We sought to identify circadian regulation of alternative splicing in mouse liver. The circadian phenotype of liver provides a paradigm for both the circadian control of metabolism and the regulation of a peripheral oscillator by the master-clock in the SCN of the hypothalamus 13. Thus, wild-type male mice were kept under 12 h:12 h light:dim red light (LD), in which they displayed the expected daily nocturnal pattern of locomotor activity (Figure 1a). This pattern continued when the animals were placed into constant dim red light (DD; Figure 1a), confirming their circadian behaviour. In the second day of DD, animals were harvested at circadian time (CT) 0, 6, 12 and 18 and their livers extracted for genome-wide analysis of gene expression and alternative splicing using the Affymetrix exon-array. The exon-arrays faithfully reproduced the temporal expression patterns of the canonical circadian genes Dbp and Arntl (Bmal1), as measured by quantitative PCR (QPCR; Figure S1 in Additional file 1), confirming these samples as rhythmic at the molecular level.

In order to identify circadian alternative exons, we analyzed the exon-array data using FIRMA (Finding Isoforms using Robust Multichip Analysis). FIRMA predicts alternative splicing in a single sample as a departure from the transcript-level expression estimate produced by the Robust Multichip Analysis (RMA) model 27. Thus, for each exon-probeset at each time point, a score is calculated that measures this departure (the FIRMA score), a score of 1 indicating no departure. FIRMA scores were then compared across circadian time by one-way ANOVA. Exon-probesets showing a significant effect of time (false discovery rate (FDR) corrected P < 0.05) were then screened to identify circadian patterns (see Materials and methods). This identified 55 exon-probesets from 47 genes with significant circadian variation in FIRMA score (Additional file 2). This corresponds to approximately 0.4% of the genes detectable on the array. Clustering of the circadian profiles of these exon-probesets revealed a variety of different phases (Figure 1b), suggesting that circadian regulation of alternative splicing occurs at all stages of the circadian cycle.

In order to examine the functions of a putative circadian splicing network, we sought to identify enriched functional annotations associated with the genes containing circadian exons. Despite the comparatively small number of genes examined, this revealed enrichments within the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways 28 representing the circadian clock itself, drug detoxification, caffeine and retinol metabolism and the peroxisome proliferator-activated receptor (PPAR) signaling pathway. These KEGG associations were supported by enrichments in Gene Ontology terms 29 and InterPRO domains 30 pertaining to the same processes, and the functions of the genes compromising them (Additional files 3, 4, and 5). Interestingly, hepatic drug metabolism was previously shown to be under circadian transcriptional control 31, suggesting that circadian phenotypes in peripheral organs are specified by control of gene expression at multiple levels.

Circadian exons were validated in a separate animal experiment with greater temporal resolution (every 3 hours) and sensitivity (n = 6). The proper rhythmic expression of Dbp, Per2 and Arntl was confirmed in liver, lung and kidney (Figure S2 in Additional file 1). Validation was performed by exon-specific QPCR normalized to gene expression. We selected 25 exon-probesets from across the four FIRMA score phase clusters (Figure 1b). Of these 25, 13 were confirmed as rhythmic in liver (P < 0.05, one-way ANOVA for effect of time; Figure 2; Figure S3 in Additional file 1). These exons comprised nine cassette exons (Pcsk4, Npas2, Usp2, Fbxo21, Nr1d1, Ash2l, Loxl4, Adrbk2 and Clock), two alternative initial exons (Hlf and Cabc1) and two alternative polyadenylation sites (Pnp1 and Decr1). In 10 of 13 cases, the FIRMA score correlated either positively or negatively with the exon-specific QPCR results (r² ≥ 0.6).

Tissue-dependent alternative splicing is most often regulated independently of transcript-level expression, such that tissue-specific exons are rarely present in genes with transcript-level regulation in the same tissue 7. In contrast, the vast majority of our validated exons were present in genes with transcript-level circadian regulation in liver (Figure 2; Figure S3 in Additional file 1). For example, Pcsk4 circadian alternative splicing displayed a large peak at CT12, and its temporal pattern was highly concordant with that of the FIRMA score (Pearon's r = 0.99; Figure 2b, left panel). Strikingly, Pcsk4 transcript-level expression was also rhythmic, but with a different phase (peaking at CT9) and amplitude (Figure 2b, right panel). The exception was Decr1, where the exon-specific expression (in this case, an alternative portion of the 3' UTR) was rhythmic, but the transcript-level expression was not (Figure S3j in Additional file 1). In spite of the evident co-regulation of circadian alternative splicing and transcript-level expression, we could observe no systematic relationship between the acrophases (circadian time of peak expression) of alternative splicing and transcript-level changes across all exons (Figure 2c). Thus, it would appear that these two modes of gene expression are regulated independently in our validated exons.

The regulation of alternative splicing is frequently tissue-dependent 2. Moreover, the same is also true for circadian transcription 32. To examine the tissue dependence of circadian alternative splicing, we analyzed some of our validated exons in lung and kidney (Figure 3; Figure S4 in Additional file 1). An example of the effects observed is illustrated with Pcsk4 (Figure 3a). The amplitude of the rhythm in alternative splicing observed in the liver was much greater than that observed in the kidney, while alternative splicing in the lung was essentially arrhythmic (Figure 3a, left panel). Indeed, seven out of nine exons showed a splicing change of greater amplitude in the liver compared to the lung and kidney (Figure 3b, right panel). Furthermore, the decrease in amplitude of alternative splicing from liver to kidney was not associated with an equivalent decrease in the amplitude of the rhythm in the transcript-level expression, while, similarly to the alternative splicing, transcript-level expression in the lung was also essentially arrhythmic (Figure 3a, right panel). Strikingly, the acrophases of the splicing changes were also frequently tissue-dependent (Figure 3b, left panel). All events showed a tissue-dependent difference in splicing acrophase. Two of the nine events (Hlf and Fbxo21) were entirely liver specific, being arrhythmic or entirely absent from lung and kidney. Of the remaining seven, three were different in both tissues compared to liver, whereas four were only different in one of the two tissues (Figure 3b, left panel). For these seven exons the individual time-differences in acrophase between liver and each of lung and kidney were examined. Seven time-differences were of 3 hours and two were of 9 hours, while four had no difference in acrophase. Taken together, these data indicate that both the acrophase and amplitude of circadian alternative splicing is tissue-dependent.

The tissue-dependence of the temporal relationship between alternative splicing and transcript-level changes grouped our validated exons into two categories. For seven out of nine exons, this temporal relationship was not consistent across tissues. Either the transcript-level regulation was present in all tissues and circadian alternative splicing was present predominantly in liver - for example, Npas2 (Figure S4 in Additional file 1h) - or a variety of patterns in both alternative splicing and transcript-level expression was observed - for example, Nr1d1 (Figure S4 in Additional file 1e) and Cabc1 (Figure S4 in Additional file 1g). By contrast, two out of nine exons had a temporal relationship between alternative splicing and transcript-level changes that was preserved across tissues. For example, Usp2 (Figure S4 in Additional file 1a) showed an approximate 6 hour delay between the acrophases of alternative splicing and transcript-level expression, which was preserved in kidney and lung. In conclusion, we have observed that the circadian regulation of alternative splicing is tissue-dependent, in terms of both phase and amplitude. Moreover, we have identified exons where the temporal relationship between alternative splicing and transcript-level expression appears to be preserved across tissues, suggesting that these two processes may be coupled in these particular cases. Thus, as is the case for circadian transcription, circadian regulation of alternative splicing adds an important temporal dimension to the gene expression profile of a tissue.

We sought to investigate the physiological pathways regulating circadian alternative splicing. The timing and availability of food can affect circadian patterns of mRNA abundance in the liver, including those of the core clock genes, independently of the master clock of the SCN 333435. This ability of the liver to be independently entrained by food illustrates the importance of circadian timing in the liver's role in the regulation of energy metabolism and xenobiotic detoxification 13. In order to understand the relevance of this effect to the regulation of circadian alternative splicing, we examined the effect of fasting on the circadian alternative splicing of a number of our validated exons in liver.

Wild-type mice kept under LD were fed ad libitum until the night of day 1 of the experiment (Figure 4a). At ZT15 (Zeitgeber time; the time of day under LD conditions), food was removed or a mock food removal implemented to control for effects of disturbing the animals during the food removal 36 (Figure 4a). At the beginning of day 2, rather than the lights coming on, the mice remained in DD. Mice both with and without food were then harvested at CT5 and CT15 on day 2, and their livers extracted for analysis (Figure 4a). These times were chosen as they defined a selection of our validated exons with robust differences in alternative splicing and transcript-level expression (the exons identified within Npas2, Hlf, Usp2, and Pcsk4). Thus, we examined the response to fasting of circadian alternative splicing and transcript-level expression of these four exons. All four exons recapitulated the circadian pattern observed previously in the presence of food (Figure 4b). Under fasting conditions, however, three of the four exons showed a significant decrease in the difference between the two time points (Npas2, Hlf and Pcsk4; significant main effect of time and significant interaction effect between time and feeding by two-way ANOVA (P ≤ 0.05)). Moreover, whereas a lesser temporal difference persisted with Npas2 and Hlf, no significant difference was apparent for Pcsk4 under fasting conditions (P ≥ 0.05, Bonferroni's post hoc test). The circadian difference in Usp2 alternative splicing, however, was resistant to the fasting conditions (Figure 4b, inner right panel; significant main effect of time by two-way ANOVA (P ≤ 0.05), but no significant interaction effect between time and feeding (P ≥ 0.05)). Thus, the absence of food modulates the circadian regulation of alternative splicing in an exon-dependent manner.

The transcript-level changes observed in the presence of food faithfully recapitulated those observed previously (Figure 4c). Strikingly, however, changes in response to fasting at the transcript level did not always mirror those in alternative splicing (Figure 4c). Similar to its alternative splicing, Usp2 transcript-level expression was also resistant to fasting conditions, showing no significant change in the circadian difference in expression (significant main effect of time by two-way ANOVA (P ≤ 0.05), but no significant interaction effect between time and feeding (P ≥ 0.05)). Furthermore, Hlf transcript-level expression behaved similarly to its alternative splicing, being reduced by the fasting conditions. However, while some temporal difference in alternative splicing remained, the difference in transcript-level expression became insignificant (P ≥ 0.05, Bonferroni's post hoc test). Given the concurrent change in both alternative splicing and transcript-level expression in response to the fasting conditions for this gene, this case would be consistent with some form of coupling between transcription and splicing. However, while Npas2 and Pcsk4 both showed a significant change in alternative splicing, neither showed a significant change in transcript-level expression (significant main effect of time by two-way ANOVA (P ≤ 0.05), but no significant interaction effect between time and feeding (P ≥ 0.05)), despite an apparent downward trend in Pcsk4. This would suggest that fasting decouples any relationship between transcription and splicing that may exist for these exons. Taken together, the molecular response to fasting in liver appears to be complex: not only do fasting conditions modulate circadian alternative splicing in an exon-dependent manner, but they also modulate the temporal relationship between circadian alternative splicing and circadian mRNA abundance in a gene-dependent manner.

Besides the influence of direct entraining stimuli such as feeding, rhythmic gene expression in a peripheral organ can be controlled by both the local circadian clock and systemic signals emanating from the master clock of the SCN 37. To discern whether our validated exons were being influenced by such systemic signals, we examined their expression in mice lacking the vasoactive intestinal peptide receptor 2 (Vipr2^-/-). The neuropeptide Vip and its receptor Vipr2 are essential for the proper functioning of the SCN network; the SCN of mice lacking Vipr2 display disorganized, low amplitude rhythms in both gene expression and electrical activity 38. Behaviorally, Vipr2^-/- mice appear rhythmic under LD, but when released into DD they show a phase-advance in locomotor activity and may become arrhythmic 39. Remarkably, those mice that exhibit this locomotor phase-advance also display phase-advanced and fully rhythmic clock gene expression in liver. Thus, if alternative splicing and transcript-level expression of our validated exons were phase-advanced, we could consider that they are under the control of the local liver clock.

Phase-advanced male Vipr2^-/- mice and wild-type controls in the second day of DD were harvested every 5 hours for one day (CT0, CT5, CT10, CT15 and CT20) and their livers extracted. Expression of core clock genes was phase-advanced by 5 to 10 hours (Figure 5a), as described previously 39. Thus, we examined the alternative splicing and transcript-level expression of a selection of our validated exons that had previously shown robust changes around these time points. For Npas2, Hlf and Usp2, both alternative splicing (Figure 5b) and transcript-level expression (Figure 5c) were phase-advanced by 5 to 10 hours, indicating that both mRNA levels and alternative splicing are under the control of the local liver clock. The phase-relationships between alternative splicing and transcript-level expression observed previously were maintained in the phase-advanced context. However, this was not the case for Pcsk4; while alternative splicing was phase-advanced by 5 hours, transcript-level expression, which was circadian in wild-type mice, now appeared to be arrhythmic (Figure 5c, far right; P > 0.05, one-way ANOVA for effect of time). Although this may be an artifact of our sampling frequency, it suggests that Pcsk4 alternative splicing, but not transcript-level expression, is dependent on the local liver clock. Thus, we conclude that all of our tested exons are under the control of the local liver clock, but that Pcsk4 appears to illustrate a class of gene that integrates local and systemic circadian cues at different levels of gene expression.

As discussed, exon inclusion can be regulated by a variety of mechanisms, and amongst these is the expression and activity of the families of auxiliary splicing factors. Although the expression and activity of splicing factors can be regulated by a variety of mechanisms, we hypothesized that some splicing factors undergo circadian changes in mRNA expression. To examine whether this is the case, we examined the expression of known and putative splicing factors, as defined by Grosso et al. 40, in a high temporal resolution microarray study of circadian mouse liver 41. Of the 227 splicing factors, 62 had corresponding probesets judged to be rhythmic. Their circadian expression patterns showed a variety of phases, but with clear clusters of factors that peaked in expression during the subjective day and the subjective night (Figure 6a). Furthermore, these 62 were significantly more likely (P = 0.029, Fisher's exact test) than a random selection of 62 non-rhythmic splicing factors to contain one or more of the consensus motifs corresponding to circadian transcription factor bindings sites (E-box, E'-box, D-box and Rev response element (RRE)) in their proximal promoter region (Additional file 6), further supporting their classification as circadian.

To identify robustly circadian splicing factors, we sorted the microarray expression profiles based on their signal-to-noise ratio and the reproducibility observed between the two days. We examined the expression of the 33 best candidates by QPCR. We then compared our QPCR data to the microarray data to see whether the two temporal patterns correlated significantly (P ≤ 0.05, Pearson's r); this yielded a shortlist of 12 factors (Figure 6b; Additional file 7; Figure S5 in Additional file 1). To expand our analysis, we selected three additional genes (Pcbp2, Srsf3 and Srsf5), bringing our shortlist to 15. These additional genes possessed significant circadian expression by QPCR in our samples (P ≤ 0.05, one-way ANOVA for effect of time), but did not correlate significantly with the data from Hughes et al. 41 (Figure S5 in Additional file 1j, l). Our short-list contained a variety of genes. They include well characterized regulators of alternative splicing (Srsf3, Srsf5, Tra2b and Khdrbs1 (Sam68)), a component of U2 snRNP (Sf3b1), two RNA helicases (Ddx46 (also a component of U2 snRNP) and Dhx9), three hnRNP proteins better known for their roles regulating RNA stability and translation (Hnrnpdl, Cirbp (hnRNP-A18) and Pcbp2 (hnRNP E2)), and six other proteins with less well characterized roles in RNA processing (Gtl3, Rbms1, Thoc3, Pcbp4 and Topors).

To test if the 15 robustly circadian splicing factors were under the control of the local liver clock, we examined by QPCR whether their expression was phase-advanced in the Vipr2^-/- liver (Figure 6c; Figure S6 in Additional file 1). Thirteen of the genes showed a significant main effect of genotype and/or a significant interaction effect between genotype and time by two-way ANOVA (P ≤ 0.05), indicating the Vipr2^-/- mutation was having an effect on the splicing factor expression. Of these 13 genes, 12 showed a significant effect of time in the wild-type time-course (P ≤ 0.05, one-way ANOVA), confirming their normal circadian expression. Strikingly, we observed a variety of responses to the Vipr2^-/- mutation. Three of 13 genes showed a significant effect of time (P ≤ 0.05, one-way ANOVA) in the Vipr2^-/- time-course and were clearly phase-advanced (Ddx46, Pcbp2 and Topors; Figure 6c; Figure S6 in Additional file 1). A further four genes showed temporal patterns very suggestive of phase-advancement, but failed to achieve statistical significance (Srsf3, Cirbp, Sf3b1 and Gtl3; Figure S6 in Additional file 1). Amongst these, Cirbp is of particular interest. Circadian Cirbp expression was found to be rendered arrhythmic in Clock mutant liver 42 (Figure S7 in Additional file 1), indicating that it is under direct clock control. However, it was also found to remain rhythmic upon specific inactivation of the liver clock 37, indicating peripheral circadian cues are sufficient to drive its rhythmicity. The remaining six genes showed no significant effect of time and appeared to be arrhythmic in the Vipr2^-/- time-course, exemplified by Rbms1 (Figure 6c; Figure S6 in Additional file 1). Of these genes, Dhx9 was also found to be arrhythmic in Clock mutant liver 42 (Figure S7 in Additional file 1). Arrhythmic expression has not been observed previously in the Vipr2^-/- mutant. Taken together, this indicates that some of the examined circadian splicing factors are under the control of the local liver clock, while others are likely rhythmic in response to systemic rhythmic cues absent in the Vipr2^-/- mutant. Those that phase-advance would be candidate regulators of the exons examined in Figure 5. As most RNA binding proteins are multi-functional 43, this indicates that the liver contains at least two circadian post-transcriptional programs, differentially dependent on the functional integrity of the SCN network.

In order to identify the candidate regulators of circadian alternative splicing, we examined the correlation between the average temporal expression patterns of the robustly circadian splicing factors and the validated circadian exons (Figure S8 in Additional file 1). Several interesting associations were revealed. The strongest positive correlations were found between the Srsf3 splicing factor and the circadian exons within Pcsk4 and Npas2 (0.904 (P = 0.002) and 0.841 (P = 0.0089), respectively). Moreover, both these exons and Srsf3 expression were phase-advanced in the Vipr2^-/- mutant (compare Figure 5b and Figure S6e in Additional file 1). Strong negative correlation might indicate a repressive regulatory role for a particular factor on a particular exon. The strongest negative correlations were observed between Srsf3 expression and Cabc1 and Clock exon-specific expression (-0.929 (P = 0.0008) and -0.866 (P = 0.0054), respectively). These associations, both positive and negative, provide leads for further study of the molecular mechanisms controlling circadian alternative splicing.

Of particular interest are Khdrbs1 (Sam68) and Tra2b, which have a well-characterized role in regulating alternative splicing. To gain further insight into whether these factors may be driving circadian alternative splicing, we examined circadian splicing of a selection of their proven murine target exons. While none of the Tra2b target exons we examined were rhythmic, three of the Sam68 target exons identified by Chawla et al. 44 had circadian splicing patterns (Figure S9 in Additional file 1). All three showed highly significant circadian patterns by ANOVA. Ktn1 exon2 shows a similar circadian splicing pattern to Rev3l exon2, with higher exon inclusion during the late subjective day, while Opa1 exon2 shows an anti-phasic pattern. Of the three exons, only Ktn1 exon2 is changing with a pattern that is consistent with both the circadian expression pattern of Sam68 and the effect on the exon of Sam68 knockdown in Chawla et al. 44. This may be an effect of the different biological contexts we are examining (circadian liver tissue compared to neuroblastoma cell line). In conclusion, taken together, our correlation analysis and examination of known target exons provides a list of interesting candidates for further study of the molecular mechanisms regulating circadian exons.

Studies conformed to the Animals (Scientific Procedures) Act (1986). Adult male wild-type (C57BL/6) and Vipr2^-/- (C57BL/6 background) mice were housed under 12 h:12 h light:dim red light (LD) with ad libitum food and water. For circadian locomotor activity recording, mice were caged within light-tight ventilated cabinets and transferred from LD to continuous dim red light (DD). Locomotor behavior was analyzed using ClockLab software (ActiMetrics, Wilmette, IL, USA). The exon-array and validation experiments were performed simultaneously. Herein mice were caged in groups of four to five during behavioral analysis due to the large number of mice involved 2365. In the experiment comparing wild-types to Vipr2^-/- mutants (Figures 5 and 6), mice were individually caged during behavioral analysis. To obtain circadian timed organ samples, mice at the appropriate time were harvested under DD and extracted organs were rapidly frozen on dry ice, then stored at -80°C. Total RNA was extracted from frozen tissue by homogenization in Tri-Reagent using a rotor-stator homogenizer, followed by purification using the Ribopure kit (Life Technologies (Grand Island, NY, USA). RNA was then treated with the TURBO DNA-free kit (Ambion).