Dirección Nacional del Talento Humano

Are dynamic PPIs also important for the regulation of cellular events? To identify such processes, we first assigned to each network component one or more specific GO categories. Secondly, using the information whether a PPI is dynamic or not (see Figure 14 B) we investigated, which cellular processes are significantly connected via dynamic PPIs. In other words, we tested whether dynamic interactions were over-represented in the total set of interactions between a pair of GO categories. This resulted in a network with twelve dynamic links between eleven biological processes with `Circadian rhythm´ as the central `hub´ rhythmically connected with processes such as `DNA repair´, `Transcriptional regulation´ and `Response to external stimulus´ (Figure 17 A) (see also Appendix section).

Results 75

Figure 17. Dynamic Regulation of Circadian Output.

(A) Coupling of biological processes (as represented by the corresponding GO terms) via dynamic PPIs within the circadian PPI network. Node size represents the number of associated genes in the corresponding GO category. Nodes color shows significance of overrepresented GO category from yellow (low significance, FDR < 0.25) over orange (FDR < 0.01) to red (high significance, p < 0.0001). Edge width corresponds to the number of interactions between biological processes. Edge color represents enrichment in dynamic interactions between processes (blue: p < 0.001; green: p < 0.1). I prepared the figure. Analysis in (A) was conceptionally designed by me and performed by Dr. Matthias Futschik.

(B) KEGG pathway analysis of proteins within network neighborhood. Colors show significance of enrichment from yellow (low significance, p < 0.02) over orange (p < 0.0005) to red (high significance, p < 0.0002). Font size represents the number of components in each category.

Results 76

(C) Identification of highly connected clusters within the circadian PPI network (for details see Methods). Depicted are four representative functional modules that show significant enrichment in the following GO categrory from left to right: `Histone methyltransferase complex´, `Transcription coactivator activity´, `Response to DNA damage stimulus´ as well as `Histone acetyltranferase activity´. Grey nodes: network neighborhood components; red nodes: core clock component. Yellow borders indicated a circadian transcript profile in murine liver with peak expression time (circadian time) given as numbers in the nodes. Border width represents the significance for rhythmic expression.

(D) Left: Identification of dynamic PPIs within the proteome. Identified interactions are visualized as a heat map. Middle: Coupling of biological processes via dynamic interactions within the proteome. 20 processes are linked with 89 connections. Node color: significance of enrichment in components with dynamic interactions (yellow: p < 0.25; orange: p < 0.0001; red: p < 10-8_{); node size: number of genes per category; edge color:}

enrichment in dynamic interactions between processes (blue: p < 10-16_{; green: p < 10}-5_{); edge width: number of}

dynamic interactions. Right: A dynamic global PPI network consisting of 1,979 proteins with 2,788 dynamic interactions. Interactions/edges were colored based on their peak times: dark red (CT 18-24), light green (CT 0- 6), dark green (CT 6-12) and light red (CT 12-18). Proteins/nodes were colored with the respect to the five most significant biological processes and are associated with: `Cell cycle´ (green), `Cell death´ (red), `Protein modification´ (yellow), `Signal transduction´ (blue) and `Transcription´ (orange). The four proteins with most dynamic interactions (> 40) are highlighted. Analysis in (D) was conceptionally designed by me and performed by Dr. Matthias Futschik and Dr. Ravi Kalathur.

A strong association of the circadian clock network with these processes relevant for e.g. cancer and cell-cycle is supported by (i) KEGG pathway analysis of the network neighborhood only (Figure 17 B) and (ii) the significant (p < 10-8_{; Chi-squared test)}

enrichment of the network neighborhood with cancer-associated genes (as reported in the Cancer Gene Census; for identity of cancer associated genes see Methods section).

Are these rhythmically regulated processes mediated by individual components or rather by functional modules within the network consisting of interconnected components? To answer this question, I have explored the circadian PPI network topology for clusters of highly connected proteins (structural modules) and identified eleven different modules within the circadian network (Figure 17 C, Figure 18 A).

Results 77

Figure 18. Identification of Functional Modules within the Circadian PPI Network and Dynamic Regulation within the Global Interactome.

(A) Highly connected clusters were identified using the Cytoscape plugins MCODE or the ClusterOne algorithm (for details on analysis see Methods). Node colors: grey – network neighborhood, red – clock core, green – regulatory components. Yellow circles highlight rhythmic RNA profiles. Numbers are mRNA peak times in circadian time (CT). Modules were analyzed for enrichment of processes using GO, KEGG and Pfam family annotations.

Results 78

(C) Coupling of biological processes within the interactome via rhythmic PPIs. Significance of connections was calculated based on the comparison with randomized versions of the dynamic interactome. Connections, for which no more than ten out of 1,000 random networks show a larger number of rhythmic interactions are displayed. In total, 26 processes are linked via 52 connections. Node color: significance of enrichment in components with dynamic interactions (yellow: p < 0.25; orange: p < 10-4_{; red: p < 10}-8_{); node size: number of}

genes per category; edge color: number of random networks with larger number of rhythmic interactions between processes (blue: n = 0; green: n ≤ 10); edge width: number of rhythmic interactions. Processes, for which n ≤ 10 random networks have more internal dynamic interactions than observed in the global interaction network, were highlighted with a dark red border. I have prepared the figure. Analysis in (C) was performed by Dr. Matthias Futschik.

Modules were enriched with components e.g. relevant for the regulation of chromatin states (GO terms: `Histone methyltransferase complex´ and `Histone acetyltransferase activity´), DNA repair (GO term: `Nucleotide excision repair´) and transcriptional regulation (GO term: `Transcription coactivator activity´). Notably, the GO and KEGG terms derived from the analysis of the whole circadian network matched with the terms achieved for the structural modules. In addition, transcript levels of five of seven components of the `Histone methyltransferase´ module (consisting of ASH2L, HCFC1/2, MEN1, MLL, RBBP5 and WDR5; Figure 17 C left) are rhythmic and peak at similar circadian times. Co-expression in time was also observed within other highly connected clusters such as the `Histone acetyltransferase´ module (Figure 17 C right). Together, this suggests that modular organization within the circadian PPI network is contributing to a coherent functional regulation of cellular processes by the circadian clock.

Is a time-of-day dependent interaction of cellular processes via PPIs a general feature? To test this, we first identified 2,788 rhythmic PPIs (as described above – Figure 14 B) for the whole human proteome (Figure 17 D left) and then searched for biological processes that are significantly connected via dynamic PPIs (Figure 18 B). We constructed a network of 20 biological processes with 89 dynamic links. The central `hub´ of this `process network´ constitutes the term `Signal transduction´ (Figure 17 D middle) suggesting a time-of-day dependent modulation of a huge variety of cellular events such as `Protein transport´, `Response to stress´ and `Cell death´ by signaling pathways via rhythmic PPIs. This is also the case if we use randomized versions of our interactome as a background model (Figure 18 D). Interestingly, in this case we also observe an overrepresentation of dynamic PPIs within the individual processes (Figure 18 D).

To characterize the underlying PPI network properties, we constructed a dynamic PPI network for the whole proteome. We found that it again has `scale-free´ properties and

Results 79 identified 269 dynamic `hubs´, i.e. proteins with at least five dynamic interactions. The protein with the most rhythmic interactions (79 of 105 in total) is heat-shock protein HSP90AA1 – a factor required for proper protein folding upon heat stress. Notably, three of the four interaction-richest proteins (with more than 40 interactions) are cell-surface receptors (ESR1, PDGFRB and TGFBR1) again highlighting the central role of signaling pathways for dynamic regulation (Figure 17 D right). The construction of the process networks and dynamic interactome was conceptionally designed by me and the analysis was performed by Dr. Matthias Futschik and Dr. Ravi Kalathur.