Abstract
Circadian transcriptome studies identified a novel transcript at the Period2 (Per2) locus, which we named Per2AS. Per2AS is a long non-coding RNA transcribed from the antisense strand of Per2, and is expressed rhythmically and anti-phasic to Per2 mRNA. Previously, we mathematically tested the hypothesis that Per2AS and Per2 mutually inhibit each other’s expression by forming a double negative feedback loop, and found that Per2AS expands the oscillatory domain. In this study, we have experimentally tested this prediction by perturbing the expression of Per2AS in mouse fibroblasts. We found that Per2AS represses Per2 pre-transcriptionally in cis and regulates the amplitude of the circadian clock, but not period or phase. Unexpectedly, we also found that Per2 positively regulates Per2AS post-transcriptionally, indicating that Per2AS and Per2 form a single negative feedback loop. Because knock-down of Per2 does not recapitulate the phenotypes of Per2AS perturbation and Per2AS also activates Bmal1 in trans, we propose that Per2AS regulates the amplitude of the circadian clock without producing a protein by rewiring the molecular clock circuit.
Introduction
Eukaryotic genomes are pervasively transcribed, and non-protein coding portions of the genome dominate the transcriptional output of mammals (Bertone et al. 2004; Rosok and Sioud 2004; Carninci et al. 2005; Katayama et al. 2005; Sun et al. 2006). RNA species beyond mRNA are known as non-coding RNAs (ncRNAs) and have been categorized into many subtypes, such as ribosomal RNA (rRNA), transfer RNA (tRNA), small nucleolar RNAs (snoRNAs), small nuclear RNAs (snRNAs), microRNAs (miRNAs), circular RNAs (circRNAs), and long noncoding RNAs (lncRNAs) (Mercer et al. 2009; Panda et al. 2017). Despite their pervasive transcription, ncRNAs, particularly lncRNAs, were originally considered to be mere transcriptional noise and to lack defined functions. Poor evolutionary conservation in their primary sequences between species also raised concerns about their functional significance (Ponjavic et al. 2007; Guttman et al. 2009; Johnsson et al. 2014). More recent studies, however, have revealed an intriguing conservation of the genomic positions of lncRNAs (i.e., synteny) as well as conservation of their promoter and exon sequences, compared to intron or nontranscribed intergenic regions of protein-coding genes (Khaitovich et al. 2006; Pang et al. 2006; Yassour et al. 2010; Rhind et al. 2011; Derrien et al. 2012; Goodman et al. 2013; Anderson et al. 2016; Engreitz et al. 2016; Groff et al. 2016). These observations raise the possibility that some lncRNAs, if not all, are biologically relevant and have important and conserved functions. Indeed, a few dozen examples have highlighted the importance of lncRNAs in a variety of biological processes, such as X chromosome inactivation, imprinting, cell cycle regulation, and stem cell differentiation (Lee et al. 1999; Smilinich et al. 1999; Sleutels et al. 2002; Feng et al. 2006; Dinger et al. 2008; Morris et al. 2008; Zhao et al. 2008; Bond et al. 2009; Sopher et al. 2011; Modarresi et al. 2012).
Circadian rhythmicity, which modulates daily biochemical, physiological, and behavioral cycles, is a fundamental aspect of life on Earth. In mammals, essentially every cell is capable of generating circadian rhythms, and, within each cell, a set of clock genes forms a network of transcription-translation feedback loops that drive oscillations of approximately 24 hours (Lowrey and Takahashi 2004; Takahashi et al. 2008; Takahashi 2017). In one of these loops, the heterodimeric transcription activators (BMAL1/CLOCK and its paralogue BMAL1/NPAS2) activate transcription of the Period (Per) 1-3 and Cryptochrome (Cry) 1-2 genes, resulting in high levels of these transcripts. The resulting PER and CRY proteins then heterodimerize in the cytoplasm, translocate back to the nucleus, and interact with CLOCK/BMAL1 to inhibit transcription of Per and Cry genes. Subsequently, the PER/CRY repressor complex is degraded, and BMAL1/CLOCK can now activate a new cycle of transcription. In a second feedback loop, BMAL1/CLOCK activates the expression of orphan nuclear receptor genes Rev-erbα/β (Nr1d1/2). REV-ERB proteins, in turn, repress the expression of Bmal1, Clock, Npas2, Cry1 and Nfil3. ROR proteins, also orphan nuclear receptors, recognize the same DNA motif as REV-ERB proteins, compete with their binding, and activate the expression of the same target genes. In the last loop, BMAL1/CLOCK activates the expression of Dbp, while REV/ROR proteins activate and inhibit the expression of Nfil3, respectively. Both DBP and NFIL3 are transcription factors: DBP activates while NFIL3 represses the transcription of target genes, such as Rev-erbs, Rors, and Pers. These feedback loops constitute the molecular mechanism, as currently understood, of circadian rhythms in mammals (Takahashi 2017).
Several circadian transcriptome studies discovered a new RNA molecule, which we named Per2AS, that is transcribed within the Per2 locus in mouse liver, lung, kidney and adrenal gland (Koike et al. 2012; Menet et al. 2012; Vollmers et al. 2012; Fang et al. 2014; Zhang et al. 2014). In mammals, 25-40% of protein-coding genes have antisense transcript partners (Rosok and Sioud 2004; Katayama et al. 2005; Sun et al. 2006), and some antisense transcripts have been shown to exert functions in a variety of processes, such as cell cycle regulation, genome imprinting, immune response, neuronal function and cardiac function (Faghihi and Wahlestedt 2009; Khorkova et al. 2014; Wanowska et al. 2018). However, the physiological roles of most antisense transcripts remain uncertain.
In an earlier publication, we constructed mathematical models of Per2-Per2AS interactions, assuming that Per2AS and Per2 mutually inhibit each other’s expression in either pre-transcriptional (i.e., transcriptional interference) or post-transcriptional (i.e., in trans and degradation of RNA duplexes), or a combination of effects. The model predicted that all three mechanisms are consistent with the basic molecular details of circadian rhythms in mouse, but the pre-transcriptional model gives a more robust account of the circadian, anti-phasic oscillations of Per2 and Per2AS, compared to the post-transcriptional model (Battogtokh et al. 2018). To test predictions of these models, we have experimentally perturbed the expression of Per2AS both pre- and post-transcriptionally in this study. We found that Per2AS pre-transcriptionally represses Per2, however, Per2 positively regulates Per2AS post-transcriptionally. We also found that Per2AS regulates the amplitude of the circadian clock, although this effect is not solely dependent on its interaction with Per2. Overall, we propose that Per2AS serves as an important regulatory molecule not only to regulate the amplitude, but also to limit the level of Per2 within the oscillatory range. These conclusions are significant because Per2 is the only core clock gene for which the abundance, rhythmicity, and the phase of its expression are critical to maintain circadian rhythmicity in mouse (Chen et al. 2009).
Results
Characterization of Per2AS
Recent circadian transcriptome studies have identified a novel transcript, Per2AS, at the Per2 locus that is transcribed from the opposite strand, and expressed anti-phasic to the sense Per2 transcript in mouse liver (Koike et al. 2012; Menet et al. 2012; Vollmers et al. 2012; Fang et al. 2014; Zhang et al. 2014). Our quantitative PCR (qPCR) analyses demonstrated that the Per2AS is indeed expressed rhythmically and anti-phasic to Per2 mRNA in mouse liver, and it peaks at ZT 4 (Zeitgeber time, where ZT 0 and ZT 12 are defined as time of lights on and lights off, respectively) (Fig. 1A). Rhythmic and anti-phasic expression patterns of Per2AS and Per2 have been observed, as well, in NIH3T3 cells (Bmal1-luc) and mouse embryonic fibroblasts (MEFs) derived from PER2::LUCIFERASE knock-in mouse (Fig. 1A) (Yoo et al. 2004).
Per2AS is spliced and polyadenylated, similar to mRNAs (Figs. 1B, S1A). Rapid Amplification of cDNA Ends (RACE) analysis revealed that the transcription start site (TSS) of Per2AS is located in intron 6 of Per2 (Fig. 1B), which is consistent with the most 5’-signals of Per2AS (Fig. S1A) (Koike et al. 2012; Menet et al. 2012; Zhang et al. 2014). Strong and rhythmic recruitment of RNAPII-Ser5P (RNA polymerase II whose serine 5 in the C-terminal domain is phosphorylated) has been observed just upstream of Per2AS TSS, indicating an active and rhythmic initiation of transcription at this site (Phatnani and Greenleaf 2006) (Fig. S1A). RNAPII’s recruitment pattern also coincides with the expression pattern of Per2AS (Fig. S1B), indicating that Per2AS is transcribed by RNAPII.
Per2AS consists of at least three variants ranging in size from ~2000 to 3500 nt (Fig. 1B), while Northern Blot analysis yielded a strong signal around ~5000 nt (Fig. 1C). Given that the signal from Northern Blot was smeared, ranging from ~1500 to 5000 nt (Fig. 1C), and the Per2AS signals from circadian transcriptome studies were detected in a 10+ kb range (Fig. S1A) (Koike et al. 2012; Menet et al. 2012; Zhang et al. 2014), Per2AS presumably has many variants that are different in splicing patterns and transcription termination sites.
All three Per2AS variants that we detected by RACE have a potential to encode a small protein (the longest ORF common to all three is 291 nt; 97 amino acids); however, the potential polypeptide has no sequence similarity to any existing or predicted polypeptides in the GenBank protein database or the Conserved Domain Database. Furthermore, the Coding Potential Calculator (http://cpc.cbi.pku.edu.cn/) predicts that Per2AS does not encode a protein, as the coding score of Per2AS is much smaller (from −0.97 to −0.99 depending on variant) than that of Per2 (17.53). Furthermore, Per2AS transcripts are predominantly detected in the nucleus in mouse liver, NIH3T3 cells, and MEFs (Fig. 1D), similar to Neat1, a lncRNA known to localize in the nucleus (Clemson et al. 2009). In contrast, protein-coding transcripts, such as Gapdh and Per2, were localized largely in the cytoplasm in order for them to be translated (Fig. 1D). In addition, Per2AS is not bound to polysomes and therefore not actively translated, in contrast to Per2 that shows strong signals both in ribosome profiling and transcriptome data and those signals are higher at ZT16 compared to ZT4 (Fig. 1E, Fig. S1C) (Atger et al. 2015; Janich et al. 2015). The neighboring protein-coding transcript, Traf3ip1, that is transcribed from the same strand as Per2AS, also showed a clear signal both in ribosome profiling and transcriptome data, eliminating the possibility that lack of Per2AS signals in the polysome-bound fraction is due to its strand. These data collectively indicate that Per2AS is a long non-coding antisense transcript and does not produce a protein.
Per2AS regulates Per2 and the amplitude of the circadian clock
To understand the functional relevance of Per2AS in the mammalian circadian clock, we first used CRISPR technology to perturb Per2AS expression. We targeted the putative Per2AS promoter region, defined by the strong and rhythmic recruitment of RNAPII-Ser5P spanning approximately 900 bp (Fig. S1A-B). We introduced CRISPR mutagenesis in two independent cell lines: MEFs from PER2::LUCIFERASE (PER2::LUC) knock-in reporter mice, in which a luciferase gene is fused to the 3’-end of the endogenous Per2 gene (Yoo et al. 2004), and Bmal1- luc, an NIH3T3-derived luciferase-reporter cell line that has been stably transfected by a luciferase gene driven by the Bmal1 promoter (Morf et al. 2012). Following single cell sorting, we successfully isolated two PER2::LUC MEF (5D8 and 6F8) and one Bmal1-luc (mut8) mutant clones that had similar but distinct mutations at approximately 260 bp upstream of Per2AS TSS (Fig. S1D). Because these cell lines were near tetraploid on average (Fig. S1E), we not only identified the types of mutations, but also calculated the frequency of each mutant allele in each mutant (Fig. S1D). All the alleles were mutated in PER2::LUC 5D8 and 6F8, with which various deletions were detected (i.e., 11bp, 3bp, 2bp, and 1bp). Meanwhile in Bmal1-luc (mut8), only 29% of alleles had a mutation (6 bp deletion) and the remaining 71% were intact (Fig. S1D). We also targeted five other regions within the Per2AS promoter, including TSS and the TATA-boxlike sequence (TATAATCAA) located 63 bp upstream of the TSS; however, we were unable to isolate mutant clones, probably because these targeted regions were essential for cell survival or could not be easily accessed by the CRISPR machinery.
In these mutant cell lines, the level of Per2AS was upregulated to 138% (5D8) and 325% (6F8) compared to the control (parental) cell line, as opposed to our expectations that Per2AS would be downregulated. Nonetheless, the level of Per2 was reduced to 57% (5D8) and 70% (6F8), respectively (Fig. 2B). Similarly, in Bmal1-luc mut8, the Per2AS level was upregulated to 145%, while the Per2 level was reduced to 76% (Fig. 2F). These data indicate that the DNA region approximately 260bp upstream of the Per2AS TSS is important for Per2AS expression and that Per2AS represses Per2. Some long non-coding antisense transcripts modulate the expression of not only their target sense gene, but also neighboring genes (Halley et al. 2014; Villegas et al. 2014). However, this was not the case for Per2AS, as the levels of Ilkap, Hes6 and Traf3ip1, the three closest genes, remained unchanged in all of our mutants (Fig. S1F-G).
When we monitored the bioluminescent output from these mutant cells, the luminescent levels of both 5D8 and 6F8 were lower and their rhythmicity was less robust compared to the control (Fig. 2B, S2A-B). Quantification analyses further revealed that the amplitude of 5D8 and 6F8 was decreased to 25% of the control, whereas the period and phase remained unchanged (Fig. 2C). In contrast, the bioluminescence signal from Bmal1-luc mut8 was markedly higher (Fig. 2E, S2C-D), and the amplitude was increased to 204%, while the period and phase remained unchanged, compared to the control cell line (Fig. 2F). These data indicate that Per2AS regulates Per2, as well as the amplitude of the circadian clock. It is unlikely the observed phenotypes are due to an off-target effect, because mutants from two independent cell lines (three independent clones) with similar but distinct mutations all led to the same phenotype.
To further characterize the Per2AS mutants as well as to gain insights into the underlying mechanisms of circadian amplitude regulation, we next measured the mRNA expression patterns of 13 core clock genes after synchronizing these cells by serum shock (Balsalobre et al. 1998). We found that the mRNA expression of Bmal1, Cry1, and Rorα were elevated, while the expression of Per2 was decreased in all three mutants (Fig. 3). The up-regulation of Bmal1 mRNA level was consistent with the increased bioluminescence levels in Bmal1-luc cells (Fig. 2E). The mRNA expression of Nfil3 was up-regulated in PER2::LUC 5D8 and Bmal1-luc mut8, while that of Cry2 and Dbp was down-regulated only in PER2::LUC mutants (5D8 and 6F8) (Fig. 3). Interestingly, the expression of Npas2 was down-regulated in the PER2::LUC mutants but up-regulated in the Bmal1-luc mutant. No significant changes were observed for Rev-erbα and Rev-erbβ (Fig. 3). These data suggest that Bmal1, Cry1, Rora, Per2 (changed in all three mutants), but not Rev-erbα/β, are involved in the Per2AS-mediated amplitude regulation.
Transcripts of Per2AS do not play a major role in the circadian clock machinery
In contrast to coding genes whose functional unit is an encoded protein, the functional unit of lncRNAs is often unknown. It can be the transcript itself (i.e., RNA molecule and/or the small ORF embedded in the sequence that has the potential to encode a protein) that regulates target mRNAs post-transcriptionally by affecting splicing, mRNA stability, mRNA localization, or epigenetic marks (Wight and Werner 2013; Khorkova et al. 2014). It can also be the very act of transcription, in which transcription of one strand suppresses transcription of another in cis, a process called transcription interference (Wight and Werner 2013; Khorkova et al. 2014).
To distinguish whether the functional unit of Per2AS is either a transcript (i.e., post-transcriptional model) or the act of transcription (i.e., pre-transcriptional model) (Battogtokh et al. 2018), we post-transcriptionally decreased the level of Per2AS using “gapmers”, chimeric antisense nucleotides containing modified nucleic acid residues to induce RNase H-mediated degradation of nuclear-retained RNA (Lee et al. 2012), as the majority of Per2AS RNA remains in the nucleus in fibroblasts (Fig. 1D) and RISC-mediated RNA cleavage triggered by siRNAs occurs mainly in the cytoplasm (Carthew and Sontheimer 2009). We also targeted gapmers to exon 1 of Per2AS (i.e., intron 6 of Per2) that was shared by all three variants (Fig. 1B). Gapmers 5 and 8 successfully reduced the level of Per2AS to 61% in Bmal1-luc cells (Fig. 4A); however, the level of Per2 remained unchanged (Fig. 4A). Even though gapmers can not only induce RNA duplex-mediated degradation but also premature transcriptional termination of target mRNAs leading to reduced transcriptional activity and thereby confounding the interpretation of the results (Lee and Mendell 2020), this is unlikely the case for Per2AS gapmers, as the level of Per2 is unaffected. Unexpectedly, however, gapmer-mediated Per2AS knock-down led to an approximately 20% reduction in the Bmal1 level in Bmal1-luc cells (Fig. 4A).
We also post-transcriptionally overexpressed Per2AS using variant 2 (i.e., the longest variant) (Fig. 1B) in Bmal1-luc cells. This led to a marked increase in the Per2AS level by ~75,000-fold. Nevertheless, the Per2 level remained unchanged, while the Bmal1 level showed an approximate two-fold increase (Fig. 4B). Knock-down or overexpression of Per2AS did not alter period, phase, or amplitude of bioluminescence output in Bmal1-luc cells (Fig. 4C). The change in Bmal1 level was not due to comparable changes in the levels of Rev-erbα/β or Rorα, transcription repressor and activators of Bmal1, respectively (Sato et al. 2004) (Fig. 4A-B). Rather, Per2AS appears to directly regulate the Bmal1 transcription, as the increasing amount of Per2AS RNAs led to upregulation of reporter activity of Bmal1-luc, but not Per2-luc (Fig. 4D). Interestingly, however, the changes of Bmal1 by Per2AS RNA was also cell type-specific, as it was not observed in PER2::LUC MEFs, even though the efficiency of Per2AS knock-down was higher (gapmer 5: 63%, gapmer 8: 60%) in PER2::LUC MEFs, compared to Bmal1-luc cells (both gapmer 5 and 8: 39%) (Fig. S3). Overall, these data indicate that the transcript of Per2AS has little effect on the circadian control system (i.e., that its effects are not post-transcriptional). However, we also observed a small effect of Per2AS on Bmal1 at least in Bmal1-luc cells, suggesting that Per2AS post-transcriptionally regulates Bmal1 in a cell-specific manner.
Per2 knock-down downregulates Per2AS and does not replicate the phenotypes of Per2AS CRISPRI mutants
If Per2AS and Per2 indeed form a double negative feedback loop and mutually inhibit each other’s expression, then Per2 mRNA knock-down would result in up-regulation of Per2AS level in the post-transcriptional model but no changes in the pre-transcriptional model. To test this, we first reduced the level of Per2 post-transcriptionally using shRNAs. Our Per2 knockdown successfully reduced the level of Per2 to 65% in AML12 cells, a mouse hepatocyte cell line, and 45% in Bmal1-luc cells (Fig. 5A, F). Per2 knock-down also led to a decrease in the amplitude of bioluminescence output in PER2::LUC MEFs but not in Bmal1-luc cells (Fig. 5B-D). Period remained unchanged upon Per2 knock-down in both cells, despite that previous studies in murine fibroblast and hepatocyte cell lines reported that circadian period became shorter upon Per2 knock-down (Ramanathan et al. 2014). The residual Per2 level is higher in our system compared to those reports (45% vs 20%), and this could have contributed to the difference observed in our study. Nevertheless, Per2 knock-down resulted in a decrease of the Per2AS level to 55% (Fig. 5A), contrary to either of our expectations (i.e., pre-transcriptional or post-transcriptional interference).
We also measured the expression patterns of the core clock genes upon Per2 knock-down in Bmal1-luc cells to evaluate whether the changes observed in the Per2AS mutant cells (Fig. 3) were mediated entirely through Per2AS’s effect on Per2. Per2 knock-down did not result in changes of Bmal1 mRNA levels (Fig. 5F), consistent with the bioluminescent output from Bmal1-luc cells (Fig. 5D), but inconsistent with our observations in Bmal1-luc mut8 cells, in which Bmal1-luc bioluminescence and Bmal1 mRNA level were both significantly increased (Fig. 2E, 3). In addition, changes in mRNA expression patterns were observed only for Nr1d2 and Nfil3 upon Per2 knock-down (Fig. 5F), in contrast to the changes in the mRNA expression patterns of Cry1, Bmal1, Clock, Npas2, Rora, and Nfil3 observed in Bmal1-luc mut8 cells (Fig. 3). Interestingly, we did not observe any changes in mRNA levels of the core clock genes upon Per2 knock-down in AML12 cells (Fig. S4A), in which the level of Per2AS is approximately 100-fold higher compared to Bmal1-luc cells (Fig. S4B). As the level of Per2 was lower in Bmal1-luc with Per2 knock-down (45%) compared to Bmal1-luc mut8 cells (75%), it is highly unlikely that changes in the mRNA expression patterns observed in Bmal1-luc mut8 cells (Fig. 3) are solely due to the Per2AS’s effect on Per2. Rather, these data support the idea that Per2AS has a distinct role in the mammalian circadian system, independent of Per2.
Discussion
There is only limited experimental evidence about the functions and regulatory mechanisms of antisense transcripts, particularly in mammals. Our study contributes to an understanding of the possible functional roles of antisense transcripts by demonstrating that Per2AS, a natural antisense transcript to a core clock gene Per2, regulates the amplitude of the circadian clock. Although the precise regulatory mechanism remains unclear, we think Per2AS does not solely rely on Per2 for its functions. Per2 knock-down leads to decreased levels of RORE-controlled genes, such as Bmal1 and Nfil3 (Shearman et al. 2000; Schmutz et al. 2010; Ramanathan et al. 2014); however, Per2AS mutants that upregulate antisense transcription do not exhibit these changes. Of potential interest in understanding the amplitude-regulatory mechanism of Per2AS would be Rora, as Rora is one of the core clock genes whose expression is elevated in all the Per2AS mutants in our study (Fig. 3). Interestingly, Ror genes have been shown to increase the circadian amplitude of molecular and behavioral rhythms (He et al. 2016). Furthermore, our recent analysis suggested that Rorc and Per2AS are potential amplitude regulators of circadian transcriptome output, as the levels of Rorc and Per2AS correlate with the percentage of rhythmic transcripts in various mouse tissues (Littleton and Kojima 2020). Notably, existing PER2 null animals all lack the Per2AS locus (Zheng et al. 1999; Bae et al. 2001), and thus complicate the interpretation of the data from these animals in understanding whether the observed phenotypes are solely due to Per2, or a combined effect of Per2AS and Per2.
The experimental observations from this study bear on the assumptions underlying our mathematical models of Per2-Per2AS interactions (Battogtokh et al. 2018). In building our models, we considered that Per2AS and Per2 mutually inhibit each other’s abundance either pre-transcriptionally or post-transcriptionally, or a combination of both effects (Battogtokh et al. 2018). Our experimental interrogation clearly demonstrated that Per2AS represses Per2 (Fig. 2A, D), presumably via a pre-transcriptional mechanism, as the knock-down or overexpression of Per2AS transcripts did not alter the level of Per2 mRNAs (Fig. 4, 6). Contrary to our expectation, however, Per2 positively regulates Per2AS, as the knock-down of Per2 led to downregulation (rather than upregulation) of Per2AS (Fig. 5A). This effect is post-transcriptional, because the level of Per2 pre-mRNA remained unchanged upon Per2 knock-down (Fig. S4C). We think it is unlikely that Per2 RNA and Per2AS RNA post-transcriptionally form an RNA duplex and stabilize each other, because (1) the expression of Per2AS and Per2 are anti-phasic in at least some tissues (Fig. 1A) (Zhang et al. 2014), (2) Per2 is ~25x more abundant than Per2AS (Fig. 1A) (Koike et al. 2012), (3) Per2 predominantly localizes in the cytoplasm while Per2AS localizes in the nucleus (Fig. 1D), and (4) Per2AS consists of many variants and their nucleotide sequences vary (Fig. 2E). Rather we favor the hypothesis that PER2 indirectly or directly regulates Per2AS transcription. In fact, REV-ERBα/β and BMAL1/CLOCK/PER1/PER2/CRY2 are recruited to the vicinity of the Per2AS TSS (Koike et al. 2012), potentially involved in regulating Per2AS transcription.
These experimental observations clearly indicate that our mathematical models need to be revised to better understand the functions and regulatory mechanisms of Per2AS. Specifically, we need to alter the description of the Per2AS-Per2 relationship, include the effect of Per2AS on Bmal1, and add the potential effect of circadian transcription factors on Per2AS transcription (Fig. 6). Nevertheless, our original model predicted that the region of circadian oscillations is greatest in the pre-transcriptional model (Battogtokh et al. 2018). This conclusion is also true in a simpler mathematical model, involving only the core negative feedback loop, as we show in Fig. S5.
Our mathematical models were also constrained by the requirement that the expression of Per2AS and Per2 are both rhythmic and anti-phasic (180° out of phase). These patterns have been observed not only in liver but also in adrenal gland, lung and kidney (Zhang et al. 2014). Analysis of mouse ENCODE datasets demonstrated that Per2AS is also expressed in many other tissues, such as genital and subcutaneous fat pad, bladder, kidney, colon, duodenum, large and small intestines, stomach, lung, and potentially in mammary gland, ovary and testis (Consortium 2012; Davis et al. 2018) (Fig. S6A), although it is unclear whether Per2AS expression is rhythmic and antiphasic to Per2 in these tissues. A Per2AS signal has not been detected in the central nervous system and a few other peripheral tissues including heart, placenta, spleen, and thymus (Consortium 2012; Davis et al. 2018) (Fig. S6A). These results could be due to tissue sampling times, i.e., little or no Per2AS may have been expressed when these tissues were harvested, or the Per2AS level was under detection threshold, as the expression of lncRNAs is generally low (more than 10-fold lower than sense transcripts on average) and transcriptome analyses sometimes lack the sensitivity to detect all lncRNAs (He et al. 2008; Faghihi and Wahlestedt 2009; Xu et al. 2009; Xu et al. 2011; Djebali et al. 2012). An antisense transcript of Bmal1 was also detected in a few tissues, such as subcutaneous fat, bladder, kidney, colon, lung, and potentially in duodenum, stomach, mammary gland, and ovary (Fig S6B) (Consortium 2012; Davis et al. 2018), as was also reported previously (Zhang et al. 2014). It would be of great interest to analyze whether the antisense transcript of Bmal1 exert some functions in the mammalian circadian clock system.
Antisense transcripts of a core clock gene have also been reported in other organisms. In Neurospora, the sense frequency (frq) and antisense (qrf) transcripts are both located on the same chromosome and overlap almost completely (Kramer et al. 2003; Xue et al. 2014). In Antheraea, the sense-antisense transcripts for the homologue of Drosophila’s period (per) gene are located on different chromosomes (Sauman and Reppert 1996). Interestingly, human PER2 also has an antisense transcript (PER2AS), but, based on human ENCODE datasets, the transcription of PER2AS and PER2 diverges from their respective TSS, most likely using a bidirectional promoter (Consortium 2012; Davis et al. 2018) (Fig. S6C). Although it is unclear whether all these sense-antisense transcripts have rhythmic and antiphasic expression patterns, the evolutionary conversation of sense-antisense pairs of a core clock gene suggests that antisense transcripts are part of a common mechanism for circadian clock regulation.
Even though the expression of qrf is rhythmic and anti-phasic to frq, similar to Per2AS and Per2 (Kramer et al. 2003; Xue et al. 2014), their functions in the core clock machinery appear to be different. While the primary function of qrf is to regulate phase and light entrainment using both pre- and post-transcriptional regulation (Kramer et al. 2003; Xue et al. 2014), primary function of Per2AS appears to regulate amplitude using a pre-transcriptional mechanism (Fig. 2). Because the molecular clock circuitry is markedly more complex in mammals than Neurospora, it is possible that Per2AS acquired additional or different functions, such as its apparent ability to regulate Bmal1 in trans.
Per2 is the only core clock gene whose rhythmicity, proper phase, and expression levels are all critical to sustain rhythmicity (Zheng et al. 1999; Bae et al. 2001; Chen et al. 2009). Per2 is also under multiple layers of regulation both transcriptionally and post-transcriptionally to sustain robust rhythmicity (Yoo et al. 2005; Chen et al. 2013). This study demonstrates that Per2AS serves as an additional regulatory mechanism, to further ensure that Per2’s expression pattern stays within a certain window for sustaining robust circadian rhythmicity. It would be of great future interest to explore the functions of Per2AS at the tissue and organismal levels.
Materials and Methods
Tissue harvesting and Cell culture
Male C57BL/6J mice were maintained on a 12:12 LD cycle and fed ad libitum, then livers were collected at the time indicated. All the animal experiments were conducted and the protocols were approved by the Institutional Animal Care and Use Committees at the University of Texas Southwestern Medical Center.
NIH3T3, HEK293/T17, Mouse Embryonic Fibroblasts (MEFs), PER2::LUC MEFs, and Bmal1-luc cells were grown in Dulbecco’s Modified Eagle Medium (Life Tech) with 10% fetal bovine serum (FBS) (ATLANTA biologicals) at 37°C with 5% CO2. AML12 cells were grown in Dulbecco’s Modified Eagle Medium/F12 (1:1) with 10% FBS and 1% Insulin-Transferrin-Selenium supplement (Gibco) at 37°C with 5% CO2.
Chromosomal number was counted as previous reported (Nicholson et al. 2015) with minor modifications. Briefly, cell cultures were incubated in 1000 ng/ml Colcemid (Karyomax, Invitrogen) at 37°C for 3–6 hr to enrich in mitotically arrested cells. The cells were then collected by centrifugation, and pre-warmed hypotonic solution (0.075M KCl) was added dropwise to the cell pellet. After incubated for 18 min at 37°C, cells were fixed with an ice-cold 3:1 methanol:acetic acid solution for 5 min followed by centrifugation. After repeating this last step twice, fixed cells were dropped on microscope slides.
CRISPR mutagenesis
The sgRNAs were designed by CRISPR design tool (http://crispr.mit.edu), and complementary oligonucleotides were annealed, phosphorylated and cloned into the BbsI sites of the pSpCas9(BB)-2A-GFP (Addgene plasmid # 48138) (Ran et al. 2013). After the nucleotide sequences of the plasmids were verified by Sanger sequencing, DNA was transfected into PER2::LUC MEFs and/or Bmal1-luc cells using FuGENE6 according to the manufacturer’s instructions. After 48 hrs, cells were trypsinized and sorted using FACSAria I (BD Biosciences) with GFP signals. Subsequently, cells were cultured in DMEM supplemented with 10% FBS and 1x Penicillin/Streptomycin (Gibco). When each clone reached confluency, cells were lysed with 90 μL of 50 mM NaOH and 10ul of 100 mM Tris–HCl (pH 8.0), followed by incubation at 95C for 10 min. The mutated genomic region was first amplified by PCR, then cloned into pGEM-T vector (Promega), to reveal the nucleotide sequence of each clone by Sanger sequence. All the primer sequences used in this study can be found in Supplemental Table S1.
Real-time bioluminescence recordings and luciferase assay
Cells were first plated into 35 mm dishes and then allowed to become confluent. The medium was then replaced with DMEM supplemented with 50% horse serum (gene expression analyses) or dexamethasone (1 μM) was added to the media (real-time bioluminescence recordings) for 2 hr. After synchronizing cells, medium was changed to phenol red-free DMEM (Cellgro 90-013-PB) supplemented with 100 μM luciferrin, 10 mM HEPES pH 7.2, 1 mM sodium pyruvate, 0.035% sodium bicarbonate, 2% FBS, 1x Penicillin/Streptomycin, and 2 mM L-glutamine. Real-time bioluminescence recordings were performed using a LumiCycle (Actimetrics, Inc. Wilmette, IL). Quantification analyses was performed by JMP software (Oklejewicz et al. 2008). The recordings from the first 24 hrs were eliminated for quantitative analyses because these data are unreliable.
Luciferase assays were performed as previously reported (Kojima et al. 2010). Briefly, a mixture of plasmid DNAs containing 100 ng Per2E2-luc or Bmal1-luc firefly luciferase reporter genes, 10 ng Renilla luciferase reporter genes, and increasing amounts of Per2AS (variant2) - expressing plasmids (50, 100, 200, and 400 ng) were co-transfected in NIH3T3 cells. Luciferase activities were measured approximately 48 hrs after transfection. Per2E2-luc reporter gene was constructed by inserting the mouse Per2 promoter (−83 to +156 in reference to the Per2 TSS) into pGL4.12[luc2CP] (Promega), whereas Bmal1-luc was constructed by inserting the mouse Bmal1 promoter (−779 to +127 in reference to the Bmal1 TSS) to pGL4.11[luc2P] (Promega).
RT-qPCR
Total RNA was extracted with TRIZOL reagent (Life Tech) according to the manufacturer’s instructions and treated with TURBO DNaseI (Life Tech). RNAs were then subjected to reverse transcription using SuperScript II (Life Tech) or High Capacity cDNA Reverse Transcription Kits (Applied Biosystems). For Per2AS transcripts, cDNA was synthesized using strand-specific primers (Table S1), except for Fig. 1A, in which the oligo(dT) was used. qPCR was performed using QuantiStudio 6 (Life Tech) with SYBR Power Green (Applied Biosystems). All the primer sequences used in this study can be found in Table S1.
Rapid amplification of cDNA ends (RACE) assays
RACE assay was performed with SMARTer RACE 5’/3’ kit (Clontech) according to the manufacturer’s instructions. Total RNAs from mouse liver (C57BL/6J) harvested at ZT 4 was subjected to 5’- and 3’-RACE cDNA synthesis. Each cDNA was then amplified by Per2AS specific primers (Table S1) together with the Universal Primer (Clontech). The first PCR products were subsequently amplified by the nested Per2AS specific primers with Universal Primer (Clontech). The nested PCR products were cloned into pGEM-T vector (Promega) and the nucleotide sequence was determined by Sanger sequencing. Two independent 5’RACE products had the identical sequence, while four 3’RACE products yielded three different sequences, giving rise to the three variants of Per2AS (Fig. 1B, Fig. S1C).
Northern Blotting
Total RNA (~500ug) extracted from mouse liver harvested at ZT 4, at which Per2AS expression is the highest (Koike et al. 2012), was subjected to poly(A)+tract isolation kit (Promega). Poly(A)+ enriched RNA, as well as RNA ladder (Life Tech), was further separated on a 1% agarose gel, transferred to Hybond-N+ membrane (Amersham), and UV cross-linked (Stratagene) using a standard protocol. The membrane was further hybridized at 68°C for overnight using PerfectHyb PLUS (Sigma) with a P32UTP-labeled probe that has a complementary sequence against Per2AS (1-1812 nt, Fig. 1B, Fig. S1C), where all the variants share its nucleotide sequences. After extensive washing of the membrane, the radioactive signal was analyzed with a Storm image analyzer (GE Healthcare).
Subcellular fractionation
Livers from mice (C57BL/6J) were immediately rinsed with ice-cold 0.9% NaCl solution, then homogenized on ice with 5 volumes of homogenization buffer (10 mM Tris-HCl buffer, pH 7.7, containing 10 mM NaCl, 0.1 mM EGTA, 0.5 mM EDTA, 0.5 mM spermidine, 0.15 mM spermine, 0.5% Tergitol NP-10, 1 mM dithiothreitol, and 1 mM phenylmethylsulfonyl fluoride) in a Dounce homogenizer. The homogenates were filtered through two layers of cheesecloth and, following dilution with 8 tissue volumes of 2.2 M sucrose in the homogenization buffer, applied to a 10-ml cushion of 2 M sucrose in homogenization buffer and spun at 24,000 rpm for 60 min at 2 °C in a prechilled SW28 rotor. The resultant nuclear pellet as well as supernatant were subjected to RNA extraction. For NIH3T3 and MEFs, cells were lysed in an ice-cold Hypotonic Lysis buffer (10mM Tris-HCl, pH 7.4, 10mM NaCl, 3mM MgCl2, and 0.3% NP-40) supplemented with protease inhibitor cocktail (Sigma). After incubated for 5 min on ice with occasional pipetting, lysates were separated to nuclear and cytoplasmic fractions by centrifuging 600xg for 5 min at 4°C. RNAs were extracted from each fraction using TRIZOL reagent (Life Tech) from each fraction and subjected to RT-qPCR.
Per2AS overexpression and knockdown
Per2AS variant 2 overexpression plasmid was generated from 5’RACE and 3’RACE products that were cloned in to pBluescript (Agilent), as well as Per2AS qPCR product cloned into pGEM-T (Promega). After combining 5’RACE and qPCR products in pBlueScript using Hind III and Xho I sites, the 3’RACE product was also inserted using Xho I and Kpn I sites. The full length was then transferred to pcDNA3.1-mycHis (Invitrogen) and its sequence was verified by Sanger sequencing. DNA transfection was performed with either FuGENE6 Reagent (Promega) or Lipofectamine 2000 (Life Tech) using Opti-MEM (Gibco), while gapmers were introduced via gymnosis (Stein et al. 2010) with the final concentration of 100nM (Table S1).
Lentivirus generation and transduction
For virus production, either control or Per2 shRNA as well as viral packaging vectors were transfected into HEK293/T17 cells according to the manufacturer’s instruction (ViraPower Lentiviral Expression System, Life Tech). Cell culture media was collected after 48 hours, ultracentrifuged at 70,000xg for 2 hours at room temperature, then resuspended in target cell-specific media. Viral media and 20μg/ml polybrene (Millipore) were added to target cells, and cells were harvested 48-72 hours after viral transduction.
Mathematical Model
Our model of Per2-Per2AS interactions (Battogtokh et al. 2018) was based on an earlier model (Relogio et al. 2011) of the mammalian circadian clock, which did not take Per2AS into account. Our model of the basic negative feedback loop, whereby PER2:CRY inhibits BMAL:CLOCK (without additional feedback loops through REV-ERB and ROR), consisted of 15 ordinary differential equations (ODEs) with 44 parameter values (rate constants, binding constants, etc.). To illustrate some of the results of this detailed model, we present here a simpler model of the negative feedback loop (Fig. S5A) is based on Goodwin’s original model (Goodwin 1965), with an important modification suggested by Bliss, Painter and Marr (Bliss et al. 1982). Per2AS RNA was added to the basic Goodwin model by assuming that Per2AS and Per2 mutually inhibit each other pre-transcriptionally. The ODEs describing the model are presented in Fig. S5B. The model with Per2AS consists of all four ODEs, as written. The model without Per2AS consists of the first three ODEs only, and the blue factor in the first ODE is set = 1. These two models were simulated with the XPP-AUT program (http://www.math.pitt.edu/~bard/xpp/xpp.html) using the parameter values given in Table S2. The parameter values were chosen to give oscillations with a period close to 24 hours and reasonable amplitudes and phase relations of the variables (Fig. S5C. In particular, the model with Per2AS was constrained to fit experimental observations in mouse liver that the level of Per2AS is approximately 5% of Per2, and that Per2AS and Per2 are expressed anti-phasically (Koike et al. 2012). In Fig. S5D we compare the robustness of oscillations in the models with and without Per2AS, in terms of the range of a1 values (the maximum rate of synthesis of Per2 mRNA) over which the models exhibit oscillations. By this measure, the model with Per2AS is considerably more robust than the model without Per2AS. In Fig. S5E, we show the domains of oscillation for both models on a two-parameter diagram: a1 and b3, where b3 is the maximum rate of degradation of phosphorylated PER2 in the nucleus. This diagram also shows a larger region of oscillations in the model with Per2AS. Even if we restrict our attention to the regions of “circadian” oscillations (22 < period (h) < 26) in parameter space, the model with Per2AS (blue lines in Fig. S5E) is noticeably more robust than the model without Per2AS (black lines).
Acknowledgement
We thank Dr. Ueli Schibler (Université for Genève) for providing Bmal1-luc cell line, Dr. Seung-Hee Yoo (University of Texas Health Science Center at Houston) for PER2::LUC MEFs, and Dr. Andrew Liu (University of Florida) for Per2 shRNA vectors (Ramanathan et al. 2014). We also thank Melissa Makris (Flow Cytometry Resource Laboratory, Department of Biomedical Sciences and Pathobiology, Center for Molecular Medicine and Infectious Diseases, Virginia-Maryland Regional College of Veterinary Medicine, Virginia Tech), Drs. Nicolaas Baudoin and Daniela Cimini (Department of Biological Sciences, Virginia Tech), and Tsubasa Toda and Akari Ueta (Department of Environmental and Life Sciences, Toyohashi University of Technology) for technical assistance. The authors also thank all the past and current members of the Green, Takahashi, Kojima laboratories for invaluable discussion. This work was supported by the Luther and Alice Hamlett Undergraduate Research Support (to C.T.S), Howard Hughes Medical Institute (J.S.T), and National Institutes of Health GM127122 (to C.B.G), and GM126223 (to S.K.). J.S.T. is an Investigator in the Howard Hughes Medical Institute.