Figure 1.
Systematic Interaction Mapping between 46 Circadian Clock Proteins and Associated Components.
(A) Matrix based high-throughput yeast-two-hybrid interaction screen. (B) CLOCK interactors: Mating controls (top left); upon PPI reporter genes are activated (top middle: HIS, URA for growth selection, top right: lacZ for β-galactosidase activity). Bottom: Detected interactions with CLOCK; red lines: interactions previously discovered in yeast (see also Figure S1). (C) Clock protein interaction matrix. Circles: interactions between two components not differentiating between bait and prey configuration. (D) Validation of new CLOCK and BMAL1 interactions in mammalian cells. HEK293 cells expressing CLOCK- or BMAL1-luciferase fusions were transfected with MYC-tagged components. Luciferase activity in anti-MYC co-immunoprecipitates is presented for one representative result of at least two independent experiments with similar results (for method and input controls also see Figure S2). MYC-β-galcactosidase fusions served as negative, MYC-BMAL1 and MYC-CRY1 as positive controls, respectively.
Figure 2.
The Circadian Protein–Protein Interaction Network.
The circadian interaction network integrates different interaction sources and visualizes 134 proteins with 625 interactions. Red lines: interactions discovered in yeast (see Figure 1); green lines: previously described (and detected in our Y2H screen) interactions (source: UniHI database and/or literature (MAN)); blue lines: interactions in network extension (EXT - stored in UniHI), i.e. between clock core and regulatory components and neighborhood components (see also Figure S3A). Yellow border: components with a rhythmic transcript in mouse liver [6]. Border width: significance for rhythmic expression.
Figure 3.
Network Neighborhood Contains Clock Modulating Components.
Systematic RNAi-mediated downregulation of network neighborhood genes in dexamethasone-synchronized U2OS cells harboring a Bmal1-promoter luciferase reporter. Shown are altered oscillation dynamics (red dots with corresponding fit lines) for 16 genes achieved by individual RNAi constructs (see Table S2). For twelve genes, two RNAi constructs resulted in similar phenotypes, for nine genes only one construct was available in our laboratory library. Black dots with corresponding fit lines are controls representing the mean values of at least 80 irrelevant constructs. Period deviations from controls are shown.
Figure 4.
Interaction Dynamics within the Liver Circadian Protein–Protein Network.
(A) Interacting proteins are more likely to be co-expressed in time. Left: Co-expression of interacting proteins was calculated using the Pearson correlation coefficient (PCC) of circadian expression profiles in liver [6] and compared to randomly selected protein pairs. Among interacting proteins co-expressed proteins (i.e. PCC>0.5) are significantly overrepresented (Chi squared test: p<10−15). 13% of interacting proteins have a PCC>0.5 compared to 4% for random pairs. Right: analogous analysis for the circadian PPI network. Co-expressed (PCC>0.5) interacting proteins are highly overrepresented (Chi squared test: p<10−15; 22% compared to 4% with PCC>0.5). (B) Left: heat map representing the predicted dynamics of protein–protein interaction based on their liver expression profiles. Interactions were classified as rhythmic if the product of their expression vectors shows highly significant periodic expression (FDR<10−5). Right: examples for interaction pairs and their predicted interaction phase. Red lines: products of individual transcript profiles of two interacting proteins. Dotted rectangles highlight predicted phase of interaction.
Figure 5.
Importance of Dynamic Interactions for Circadian Rhythmicity.
(A) Systematic RNAi-mediated silencing of circadian clock core and regulatory components. RNAi constructs were lentivirally delivered into U2OS cells harboring a Bmal1-promoter luciferase reporter and oscillation dynamics were monitored for several days (see also Figure 3). Circles represent the difference in period (± s.e.m.; n = 3 independent experiments) relative to non-silencing controls (n>10) for two RNAi constructs (if available). Filled circles show additional amplitude and/or damping phenotypes. Cells were classified as arrhythmic (ar) if the fit to a cosine function resulted in a low correlation coefficient (see Text S1). Period deviations of more than 2 hours are given (see also Figure S4A). (B) Systematic overexpression of circadian clock core and regulatory components. Experiments were performed with lentivirally delivered overexpression constructs as described in (A) (see also Text S1). Results of three independent experiments (± s.e.m.) are given (also see Figure S4B). (C) Correlation of circadian phenotype with number of dynamic interactions. The combined phenotypic score from silencing and overexpression experiments is significantly different for components with many dynamic interactions (≥5) compared to those with few (<5) (t-test: ** p<0.01; Mann Whitney test: ** p<0.01).
Figure 6.
Prediction of Circadian Output Regulation.
(A) Coupling of biological processes via predicted dynamic PPIs. Node size: number of genes in GO category (significance (FDR<0.25, <0.01 and <0.0001 from yellow to red). Edge width and color: number of interactions and enrichment in dynamic interactions (blue: p<0.001; green: p<0.1) (for details see main text and Text S1). (B) KEGG pathway analysis of network neighborhood. From yellow to red (p<0.02, <0.0005 and <0.0002). Font: number of components in each category. (C) Highly connected clusters. Modules with histone methyltransferase complex, transcription coactivator activity, response to DNA damage stimulus and histone acetyltranferase activity as significant GO terms (left to right). Peak expression times are given within circles (see Figure S6A for all modules identified). (D) Left: Predicted dynamic PPIs within the liver (see Figure S6B). Middle: coupling of biological processes via predicted dynamic interactions in the liver. Node color: significant dynamic interactions (p<0.25, <0.0001, p<10−8 from yellow to red); edge color: enrichment in dynamic interactions (blue: p<10−16; green: p<10−5). Right: dynamic global PPI network. Node color: most significant biological processes, i.e. cell cycle (green), cell death (red), protein modification (yellow), signal transduction (blue) and transcription (orange).
Figure 7.
Protein Phosphatase 1 Modulates CLOCK/BMAL1 Function.
(A) CLOCK and BMAL1 interactors identified in yeast and their paralogs were co-transfected with CLOCK/BMAL1 and an E-box containing luciferase reporter (see also Figure S7A). Shown are means ± s.d. of CLOCK/BMAL1 modifiers (n = 3; *** p<0.001, t-test). (B) PPP1CA dose-dependently reduces CLOCK/BMAL1 transactivation (n = 3; means ± SD.). (C) PPP1CA is present in the CLOCK/BMAL1 complex. Murine livers were harvested at indicated times. Dashed lines: longer exposure. (LC: light chain; HC: heavy chain). (D) PPP1CA destabilizes BMAL1 protein. Left: Stability is reported by the change of EGFP to DsRed ratio [30], [31]. Right: PPP1CA co-expression with BMAL1, CLOCK or short-lived EGFP fusion proteins in U2OS cells reduces BMAL1 stability (mean ± s.d.; ***p<0.001; t-test; n = 3; (see also Figure S7B, S7C). (E) Endogenous BMAL1 levels are reduced upon PPP1CA overexpression in U2OS cells. Depicted are two independent experiments. (F) PPP1CA reduces BMAL1 stability. U2OS cells stably expressing PPP1CA or GFP were harvested at the indicated time points after cycloheximide (CHX) application and protein levels were analyzed by Western blot. Shown is one representative of two independently performed experiments (see also Figure S7D).