|Home | About | Journals | Submit | Contact Us | Français|
Two decades of research identified more than a dozen clock genes and defined a biochemical feedback mechanism of circadian oscillator function. To identify additional clock genes and modifiers, we conducted a genome-wide siRNA screen in a human cellular clock model. Knockdown of nearly a thousand genes reduced rhythm amplitude. Potent effects on period length or increased amplitude were less frequent; we found hundreds of these and confirmed them in secondary screens. Characterization of a subset of these genes demonstrated a dosage-dependent effect on oscillator function. Protein interaction network analysis showed that dozens of gene products directly or indirectly associate with known clock components. Pathway analysis revealed these genes are overrepresented for components of insulin and hedgehog signaling, the cell cycle, and the folate metabolism. Coupled with data showing many of these pathways are clock-regulated, we conclude the clock is interconnected with many aspects of cellular function.
Circadian rhythms in mammals are exemplified by the sleep/wake cycle and regulated by an internal timing system. A central clock in the suprachiasmatic nuclei (SCN) of the hypothalamus coordinates input and, through a complex signaling cascade, synchronizes local clocks in the brain and throughout the body (Reppert and Weaver, 2002; Panda et al., 2002b; Liu et al., 2007a). Over the past two decades, extensive genetic, genomic, molecular, and cell biological approaches identified more than a dozen clock genes that collectively comprise a biochemical feedback loop that drives circadian oscillations (Reppert and Weaver, 2002; Panda et al., 2002b).
Recent models consider the clock a biochemical and cellular oscillator, and also a genetic network. A highly conserved negative feedback loop was discovered and elucidated at biochemical, cellular, and organismal levels in mammals and Drosophila; transcription factors play a prominent role in the clock mechanism. The bHLH-PAS transcriptional activators BMAL1 (official name: ARNTL, an ortholog of the Drosophila cyc gene) and its heterodimeric partner CLOCK (an ortholog of the Drosophila Clk gene) interact to bind to E-box cis-elements present in the promoter regions of their target genes. These targets include two families of repressor proteins, the PERIODs (PER1, PER2, and PER3) and the CRYPTOCHROMEs (CRY1 and CRY2), which interact in a protein complex that translocates from the cytoplasm to the nucleus. In the nucleus, this repressor complex physically associates with the BMAL1-CLOCK complex to inhibit E-box-mediated transcription. This process results in the cyclic transcription of these repressor genes as well as thousands of transcriptional output genes elsewhere in the genome (Hughes et al., 2009; Panda et al., 2002a; Ueda et al., 2002).
In addition to rhythmic transcription, PER and CRY protein levels and subcellular localization also oscillate. Protein level cycling is a consequence of the transcriptional regulation mentioned above, but also through post-transcriptional and post-translational mechanisms that regulate the stability and degradation of messages and proteins. These processes are mediated by kinases (e.g., CSNK1D, CSNK1E, CSNK2, and GSK3B) (Vanselow et al., 2006; Maier et al., 2009; Hirota et al., 2008; Etchegaray et al., 2009) and the proteasomal machinery including the E3 ligase, FBXL3 (Siepka et al., 2007; Busino et al., 2007; Godinho et al., 2007; Reddy et al., 2006). Thus, while transcriptional regulation generates rhythmic RNA levels, regulated post-translational modifications control protein abundance, subcellular localization, and repressor activity of PER and CRY. Importantly, these additional regulatory steps introduce a delay, critical for rhythm generation and period regulation, in the clock mechanism (Gallego and Virshup, 2007).
The circadian oscillator is also a highly interconnected genetic network that uses other transcription factors and response elements. In addition to the biochemical feedback loop that regulates cycling at the E-box (termed the "core loop”), circadian gene expression is mediated by transcription at the ROR/REV-ERB (RORE) and the DBP/E4BP4 (D-box) binding elements. Two subfamilies of nuclear hormone receptors, the NR1Ds (NR1D1 and NR1D2, or REV-ERBα and β) and RORs (α, β and γ, or RORa, RORb and RORc), either repress or activate gene transcription from the ROR elements in several clock genes (Ukai-Tadenuma et al., 2008). The bZIP transcription factors, DBP, TEF and HLF, perform similar functions on the D-box element (Gachon et al., 2006).
The role of these genes was examined in vivo and in vitro. Knockout mice for Rev-erbα, Rorα or Rorβ, have lower amplitude rhythms, abnormal period lengths, and inter-individual variability in both phenotypes (Andre et al., 1998; Preitner et al., 2002; Sato et al., 2004). A coactivator shown to interact with the ROR proteins, PGC1α, modulates Bmal1 expression, and Pgc1α knockout mice display long period locomotor activity behavior (Liu et al., 2007c). Furthermore, dose-dependent knockdown of ROR and REV-ERB genes in vitro has a potent effect on the baseline and the amplitude of circadian gene expression (Baggs et al., 2009). Taken in sum, these data demonstrate the important role of the clock gene network in regulating circadian amplitude, resistance to perturbation, and, in several cases, modulation of period length.
Although the clock can function in a cell-autonomous manner (for discussion purposes, we extend the definition of autonomy to include potential auto regulation in trans), in vivo rhythms in physiology and behavior result from more complex cellular and humoral interactions. Genetic dissection of the circadian system showed the important role of intercellular coupling in generating physiological and behavioral responses; these explain the apparent discordance among observations from cells, organotypic SCN cultures, and whole animals (Liu et al., 2007b). For example, although Cry1 deletion in mice causes short period length of locomotor activity rhythms, dissociated SCN neurons and peripheral cells derived from these mice are arrhythmic (Liu et al., 2007b). These results demonstrate that intercellular coupling in the SCN compensates for the loss of Cry1 function and, as a result, masks the more severe cellular clock deficits. Similarly, while Clock gene knockout mice are nearly wild-type in their behavioral rhythms, lung and liver preparations from these mice display arrhythmic gene expression (DeBruyne et al., 2006; DeBruyne et al., 2007). In fact, cellular clock defects are often more severe than behavioral in clock gene knockout mice (Brown et al., 2005; Liu et al., 2007b). Thus, although transcriptional rhythms are cell-autonomous and reflect the cellular nature of the circadian signaling, physiology and behavior result from more complex in vivo interactions.
The circadian research community took nearly four decades to identify the above clock components. In metazoan models, successful approaches included forward mutagenesis in flies and mice, and cellular and biochemical methods followed by confirmatory reverse genetics. Although these efforts were fruitful, they remain incomplete as evidence, including quantitative trait loci (QTL) studies, suggests that additional clock components and modulators exist (Takahashi, 2004).
Remarkably, circadian clocks exist in cell lines (Balsalobre et al., 1998). Some of these lines are amenable to cDNA over-expression and RNAi techniques, and have been used to identify new components of many signal transduction pathways (for example see (Conkright et al., 2003; Iourgenko et al., 2003; Aza-Blanc et al., 2003)). To conduct a whole genome screen for circadian clock modifiers, we used the U2OS (human osteosarcoma) cell line that was successfully employed in chemical screens (Hirota et al., 2008), in dissection of network properties of the oscillator (Baggs et al., 2009), and in a limited-scale RNAi screen (Maier et al., 2009). We screened ~90,000 individual siRNAs and measured luciferase reporter gene expression every two hours for four days. We applied statistical algorithms to find modifiers of rhythm amplitude and period length. Knockdown of nearly a thousand genes reduced rhythm amplitude. On the other hand, knockdown of hundreds of genes by multiple independent siRNAs generated strong circadian phenotypes. Characterization of a subset of these genes demonstrated a dosage-dependent effect on clock function. Protein interaction network analysis indicated that some candidates directly or indirectly interact with known clock components. Gene expression studies showed that some affect the clock by regulating clock component expression levels. Finally, to disseminate this data and integrate with existing public resources, we constructed a database (http://biogps.gnf.org/circadian/). This resource will serve the research community in genetic and biochemical investigations of circadian regulation of physiology and behavior.
Because circadian rhythms are dynamic, we employed kinetic, rather than steady-state, bioluminescence detection to assess the persistence and character of rhythms following perturbation. To conduct such a screen, we required a cell line amenable to efficient reverse transfection, where siRNAs are deposited on plates, dried, mixed with transfection reagents, and, finally, cells are added to complete the transfection process. We chose U2OS cells for assay development as this model was extensively characterized by functional studies examining knockdown effects of all known clock components with cellular and behavioral phenotypes in knockout mice (Baggs et al., 2009; Liu et al., 2007b). In addition, this model was used in small molecule screening and limited-scale RNAi screening to find modifiers of amplitude and periodicity (Hirota et al., 2008; Maier et al., 2009; Baggs et al., 2009).
We established cell lines harboring a rapidly-degradable form of luciferase, dLuc, driven by the mouse Bmal1 or Per2 gene promoter (Liu et al., 2008) (Figure S1). Knockdown of CRY1, CRY2 and BMAL1 in both lines resulted in phenotypes consistent with previous knockout mouse and cellular knockdown studies (Liu et al., 2007b; Liu et al., 2008; Maier et al., 2009; Baggs et al., 2009) (Figure S1). For instance, CRY1 knockdown shortens period length and compromises rhythm persistence, CRY2 knockdown lengthens period length, and knockdown of both results in arrhythmicity.
We next adapted these assays for high-throughput screening (HTS) in 96- and 384-well plates. By optimizing growth conditions and transfection efficiency, we obtained consistent bioluminescence rhythms in a HTS format (e.g., ± 0.5 hr in control wells of 384-well plates, n = 7680 wells) with knockdown effects for period and amplitude (see below). Knockdown of known clock components such as CRY1, CRY2, and BMAL1 in this format produced similar cellular clock phenotypes to those in LumiCycle assays using 35-mm dishes (Figure S1). These conditions enabled large-scale siRNA screens.
The screening protocol and logic are outlined in Figure 1A and Figure S2, and describe the primary screen, data mining, hit selection, secondary screen, and validation studies. The Bmal1-dLuc line was more robust than our Per2-dLuc line and therefore used for primary screens in a 384-well format. CRY2 siRNAs were used as positive controls in each plate. We screened an siRNA library (Qiagen) targeting 17,631 known and 4,837 predicted human genes. In these libraries, 4 siRNA constructs were designed for each gene with a pool of 2 siRNAs per well (2 siRNAs/well, 2 wells/gene) for a total of 89,872 siRNAs. These siRNAs were pre-spotted on 384-well plates. After addition of transfection reagent, we seeded approximately 2,000 cells into each well to complete the transfection. We measured bioluminescence every two hours for four days for a total of 48 time points for each well; this temporal resolution was chosen based on simulation studies using data collected from known clock gene perturbation. The primary screen produced a total of > 4.3 × 106 data points (supplemental Data Files S1 and S2), which are also deposited in the PubChem Bioassay database (AID: XXXX).
We used the MultiCycle Analysis program (Actimetrics, Inc.) to analyze period length and rhythm amplitude for each well. Plate-to-plate variation in period length and amplitude was modest in control wells (period length or τ = 25.20 hr ± 0.55, mean ± SD, n = 1176; amplitude or A = 4380 ± 1010, n = 1176) (Figures 1B and S3). Knockdown of CRY2, a period length control, resulted in robust lengthening (29.43 hr ± 0.65; n = 1176), while knockdown of BMAL1, an amplitude control, potently reduced amplitude (692 ± 626 arbitrary luminescence units; n = 1176).
Gratifyingly, many known clock components had phenotypes in this screen consistent with those seen in knockout animals (Liu et al., 2007b; Liu et al., 2008) or dose-dependent siRNA knockdown (Baggs et al., 2009) (Figure 1C). For example, CRY2 knockdown lengthened period, CRY1 knockdown led to rapid loss of rhythmicity. Knocking down BMAL1 or CLOCK genes resulted in low amplitude/arrhythmicity, and knockdown of PER1 or PER2 also caused almost immediate arrhythmicity. Consistent with its role as a repressor for RORE-mediated transcription, knocking down NR1D1 raised overall reporter expression levels and lengthened the period (see below), consistent with previous reports (Baggs et al., 2009). We also saw knockdown effects for CSNK1D and CSNK1E (Xu et al., 2005; Meng et al., 2008; Gallego and Virshup, 2007; Etchegaray et al., 2009), GSK3B (Hirota et al., 2008), and FBXL3 (Siepka et al., 2007; Busino et al., 2007; Godinho et al., 2007), FBXW11 (or β-TRCP2), and CSNK2 (Maier et al., 2009). Collectively, these results support the suitability of our primary screen.
To identify genes whose knockdown modulates the circadian clock, rather than the overall health of the cells, we focused on period length deficits and increased amplitude of rhythmicity. Low-amplitude traces usually exhibit poor curve fitting and generate inconsistent period length data between replicate wells; they constitute less than 4.5% of all wells and were excluded from the analysis (Figure S4). We found 1028 short-period hits including 76 genes there were hit by two independent siRNA pairs (double hits), 4230 long period hits (274 double hits), and 493 high amplitude hits (18 double hits) (Figure1B). Consistent with data from small molecule screens (Hirota et al., 2008; Hirota and Kay, unpublished results) and previous model organism screens (Takahashi, 2004), many more genes generated long period length than short.
As siRNAs have well described potential "off target" effects, we focused on genes where two or more independent siRNAs generated a consistent phenotype (Echeverri et al., 2006). The plotted traces for these genes were individually inspected to remove false positives caused by poor curve fitting. We selected 254 genes whose knockdown resulted in strong circadian phenotypes, defined by three or more standard deviations from the mean, in period length and rhythm amplitude (Figure S2). We also selected 89 single siRNA-pair hits that showed strong circadian phenotypes in duplicate wells (Figure S2). Thus, our primary hit selection list contains a total of 343 genes including known clock components (Table S1 and Figure 1).
Next, we performed a secondary screen to confirm the knockdown phenotypes identified in the primary screen (Figure S2). We tested at least 4 siRNA constructs (1 siRNA/well, ≥ 4 wells/gene) for each gene in both the Bmal1-dLuc and Per2-dLuc reporter cell lines (Figure S1). We hypothesized that consistent changes in both assays indicates that resultant phenotype is not reporter- or response element-specific, but rather reports an impact on clock function in cells.
As expected, the secondary screen using Bmal1-dLuc reporter assay generated highly concordant results with the primary screen (Figure 2A and Figure S2): 222 of 238 genes that were identified in the primary screen with two siRNAs and 47 of 83 with a single siRNA were confirmed (Table S1). In Per2-dLuc reporter assay, 219 were also independently confirmed (Figures 2B–2D, and Figure S2). The remaining unconfirmed genes either failed to have more than one independent siRNA conferring a consistent phenotype or confirmed in the Bmal1-dLuc but not the Per2-dLuc assay. In addition, a few genes altered both period length and amplitude in these assays (Figure S2). In summary, we observed a high level of concordance between our primary and secondary screens, and we identified hundreds of genes whose knockdown generates strong circadian phenotypes.
To determine those genes that most sensitively impacted period length or increased amplitude (cellular rheostats of clock function), we conducted a dosage-dependent knockdown analysis as per our previous work (Baggs et al., 2009). In this study, we selected 17 genes from the primary screen with extreme phenotypes and generated an 8-point dose response. All but one showed dose-dependent effects on circadian rhythms (Figure 3A). Gene knockdown was confirmed by quantitative PCR (Q-PCR) (Figure 3B). Genes whose knockdown lengthened period included HCFC1, POLR3F, PRPF4, SEC13, UNC119, and ZMAT3, similar to CRY2 or FBXL3. Knockdown of ACSF3, B4GALT2, CEACAM21, TBCB, or MPG led to dose-dependent period-shortening, similar to CRY1. Finally, knockdown of COX4NB, FHIT, HIST1H1B, NMNAT1, or PDE1B caused dose-dependent increases in rhythm amplitude. These data confirm primary screening results and show that many genes from our screen impact clock function equivalently to or better than known clock components.
Because of the complex regulatory architecture of the clock gene network, perturbation of one clock component can lead to dynamic changes in the levels of others (we describe these as "network effects"). To determine whether genes from our screen had these effects, we selected a subset of genes that confirmed in dose response (Figure 3): POLR3F, PRPF4 and SEC13 (long-period hits), ACSF3 and MPG (short-period hits), and COX4NB (high-amplitude hit) to analyze their potential for network effects. (The technical demands of this experiment prevented us from extending this analysis to all genes from Figure 3.) Knockdown of known clock genes had potent effects on circadian phenotypes as well as clock gene expression (Figure 4), consistent with a previous study (Baggs et al., 2009). Interestingly, knockdown of most screen hits led to dose-dependent reduction of NR1D1 and DBP transcript levels (Figures 4), consistent with E-box-driven regulation (Ueda et al., 2005; Liu et al., 2008; Baggs et al., 2009). These observations provide additional support for the notion that regulation of E-box-mediated transcription represents a topological vulnerability in mammalian circadian clocks (Ueda et al., 2005).
Similar to NR1D1, SEC13 knockdown up-regulated BMAL1 transcription; like NR1D1, SEC13 may impact clock function through network effects, rather than direct physical association with clock components. Intriguingly, knockdown of ACSF3 and POLR3F did not impact expression of core clock genes despite dose-dependent functional consequences on the clock (Figure 4). We note that CRY2 knockdown also does not alter the mRNA levels of known clock genes. These genes may regulate post-translational modification of clock components and impact protein abundance, localization, or function to perturb the oscillator.
We hypothesized that some hits from our screen directly or indirectly associate with known clock components. Using the Entrez Gene and Prolexys protein-protein interaction databases (Prolexys Pharmaceuticals, Utah), we identified a comprehensive set of proteins that interacted with gene products of screen hits or known clock components (Table S2). To visualize the global topography of these interactions, we constructed an expanded clock gene interaction network (Figure 5). Most genes identified in our screen are present in a highly connected cluster of proteins centered on the core clock components: BMAL1, CLOCK, PERs, and CRYs. While ZMAT3, BLNK and RRP12 directly interact with core clock components, most others associate indirectly through bridging molecules. (Many of these bridging molecules have phenotypic effects in our screen, but fell short of our reporting threshold.) For example, P53 (official name: TP53, tumor protein 53) physically interacts with CDK9, MAPK8, NCL, ABL1, BCR, DHFR, and BRCA2, as well as PCAF and CSNK1E; knockdown of these genes generate short, long, and high amplitude phenotypes in our screen. Knockdown of P53 itself generated low amplitude rhythms in our screen; U2OS cells are wild type for P53 expression (Florenes et al., 1994). Interestingly, a recent RNAi screen showed that BMAL1 potently regulates P53 pathway function (Mullenders et al., 2009). Taken together, these observations provide evidence of the interconnectedness between the circadian clock and many other cellular pathways.
Although the protein interaction network analysis gave some sense of pathway regulation, we sought to directly test for functional interconnectedness between the clock and other biological processes using the NIH David pathway analysis tool (Huang et al., 2009; Dennis, Jr. et al., 2003). We found that hits from our screen were members of dozens of cellular pathways. In addition, several pathways were overrepresented in this analysis, including the insulin signaling pathway (p < 0.013), hedgehog signaling (p < 0.0088), cell cycle (p < 0.054), and folate metabolism (p < 0.014) (Figures 6A and S5). While inhibition of multiple components of the insulin, hedgehog, or cell cycle pathways resulted in long period length of circadian oscillations, knockdown of components of folate metabolism resulted in short and long period, as well as high amplitude phenotypes. (We note that components of these pathways often play important roles in other pathways; for example, several components in insulin signaling are shared in B cell receptor, IL-4, IL-8, PDGF and IGF-1 signaling.) Our previous work has demonstrated that many components in these pathways are themselves under clock control and expressed in a circadian fashion in various peripheral tissues such as the liver (Panda et al., 2002a; Hughes et al., 2009). Collectively, these data show functional interconnectedness between the circadian clock and many other biological pathways.
One dramatic example of this regulation is the insulin signaling pathway (Figure 6A). (Insulin signaling and many other pathways have inherent feedback mechanisms; in this regard, these pathways may not function in a wholly cell-autonomous manner, and instead may auto-regulate in trans.) Down-regulation of multiple components of the insulin pathway resulted in period changes, including JNK (MAPK8, long period), IKK (IKBKB, long period), PI3K (PIK3R5, long period), MTOR (FRAP1, long period), APKC (PRKCI, long period), PFK (PFKP, short period), and PYK (PKLR, long period) (Figure 6B). In addition, at least 19 components of the insulin pathway are regulated at the transcriptional level by the circadian clock (Figure 6A). These include IKK, PI3K, and MTOR, whose perturbation also impacts clock function. To provide additional validation, we employed small molecules targeting multiple components of the insulin pathway in the U2OS model. SP600125 (a JNK inhibitor) and the Dequalinium analog C14 linker (a PKC inhibitor) led to long period length of circadian oscillations, PMA (a PKC activator) resulted in short period length, and wortmannin and LY294002 (PI3K inhibitors) resulted in a phase delay (Figure 6C). These results are all consistent with genetic perturbation of their intended targets. These results highlight the functional intersection between two important biological pathways, insulin signaling and the circadian clock.
To facilitate use of this resource, we constructed a database to enable visualization and exploration. Our screening data are displayed as a plug-in in BioGPS, an open-access, searchable database for aggregating gene annotation from multiple online sources (http://biogps.gnf.org). We created a custom BioGPS circadian layout focused on the siRNA screen described in this manuscript (Figure S6) and accessible at http://biogps.gnf.org/circadian/. This BioGPS layout displays plug-ins from our siRNA circadian screen and a recently published 1 hr-resolution circadian gene expression databases from mouse liver and pituitary tissues (Hughes et al., 2009). Additional online resources are also available including the UCSC Genome Browser (http://genome.ucsc.edu/), our previously published reference gene expression data obtained from various tissues and cell lines (Su et al., 2002; Su et al., 2004), and Wikipedia, which describes the expanded gene annotations (e.g., PER2 as shown here). Furthermore, BioGPS provides a flexible platform for building customized layouts that can be modified to include a selection of the more than 100 other online data sets and/or resources in the BioGPS plug-in library.
To identify additional clock genes or modulators, we carried out a genome-wide siRNA screen using a robust reporter gene assay in human U2OS cells. We found nearly 1000 genes whose knockdown resulted in low amplitude circadian oscillations, which may indicate their potential in regulating the general health of cells. In addition, we found hundreds of genes whose knockdown led to long or short period length of oscillation, or increased amplitude. Many of these factors had phenotypic effects on the oscillator similar to known clock components including dose-dependent perturbation. These genes are excellent points of intervention for perturbation of autonomous clock properties and may later prove important in regulation of physiology and behavior in the whole organism. Molecular analysis of these dose-dependent genes suggests that most of these genes function by regulating clock gene expression levels. These observations point the way to new aspects of genetic network architecture and possibly new feedback loops in the clock.
Protein interaction network analysis showed that some factors physically interact with core clock components, while others interact with bridging proteins that physically interact with the clock (one degree of separation). This analysis also showed that the P53 pathway is functionally interconnected with the clock. Focused pathway analysis of screen hits dramatically extend this observation—genes from dozens of pathways are represented on our hit list, and several pathways, including insulin and hedgehog signaling, the cell cycle, and folate metabolism, are overrepresented. Previous data has shown that these pathways are clock regulated. Collectively, we conclude the clock is massively interconnected and functionally intertwined with many biological pathways. As a cautionary note, observations from this cellular model may not be directly applicable to clock function in the whole organism as many genes identified in our screen are tissue specific.
Why is this interconnectedness important? Research in other fields has shown perturbation of one pathway often has deleterious and unintended consequences on another. For example, COX2 inhibitors such as Vioxx were designed to inhibit pain and inflammation; however, intermediates produced by COX2 (and inhibited by this class of small molecules) also afford cardio protection (FitzGerald, 2004). Alternatively, this interconnectedness may provide advantages—many of the above pathways have specific inhibitors that may prove useful in altering circadian phenotypes. Knowledge of this interconnectedness will be valuable in exploiting existing tools while avoiding the potentially deleterious consequences of doing so blindly.
Finally, we provide a publicly accessible resource to explore and visualize this data. It is our hope that this resource will serve as a useful launching point for colleagues studying the circadian clock in interpretation of their biochemical and genetic results to accelerate understanding of how the clock regulates physiology and behavior.
U2OS cells were obtained from the American Type Culture Collection (ATCC). We generated Bmal1-dLuc and Per2-dLuc reporter cell lines. The siRNA libraries used in the primary and secondary screens were purchased from Qiagen. Additional siRNAs were purchased from Invitrogen and Dharmacon and used as controls in target validation experiments. GL2 and GL3 siRNAs designed against the Luc in pGL2 and pGL3 vectors, respectively, were purchased from Qiagen. The GL2 siRNA was used as a negative control. Screen controls purchased from Invitrogen included: CRY1-HSS102308; CRY2-HSS102311; BMAL1 or ARNTL-HSS100703. CRY1 and NR1D1 siRNAs used in dose-dependent experiment were previously described (Baggs et al., 2009). Other siRNAs used in the dose experiment are listed in Table S3 and used as pools.
U2OS cells were grown in regular DMEM supplemented with 10% FBS and antibiotics. Lentiviral Per2-dLuc and Bmal1-dLuc reporters (Liu et al., 2008), were introduced into U2OS cells via lentivirus-mediated infection as described (Liu et al., 2008). We selected stable cell lines with blasticidin and clonal lines by FACS-based single cell sorting in 96-well plates; we tested these lines as described in the LumiCycle (Liu et al., 2008). The clonal lines are genetically and morphologically indistinguishable from parental cells, and represent the average period length of the infected cell population.
For primary siRNA screens, U2OS circadian reporter cells were reverse transfected with siRNAs in 384-well plates. Briefly, we trypsinized rapidly growing cells and re-suspended them in DMEM containing 20% FBS without antibiotics at 0.1×106 cells/ml. We next added 20 µl of transfection reagent mixture (3.3 µl/ml or 0.066 µl/well Lipofectamine 2000 in Opti-MEM) to each well containing pre-spotted siRNA constructs (1 pmol; 0.5 pmol for each siRNA; final concentration of 12.5 nM), incubated at room temperature for 20 min, and added 20 µl of cells (2000 cells/well) with our robotic system. Approximately 18 hr after transfection, we replaced this media with 60 µl pre-warmed fresh DMEM containing 10% FBS and antibiotics and allow the cells to grow for an additional two days.
Three days post-transfection, we replaced this media with 60 µl HEPES-buffered explant medium supplemented with luciferin (1 µM) (Promega) and B-27 supplements (Invitrogen), and the plates were sealed with an optically clear film (USA Scientific). We next loaded these plates in a 36°C incubator and recorded bioluminescence expression with a ViewLux (Perkin Elmer). We measured bioluminescence for 30 seconds every two hours for four days. Secondary screens were performed in the same manner. For technical reasons, 20 out of the 292 plates in the primary screen were recorded for 3 days.
We used the MultiCycle circadian data analysis program (Actimetrics, Inc.) to analyze recorded bioluminescence data. In brief, data was detrended by subtracting a best fit line (first order polynomial), and, subsequently, fit to a sine wave to obtain circadian parameters such as rhythm period length and amplitude.
The Bmal1-dLuc cells were transfected as described (Hirota et al., 2008) with a small modification, 0.4 µl/well of Lipofectamine 2000. Parallel transfection experiments were performed for Q-PCR and functional analyses. For Q-PCR analysis, the cells were harvested prior to medium change for rhythm recording and therefore unsynchronized. Total RNA preparation and Q-PCR were performed as described (Hirota et al., 2008). SYBR Green PCR Master Mix (Applied Biosystems) or QuantiTect SYBR Green PCR kit (Qiagen) was used for Q-PCR. The primers used in Q-PCR analysis are listed in Table S4.
We thank Buu Tu for technical support in the screens, Jia Zhang and Tony Orth for help with Qiagen siRNA libraries, Chunlei Wu and Ghislain Bonamy for help with BioGPS, and Richard Glynne and Peter Schultz for their institutional support and encouragement. We also thank Jose Pruneda-Paz, Elizabeth Hamilton, Dmitri A. Nusinow, Michael Hughes, Julie Baggs, and Jason DeBruyne for critically reading the manuscript. This work is supported by NIH grants to S.A.K. (MH51573 and GM74868). J.B.H. is supported by the Pennsylvania Commonwealth Health Research Formula Funds, the National Institute of Neurological Disease and Stroke (1R01NS054794), and the National Institute of Mental Health (P50 MH074924-01, awarded to Joseph S. Takahashi, Northwestern University). A.C.L. is supported by The University of Memphis startup fund and the National Science Foundation (IOS-0920417). S.A.K. is a founder of ReSet Therapeutics and is a member of its Scientific Advisory Board. BioGPS development is supported by NIGMS (1R01GM083924 to A.I.S.). This is manuscript # 090513 of the Genomics Institute of the Novartis Research Foundation.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.