Cataloging and organizing p73 interactions in cell cycle arrest and apoptosis

Data compilation

Several databases (DIP, STRING, Biocarta, e-MIM, DOQCS, Biopax, Bind, Patika, Puma2, Biocyc, Mpact, Transpath, Predictome, Indigo, MINT, HPRD, BioGRID) of protein–protein interactions (PPI) and protein interaction networks are scanned for interactions of p73. In addition, a detailed literature search on article databases has been carried out, independent of the results obtained at the first step. This data compilation process is until August 2007. The interactions are verified by more than one independent source, where available, before being incorporated into the map. The independent sources could either be two articles that do not take the information from the same reference or one interaction database and one article. In those cases where the interaction presented in the article is validated through a small-scale experiment such as yeast two-hybrid or affinity purification, one reference was taken as sufficient. However, if the interaction obtained from a single reference is produced as a result of a large-scale experiment, the information was not incorporated into the map, since high-throughput data does not necessarily point out to a real physical association. We also looked for additional support before incorporating the single cited interactions into the network. Based on the supportive information, single referenced nodes can be classified into three categories: (i) proteins with available information related to specific binding sites; this could indicate an assurance of the existence of binding interaction (e.g. Itch in D6); (ii) proteins that are cited as part of the p53 network (e.g. c-Jun in A53), which may suggest that they also have a function in p73 network and (iii) proteins whose interaction sites are known and are also cited as existing in the p53 network (e.g. Chk1 in A31).

When contradictory information was encountered, the publication years and sources were compared. Under such circumstances, a more detailed search on these proteins has been performed by investigating their properties and interactions apart from the pathways related to p73 function. The decision as to the incorporation of the interactions has been reached only after this detailed analysis was made.

PPI network diagrams

Network diagrams can be classified into three main categories: explicit, heuristic and combinatorially complex. Heuristic and combinatorial diagrams can display large networks in a compact form, covering all possible interactions, without space and time restrictions. On the other hand, explicit diagrams include also the directed and sequential relationships of the pathways (25). Thus, heuristic and combinatorial diagrams can present the logic of the network in a human-readable format, while explicit diagrams are more suitable for computer simulations (26).

Several notations have been proposed for the representation of processes in interaction networks. Two well-known notations are the molecular interaction maps (MIM), proposed by Kohn and coworkers (24,27) and process diagrams, proposed by Kitano et al. (28,29). While being able to represent explicit diagrams, the MIM notation is prepared mainly for construction of heuristic and combinatorially complex diagrams (26). On the other hand, process diagrams are generally used for the construction of explicit diagrams (25). Detailed comparisons of notations can be found in the literature (25,30). Considering the number and characteristics of p73 interactions, we have decided to use the MIM notations for the representation of the p73 interaction network.

MIM notation and symbols

In a MIM each species is represented at one place only. The interactions, complexes and effects of the proteins or interactions are represented using defined symbols. In the MIM notation, the symbols representing the interactions are mainly grouped into two types, reaction symbols and contingency symbols (Figure 2b). By classifying the relations among nodes (proteins or promoters), the reaction symbols define physical interactions, protein complexes, covalent binding and other such events. Contingencies represent effects like inhibition and stimulation (25), which may emerge from and be directed to nodes or edges (reactions or contingencies of the MIM). The notation of MIM can be found in the literature together with examples (27).

Extensions made on the Kohn notation

The domain representation that is already available in the Kohn notation (24) has been extended by a line, which is drawn above the node, representing the whole protein, regardless of the domains. The distinction between the interactions known to take place between specific domains of the involved protein, and interactions without such available detailed information, is incorporated into the map (Figure 3). Interaction number 1 is specific to domain 1, but interaction number 2 implies no currently known domain specificity. Therefore, an isoform lacking domain 1 will not have interaction 1, whereas the status of interaction 2 is unknown, since the place of the binding site involved in the interaction is not known.

As discussed in the Introduction section, the ratio among ΔN and TA isoforms is the key factor in cancer research on p73 (17,22). Thus, it is reasonable to represent the ΔN and TA isoforms separately on the p73 MIM. Almost all interactions are common for the ΔN and TA isoforms except for the ones taking part through the TA domain. These common interactions are not shown on the ΔN isoform to avoid redundancy.

Assessment of data

The nodes of p73 MIM are grouped into functional categories, and the interactions are labeled accordingly. Each label is a combination of a letter, representing the category interaction it is associated with, and a number. While there is no such restriction, in most cases the numerical sequence of interaction labels represent the occurrence of interactions on the pathways (see Table 3, Annotations).

In the majority of the publications aiming to test the p73 activity level, the transcription levels of one selected protein, mostly p21, is measured experimentally and the effect is stated as transcriptional activation or inhibition of p73. Unless stated otherwise, these effects are accepted as related to transcriptional activation of all target promoters (combined promoter). Mapping it on the p73 MIM, the stimulation or inhibition effect is directed to the interaction labeled ‘A29’.

Clustering

In agglomerative hierarchical clustering, the clustering starts at the binary level and attempts to get higher degree clusters until a united cluster is obtained (31). The main criterion of the algorithm is the distance between the members. The nearer the members are, the higher the probability that they will get under the same tree branch. Keeping the distance measure constant, groups are constructed according to three norms: single-link (farthest neighbor), complete-link (nearest neighbor) and average-link. In single-link clustering, the distance is designated as the smallest; in complete-link as the largest; and in average-link as the mean distance between the clustered elements of each iteration (31).

The input for the clustering algorithm was a connectivity matrix (CM). CM is binary, square and its dimension is n × n, where n is the number of the nodes in the network. Each element (i,j) of the CM corresponds to the availability of an interaction between any proteins in the network. The interacting nodes are recorded as CM (i,j) = 1 and noninteracting nodes as CM (i,j)= 0.

For clustering purposes, the tool EXpression Analyzer and DisplayER (EXPANDER) is used (32). Average-link agglomerative hierarchical clustering is employed with a distance expression inversely related to the Pearson correlation coefficient.

MIM of p73 is a heuristic MIM constructed for p73 mediated cell cycle arrest and apoptosis

Using the MIM notations proposed by Kohn et al. (27) with the additional notations including the extended domain representation, distinct representation of isoforms (see Materials and methods section), and organizing the information related to the interactions, the p73 MIM is constructed. At the data gathering stage only the interactions which are directly or indirectly related to the function of p73 in cell cycle arrest and apoptosis are taken into account. The interactions of p73 related to its role in differentiation are excluded from the map. For ease of use in future studies, the interaction information is stored in both computer readable (using Systems Biology Markup Language) and human readable formats.

In the p73 MIM, p73 is placed at the center of the map. The interactions are organized around the p73 node, without time and space constraints. The diverse roles of the p73 domains in the p73 functions are presented in detail. All interactions are directed to the relevant domain when information is available. The resulting p73 MIM consists of 53 nodes (proteins) and 176 edges (physical interactions or functional relations) (Figure 2a). In addition to the nodes and edges, the p73 MIM also contains several promoters and definition boxes. The boxes represent states or events, such as DNA damage or degradation. The p73 MIM includes only end points of some pathways, where the leading path is not directly related to p73 function, such as cleaved PKCδ, PKCδCF. We excluded the details of each box and such pathways, since they are beyond the scope of this work. The proteins represented as nodes of the p73 MIM may have interactions other than those displayed in the map, related to their individual functions which are apart from p73 mediated apoptosis.

A list of proteins represented in the map with their accession codes is also provided (Supplementary Table 1). A softcopy of the MIM of p73 can be downloaded from http://www.prc.boun.edu.tr/PRC/software.html.

Proteins of p73 MIM are grouped into functional categories

The proteins involved in the p73 MIM are grouped into functional categories, according to their effects on the network (Table 1). The defined categories are Upregulation, Activation, Suppression, Degradation, and Transcriptional Activation. A protein may have roles in more than one pathway, belonging to different functional categories. In such cases, the protein is located in a category according to its major function based on current literature. However p73, its ΔN isoform and pRB could not be assigned to a single functional category and are represented as separate nodes, since their roles are critical in several paths. The edges are labeled according to functional categories so that the letters used in the labels of the edges are the initials of their related categories (see Materials and methods section). The details of all interactions and their corresponding references can be found at the end of the article, in the Annotations (Table 3). The Annotations are organized with respect to the defined categories and the labels of the edges.

In the related Results section, we investigated the following characteristics of each functional category: (i) the number of members of each category; (ii) important nodes which are assigned to each category and (iii) interrelationships between members of different categories.

The Upregulation functional category

The Upregulation category contains the nodes and edges that are related to the transcription of the p73. In this category, there are 6 nodes, 24 edges and 3 promoters. This group is directly related to the Activation category, since both categories are essential for p73 function. The key node of this unit is E2F1 which has the highest number of interactions and is of crucial biological importance in the activation of the p73 promoter. E2F1 is a transcription factor which controls the progression of the cell-cycle from the G1 to the S phase. E2F1 is activated via phosphorylation (U7; see label U7 in the Map and its details in the Annotations, Table 3). Active E2F1 can stimulate either cell cycle arrest or apoptosis, as in the case of the DNA damage. By enhancing the activity of ATM on Chk2 (labels U2, U11), E2F1 forms a positive feedback loop, where ATM activates Chk2 and Chk2 further increases the activity of E2F1 by phosphorylation. Additionally, E2F1 can bind to either p73 (U9) or the Apaf-1 promoter (T6) to stimulate the transcription of p73 and Apaf-1 (U10, T8). Furthermore, the activity of E2F1 is inhibited when it forms a complex with pRB. Under such circumstances, the concentration of E2F1 is regulated by pRB (U17–18).

There are controversial data in the literature regarding the viral oncogene E1A. According to some sources, E1A hinders the activation of p73 (33). In contrast, other studies (34) point out that E1A can activate p73, by binding to pRB and thus releasing E2F1. Following a comprehensive literature search, examining the sources of the arguments, and the dates of the relevant publications, it appears to us that E1A plays a role in the activation of p73 (U21–22). Consequently, only this aspect of the behavior of E1A is mapped on the p73 MIM.

The Activation and Suppression functional categories

The Activation and Suppression categories can be grouped under the same heading, since the nodes (proteins) belonging to those categories function by modifying p73 at the posttranscriptional level.

The nodes included in the Activation functional category are related to the posttranscriptional activation of the p73 protein. This category consists of 18 nodes, 2 promoters and 59 edges. Within this category DNA damage is represented by a box. p73 is activated when DNA damage results as an outcome of γ-IR or some chemotherapeutic drug (1). The main activation pathway of p73 is the signaling cascade initiated by the recognition of the damaged DNA by ATM (A3). After the activation of ATM, it activates c-Abl by phosphorylation (A8–10). c-Abl binds to the PXXP motif of p73 (A12) and causes its phosphorylation at the tyrosine (Y99) (A14) and threonine (A16) residues, directly and by activating p38, respectively. Phosphorylated p73 is recognized by Pin-1 isomerase with a consequent conformational change of p73. In the next step, p73 is recruited to PML-NBs where it is acetylated by p300/CBP and becomes fully active (A21, A25–27). In addition to the main pathway, there are several proteins which can lead to an increase in the activity of p73 either by direct binding or via posttranscriptional modifications.

The Suppression functional category contains 15 nodes, one promoter and 54 edges, which are related to the suppression of p73 function. The most important node in this category is MDM2, which is also a crucial node in the p53 network. MDM2 leads to the inactivation of p73, but does not degrade it. In addition, the MDM2 promoter is activated by p73, forming a negative feedback loop. Most of the proteins involved in p73 suppression are p73's first-degree interactions. The inhibitory effects of p73 SAM and ID domains, on the p73 protein itself are also assigned to the Suppression functional category (S41).

The Activation and Suppression functional categories are interrelated through few direct interactions, as can be seen from the p73 MIM. (Figure 2a) The functional relationship between these categories occurs largely at two successive steps, such that a member of one category blocks the effect of a member of the other, through p73. The Activation functional category is interrelated to the Upregulation by first-degree interactions, with a larger number of direct interactions as compared with the Suppression category.

The Degradation functional category

This category contains the nodes that are related to the degradation of p73. In this category, there are 7 nodes, 1 promoter and 22 edges. Among these seven, three regulate the ‘standard’ degradation mechanism via ubiquitination (D6–9, 11–14); one by sumoylation (D4–5, S22); three by forming a complex with p73 (D15–20).

Neddylated Cullin:Roc1 complex (D10–11) constitutes the core module of E3 ubiquitin ligase. The formation of the module stimulates the ubiquitination of p73 resulting in the proteasomal degradation of p73 (D12–14). In a similar manner, another ligase Itch, promotes an increase in the ubiquitination level of p73 (D6–9). Other than these, the SUMO conjugating enzyme, Ubc9 stimulates the sumoylation of p73 at the C-terminus (D4–5, S22). This is another way for p73 to be flagged for rapid degradation. Additionally, there is a negative feedback loop between p73 and Cyclin G. p73 stimulates the transcription of Cyclin G, while Cyclin G enhances its degradation (D19–22).

The nodes involved in this category exert their effects on p73 directly, without forming complex paths. There are no direct interrelationships with other categories.

The Transcriptional Activation functional category

The Transcriptional Activation category contains 6 nodes, 3 promoters, 17 edges and an Apoptosis box. In this functional category, the promoters which are activated by p73 are represented by one combined promoter (T1, T2) (see Materials and methods section). Currently available data on the interactions between p73 and specific promoters are provided in the Annotations (Table 3). If the edges directed to T1 are traced in the map, specific information about each promoter can be reached through the Annotations. Following TA of the promoters by p73, the translated proteins can promote cell cycle arrest and apoptosis.

The targets of p73 that are involved in the differentiation are not included in the list, since in principle the map is constructed for apoptosis. If a promoter is significant in the p73 MIM, other than being a target of p73, such as MDM2 promoter, then it is also represented in other relevant categories (e.g. S10). Other than the activated promoters, the apoptosis pathway that is induced by Bax and PUMA proteins (T9–10) and the inhibition of hTert transcription (T3–5) are of importance in this category. In the apoptosis pathway, PUMA and Bax are transcribed by p73 (T2). After being transcribed, PUMA causes conformational change of Bax, and induces its translocation inside the mitochondria (T9). Once Bax is located in the mitochondria, it causes cytochrome c release. Binding of Apaf-1 to the released cytochrome c leads to the formation of the caspase cascade and eventually apoptosis (T12–15). The hTert upregulation is a well-known cause of the maintenance of the telomerase length in most of human cancers. The binding of p73–Sp1 (T6), leads to inhibition of the hTERT transcription; thus is also important in the p73 mediated tumor suppression.

The interrelation of this category with other categories is limited to the interaction of p73 with promoters (T1) and activation of specific promoters by E2F1 (T3/6).

The MIM of p73 includes anticoherent interactions

To be preferred by evolution, the function of a protein should provide robustness to the cell. And mechanisms involved in a robust biological network should be regulated in a coherent manner to maintain homeostasis at any level of an organism (28). Therefore, if a specific form of a protein is a positive regulator of a particular event in one path, then its effect via other paths toward that outcome should also be positive. Similarly, a negative regulator should have negative effects on the same node through different interactions. In the literature, interactions of a protein may be stated to lead to opposing end points by using different paths. In such cases, the node is defined as anticoherent. When related edges of each node are tracked on the p73 MIM, two important proteins can be determined as anticoherent nodes of the system. These proteins are pRB, which is not assigned to any functional category, and c-Jun in The Activation and Suppression functional categories.

The activity of E2F1, which is discussed in detail under The Upregulation functional category section, is suppressed when E2F1 forms a complex with pRB (U17–18). With this interaction, pRB has a negative effect on the p73 activity. However, binding of pRB to RACK1 (S17) appears to have a positive effect on p73. RACK1 is a repressor of p73; when there is a physical interaction between p73 and RACK1, the p73 transcriptional activity is inhibited (S15–16). The pRB interaction with RACK1 frees the p73 from repression (S18). These two opposing actions suggest that pRB is an anticoherent node in the p73 tumor suppression network.

c-Jun appears to be a positive regulator of p73. It inhibits p73 degradation (D3) and stimulates p73 transcriptional activity (A53). On the other hand, c-Jun also binds to the MDM2 promoter and enhances its transcription (S8, S10). As mentioned in The Activation and Suppression functional categories section, MDM2 suppresses the p73 activity. These two opposing actions indicate that c-Jun is an anticoherent node in the p73 network. Interestingly, there is a direct relationship between those two anticoherent proteins. pRB binds to the c-Jun promoter and stimulates its transcription (A57–58).

The p73 MIM and the network of p53 share proteins with similar and opposing roles

A good starting point to understand the logic of the p73 tumor suppression mechanism would be a comparison between the p53 and p73 networks. The common and distinct features of the p53 and p73 networks may describe parallel or diverse functions of these tumor suppressors. Expression of the compared features of both p53 and p73 networks in the p73 MIM would require implementation of the p53 protein and p53 related pathways into the MIM. Such an implementation would cause divergence from the focus on p73 interactions in cell cycle and apoptosis in the MIM. Therefore, a brief comparison among the nodes in the p73 MIM and the members of the p53 network is made. For this purpose, a preexisting p53 MIM (24) and the data obtained from the literature are used. Among the compared nodes, 12 proteins (ATM, p300, Cyclin A/Cdk2, HIPK2, Chk1, Chk2, p21, 14–3–3σ, Gadd45, Bax, PUMA, Cyclin G) function in the same manner or lead to similar effects using different mechanisms, 2 (MDM2, c-Abl) have different roles and two nodes (c-Jun, CTF2) have diverse end effects on p53 and p73. A summary of the comparison is given in Table 2.

Following the interactions of ATM in the p53 network and the p73 MIM, it can be seen that ATM directly activates p53 by phosphorylation at Ser15 (35) and leads to activation of p73 indirectly (36), (A8). Pin1 imposes a conformational change on p73 and p53 (A17) (17). p300 binds to the TA domains of p53 and p73, to enhance their functional activities (24,37), (A23). Cyclin A/Cdk2 can phosphorylate p53 and p73 (38,17), (S43). Serine/threonine kinase HIPK2 physically interacts with the ODs of both p73 and p53, resulting in their colocalization into the nuclear bodies (A41). Chk1 targets the TA domains of p53 and p73 to improve the TA of p53 and p73. Chk2 mediates the posttranslational modification of p53 and p73, by direct and indirect means, respectively (40). Cyclin G interacts both with p53 and p73, causing downregulation of these proteins. (D19) The mechanisms differ such that the negative effect of Cyclin G on p53 is MDM2-dependent. In addition, some promoters e.g. p21, 14–3–3σ, Gadd45, Bax, PUMA (41), (T2) are also transactivated by both p73 and p53.

MDM2 is a significant protein in the p53 network, since it mediates ubiquitination and degradation of p53 (42). It is also associated with p73 by direct binding and instead of its degradation, binding of MDM2 results in stabilization of p73 (42), (S1). This binding also leads to the inactivation of p73, by disrupting the interaction between p73 and p300/CBP (43). The tyrosine kinase, c-Abl stimulates DNA binding by p53 through direct interaction, but it stabilizes p73 by posttranslational modification (24,44), (A12). c-Jun stabilizes p73, increasing its pro-apoptotic activity (A53), while it negatively regulates p53 (45). Another protein having opposing affects on p73 and p53 is CTF2. It enhances p53 activity but inhibits DNA binding to p73 (46), (S30).

There are also proteins that affect p73 without having any interaction with p53, such as NEDL2 (47), (A45); MM1 (45), (A37); adenovirus protein E4orf6 (48), (S23) and Itch, (D6). Among these proteins, Itch stands out as being the protein having a role in the degradation of p73. Itch functions in a manner similar to that of the MDM2 in p53 network (47).

Hierarchical clustering can define paths with small number of interactions, but not large functional categories

The interactions of the p73 MIM are defined at a binary level by two different approaches. In the first approach, only direct (first degree) physical interactions between proteins are taken into consideration. The results were unable to define meaningful clusters. They also excluded many proteins, which are related to the network via contingencies only. In the second approach, first-degree contingencies where the two proteins having an interaction defined by a contingency are also counted as binary interactions. However, contingencies affecting other contingencies (higher degree) are not included.

As a result of the second approach, the clustering program was able to define paths with small number of interactions in the network, although large clusters corresponding to functional categories could not be obtained (See Supplementary material). For relatively small categories (Figure 2a), namely Transcriptional Activity and Degradation, the proteins involved could mostly be distinguished in the hierarchical tree. The main pathway leading to apoptosis in the Transcriptional Activation functional category, starting from the PUMA and Bax transcription (T2), and ending at the apoptosis box (T15) is defined as a branch, where Bax and PUMA have a closer relation, merging with cytochrome c, Apaf-1 and the Apoptosis box. In the case of the Degradation functional category, the proteins influencing degradation of p73 by direct interactions are also defined as a branch including Cyclin G, Cyclin G promoter, NQO1, UFD2a and the degradation of p73. However, other members of this category which lead to the degradation of p73 by covalent modifications could not be mined from the given input information. In the Suppression and Activation functional categories, theÿ 11 nodes—including theΔNp73 isoform and its promoter—that affect the p73 activity in the same path, acting on T1, are well clustered. In addition to those mentioned clusters, there are also small groups (3–4 nodes) of interrelated proteins placed in close relation in the hierarchical tree, such as hTERT, hTERT promoter and Sp1.

Biological information on the interactions of p73 are organized in the p73 MIM with special emphasis on domain specificity

p73 has major roles in the embryonic differentiation and in the tumor suppression networks. Due to the scarcity of data, p73 interactions related to differentiation are not considered in this work. Here, we focus on the apoptotic function of p73. MIM notation (Figure 2b) is selected for the network representation. We added some extensions for the representation of domains (Figure 3) and isoforms. The extensions which emphasize the domain specificity made it possible to demonstrate whether an interaction is valid for a specific p73 isoform of interest. The data are stored in both human and computer readable formats for ease of use in future studies.

The constructed p73 MIM organizes the information related to the biology of the p73. While constructing the map, special attention was paid to the domain specificity of the p73 interactions. This feature makes the p73 MIM distinguishable as compared with other PPI maps. The organization of the 53 nodes and 176 edges of the p73 MIM into five functional categories provide a clear and compact representation of the biological mechanisms related to p73 function.

Functional categories demonstrate biologically significant interrelations

The functional categories identified from the nodes and edges of the p73 MIM provide a compact structure of the network, yielding five functional groups, Upregulation, Activation, Suppression, Degradation, Transcriptional Activation and their interrelationships.

The Upregulation and Activation categories are related via the node E2F1 of the Upregulation category, and the major pathway for p73 activation (A54/U11). When ATM (A1–7) is activated after DNA damage, it activates the stimulators of p73 (A8) and E2F1 (U2). In turn, E2F1 increases the activity of ATM (U11), forming a positive feedback loop. This loop is biologically significant and it can be followed from the p73 MIM. The PML-NB:p73 complex and c-Jun can inhibit the degradation of p73 (D2-3), relating the Activation and Degradation functional categories. Furthermore, recruitment of p73 into PML-NBs complements paths, which are stimulated by ATM upon recognition of DNA damage (A3). The existence of the interactions in different paths initiated by the same signal, DNA damage in this case, demonstrates the robust structure of this biological network.

In the Suppression functional category, instead of forming complex pathways for p73 suppression, the proteins mostly bind directly to p73 or inhibit the interactions that activate p73. When p73 is no longer required for the cell activity, it is degraded via sumoylation (D4–5, S22), ubiquitination (D6–9, D11–14) or direct association with proteins such as UFD2a or Cyclin G (D16–20) which may not need proteasomal activity. The Transcriptional Activation and the Upregulation functional categories are related via the activation of the Apaf-1 promoter by E2F1 (T3).

Through the map, we observe numerous paths for p73 degradation, and a rather simple composition for interactions relating to p73 suppression. p73 is kept at very low concentrations when there is no DNA damage (49), which demonstrates the biological priority of the degradation invoked by the suppression mechanism. Here, we illustrate that the structure of the Suppression and Degradation functional categories in the p73 network correlates with this biological preference.

Anticoherent interactions of the p73 MIM suggest further experimental investigation

Construction of the interaction networks allows identification of the regions where there are inconsistencies in the functions of proteins. By tracing the interaction paths, we point out two nodes, pRB and c-Jun, which are anticoherent in their tasks in the p73 MIM. pRB is considered as anticoherent, since it regulates p73 both negatively and positively, by suppressing the activities of E2F1 and RACK1 (U17–18, S17), respectively. c-Jun has strong effects on the positive regulation of p73, by repressing its degradation and stimulating its transcriptional activity (D3). In contrast to these positive effects, transcription of MDM2, a p73 suppressor, is also stimulated by c-Jun (S8). Furthermore, those two anticoherent nodes are interrelated: pRB binds to the promoter of c-Jun and enhances its transcription (A56–57).

The observed apparent inconsistency in the p73 network for pRB and c-Jun may indicate the existence of interactions at different spatial or temporal cell states, an experimental misidentification or missing data regarding these proteins. Either way, based on our analysis of the p73 interaction network, we propose that the pathways involving pRB and c-Jun, should be further explored experimentally to clarify the regulatory mechanisms in p73 network.

Comparison of the p73 MIM and the p53 network presents similarities and differences in the interactions

The compared nodes of the p73 MIM and of the p53 network can be classified into four categories: proteins (i) with the same effects in both networks; (ii) with similar effects on p73 and p53 via different mechanisms; (iii) with different, or in some cases, opposing roles in the networks and (iv) existing in the p73 network which have no effect on p53, as explained in the Results section. This classification indicates that p73 and p53 can use parallel or alternate interactions leading to apoptosis.

Eleven nodes are shared by the networks of p73 and p53. This fact emphasizes the crucial importance of p73 in cancer research. The common pathways also indicate that tumor formation caused by malfunctioning of the shared proteins could affect both p53 and p73, and that p73 could be an alternative tumor suppressor in cases of cancer formation related to p53 mutations. Finally, they indicate a competition between p73 and p53 in cases where they target the same promoters. Furthermore, proteins that do not exist in the p53 network and are related to the transcription mechanism of p73 can be good candidates for further research, since proteins related to the transcription of p73 have important roles in both activation and inhibition of apoptosis pathways.

Pathways with fewer interactions are extracted from the network by hierarchical clustering with introduction of first-degree contingencies

For cross-validating the manually organized functional categories, the p73 MIM is analyzed by a clustering algorithm, using two different input matrices. The first approach, which excluded the contingencies, was unable to define the paths of the network. In the second approach, which included both the first-degree physical interactions and the first-degree contingencies, the large categories, e.g. Activation cannot be thoroughly extracted from the network. These categories include many high-order contingencies; more importantly, common ending points of the pathways are mostly contingencies. However, via the additional information present in the second approach, nodes which construct pathways with small number of interactions can be clustered together and most members of small functional categories can be extracted. This improvement suggests that by using a proper representation of the MIM notation in a computer readable format, computer-based analyses can supply information. Such knowledge can be used for improvement of the maps, assisting in the construction of functional categories.

Tracing the edges of the map and considering the results of the network analysis will provide an organized way to review and analyze the underlying apoptosis machinery of p73. Such a study of p73 is essential both for experimentalists and theoreticians working with p73 or proteins in its network. Clearly, pointing to specific regions in the network may help in avoiding some misleading effects which are the outcome of the sparse data, an evident fact in cancer research.