A genome-wide RNAi screen for modifiers of the circadian clock in human cells

Abstract

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 small interfering RNA screen in a human cellular clock model. Knockdown of nearly 1000 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.

Figure 1. A Cell-Based Genome-Wide siRNA Screen for Circadian Clock Modifiers

(A) A schematic diagram of the genome-wide siRNA screen including the primary screen, data mining, hit selection, secondary screen, and validation of several selected targets. In the primary screen, reporter cells were transfected with siRNA in 384-well plates followed by kinetic bioluminescence recording (see Experimental Procedures for details). Luminescence data were analyzed to obtain circadian parameters and select primary hits. Secondary screen and validation studies were performed to confirm circadian phenotypes of hits and to demonstrate the validity of the primary screen. To catalyze the use of this dataset by the research community, we constructed a comprehensive circadian genomic screen database in BioGPS (see Figure S6 for details). (B) Distribution of circadian parameters of the entire primary screen. Dots represent normalized period (upper) and amplitude values (lower). For period length, the average of duplicate wells was divided by the mean of the entire screen and indicated in Log2 space. The cut-off was −0.1 and +0.1 (corresponding to raw data 23.55 hr for short- and 26.85 hr for long-period hits). Traces that lack apparent ~24 hr bioluminescence oscillation usually returned as a period length of 48 hr and are considered as arrhythmic. In addition, Log2 values above 0.4 (corresponding to 38 hr), for example, display poor curve fitting and are also considered as arrhythmic. For rhythm amplitude, average of duplicate wells was divided by the mean of the entire screen, and the cut-off was 2.20 (corresponding to raw data 7390) for high-amplitude hits. The knockdown phenotypes of several representative clock genes are shown in colored dots. (C) Cellular clock phenotypes of siRNA knockdown of known clock genes. Plots of cellular oscillations upon knockdown of BMAL1, CLOCK, PER1, PER2, CRY1 or CRY2 by 2 independent pairs of siRNAs in the primary screen are presented. The spikes of initial 10 hr bioluminescence readings resulted from media change and were removed from the plot.

Figure 2. Secondary Confirmation Assay of Cellular Clock Phenotypes

(A) A heat map of the secondary screen using Bmal1-dLuc cells. 872 independent siRNAs were tested against 154 genes in duplicate, along with 20 wells for each of the controls (GL2, CRY2, BMAL1 and GL3 siRNAs) for a total of 1,824 wells. Bioluminescence intensity for each well was plotted against time (hr), with each horizontal line representing luminescence recordings from a single well. The circadian profiles from each well were classified by hierarchical clustering (clustering method: maximum complete linkage; similarity measure: correlation; ordering function: average value). (B–D) Circadian parameters for 17 genes generating long (B), short (C), and high amplitude phenotypes (D) in both Bmal1-dLuc and Per2-dLuc cell lines. CRY2 siRNA was a positive control. For the Bmal1-dLuc cell line, the period length of the control wells was 25.07 hr ± 0.59 and amplitude was 4120 ± 1285 (n = 768). For the Per2-dLuc cell line, the period length was 24.18 hr ± 0.55 and amplitude was 6438 ± 1140 (n = 768). Cellular clock phenotypes are color coded, with darker colors in each category representing stronger alteration of circadian parameters (mean ± 3 × SD) than lighter colors (mean ± 2 × SD). Four different siRNAs (Y axis) were tested for each gene, and the assay was conducted in duplicates (X axis) in each screen using either Bmal1-dLuc or Per2-dLuc cells.

Figure 3. Dose-Dependent Phenotypic Validation

(A) Dose-dependent effects on circadian phenotypes. Bmal1-dLuc cells were transfected with the indicated amounts of siRNAs (8 point, 2-fold dilution series; from 8 to 1000 fmol/well) against 17 genes, and bioluminescence oscillations were recorded. Representative bioluminescence profiles (left) and circadian parameters (right) are indicated. Data represent the mean ± SD (n = 3). (B) Knockdown of target genes. Bmal1-dLuc cells were transfected with 3000 fmol/well siRNAs (corresponding to 1000 fmol/well on 384-well plate) and subjected to Q-PCR analysis under unsynchronized conditions. mRNA levels of target genes relative to GAPDH are indicated. Data represent the mean ± SD (n = 2). In parallel experiments, bioluminescence rhythms were recorded and circadian phenotypes were confirmed.

Figure 4. Dose-Dependent Effects of siRNAs on Clock Gene Expression

(A) Dose-dependent effects on circadian phenotype. Bmal1-dLuc cells were transfected with siRNAs (8 point, 2-fold dilution series from 8 to 1000 fmol/well) designed against 11 genes including clock gene controls, and bioluminescence rhythms were recorded. Representative bioluminescence profiles (left) and circadian parameters (right) are indicated. Data represent the mean ± SD (n = 3). (B and C) Dose-dependent knockdown of target genes (B) and effects on known clock gene expression (C). Bmal1-dLuc cells were transfected with siRNAs (8 point, 2-fold dilution series from 24 to 3000 fmol/well) and analyzed by Q-PCR in unsynchronized conditions. mRNA levels of target gene (left) and ACTB (right, as control) relative to GAPDH are indicated in (B), and mRNA levels of 9 clock genes relative to GAPDH are indicated in (C). Data represent the mean ± SD (n = 2). In parallel experiments, bioluminescence rhythms were recorded and circadian phenotypes were confirmed. (D) Summary of dose-dependent effects on known clock gene expression.

Figure 5. The Expanded Clock Gene Network

Clock components in the core feedback loop (blue), as well as other clock components known to regulate clock mechanism (light blue), were used to identify a list of interactors from the hits identified in the primary siRNA screen (green, short period; red, long period; purple, high amplitude) (Table S2). Protein-protein interactions were collated from Entrez Gene and the Prolexys protein-protein interaction databases. Common interacting proteins (pink) are depicted as nodes (circles). Edges are depicted as protein-protein interactions (black), phosphorylation reactions (blue), transactivation (green), and transrepression (red). The graph was generated in Cytoscape (http://www.cytoscape.org/).

Figure 6. Regulation of Cellular Circadian Clock by Components in Insulin Signaling Pathway

(A) Insulin signaling pathway genes that impact circadian function. Using the NIH David Pathway Analysis tool, we identified many components of the insulin signaling pathway represented in our screen hits. Genes that impact clock function and genes regulated by the clock are indicated with colored boxes. (B) Effects of siRNAs against genes involved in insulin signaling. The representative results from the secondary screen using individual siRNA are presented. Data represent the mean ± SD (n = 2). (C) Effects of chemical inhibitors against protein kinases involved in insulin signaling. Bioluminescence rhythms of Bmal1-dLuc cells were monitored in the presence of various concentrations of compounds (8 points of 3-fold dilution series; final 3 nM - 7 µM). Period or phase parameter was plotted against final concentrations of the compound. Data represent the mean ± SD (n = 4). The results for SP600215 and PMA are consistent with our previous observations (Hirota et al., 2008).