RESULTS

3.1 Data quality status monitors changes over the cell cycle

After DynaPho preprocesses raw data, 14,703 phosphorylation events (class I phosphosites) are remained and 5,740 ones were filtered. The proportion of each phosphosites to these events is presented in Figure 12. The proportion of serine, threonine and tyrosine are respectively 78%, 19% and 3%. The phosphosite number of serine, threonine and tyrosine are 11,526, 2,742 and 435 respectively (no ambiguous phosphosite exists). The ratio between three phosphosites is similar with previous study [5], but tyrosine-based peptides are discovered more than it. Different phosphosites play discrepancy roles in phosphorylation signaling; for instance, phosphorylation of tyrosine was stringently regulated than others. Its phosphorylation was related to cellular regulatory function and its signaling pathway which is the major role in complex organisms [28].

All labeling ratios from these phosphorylation events are further transformed into log2 scale in a distribution chart (Figure 13). There are total of 88,218 labeling ratios, 81,913 are under 2 S.D., 4,216 are between 2 S.D. and 3 S.D., and 2,089 are over 3 S.D. About seven percentages of total ratios is potential analyzing and distributes over 14,703 phosphorylation events. These ratios are under -3 (labeling change is 0.125) and over 4 (labeling change is 16) in log2 scale. The parameter (interquartile range (IQR) / 1.35) is about 0.541 and is dissimilar with standard deviation. These descriptive statistics parameters represent the analyzing potential in the cell cycle.

3.2 Dynamic phosphorylation profiles reveal unified biological information

Eight co-expression clusters are identified by the analysis of profile clustering module.

(Figure 14) Eight clusters stand for eight different purposes and signaling systems in the cell cycle. More precisely, there are nine clusters because preprocessing procedure filters phosphorylation events which all labeling ratios are not changed. Ninth cluster may relate to

18

housekeeping, homeostasis functions or is unallied to the cell cycle. The result of each cell cycle stage contains the dynamic phosphorylation profile, the member number of phosphorylation events in each cluster, and the analysis of biological process (GO) by function enrichment module. The table of each cluster presents biological processes whose p-values in log10 scale are the top. The member number among these eight clusters is similar except that fifth cluster contains 6,394 phosphorylation events (43.488%). The dynamic profile of this cluster is that labeling ratios are highly changed in mitosis stage, but no changed in the other ones. The biological process analysis of the cluster presents most phosphorylation events are involved in mitosis cell cycle process, including nucleic acid organization (chromatin organization) and cytoplasmic component organization (organelle organization, cytoskeleton organization, macromolecular complex assembly, cytoplasmic transport, protein complex assembly, single-organism intracellular transport). These processes coincide with not only mitosis stage but also the original article. Both show that about half the peptides are phosphorylated in mitosis phase.

The dynamic profile of each cluster perfectly coincides with each cell cycle stages do not include first and eighth cluster. The mitosis, G1, G1/S, early S, late S, and G2 stages are respectively to fifth, second, fourth, third, seventh, and sixth clusters. In function enrichment analyses of these six clusters, it is not hard to understand biological processes involved in specific cell cycle stage. For example, fourth cluster (in G1/S) is mainly related to synthesis processes that prepare for DNA replication, including chromosome organization, gene expression and chromatin organization. The dynamic profile of eighth cluster is a sub-group which should be a part of mitosis stage (fifth cluster) in the original article because the labeling is highly changed in mitosis stage but a lightly down changed on early S and late S stages. And its biological process analysis presents highly related to the mitosis stage and homeostasis (including regulation functions). It means that some phosphorylation events are up-regulated in mitosis stage but down-regulated in S stage for homeostasis or regulation

19

purpose. The dynamic profile of the first cluster is that labeling changes are high in both mitosis and G1 stages. This cluster is also a sub-group which should be a part of G1 stage (second cluster) in the original article due to its result in Fig 3A. The biological process analysis indicates functions involved in both stages, including mitotic cell cycle process, microtubule-based process, and cytoskeleton organization. It implies that some phosphorylation events are in transition functions from mitosis to G1 stage. DynaPho strengthens the temporal resolution of phosphorylation events rather than transitional analyses. Also, DynaPho provides users more precise analysis algorithms to perform better clustering so that two more detailed co-expression sets are discovered.

3.3 Cellular signaling in temporal function profiles

After the analysis of function enrichment module, DynaPho summarizes core processes over all cycle stages in a functional network that nodes and edges are respectively biological processes (GO terms) and the proportion of intersection proteins. (Figure 15) Detailed biological processes with their adjusted p-values are listed in Table 4. In the functional network, each sub-network represents a set of biological processes for one or more cellular signaling (purposes). More precisely, these sub-networks correspond with stages in cell cycle.

For example, the sub-networks located on the bottom and right are mitotic chromosome condensation, mitotic nuclear division and chromatin organization. These processes present biological functions involved in the mitosis stage. Besides, sub-network in the center is mainly for homeostasis and contains lots of biological processes.

DynaPho analyzes core and detailed processes which are presented respectively in a network and a list on each temporal profile (on each cell cycle stage). The list contains biological processes which their adjusted p-values in log10 scale are the top. (Figure 16A - F).

It is easy to map the detailed processes into the summarized network on each temporal profile.

For example, in Figure 16D, several biological processes, including cellular protein complex assembly, cellular protein localization, DNA packaging, and regulation of chromosome

20

segregation are highly related to synthesis stage. These processes are further summarized into two functional networks located on the upper right and bottom.

DynaPho reinforces the dynamic analysis with a temporal heatmap profile. Summarize all biological processes with their adjusted p-values in a heatmap. The adjusted p-values are first transformed into log10 scale and further normalized by z-score. (Figure 17) In the temporal heatmap, DynaPho embraces all biological processes and their dynamic signaling among all cell cycle stages. For example, the processes which the regulation of chromosome segregation and regulation of microtubule polymerization or depolymerization are mainly involved in early S stage (time 4). The processes for nuclear envelope disassembly, chromatin organization and cytoskeleton organization are mainly in mitosis (time 1). DynaPho provides comprehensive analyses rather than transitional bioinformatics analyses.

3.4 Regulated phosphoproteome by potential kinases

DynaPho uncovers dynamic activation profiles of kinases after the analysis of kinase activation profile module. The phosphorylation sequences are sent to motif-x service based on different phosphosites and then DynaPho fetches several conserved motifs. (Table 3) Seventy-three serine-based motifs are found, twelve threonine-based ones are found, but no conserved motif exists when the phosphosite is tyrosine. Tyrosine-based motif is not found are the parameter settings (occurrences and significance) in motif-x due to maintenance of a low false positive rate.

The conserved motifs with their contribution sequences are valuable information for kinase sequence profiles. The evidenced sequence profiles maintained by PhosphoNetworks are further compared with ones generated by DynaPho from motif-x. The kinase similarity between PhosphoNetworks databases and sequence profiles from motif-x is presented in a heatmap. (Figure 18) The x-axis is the conserved motif whose phosphosite is serine or threonine, and the y-axis is the kinase. The text in white with grey background on the top of heatmap presents the cluster of conserved motifs whose kinase profiles are similar. The

21

darker red represents higher similarity between conserved motifs and kinases; in the contrary, the light green represents lower similarity. Three sections on the heatmap are relatively higher similarity. The corresponding conserved motifs and kinases are listed in Figure 19.

In the top of Figure 19, the temporal profiles of kinase activation and deactivation are also presented (the same with Figure 11). The adjusted p-values are transformed into log10 scale and the value is further multiplied by -1. The cluster 1 contains sixteen kinases, including cell cycle-related CDK family. The dynamic profile presents that kinases in cluster 1 are potential activation in G1, early S and G2 stages. The result is similar to the previous study that CDK1 was involved in G1 and G2 stages [44]. The cluster 2 contains thirteen kinases and most of them are related with cellular homeostasis or regulation. For example, both AKT1 and PAK4 were involved in homeostasis functions [45, 46]. Therefore, the activation and deactivation profile of cluster 2 are not significantly changed. DynaPho strengthens the analysis of phosphoproteome on potential kinases rather than transitional pathway analyses.

3.5 Phosphorylation signaling in cell cycle by protein interaction network

In interaction network module, DynaPho links sequential phosphorylation events across all cell cycle stages for the comprehensive signaling. Networks on Figure 20 A, B, C, D, E, and F are respectively the interaction network on mitosis, G1, G1/S, early S, late S, and G2 stage. Four shapes, including triangle, rectangle, hexagon, and circle, are respectively transcription factor, phosphatase, kinase, and the other protein type. The nodes and edges are respectively proteins (in gene name) and interactions. The interaction is composed of two types that are linking in the same stage (solid line) and linking across different stages (dashed line). The color of protein stands for its labeling ratio. The proteins in grey represented their labeling ratios that are filtered in the stage.

The interaction means the signaling event; for example, in G1 stage (Figure 20B) kinase EGFR phosphorylates transcription factor STAT3. Such signaling was evidenced by previous

22

study that EGFR-STAT pathway is involved in liver regeneration on G1/S stage [47]. In advanced, DynaPho links signaling information across cell cycle stages. For example, the interaction is linked between RanBP2 (Nup358) and ErbB2 across mitosis (Figure 20A) and G1 (Figure 20B) stage. In previous study, the cell membrane-embedded ErbB2 activates PI3K-signaling pathways which constitute important regulation in G1 stage. It migrates from the cell surface to the nucleus through endocytosis process by interacting with a nuclear pore protein RanBP2 as a traffic light [48-50]. DynaPho reinforces the analysis to construct a dynamic network across different cell cycle stages for further validation rather than the analysis in single stage.

23