Cell Type-Specific Functions of Period Genes Revealed by Novel Adipocyte and Hepatocyte Circadian Clock Models

As an initial effort to develop new cellular clock models pertinent to metabolism, we screened cell lines for robust rhythms and chose 3T3-L1 adipocytes and MMH-D3 hepatocytes. We introduced a lentiviral reporter harboring the rapidly degradable firefly luciferase (dLuc) gene under the control of either mouse Per2 or Bmal1 gene promoters into cells [23]. Whereas the 3T3 reporter cells were directly used in bioluminescence recording, 3T3-L1 and MMH-D3 cells were first differentiated into mature adipocytes and hepatocytes, respectively, prior to recording. These cells displayed persistent bioluminescence rhythms in 35 mm culture dishes monitored in a LumiCycle luminometer (Figure 1A). In each cell line, Per2-dLuc and Bmal1-dLuc reporters displayed anti-phasic rhythms of bioluminescence, consistent with the function of E-box- and RORE-containing promoters in regulating distinct and opposite phases of gene expression.

Next, we adapted the LumiCycle reporter assay to high-throughput screening (HTS) formats on 96 well plates. For this, we performed single cell cloning and selected clonal cell lines that expressed high levels of bioluminescence. These reporter lines displayed persistent rhythms under optimized growth conditions when monitored on a microplate reader (Synergy 2 SL) with highly consistent period lengths (Figure 1B). These highly reproducible rhythms seen in 96 well plates were similar to those in the LumiCycle, a lower throughput but much more expensive recorder. Therefore, these lines represent a tangible advantage to many labs interested in exploring circadian biology in these metabolically relevant cell lines.

For genetic perturbations, we developed a pipeline to produce high-quality, validated lentiviral shRNA vectors to knock down any mouse gene. We chose lentiviral shRNAs over transfected siRNAs because lentivirus-mediated delivery mediates potent transduction and stable integration in both dividing and non-dividing cells of various types in vitro and in vivo, thus circumventing low transfection efficiency for certain cells. We designed 6 target oligonucleotide sequences for each gene and cloned the shRNA expression cassette into the lentiviral pLL3.7GW Gateway vector, in which shRNA expression is under the control of the mouse U6 promoter, as we reported previously (Figure S1) [17]. Infectious lentiviral particles were produced in 293T cells using standard procedures and used to infect reporter cells. Infection efficiency was estimated by observing GFP co-expressed from a separate expression cassette under control of the CMV promoter (Figure S1).

We used this pipeline to generate a panel of shRNA constructs targeting the following selected 13 clock genes: Bmal1, Bmal2, Clock, Npas2 (core loop activators); Per1, Per2, Per3, Cry1, Cry2 (core loop repressors); Fbxl3 (core loop post-translational modifier); Nr1d1, Nr1d2 (RORE repressors); and E4bp4 (D-box repressor). Because of the more prominent roles of repressors in clock function, we chose to examine Nr1d1, Nr1d2, and E4bp4, the RORE and D-box negative factors, rather than the corresponding activators [17], [40], [41].

We tested shRNA knockdown (KD) efficiency for the 13 clock genes. Co-transfection of shRNA with Flag-tagged cDNA in 293T cells followed by Western blot analysis showed efficient KD of each gene at the protein level by at least two shRNAs (Figure S2). To check the KD efficiency for endogenous gene expression, mRNA levels of targeted clock genes were also measured using qPCR. For each gene, at least two shRNAs were effective in knocking down gene expression, a requirement to filter out off-target effects of shRNAs [42]. The versatility and efficiency of lentiviral shRNA allowed us to study all 13 known clock genes in all three cell type-specific clock models in parallel, which allows direct phenotypic comparison.

Knockdown of Bmal1, Clock, Cry1, Cry2, and Fbxl3 in all three cellular models resulted in expected phenotypes similar to those in LumiCycle assays using 35 mm dishes and consistent with previous knockout and knockdown studies using human and mouse cellular models [17], [27], [31], [43]–[45]. Specifically, KD of Bmal1 or Clock results in rapid damping or arrhythmicity (Figure 2A and Tables 1, S1, S2, S3); Cry1 KD leads to low amplitude or rapid damping depending on KD efficiency, whereas Cry2 KD lengthens period and increases rhythm amplitude (Figure 2B). The phenotypic defects correlate with KD efficiency of the endogenous genes by the individual shRNAs as determined by qPCR analysis. Taken together, our data demonstrate that Bmal1, Clock, Cry1, and Cry2 play similar roles in the clock mechanism across tested cell types, which provides validation for the three cellular models.

Knockdown of Bmal2 and Npas2 did not exhibit any obvious circadian phenotypes in any of the three cellular models (Figure S3), even though their expression was knocked down to levels similar to those for Bmal1 or Clock (Figure 2A and Table 1). These results are consistent with absence of observable circadian phenotypes of liver and lung tissues from Npas2^−/− mice [46], though these mice do have deficits in circadian behavior and sleep homeostasis [47]. Despite potential functional redundancy between Bmal1 and Bmal2[45], [48], and between Clock and Npas2[46], our data suggest that Bmal2 and Npas2 are not necessary for the clock to operate in these cells.

We show that Fbxl3 KD caused long period and low amplitude in 3T3, long period in 3T3-L1, and low amplitude and rapid damping in MMH-D3 cells (Figure 2C). This cell-autonomous phenotype is much more extreme compared to the relatively modest period-lengthening phenotypes seen at the SCN tissue and behavioral levels [10]–[12], or in human U2OS cells [27], [44]. Notably, although KD of Fbxl3 or Cry2 both produced long periods, Cry2 down-regulation increased rhythm amplitude, whereas Fbxl3 silencing resulted in low amplitude, consistent with its dual role in ubiquitin-mediated degradation of CRY proteins and in regulation of NR1D1-mediated transcriptional suppression [10]–[13].

Nr1d1 and Nr1d2 play overlapping but essential functions in regulating RORE-mediated transcription, and knockdown of either gene results in low amplitude rhythms, and in some cases, short period [17], [27]. We examined the effects of Nr1d1 or Nr1d2 KD on clock function in our cellular clock models. Knockdown of Nr1d1 resulted in largely normal rhythms in these cells (Figure 3A and Table 1), indicating potential overlapping functions of Nr1d2[17]. This is different from the period lengthening produced by siRNA knockdown in U2OS cells [27], [31] or the period shortening and greater variability seen in behavioral rhythms of Nr1d1^−/− mice [18], [49]. Nr1d2 KD, on the other hand, resulted in period shortening in MMH-D3 and 3T3-L1 cells but low amplitude in 3T3 cells (Figure 3A and Table 1). The Nr1d2 impact on overall rhythms was at least similar to, if not substantially stronger than, for Nr1d1. This is consistent with the reported redundant functions of Nr1d1 and Nr1d2[17], but contrasts with previous studies showing that NR1D2 deficiency has no observable rhythm phenotype in U2OS cells [27] or in mice at the behavioral level [18]. Thus, Nr1d1 and Nr1d2 play different roles in clock function depending on tissue or cell type, and Nr1d2 may be more important than previously recognized.

Despite the widely accepted role of E4BP4 as the repressor of D-box-mediated transcription, definitive genetic evidence of clock function has been lacking. We show here that E4bp4 KD resulted in short period and rapid damping in 3T3 fibroblasts, low amplitude in 3T3-L1 cells, and short period in MMH-D3 cells (Figure 3B and Table 1). These data are in line with recent studies suggesting a prominent role of E4bp4 in regulating the phase of Cry1 transcription in mouse embryonic fibroblasts [21] and period length in Rat-1 fibroblasts [41]. Studies of E4BP4's function in the clock mechanism using E4bp4^−/− mice are therefore needed to validate our findings in cellular clock models.

The shRNA constructs against Per1, Per2, and Per3 down-regulated mRNA and protein expression (Figures S2 and 4). However, unlike Cry1 and Cry2 KDs, knockdown of the Per genes in our clock models resulted in cell type-specific clock phenotypes (Figure 4). First, compared to the dramatic circadian defects observed in peripheral tissue explants and fibroblasts of Per1^−/− mice, or upon siRNA-mediated Per1 KD in U2OS cells [31], [43], [50], Per1 KD in our clock models had milder effects on clock function. Interestingly, these less dramatic phenotypes are cell type-specific: WT phenotype in 3T3 and 3T3-L1 cells, but significantly shorter period and low amplitude in MMH-D3 cells (Figures 4A and 4D; Table 1).

Similarly, Per2 KD did not cause arrhythmicity in any of the three cellular clock models: only a modest reduction of amplitude in 3T3-L1 cells, but significantly shorter period in 3T3 and short period and low amplitude in MMH-D3 cells (Figures 4B and 4D; Table 1). The amplitude reduction in MMH-D3 cells was evident in both the subtracted data and raw data (compare Figure 4 with Figure S5). Although it is possible that the more modest phenotypes may be due to incomplete silencing, the knockdown levels were comparable to those of Bmal1, Clock, and Fbxl3 (Figures 2 and 4B); and as in the case of Per1 KD, the different phenotypes resulted from similar Per2 KD efficiency in different cells. This is unexpected given its essential role in circadian rhythms of mice at the behavioral level and in cultured fibroblasts and U2OS cells [27], [43], [51]–[53]. Interestingly, however, our finding of non-essential but cell type-specific role of Per2 is in line with a recent report showing that Per2^−/− SCN explants displayed persistent rhythms with short periods, whereas Per2^−/− pituitary explant rhythms were normal and lung explants displayed slightly long periods [54]. Thus, even though genetic knockout and knockdown (incomplete silencing) may cause variations in phenotypes, the loss-of-function phenotypes of Per1 and Per2 are largely consistent and are cell or tissue type specific.

While Per2 appears to be more important than Per1 for normal clock function in mice [52], [55], deletion of Per3 has only subtle effects on the SCN clock and is often not considered part of the core clock mechanism [43], [54], [56]. However, we show here that KD of Per3 in all three models produced significantly shorter periods than in control cells (Figures 4C and 4D; Table 1). These results are consistent with recent reports showing that tissue explants of Per3^−/− mice, including liver, lung, and pituitary, also displayed short periods [43], [54]. In addition, knockdowns of Per1, Per2, and Per3 showed similar phenotypes in cells expressing a different reporter (Figures 4 and S4), confirming that the knockdown effects are reporter independent. Furthermore, the Per1 and Per3 knockdown effects are largely consistent with data of liver explants from Per1 and Per3 knockouts [43], [57], [58].

The prominent role of Per3 in peripheral clock function led us to examine its function more carefully in both intact SCN explants and dissociated SCN neurons derived from Per3^−/−:mPer2^Luc mice. We detected persistent mPer2^Luc rhythms in Per3^−/− SCN explants with a slightly shorter period than WT (Figure 4E). This is consistent with the original study of Per3^−/− mice, where a slightly shorter period of behavioral rhythms was reported [56]. We then dissociated Per3^−/− SCN neurons and examined mPer2^Luc bioluminescence from dispersed neurons at the single-cell level, as we have done previously for other genotypes [43]. We found that dissociated WT and Per3^−/− neurons generally exhibited persistent rhythms with high amplitude. However, the mean period of rhythms in Per3^−/− neurons was substantially shorter than in WT cells (Figure 4E). The weaker circadian defect at the SCN tissue level than in cell-autonomous preparations (either our cellular models or dissociated SCN neurons) is consistent with the principle that the SCN network confers robustness against genetic perturbations [43], [59], such that Per3 plays a less prominent role in the intact SCN due to compensation by the SCN network. Based on these results, we conclude that Per3 plays an important role in the SCN cellular clock as well as in peripheral oscillators and thus represents a bona fide clock component.

The more prominent role of Per3 in peripheral oscillators is in line with several recent studies showing that disruption of Per3 resulted in internal phase misalignment or desynchrony, or aberrant metabolic and sleep phenotypes [58], [60]–[62], all pointing to the role of Per3 in coherence of circadian organization. Thus, our findings suggest that tissue-specific function or dysfunction of clock genes in peripheral tissues can be an important contributing factor to human diseases, even when the behavioral effect of gene knockout is subtle. In this context, it is interesting to note that human polymorphisms in PER3 are associated with sleep and metabolic disorders [63], [64].

Taken together, our study expands our knowledge of the distinct functions of known clock genes across tissues. In particular, Per3 plays a more important role in both SCN and non-SCN cells than previously appreciated, and Per1 and Per2 appear to have different roles in different cell types. Results from this study are broadly consistent with previous findings from loss-of-function studies and collectively point to the previously under-appreciated cell type specificity of Per gene function in circadian physiology (Figure S6).

Compared to other cell-autonomous models, the short period length in MMH-D3 cells after knockdown of each of the Per genes is unique (Figure S6 and Table 1), and therefore we sought to perform single and composite knockdowns for further phenotyping using the LumiCycle assay. Consistent with the Synergy assay, Per1, Per2, and Per3 single gene KD each gave short period phenotypes, about 2 hrs shorter than the control cells (Figures 5A and 5B; Table S4). Composite Per1/Per2 double KD and Per1/Per2/Per3 triple KD caused complete arrhythmicity (Figure 5A), indicating the prominent roles of Per1 and Per2 in the hepatocyte clock. Interestingly, Per1/Per3 and Per2/Per3 double KDs did not cause any further period shortening over single Per gene KDs (Figures 5A and 5B).

As an initial effort to probe the network features of the hepatocyte clock, we examined effects of Per KD on the expression of endogenous clock genes by qPCR. Per1, Per2, and Per3 were each knocked down in both single and composite KDs (Figure 5C). Compared to the U2OS model, paralog compensation among the Per genes in MMH-D3 cells is more pervasive. In our MMH-D3 model, both Per1 KD and Per2 KD upregulated Per3, and Per3 KD upregulated Per2. While composite Per1/Per2 KD upregulated Per3 and Per1/Per3 KD greatly upregulated Per2, Per2/Per3 KD did not increase Per1 expression. Interestingly, Per1 and/or Per2 KD had milder effects on the expression of E-box-containing genes (e.g., Nr1d1 and Nr1d2) than on RORE-containing genes such as Bmal1, E4bp4, and Cry1 (Figure 5C), in line with the notion that the PERs can directly and indirectly affect Bmal1 transcription [65], [66]. Overall, the network interactions in MMH-D3 cells appear to differ from those in the U2OS model in which PER1 plays a more dominant role than PER2 and PER3[27], and is expected to differ from those in 3T3 and 3T3-L1 models. Thus, extensive investigation into the network features of these cellular models will require additional experiments and is warranted in future studies, as we have done with the U2OS model [27].

Cell type-specific function of clock genes may result from their differential tissue expression and activity (i.e., expression levels, ratio of repressors to activators, rhythmicity, and relative amplitudes), compensatory mechanisms, alternative splice variants, and post-translational modifications (PMTs), all of which can be rendered cell type specific by local physiology. Recent studies have suggested a role for stoichiometric balance among clock proteins in circadian clock robustness and periodicity, and call for mechanistic studies in a tissue specific manner [67]–[69]. In the context of the liver and adipose tissue function, it is plausible that the basic core clock mechanism incorporates cell type-specific factors and forms distinctive functional networks to regulate (and in turn be regulated by) different local physiologies. It is interesting to note, as PTMs of clock factors (e.g., phosphorylation, ubiquitination, ADP-ribosylation, acetylation, and O-GlcNAcylation) represent critical regulatory mechanisms [33], [70], tissue-specific cellular functions and metabolic states that affect the PTMs would provide important inputs to adjust local circadian clocks, and vice versa. Thus, cell type specific clock gene function starts to make sense when local physiology is considered as inputs to the clock. This challenge surely provides an opportunity for deeper insights into mechanism of tissue specific clocks, akin to the recent realization that cyclin-dependent kinase networks in the cell cycle control program are tissue specific [71].

In summary, we established three new mouse cellular clock models: fibroblasts, adipocytes, and hepatocytes. These cellular clock models offer experimental tractability, efficiency, and versatility, which are more difficult or impossible to apply to traditional tissue or animal models. Of note, in contrast to previous cellular clock models, the new clock models are amenable to high throughput experiments with inexpensive off-the-shelf recording systems, making these lines especially suitable for screening small molecules or genomic entities for impacts on cell autonomous clocks relevant to metabolism. We validated these models by developing and testing an shRNA panel of selected known clock genes. Results from this study and others point to the previously under-appreciated cell type specificity of clock gene function in circadian physiology (). The prevalence of tissue-specific clock gene function will have important implications for future studies of clock factors that affect local clock function. It is our hope that our findings in this study, along with the new cellular clock models, approaches, and tools developed here, can be applied to a greater variety of cell types in future studies, to reveal the full range of tissue-specific clock properties underlying local circadian biology. Figure S6

Per3^−/− mice were obtained from David Weaver at the University of Massachusetts. Knockout mice were bred with mPer2^Luc reporter mice to obtain homozygous knockouts harboring the mPer2^Luc reporter. Wheel-running assays were performed and analyzed as described previously [8]. All animal studies were conducted in accordance with the regulations of the Committees on Animal Care and Use at University of Memphis and UCSD.

All cell culture media were from HyClone. 3T3 (also known as NIH 3T3) and 3T3-L1 cells were cultured in regular medium in which DMEM was supplemented with 10% FBS and 1× penicillin-streptomycin-glutamine (PSG). For 3T3-L1 differentiation, pre-adipocytes were first grown to confluence (Day 0). On Day 2, cells were fed with induction medium (regular medium with 1 µM dexamethasone, 0.5 mM isobutylmethylxanthine, and 2 µg/ml insulin). On Day 4, cells were changed to maintenance medium (regular medium containing 2 µg/ml insulin). From day 6 onward, cells were grown in regular medium until use. For bioluminescence recording, 3T3 and 3T3-L1 cells were grown in 25 mM HEPES-buffered regular medium (pH 7.4) containing 1 nM forskolin and 1 mM luciferin. Fully differentiated 3T3-L1 cells were used in all experiments.

MMH-D3 cells were grown in regular medium in which RPMI medium was supplemented with 10% FBS, 1× PSG, 10 µg/ml insulin, 55 ng/ml epidermal growth factor (EGF), and 16 ng/ml insulin like growth factor-II (IGF-II). For differentiation, pre-hepatocytes were first grown to 100% confluence. Two days later, cells were replaced with differentiation medium (regular medium with 2% DMSO). Medium change was repeated every 48 hours for 6–8 days for cells to be fully differentiated for use. Circadian rhythms of differentiated cells were synchronized with 200 nM dexamethasone followed by bioluminescence recording in 25 mM HEPES-buffered serum-free explant medium (pH 7.4) containing B-27 and 1 mM luciferin, as we have done previously [43]. Fully differentiated MMH-D3 cells were used in all experiments.

SCN explants and dissociated neuronal cells were prepared and cultured as previously described [43]. Bioluminescence recording of explants, single cell-imaging of individual SCN neurons, and respective data analysis were performed as previously described [43].

Lentiviral luciferase reporters of the Per2 or Bmal1 promoter were described previously [17], [43]. Reporter cells and clonal lines were generated as previously described [23]. Briefly, reporter viral particles of high titer (>10⁸ viral particles/ml) were obtained by ultracentrifugation and used to infect 3T3, 3T3-L1, and MMH-D3 cells. Clonal cell lines of homogenous cell populations were obtained by single cell sorting and cloning in 96 well plates. We then selected the clones that expressed high levels of luciferase and exhibited circadian properties comparable to infected parental cell populations. These brighter cells were used in high-throughput assays on 96 well plates. Knockdown phenotypes were confirmed to be independent of the reporter, either Per2-dLuc or Bmal1-dLuc, and thus phenotypic differences across cell types are unlikely due to different reporter insertion sites.

We used an optimized shRNA design algorithm adapted from Aza-Blanc et al. [72] for target sequence prediction. This adapted algorithm selects for optimal target sequence for knockdown, and against homologous sequences to minimize off-target effects. We selected 6 target sequences for each gene as listed in Table S5. Each shRNA construct contained a sense and an antisense target sequence of 19 nucleotides (nts) in length, separated by 9 nts for a hairpin loop, and flanked by TTTG at 5′ and GATC at 3′ ends for cohesive end cloning. All oligos (55 nts) were synthesized by Integrated DNA Technologies (IDT). The annealed oligonucleotides were first cloned into the BbsI and SpeI sites of a pGWL-si2/U6 vector, in which the shRNA expression cassette is driven by an RNA polymerase III- based mouse U6 promoter. Subsequently, the U6-shRNA cassette was cloned into the lentiviral pLL3.7GW vector (modified from pLL3.7) [17], [73] in a Gateway LR Clonase reaction (Life Technologies), according to manufacturer's instructions.

Viral particles were prepared using standard methods in 293T cells on 12-well plates as previously described [23], [74]. Culture medium containing viral particles (∼10⁶ viral particles/ml) were collected and used for subsequent infection of reporter cells. Transfection and infection efficiency were estimated by observing GFP co-expressed from a CMV promoter. To produce high titer viruses, crude viral particles were concentrated through ultracentrifugation and appropriate titers were used, as described previously [74]. This pipeline allowed us to generate a panel of shRNA constructs against all known clock factors for genetic perturbation and phenotyping.

We used a LumiCycle luminometer (version 2.31, Actimetrics) for bioluminescence recording of cells grown on 35 mm culture dishes, as described elsewhere [17], [23], [43]. The LumiCycle Analysis program version 2.53 (Actimetrics) was used to determine circadian parameters. Briefly, raw data were fitted to a linear baseline, and the baseline-subtracted data were fitted to a sine wave (damped), from which period length and goodness of fit and damping constant were determined. For samples that showed persistent rhythms, goodness-of-fit of >80% was usually achieved. Due to high transient luminescence upon medium change, the first cycle was usually excluded from rhythm analysis. Damping rate = 1/damping constant. For amplitude analysis, raw data from day 3 to day 5 were fitted to a linear baseline, and the baseline-subtracted (polynomial number = 1) data were fitted to a sine wave, from which the amplitude was determined.

We used a Synergy 2 SL microplate reader (Bio Tek) for bioluminescence recording of cells grown on 96 well plates, as previously described [23]. Synergy data were analyzed with the MultiCycle Analysis program (Actimetrics), in which bioluminescence data were baseline-subtracted and fit to a damped sine wave to determine period length, goodness of fit, and amplitude, as with LumiCycle Analysis. Due to the various reporter expression levels and for direct comparison of different rhythms, baseline subtracted data were plotted. Because there is no damping rate output function in the MultiCycle Analysis, we used a curve fitting program of “CellulaRhythm” to determine damping rate from Synergy data as previously developed [75].

For testing shRNA knockdown efficiency targeting each clock gene, the cDNA was cloned into a p3XFlag-CMV-14 vector. Flag-tagged cDNA was co-transfected with the indicated shRNA in 3T3 or 293T cells. Forty eight hours post-transfection, cells were lysed in RIPA buffer containing complete protease (Roche) and phosphatase inhibitors (Sigma). Protein expression was determined by Western blot analysis using an anti-Flag monoclonal antibody (Sigma). For all Western assays, PVDF membrane was used in protein transfer, and SuperSignal West Pico substrate (Thermo Scientific) was used for chemiluminescent detection.

For qPCR analysis, parallel infection experiments were performed as with bioluminescence recording. Cells were harvested prior to medium change and were therefore unsynchronized. Total RNAs were prepared using the RNeasy 96 kit (Qiagen), as previously described [31]. Reverse transcription was performed using a high-capacity RNA to cDNA kit (Applied Biosystems), and qPCR was performed using SYBR Green PCR master mix (Thermo Scientific) on an iCycler thermal cycler (BioRad). The primers used in qPCR analysis are listed in Table S6. Transcript levels for each gene were normalized to Gapdh and values were expressed as percentage of expression in NS control cells, as previously described [17].