of CTD peaked at the same position in the middle of the transcribed frq region (Figure 4b, Extended Data Figure 8). Mutation of the qLRE, which reduces qrf expression, resulted in decreases of both Ser 2 and Ser 5 phosphorylation. Phosphorylation of pol II can trigger histone H3K36 methylation[23]. H3K36me3 enrichment at the frq locus peaked at the same position as did CTD phosphorylation, and the qLRE mutation reduced H3K36me3 (Figure 4b, bottom panel). In the frq;frq.aq strain minus QA, in which qrf expression is completely abolished, the distributions of CTD phosphorylation and H3K36me3 on frq resemble the control locus lacking antisense transcription (Extended Data Figure 9a-c). These results suggest that pol II stalls in the middle of frq locus due to convergent transcription[9].

Extended Data Figure 7

(a) The top panel showing the strand-specific RNA-Seq results of the NCU01953 locus. The bottom panel showing the primers positions used for ChIP assays and riboprobe position used for Northern blot analysis. (b) The ChIP assays showing the relative enrichment of Pol II Ser5 phosphorylation, Pol II Ser2 phosphorylation, and H3K36me3 at the NCU01953 locus in the wild-type strain. Scale on y axis is the enrichment percentage of immunoprecipitation (IP) over input. IgG was used as the mock control for IP.

Extended Data Figure 8

(a) A diagram showing the primers positions at the frq locus used for ChIP assays and riboprobes positions used for Northern blot analyses. (b) The ChIP assays showing the relative enrichment of histone H3 and Pol II CTD in the frq locus. Scale on y axis is the enrichment percentage of immunoprecipitation (IP) over input. IgG was used as the mock control for IP. (c) The ChIP assays showing the relative enrichment of Pol II Ser5 phosphorylation and Pol II Ser2 phosphorylation (ChIP data in Figure 4b normalized by Pol II CTD ChIP results in panel b), and H3K36me3 (ChIP data in Figure 4b normalized by Histone H3 ChIP results in panel b). Asterisks indicate differences that are statistically significant (*P<0.05, **P<0.01, ***P<0.001).

Extended Data Figure 9

Complete abolishment of qrf expression changes the distributions of phosphorylated pol II CTD to a normal profile. (a) A diagram showing the primers positions used for ChIP assays. (b) The ChIP assays showing the relative enrichment of Pol II Ser5 phosphorylation, Pol II Ser2 phosphorylation, and H3K36me3 in the frq locus in the frq;frq.aq strain in medium with 2% glucose. Under such a condition, qrf expression is completely abolished (shown in Figure 2a). (c) The ChIP assays showing the relative enrichment of Pol II Ser5 phosphorylation and Pol II Ser2 phosphorylation (ChIP data in panel b were normalized by Pol II CTD ChIP results), and H3K36me3 (ChIP data in panel b were normalized by Histone H3 ChIP results).

SET-2 methylates H3K36 in Neurospora[24] and is required for clock function[25]. Even though frq and qrf levels were only modestly increased in the set-2 single mutants, their levels were markedly elevated in the frq9;set-2 double mutant (Figure 4c and Extended Data Figure 10a), suggesting that H3K36me3 contributes to the suppression of frq and qrf transcription. However, the induction of qrf still resulted in a reduction of frq in the set-2 double mutant strain (Figure 4d and Extended Data Figure 10b), indicating the existence of another mechanism that mediates the action of antisense transcription.

Extended Data Figure 10

(a) The densitometric analyses of three independent experiments shown in Figure 4c are shown. (b) The densitometric analyses of independent experiments shown in Figure 4d are shown. (c) Northern blot analysis using a N-term probe (marked in Extended Data Figure 8a) specific for the 5′half of qrf transcripts in the WT and WT;dsrrp44 strains in LL. (d) Northern blot analysis using a N-term probe (marked in Extended Data Figure 7a) specific for the 5′half of NCU01953 transcripts in the WT;dsrrp44 strain in LL.

Stalled pol II due to convergent transcription may abort transcription prematurely, which should result in truncated transcripts from their respective 5′ ends. We screened a panel of Neurospora mutants for the nuclease that degrades such transcripts. The RNA exosome is involved in degrading full-length frq transcripts[26]. Northern blot analysis using an RNA probe (frq-N term or qrf-N-term) specific for the 5′ end of frq or qrf, respectively, revealed that the silencing of rrp44, which encodes the exosome catalytic subunit, resulted in the appearance of a low molecular weight RNA smear in the wild-type strain (Figure 4e and Extended Data Figure 10c). In contrast, a probe specific for the 3′ half of frq failed to detect the RNA smear. Importantly, the amount of the RNA smear was dramatically decreased in the qLRE mutant (Figure 4f) and was not observed for a control gene without antisense transcripts (Extended Data Figure 10d). Together, these results suggest that both premature transcription termination and chromatin modifications contribute to the transcriptional inhibition of frq by qrf.

We discovered here that sense and antisense transcription of frq forms a double negative feedback loop that is interlocked with the core circadian feedback loops in Neurospora. The mutual transcription interference of frq and qrf results in antiphasic oscillations of frq and qrf that resonate to achieve robust and sustained circadian gene expression. In the silkmoth and mouse liver, antisense per RNAs exist and were also found to oscillate in antiphase to sense RNAs [27-30], suggesting that a similar mechanism may also function in animal circadian systems.

Methods

Strains, plasmid constructs, and growth conditions

The wild-type strain used in this study was 87-3 (ras-1bd, a). The frq strain is a frq-null and qrf-null mutant [15]. The frq9 strain bears a frame-shift mutation in the frq ORF [15]. The mutants wc-1, wc-2, wcc (wc-1), dcl (dcl-1, dcl-2), qde-1, qde-2, qde-3 were generated in previous studies [12,31]. The dim-2 and dim-5 strains were generously provided by Dr. Qun He [32]. The set-2 (FGSC #15505) strain was from Fungal Genetic Stock Center (FGSC). The set-2;frq9 double mutant was created in this study by crossing set-2 and frq9. The dsrrp44 strain was generated by introduction of a plasmid expressing quinic acid-driven rrp44-specific RNA hairpin into a wild-type strain [26]. The frq and frq strains were created in this study (Extended Data Figure 1b, 1e). The frq; dsrrp44 strain was created by crossing frq and dsrrp44. The frq;frqWT, frq;frqqLREmut, frq;frq.aq strains were obtained by targeting plasmids pKAJ120, pKAJ120qLRE mut, and pKAJ120.aq, respectively, into the his-3 locus of the frq mutant as previously described [15] (Extended Data Figure 1c). The frq;frqqLRE mut;qrf strain was generated by targeting the plasmid pCSR1.qrf into csr-1 locus of the frq;frqqLRE mut mutant as previously described [33] (Extended Data Figure 6a).

The pKAJ120 that contains the entire wild-type frq gene including its promoter and a his-3 targeting sequence was used as the parental plasmid [15]. The qLRE in pKAJ120 was mutated as described (Extended Data Figure 1b) by site-directed mutagenesis to create pKAJ120qLRE mut. The fragment between BssHII and SapI sites of pKAJ120 was replaced with an inverted Neurospora qa-2 promoter to create pKAJ120.aq. The PCR fragment containing entire wild-type qrf and its promoter (primer: 5′-TTCATTAAGGTGGGGCAGG-3′, 5′-TTTCCACGCCGGCCCCAGTC-3′) was inserted into vector pCSR1 between NotI and PstI sites to generate pCSR1.qrf [33].

Growth conditions were described previously [15]. Liquid cultures were grown in minimal medium (1×Vogel's, 2% glucose). When quinic acid (QA) was used, liquid cultures were grown in low glucose medium (1×Vogel's, 0.1% glucose, 0.17% arginine) with indicated concentrations of QA. For rrp44 knockdown assay, Neurospora was cultured into mats in low glucose medium with 10-2 M QA for 2 days. Afterwards, Neurospora mats were cut into discs and cultured in flasks in same medium with shaking. After 2 days, the tissues were harvested. For the mRNA decay assay, the cultured conditions were described previously [26]. For rhythmic experiments, the Neurospora cultures were transferred from LL to DD at time 0 and were collected in DD at the indicated time (hours). For light induction, Neurospora cultures were grown in DD for 24 hrs with shaking, and then treated with 1750 lux light pulse. Afterwards, Neurospora cultures were transferred back to DD and collected at the indicated time (minutes). For race tube assay, the medium contains 1× Vogel's, 0.1% glucose, 0.17% arginine, 50 ng·ml-1 biotin, and 1.5% agar with indicated concentrations of QA (for data shown in Extended Data Figures 2b, glucose was not present in the medium). Strains were inoculated and grown in constant light at 25 degree for 24 hrs before being transferred to DD at 25 degree. Calculations of period length and phase were performed as described [34].

Analysis of phase response to light

The assay was performed on race tubes containing acetate/casamino acid medium (1× Vogel's, 1.2% sodium acetate, 0.05% casamino acid hydrolysate, and 1.5% agar) [35]. Race tubes were first grown in LL at 25 degree for 48 hrs before being transferred to DD. Cultures were then grown in DD at 25 degree for 25 hrs, and different individual cultures were given a 2-min light pulse (1750 lux) at different times (2 hrs intervals) to cover an entire circadian cycle. The amount of phase shift was determined by comparing light-treated cultures (6 replicas for each time point) with those of the control cultures (kept in DD). The initial LL to DD transition was defined as circadian time (CT) 12. The phases of the cultures were calculated as the average phase for 2 consecutive days after the light treatment.

Protein and RNA analysis

Tissue harvest, protein extraction, and western blots were performed as previously described [15]. For protein separation, 40 μg total protein loaded in each lane of SDS-PAGE (7.5%). Total RNA was extracted with Trizol in accordance with the manufacturer's protocol, and then further purified by 2.5M LiCl as previously described [36]. Northern blot analyses were performed as previously described using [[32]P]UTP-labeled riboprobes [15]. Riboprobes were transcribed in vitro from PCR products by T3 or T7 RNA polymerase (Ambion) with the manufacturer's protocol. The primer sequences used for the template amplification were frq-N term (5′- TAATACGACTCACTATAGGG (T7 promoter) GGCAGGGTTACGATTGGATT-3′, 5′-GGGTAGTCGTGTACTTTGTCAG-3′), frq-C term (5′-TAATACGACTCACTATAGGG (T7 promoter) CCTTCGTTGGATATCCATCATG-3′, 5′- GAATTCTTGCAGGGAAGCCGG-3′), qrf-N term (5′-AATTAACCCTCACTAAAGGG (T3 promoter) GAATTCTTGCAGGGAAGCCGG-3′, 5′- CCTTCGTTGGATATCCATCATG-3′), and NCU01953 N-term (5′ TAATACGACTCACTATAGGG (T7 promoter) GTGCCAAAGAGTTGGCCATTC-3′, 5′- CTTGCACCACAAACTGTTGAAC-3′). Except for results shown in Figure 4e-f, riboprobes of frq-C term and qrf-N term were used for detecting frq and qrf, respectively. For Figure 4e-f and Extended Data Figure 10, the riboprobes were hydrolyzed for 25 mins in 40 mM Na2CO3, 60 mM NaHCO3, 10 mM 2-mercaptoethanol, and then the reaction was stopped with an equal volume of 0.2 M NaOAC, pH 6.0, 1% HOAC, 10 mM DTT. For western blot and northern blot, densitometry analyses were performed using Image J.

Quantitative PCR (qPCR) and quantitative reverse transcriptase PCR (RT-qPCR) were performed as previously described [26]. For strand-specific RT-qPCR, several modifications of the protocol were made. The primer sequences for strand-specific RT reactions were frq (5′-GCTAGCTTCAGCTAGGCATC (adaptor) CGTTGCCTCCAACTCACGTTTCTT-3′), qrf (5′-CCTCAGCTCGTACGAGTCGTAC (adaptor) GTCATGGAGCCCTCTGGTCTTGGT-3′), frq pre-mRNA (5′-GCTAGCTTCAGCTAGGCATC (adaptor) TTGAACGGTAGGGAGGAGGAGAG-3′), and β-tubulin (5′-CTCGTTGTCAATGCAGAAGGTC-3′). The RT reaction was performed by mixing the primers of specific strand and β-tubulin. β-tubulin was used for internal control. The primer sequences for the qPCR step of RT-qPCR assay were frq (5′- AGCTTCAGCTAGGCATCCGTT-3′, 5;- GCAGTTTGGTTCCGACGTGATG-3′), qrf (5′- CAGCTCGTACGAGTCGTACGTC-3′, 5′- ATCTTCCGATGTTGTCGAGCGT-3′), frq pre-mRNA (5′- AGCTTCAGCTAGGCATCTTGAACG-3′, 5′- ACGGCATCTCATCCATTCTCACCA-3′), and β-tubulin (5′-ATAACTTCGTCTTCGGCCAG-3′, 5′- ACATCGAGAACCTGGTCAAC-3′).

Strand-specific RNA Seq

Approximately 106 /ml conidia were inoculated into 30 ml of VMM with 2% sucrose and incubated at 25°C for 20 hrs with shaking. The mycelia was filtered through Whatman paper and washed once with ice cold phosphate buffered saline. Approximately 200 mg was put into a 2.0 ml cryo vial with silicon beads and frozen in liquid nitrogen. Total RNA was isolated from frozen mycelia in 1.0 ml TRIzol using a BioSpec beater. Five μg total RNA was run out on a 1.0% agarose gel to check for degradation. The 3′ paired-end RNAseq library was prepared from 10 μg total RNA as described [37]. Briefly, the poly(A)+ fraction was isolated on oligo(dT)25 magnetic beads [Invitrogen] then treated to chemical fractionation with ZnCl (Ambion) followed by another round of poly(A)+ enrichment. First strand cDNA was generated from an anchored oligo(dT)20 primer followed by second strand synthesis with DNA polymerase I. The addition of paired-end adaptors and size isolation with ampureXP beads (Ambion) was followed by PCR enrichment are described [37]. The library has an average insert size of 200 bp ± 36 bp. The library was sequenced to a depth of 22 million 36 bp paired-end reads on an Illumina GAIIx sequencer. The access number is SRP030415.

Nascent nuclear transcripts isolation

Liquid cultures were grown in 1L minimal medium (1×Vogel's, 2% glucose). The dried frozen tissue (∼4g) was ground in liquid N2. 8 ml of cold Buffer A (1M Sorbitol, 7% ficol, 20% glycerol, 5mM Mg(Ac)2, 3mM CaCl2, 50mM Tris.Cl pH7.5, 3mM DTT, 10μg/ml leupeptin, 10μg/ml pepstatin, 10μg/ml PMSF) was mixed with tissue powder on ice for 5-10min with stirring. The mixture was filtered through 4 layers pre-wet miracloth (Buffer A) into small flask on ice. Buffer A was added into filtered mixture till the volume is up to 8 ml. 16ml pre-chilled Buffer B (10% glycerol, 5mM Mg(Ac)2, 25mM Tris.Cl pH7.5, 10μg/ml leupeptin, 10μg/ml pepstatin, 10μg/ml PMSF) was slowly added with stirring. The obtained mixture was gently laid onto 10 ml of pre-chilled buffer A/B (2.5:4) in 50 ml of screw-cap tube. The supernatant was collected after centrifuging at 3000g for 7min at 4 degree. 33ml of the supernatant was gently laid onto 5 ml of pre-chilled Buffer D (1M Sucrose, 10% glycerol, 5mM Mg(Ac)2, 25mM Tris.Cl pH7.5, 1mM DTT, 10μg/ml leupeptin, 10μg/ml pepstatin, 10μg/ml PMSF). The pellet (purified nuclei) was collected after centrifuging at 9400g for 15min at 4 degree. The pellet was suspended by 1ml Trizol. Total RNA was extracted with Trizol in accordance with the manufacturer's protocol. The contaminated DNA was removed using TURBO DNA-free™ Kit following the manufacturer's protocol. 50ng of treated RNA was used for strand-specific RT-qPCR.

Chromatin immunoprecipitation (ChIP) assay

The ChIP assay was performed as previously described with several modifications [38]. First, 500 μg total cell lysis was used for each immunoprecipitation assay. The antibodies used in this study were WC-2 [12](2 μl/assay), Pol II CTD (Abcam #ab26721, 2μg/assay), Pol II S2P-CTD (Abcam #ab5095, 2 μg/assay), Pol II S5P-CTD (Abcam #ab5131, 2 μg/assay), Histone H3 (Abcam #ab1791, 2μg/assay) and H3K36me3 (Abcam #ab9050, 2 μg/assay). Each experiment was independently performed three times, and immuoprecipitation with IgG or wcc extract was used as the negative control. qPCR was used for measuring the immunoprecipitated DNA. Primer sequences for qPCR were frq dLRE (5′-AGAGTTTGGCCGGACAACCAGTA-3′, 5′-GCTTCGACCGAAAGTATCTTGAGCCT-3′), frq pLRE (5′-GTCGCAGAGGACCCTGAACTTT-3′, 5′-TCCCACAGATGCACAGGAATCG-3′), and qrf qLRE (5′-ATCGATCACTAGTCCCGGTTCGTT-3′, 5′-TTGCTGATAATGCGCTGAGGGTCT-3′). For experiments in Figure 4b and Extended Data Figure 8 and 9, the primer sequences were 1 (5′-AGAGTTTGGCCGGACAACCAGTA-3′, 5′-GCTTCGACCGAAAGTATCTTGAGCCT-3′), 2 (5′-GGGTAGTCGTGTACTTTGTCAG-3′, 5′-ACCGGACTTTAGGTTGTGTG-3′), 3 (5′-ACGGCCTTTCTCTGTTTACC-3′, 5′-GCAGAGTTGGGTCGGATTTA-3′), 4 (5′-CCGTTCTAGCGTGCTTCTT-3′, 5′-GGCACTAATGAGGTTCGAGATT-3′), 5 (5′-TTACTTCATCTTCCGCACTGG-3′, 5′-GGCAGGGTTACGATTGGATT-3′), 6 (5′-ACTGTTCACCGGAAAGATCAG-3′, 5′-GGGCCATGTTGGTTCCTT-3′) 7 (5′-TTGCACCGATCTTTCAGGAG-3′, 5′-CTGCAGCACATGTTCAACTTC-3′), 8 (5′-GGGCAGAGAGTGGCTATAGTA-3′, 5′-CCGCGTCTTCTTCTCACATAG-3′). For experiments in Extended Data Figure 7, the primer sequences were 1 (5′- CAAGTGGGCCCTACAGTTATT-3′, 5′- CCAATCCACTTCCCTTTCCA-3′), 2 (5′- CACCGTGTTCGTGCTTACT-3′, 5′- GGTATCCGTCTTGTCCCTTTG-3′), 3 (5′-TCAAGATACGGAGCGAAGAATG-3′, 5′- GATTGTAGCTGTTCCACCTCTC-3′), 4 (5′-TGACCGACCCAGTTGATTTC-3′, 5′- GTGGCACTTCTATCACCTTCTT-3′), 5 (5′- CGACCGTTAGGAGACCAAATAG-3′, 5′-AATGGCTTCCTTGTGGTTAGA-3′), 6 (5′- CTTAGTCGGATCAGTGGCTATT-3′, 5′- TTCGTAGATGCCGTGAGATG-3′), 7 (5′- CGTGGACTTCTGCTACTCTTC-3′, 5′- CCTTTGCAGTCCTCCTCTTT-3′), 8 (5′-GTCTGGGAGCTTCTGTTGAATA-3′, 5′- TAAAGTGGTGAACGACCTCATC-3′).

Luciferase reporter assay

The luciferase reporter assay was performed as previously described [16]. The reporter construct in Figure 2c (Pfrq-luc) was generated by inserting BamHI-NotI fragment of Pfrq-luc-I (a generous gift from J. Dunlap) into corresponding sites of pBARKS1 [16]. The reporter construct in Figure 3b (Pqrf-luc) was generated by inserting the PCR fragment containing qrf promoter and luciferase gene into NotI-EcoRI sites of pBARKS1. For the PCR fragment, qrf promoter was directly fused with luciferase gene ORF. The primer sequences were qrf promoter (5′-ATCGATTTCATTAAGGTGGGG-3′, 5′-GGTGGTGGGTGGTGGGTGGAG), and luciferase gene (5′- ATGGAGGACGCCAAGAACATCAAG-3′, 5′- TCAGAGCTTGGACTTGCCGCC-3′). The plasmids Pfrq-luc and Pqrf-luc were transformed into wild-type or frq;frq.aq strains with Ignite selection as previously described [39]. The plasmid Pmin.luc was generated by inserting the luciferase gene ORF into XmaI site of pDE3dBH. The plasmid Pmin.luc.Pfrq was generated by inserting inverted frq promoter (the same as Pfrq-luc-I) into downstream of luciferase gene ORF of Pmin.luc. The non-regulated basal promoter in pDE3dBH (upstream of EcoRI site) was used as the minimal promoter to control the expression of luciferase gene in Pmin.luc and Pmin.luc.Pfrq as previously described [17]. For Figure 3e, the plasmids Pfrq-luc-I, Pmin.luc.Pfrq, and Pmin.luc were transformed into the wild-type strain by targeting his-3 locus. Under our experimental condition, luciferasae signals are highly variable during the first 1-2 days in the LumiCycle and only become stabilized afterwards, which is likely due to an artifact caused by the light-dark transfer of the cultures. Thus, the results presented were recorded after 1-2 days in DD.

Mathematical modeling

The mathematical simulation of relative frq and qrf RNA levels in DD was performed based on a previously described model [18,40-42] with several modifications. The equations used here are listed in Extended Data Figure 4c.

The differential equations (1) through (8) were solved numerically by the Gear method and XPPAUT [43,44]. The parameters were as follows: The values of k2, k3, k6, k7, k8, k9, k11, k12, k13, k14, k15, K, K, k01 were from a previous study [18]. The values of k4 and k5 were from previous study [40,41]. The value of k10 was from a previous study [42]. The values of other parameters were determined in this study. For Figure 3g, k19 = 0 (left panel); k17 = 0.35, k19 = 0.1 (middle panel); k19 = 0.1, k17 = 0.35, k16 = 0.91 (right panel).