INTRODUCTION

TOP ABSTRACT INTRODUCTION MAP CONVENTIONS MOLECULAR INTERACTION MAP DISCUSSION APPENDIX REFERNCES

The complexity of the molecular interactions implicated in cell regulatory networks challenges human comprehension. Current diagrams of molecular interactions are often ambiguous or incomplete. The preparation of more comprehensive regulatory network diagrams is both difficult and urgent. The difficulties are not due merely to the large number of reactions, because this is a feature also of the familiar metabolic pathway diagrams. They are due rather to complexities that rarely occur in classical pathway diagrams, such as multisubunit complexes, protein modifications, enzymes that are modified by other enzymes, and protein domains whose function is regulated by other domains of the same molecule. The present work describes and applies a diagram method designed to cope with these kinds of complexity.

Why do we need molecular interaction maps? First, it is often difficult to keep in mind all of the known interactions that may be pertinent to a particular experimental or theoretical question, and a molecular interaction map can be used in much the same way as a road map or electronic circuit diagram. Second, molecular interaction maps can suggest new interpretations or questions for experiment. Third, the act of preparing a molecular interaction map imposes a discipline of logic and critique to the formulation of functional models. Finally, the diagram convention provides a shorthand for recording complicated findings or hypotheses.

Another kind of difficulty in preparing useful maps is the incompleteness and uncertainty of knowledge, as well as the limited scope of applicability of some interactions. An important aspect of molecular interaction maps, as described here, is that they are linked to an annotation list that summarizes current information relevant to particular interactions and provides references. A molecular interaction map can therefore function also as a review. The maps can be updated interactively via the Internet and thus can provide a current summary of an area.

The current work describes the mapping conventions and uses them to build a molecular interaction map of the circuitry that governs the mammalian cell cycle and DNA repair machinery. Updated and corrected versions of the map will be accessible at the internet address discovery.nci.nih.gov.

	MAP CONVENTIONS

TOP ABSTRACT INTRODUCTION MAP CONVENTIONS MOLECULAR INTERACTION MAP DISCUSSION APPENDIX REFERNCES

Desirable Characteristics of Molecular Interaction Maps

A major consideration in the design of the diagram conventions was to facilitate tracing all the known interactions of any given molecular species. Accordingly, each molecular species should ideally appear only once in a diagram, and all interactions involving that species should emanate from a single symbolic construct. A second major consideration was a concise method to represent multimolecular complexes. Multimer proteins are common components of regulatory systems and sometimes function in large-scale multimolecular assemblies. Therefore, an extensible representation of such complexes was a fundamental requirement. A third major consideration was the representation of protein modifications, such as phosphorylations. One must be able to represent various modifications of a protein by unique graphical constructs. Meeting these goals simultaneously is a significant challenge.

Additionally, one must be able to show the actions or effects of each molecular species or interaction, including enzyme action and stimulation and inhibition of activity or binding. Often there are many interaction or modification sites having diverse effects on function. The potential number of modification-multimerization combinations is staggering, and we have barely begun to explore this vast domain experimentally. Representation of all of these possible combinations in a single diagram is obviously impractical. Nonetheless, it is important to be able to represent any combinations that may be significant.

Symbols and Rules

Because each molecular species in a diagram ideally should appear only once, interactions must be indicated by several types of lines connecting the species. The different types of interaction lines are distinguished by different kinds of arrowheads or other line endings, as summarized in Figure 1.

View larger version (37K): [in this window] [in a new window]   Figure 1.   Summary of symbols.

Multimolecular Complexes

Noncovalent binding between molecular species is indicated by lines terminated at both ends by barbed arrowheads. Thus noncovalent binding in most cases is represented symmetrically. In some cases, however, it is useful to distinguish a protein that has a receptor site from a protein that donates a peptide binding to the site, for example the binding of an N-terminal peptide of p53 to a pocket in Mdm2 (Kussie et al., 1996). The receptor end of an interaction line can be represented by a double-barbed arrowhead (Figure 1). This notation can serve to indicate targets of opportunity for pharmacological intervention. Noncovalent binding is generally assumed to be reversible; when binding is unusually tight, the interaction line can be drawn heavier.

Having defined an intermolecular binding symbol, we next need a representation of the complex itself. This is accomplished by placing a small filled circle (or "node") on the connecting line. An action of the complex can then be represented by an appropriate type of line emanating from the node. See, for example, the enzymatic action line emanating from the node representing CycD:Cdk4 in Figure 5 (reaction 5), enzymatic action being indicated by the open circle at the end of the line, in accord with the symbol definition table (Figure 1).

Multiple actions of a complex can be depicted conveniently by using multiple nodes on the same line; each node then refers to exactly the same complex. See, for example, the two occurrences of node a in Figure 3.

Thus only the monomolecular species are indicated by name, and the identity of the complexes is determined by tracing the connecting lines back to the monomolecular units.

To represent alternative or competitive binding of different proteins at the same site, the lines from the competing proteins are merged before connecting to the site. Figure 2 illustrates the conciseness and flexibility of this representation. Each possible dimer (a, c, e, or g), or dimer pair for a given monomer (b, d, or f), is defined by specific placement of a node. Effects specific to any combination of interactions therefore can be represented unambiguously. For example, the actions of Cdk1 that occur regardless of whether the partner is Cyclin A or B would be indicated by an action line emanating from node f.

View larger version (16K): [in this window] [in a new window]   Figure 2.   Concise representation of alternative binding modes. Example: heterodimers formed by Cyclins E, A, and B binding to Cdk1 or 2. (a) CycE:Cdk2; (b) Cdk2 complexed with either CycE or CycA; (c) CycA:Cdk2; (d) CycA complexed with either Cdk2 or Cdk1; (e) CycA:Cdk1; (f) Cdk1 complexed with either CycA or CycB. Note that a, c, d, e, and g each lie on a unique connector line, and each represents a unique heterodimer. Nodes b, d, and f, on the other hand, represent dimer combinations. This notation greatly simplifies the representation of multiple alternative interactions: for example, the interactions of p21, p27, or p57 with various Cyclin:Cdk dimers in Figure 6A. A formal rule, required to avoid ambiguity, is that lines representing alternative interactions must join at an acute angle.

An important feature of the line-and-node representation is that it is extensible. Binding interactions involving a node are represented in the same way as binding interactions involving a monomer. To see how this works, consider the example in Figure 3, which represents the interactions of E2F1, DP1, pRb, and an E2 promoter element. Each of the naturally occurring molecular combinations of these monomers is indicated by a node, the promoter element being treated like a monomer species. The two filled circles labeled a, being on the same line segment, represent exactly the same molecular species, namely E2F1:DP1. The other node representations are as follows: b, E2F1:DP1:pRb; c, E2F1:DP1 bound to promoter element E2; and d, E2F1:DP1:pRb bound to E2. Also indicated is transcriptional stimulation or inhibition occurring when the promoter element is occupied by E2F1:DP1 (node c) or E2F1:DP1:pRb (node d), respectively. (See Figure 1 for definitions of stimulation and inhibition symbols. Also, note that connecting lines may change in direction, for example, with right-angle bends, and that crossing lines do not affect each other.) Using this type of scheme, most multimolecular contingencies can readily be depicted.

View larger version (14K): [in this window] [in a new window]   Figure 3.   Representation of multimolecular complexes: stimulatory and inhibitory complexes of E2F1, DP1, and pRb. (a) E2F1:DP1 dimer; (b) E2F1:DP1:pRb trimer; (c) E2F1:DP1 bound to promoter element E2 (transcriptional activation shown); (d) E2F1:DP1:pRb bound to E2 (transcriptional inhibition shown). Note that the promoter element can be occupied either by E2F1:DP1 or by E2F1:DP1:pRb (alternative binding represented by interaction lines joined at an acute angle).

A complex containing multiple copies of the same monomer can be represented concisely by use of a ditto symbol, consisting of an isolated filled circle at the end of a single connecting line. See, for example, the representation of a homodimer in the lower part of Figure 1. The notation can be extended to higher homopolymers, as illustrated in Figure 4 for tetramerization of p53: three nodes are placed side by side to denote the three additional copies that, together with the identified monomer, make up the tetramer. This example shows the requirement of p53 tetramerization for binding to promoter elements and for phosphorylation of p53 at Ser15.

View larger version (13K): [in this window] [in a new window]   Figure 4.   Concise representation of homopolymers: formation and effects of p53 homotetramer. (1) The three additional copies of p53 monomer required to make up the tetramer are represented by three nodes placed side by side and linked to the identified p53 monomer; a node placed internally on this line represents the homotetramer itself. (2) p53 tetramer can bind to promoter element. (3) Tetramerization stimulates (or is required for) phosphorylation of p53 Ser15 by DNA-PK.

A given protein often can form stable complexes simultaneously with two or more different protein molecules. These are of special interest because the interactions can extend to form large-scale, functionally integrated multiprotein assemblies. When there is an extensive chain of interactions, however, it is often unknown how local interactions in different parts of the chain affect each other. The map conventions allow the binary interactions to be depicted without specifying all of the possible influences between different parts of a chain of interactions while at the same time specifying those influences that are known.

Covalent Modifications

Most regulatory proteins are subject to a multiplicity of modifications, especially phosphorylations, that alter function. This presents a severe challenge to any diagram method. The difficulty is further increased when combinations of phosphorylations must be considered.

A phosphorylation (or other covalent modification) is represented by a line with a single barbed arrowhead that points to the modified protein (Figure 1). The multiplicity of phosphorylations and acetylations of p53 are included in the comprehensive diagram, Figure 6B. The modifications are arrayed along the length of the elongated, pill-shaped p53 outline from the N terminus on the left to the C terminus on the right. The amino acid positions of the modification sites are indicated. A node on a single-barb-arrowed line represents the protein modified at that site. The effects of a given modification on intra- or intermolecular actions are indicated by interaction lines emerging from the node.

Although the effects of multiple modifications on one another are largely unknown, some important interactions among modifications within the same protein molecule have been defined. To represent combinations of modifications, we need additional symbols. We use a nonarrowed connecting line to represent joint modifications; a node on such a connecting line represents the protein having the combined modifications. An example of this situation is provided by the phosphorylation states of pRb (Figure 5). pRb is multiply phosphorylated by CycD:Cdk4 and by CycE:Cdk2, but several of the sites differ. It seems that phosphorylation by CycD:Cdk4 is required before CycE:Cdk2 can phosphorylate its specific sites, and that both kinases are required to fully impair the inhibitory binding of pRb to E2F:DP complexes.

View larger version (13K): [in this window] [in a new window]   Figure 5.   Phosphorylation control of pRb: illustration of the use of conjunction symbols to denote protein modification combinations. (1) Cyclin D binds Cdk4; the filled circle (node) on the line represents the CycD:Cdk4 complex itself. (2) Similarly for Cyclin E and Cdk2. (3) The single-arrowed line linking P(D) to pRb represents phosphorylation of pRb at sites kinased by CycD:Cdk4; a node on this line represents pRb-P(D) [the P(D)-phosphorylated form of pRb]. (4) Similarly for pRb phosphorylated at sites, P(E), kinased by CycE:Cdk2. (5) CycD:Cdk4 phosphorylates pRb at sites P(D). (6) CycE:Cdk2 acts on pRb-P(D), generating fully phosphorylated pRb; the line with the filled arrowhead indicates stoichiometric conversion of pRb-P(D) to the fully phosphorylated form, which is represented by the node on the nonarrowed line connecting the pRb-P(E) and pRb-P(E) nodes. (7) E2F1 binds DP1. (8) pRb binds to E2F1:DP1. (9) Fully phosphorylated pRb can-not bind to E2F1:DP1. Thus dissociation of pRb from E2F1:DP1 requires hyperphosphorylation by Cyclin E:Cdk2, which in turn requires previous phosphorylation by Cyclin D:Cdk4/6 (Zarakowska and Mittnacht, 1997; Lundberg and Weinberg, 1998).

Multidomain Structures

Regulatory proteins often are composed of structural domains having different functions. This multidomain organization can integrate interactions with other molecules or communicate functions within a given protein molecule. To depict functional localization within a protein molecule, a horizontal, pill-shaped outline is used rather than an oval. Localized interactions or modifications proceeding from the N terminus on the left to the C terminus on the right are marked along the upper and/or lower borders of the pill shape (e.g., p53 in Figure 6B). When the locale of an interaction is unknown, it is marked at either end of the pill shape.

	MOLECULAR INTERACTION MAP

TOP ABSTRACT INTRODUCTION MAP CONVENTIONS MOLECULAR INTERACTION MAP DISCUSSION APPENDIX REFERNCES

Figure 6 presents a comprehensive molecular interaction map of regulators of cell cycle and DNA repair processes. The current map, however, is limited to events in the mammalian cell nucleus. Because of limits on what can legibly be formatted onto a journal page, the map is divided into two parts. Figure 6A maps the interactions involving E2F, pRb, Cyclin, and Cdk family members, their activators and inhibitors, as well as some important interactions with other components. Figure 6B focuses on the p53-Mdm2 subsystem and on subsystems related to DNA repair. Molecular components are grouped in putative subsystems according to mutual interactions or functional coherence. It remains to be determined whether subsystems can be identified on the basis of objective criteria.

View larger version (216K): [in this window] [in a new window]   Figure 6.   Molecular interaction map of the regulatory network that controls the mammalian cell cycle and DNA repair systems. The map is presented in two parts: A focuses on the E2F-pRb and Cyclin:Cdk subsystems and functionally related interactions; B focuses on the p53-Mdm2 and DNA repair subsystem and functionally related interactions. Some duplication between the two parts of the maps was unavoidable but was noted by using italicized species or subsystem names to indicate that their interactions are shown in greater detail on the other part of the map. Although interaction symbols are fully defined by the conventions listed in Figure 1, the following colors were used to enhance readability: black, binding interactions and stoichiometric conversions; red, covalent modifications and gene expression; green, enzyme actions; blue, stimulations and inhibitions. See Appendix for annotations.

The monomolecular species included in Figure 6 are listed alphabetically in an index (Table 1), which gives grid coordinates to help locate each species.

Some of the interaction patterns shown in the map deserve special comment. References are cited in the annotation list and can be found using the identifiers marked on the interaction lines in the map. These identifiers are italicized letter-number combinations, which are used also in the following text to specify particular interactions on the map.

We are concerned here primarily with molecular interactions rather than biological effects. The latter may eventually be understood as emerging from the former.

The E2F-pRb Box

The possible occupancy states of an E2F recognition element in a promoter comprise a multiplicity of complex patterns. The method of representing these patterns was shown by the simplified example in Figure 3. The simplest arrangement would be an E2F:DP heterodimer bound to an E2 element (E5). (The italicized letter-number combinations identify particular interactions in Figure 6 and refer to entries in the annotation list where references are cited.) Because promoters generally have two or more E2F recognition elements, however, the actual situation even in this case may be far from simple. Most E2F family proteins can activate transcription (E6) by way of a transactivation domain. E2F-6, however, has instead a transcription repressor domain (E9). E2F:DP heterodimers (other than those of E2F-6) can tether pRb family proteins onto promoters, thereby converting a potential gene activator into a repressor (E7). E2F in fact more often functions to repress rather than activate.

E2F proteins (as heterodimers with a DP protein [E1]) have individual binding preferences for pRb family members. The E2F-pRb box shows that pRb prefers E2F-1-, -2-, or -3-containing heterodimers (E2) (with lesser interaction with E2F-4), p107 binds only E2F-4 heterodimers, and p130 can bind heterodimers of E2F-4 or 5 (E3,4). All of these complexes can bind and repress E2F recognition elements (E6). E2F-6, however, does not bind any pRb family protein. E2F-1, -2, and -3 are marked together as a unit on the map, because of their mutual preference for pRb. Although there are significant functional distinctions between them, the molecular bases for these differences is not known. No major functional differences among different DP family members have been defined.

A further level of complexity arises from the ability of some E2F complexes to bind other transcription factors, such as Sp1, and to synergize the transcriptional activation.

Another mode of regulation arises from the ability of histone deacetylase (HDAC1) to bind to pRb:E2F:DP-type complexes (E13). HDAC1 could deactivate transcription that has been enhanced by histone acetylation (H1) and thus could contribute to gene down-regulation by pRb family proteins.

pRb sometimes functions as a gene activator rather than repressor. For example, it can activate the Jun family and CCAAT/enhancer-binding protein (C/EBP) transcription factors (E16,17). The detailed mechanism by which this occurs is not known.

Taken together, the interactions noted on the map add up to perhaps 20 different possible states of an individual E2F recognition element.

The interaction capabilities of the constituent proteins are subject to modulation by phosphorylation, protein binding, and regulated degradation. In addition, regulated nuclear-cytoplasmic transport is emerging as an important process. Translocations can be represented as indicted in Figure 1 but, because of space limitations, have been omitted from the current version of the map. pRb is subject to different sets of multiple phosphorylation by CycD:Cdk4/6 (C31) and cycE:Cdk2 (C32). The manner of depiction of these phosphorylations and their effects was explained in Figure 5. Only the fully phosphorylated pRb is impaired with respect to E2F binding (C33). The E2F binding of p107 and p130 is also inhibited by phosphorylation (E11). Phosphorylation of E2F and/or DP by CycA:Cdk2 (which forms stable complexes with E2F-1) (E20) inhibits the E2F-DP interaction and could serve to turn off E2F function when cycA accumulates late in S phase.

pRb family proteins may also be inhibited by binding to other proteins, such as Raf-1 (E22). This may be one of the logical connections, suggested by current work, which may communicate signals from the cell surface to the cell cycle control circuitry.

The Cyclin-Cdk Box

The activity of Cdks is intricately regulated. To begin with, Cdk activity requires binding to a Cyclin. The map shows Cdk4 or Cdk6 (which have the same molecular interactions) binding to CycD (C3), Cdk2 binding to Cyclins E or A (C4), and Cdk1 (also known as Cdc2) binding to Cyclins A or B (C5). (The subtypes of CycB and CycD are not differentiated here.)

A second class of controls on Cdks are stimulatory and inhibitory phosphorylations, which are controlled by several kinases and phosphatases noted on the map. All Cdks are activated by phosphorylation of Thr160 (or 161), carried out by CycH:Cdk7 (C14), which functions also as a constituent of the transcription factor IIH (TFIIH) complex. Cdk1, and to some extent other Cdks, can be inhibited by phosphorylations corresponding to Thr14 and/or Tyr15 (C17,19,20). These sites are phosphorylated by Wee1 or Myt1 (C16). In the case of Cdk1, these inhibitory phosphorylations are removed by dual-action phosphatase Cdc25C, which is in turn activated by phosphorylations (C18) introduced by mammalian polo-like kinase 1 (Plk1) (C37) and/or Cdk1 (C36). Cdc25C can be phosphorylated at Ser216 by Chk1 (C38) or C-TAK1 (C39). Ser216 phosphorylation generates a binding site for 14-3-3, and this binding inhibits the phosphatase (C40).

A positive feedback loop that can be traced on the map consists of just two components: Cdc25C and CycB:Cdk1. Cdc25C is activated by hyperphosphorylation (C18); activated Cdc25C dephosphorylates Thr14 and Tyr15 of Cdk1, thereby removing the inhibitory effect of these phosphates on the kinase (C17) and increasing the activating phosphorylation of Cdc25C (C36). This positive feedback could help produce switch-like behavior and may operate in the G2 to M cell cycle phase transition.

A third class of controls acts through the binding of specific Cyclin:Cdk inhibitors, including p16ink4a, p21cip1, p27kip1, and p57kip2. p16ink4a inhibits by binding Cdk4/6 in competition with cycD (C8). p21cip1, p27kip1, and p57kip2 can bind Cyclin complexes of Cdk4/6 and Cdk2 (C7,23). There may be an additional complication, however, because p21 can stabilize and enhance the activity of cycD:Cdk4 when a single p21 molecule is bound but can inhibit the same activity when a second p21 molecule binds to the complex (C22). p27 can be phosphorylated by the kinase it inhibits, CycE:Cdk2 (C21). This seemingly paradoxical relationship might be due to intermolecular action of an active CycE:Cdk2 on an inactive CycE:Cdk2:p27 complex.

The Cyclin:Cdk system can interact with elements of the DNA replication and repair systems through binding of p21 (R6) or Cyclin D (R11) to proliferating cell nuclear antigen (PCNA). This action may also involve Gadd45, which can bind simultaneously to p21 (C34) and PCNA (R10). p21, PCNA, and Gadd45 are all transciptionally activated by p53 (P43,44).

The p53-Mdm2 Box

The map shows the remarkable richness of p53 interconnections and the diversity of functionally determinative p53 modifications. Eleven phosphorylation or acetylation sites (or groups of sites) for which functionality has been surmised are shown. If all of these could occur independently, there would be ~2000 possible modification states of p53 monomers. Some interdependent modifications have been noted: phosphorylation of Thr18 requires previous phosphorylation of Ser15 (P3) (Appella, personal communication); acetylation of Lys320 requires tetramer structure of p53 and is inhibited by phosphorylation of Ser378 (P23) (Sakaguchi and Appella, personal communication). Other dependencies certainly exist, some perhaps having major functional impact, whereas many could have subtle quantitative effects, which may or may not convey a selective evolutionary advantage. Nevertheless, p53-expressing cells may contain hundreds of different modification states of p53 monomers.

Some p53 modifications and interactions are especially notable. Ser15 appears to be the site of phosphorylation responses to DNA damage signals communicated by way of the kinases ataxia telangiectasia mutated gene/protein (ATM) (P2) and DNA-dependent protein kinase (DNA-PK) (P6). Phosphorylation of Ser18 or Ser20 prevents stable binding to Mdm2 (P5), thus abrogating the Mdm2-mediated inhibition (P29) and degradation (P31) of p53. Because these sites are located within the region required for Mdm2 binding, it is plausible that their phosphorylation could inhibit this interaction. p53 forms a stable complex with p300 (P25), as a result of which p300 acetylates p53 Lys382 (P21). This acetylation, as well as the acetylation of Lys320 by PCAF (P23), enhances the sequence-specific binding of p53 to promoters, probably indirectly by inhibiting nonspecific DNA binding (P22). A similar mechanism of enhanced promoter binding (P16) may occur as a result of binding of the p53 C-terminal region to 14-3-3 (P13), which requires 14-3-3 to be dimerized (P14).

p53 can bind to a number of proteins that are involved in DNA repair functions, cell cycle control, or general control functions. In approximate order of binding location from N to C termini of p53, these include the following: Mdm2 (which has a pocket that binds a p53 N-terminal peptide) (P28); p300 C-terminal region (P25); DP1 (P26); poly(ADP-ribose) polymerase (PARP) (P46); c-Abl (A4); replication protein A (RPA) (S6); high-mobility group protein (HMG) (P52); TFIIH constituent helicases xeroderma pigmentosum complementation group B (XPB) and XPD and DNA repair protein CSB (P27); p19ARF (P40); p300 N-terminal region (P33); BRCA1 (P47); and 14-3-3 (P13). Some of these interactions (p300, BRCA1, and 14-3-3) stimulate and some (Mdm2, PARP, and RPA) inhibit the transcriptional activity of p53. (Stimulations may be indirect: 14-3-3 may block nonspecific binding of p53 to DNA [P16]; p300 may do the same consequential to acetylation of K382 [P21,22].)

p53 can form homotetramers and must be in tetramer form for sequence-specific DNA binding and transcriptional activation. The map shows the dependencies relating to tetramers; tetramerization is stimulated by phosphorylation of Ser392, and this enhancement can be inhibited by phosphorylation of Ser315 (P17). The ability to form tetramers is further modulated by other modifications and protein interactions. Influence on tetramerization may be how p53 transcriptional activity is stimulated by binding BRCA1 and inhibited by binding Mdm2, PARP, or RPA. This could be the major mechanism of regulation of p53 transcriptional activity. The activation of p53 seems to be exquisitely controlled by a large number of determinative inputs. The transition to active tetramers could be very sharp because of a possible fourth-power dependence on the concentration of tetramerization-competent monomers.

Mdm2 is an intimate part of the p53 control system. Mdm2 contains a pocket that binds a p53 N-terminal peptide (P28). Mdm2 binding blocks the transcriptional activation domain of p53 (P29) and is instrumental in p53 degradation (P31). p53, in turn, transcriptionally up-regulates Mdm2, probably forming a negative feedback loop. Mdm2 can itself activate some genes, such as Cyclin A (P37). In addition to binding p53, Mdm2 reportedly binds E2F1:DP1 (P35), pRb (P35), TATA-binding protein (TBP) (P36), TBP-associated factor II250 (TAF_II250) (P37), p19ARF(P34), and p300 (P33). Some of these interactions may compete for the same Mdm2 site, as may be the case for p19ARF and p300. Binding to p53 is abrogated by phosphorylation of Mdm2 on Ser17, perhaps through the kinase activity of DNA-PK (P49).

p19ARF, an alternate reading frame (ARF) product from the ink4a locus that also codes for p16, has recently emerged as an additional player in the p53-Mdm2 system. p19ARF binds to and inhibits the actions of Mdm2 (P34,41). It also can bind to p53 (P40). Moreover, p19ARF is transcriptionally up-regulated by E2F1:DP1 (P42). This link between p53 and E2F1 may be crucial to the control of S-phase and apoptosis.

DNA Repair

The map includes three phases of nucleotide excision repair (NER). The first phase, lesion recognition and local opening of the DNA helix, is carried out by a molecular assembly, which includes the XPC:HR23B heterodimer (N1,2), XPA (N3), and TFIIH (N13). This phase opens the DNA helix in the vicinity of the lesion and allows access to other DNA repair proteins. If the DNA is opened by another process, such as transcription, XPC is dispensable, and repair can begin with the second phase.

The second phase, excision of a short DNA strand segment containing the lesion, is carried out by an assembly of the XPG and XPF:excision repair cross-complementing 1 (ERCC1) endonucleases (N6,8,9), together with XPA, RPA, TFIIH, and PCNA (which can bind XPG) (R9). This assembly appears to be held together in part through RPA, which binds to single-stranded DNA (ssDNA) regions in the vicinity of lesion (N10), and at the same time may be able to bind XPG and XPF:ERCC1 (N7), as well as XPA (N4). In going from the recognition to the excision phase, the molecular assembly rearranges as XPC:HR23B is replaced by XPG (N13) (Wakasugi and Sancar, 1998).

In the third phase, gap filling, x-ray repair cross-complementing gene/protein 1 (XRCC1) appears to function as a platform for the assembly of DNA polymerase  (DPase ) (N20), DNA ligase III (N19), and PARP (N18). This assembly is held together, in part, via breast cancer protein 1 C-terminal module (BRCT) modules in the constituent proteins (Masson et al., 1998). The binding of PARP by XRCC1 may function to block the further action of PARP during this phase at a repair site. The assembly of XRCC1 with DPase  and DNA ligase III may also function in the single-nucleotide replacement pathway of base excision repair (Cappelli et al., 1997).

Through its binding to DNA single-stranded regions, RPA may also recruit Rad52 and Rad51 to sites of DNA damage (S14,15). Rad51 also binds to ssDNA and, together with RPA, may function in recombinational repair (N11).

Rad51 may also provide links to a network of mutually interacting components via its binding to c-Abl (N12). c-Abl may bind ATM (A1), DNA-PK (B7), pRb (E18), and p53 (A4), although it is not fully determined which of these interactions can occur simultaneously and which are mutually exclusive. Rad51, ATM, and DNA-PK bind to a Src homology 3 (SH3) domain in the c-Abl N-terminal region, whereas pRb and p53 may bind to the c-Abl C-terminal region. c-Abl may regulate Rad51 function by phosphorylating Rad51 on Tyr54, thereby abrogating the direct binding of Rad51 to ssDNA (N12).

Another set of interactions is implicated in the processing of DNA double-strand breaks. DNA double-strand ends are recognized and bound by the Ku70:Ku80 heterodimer (Ku) (B1,2), which can recruit DNA-PK to the site (B3), thereby activating the kinase (B4). DNA-PK, however, can also bind to RPA (S16). RPA is a heterotrimer (S1,2) that binds ssDNA regions (S3). DNA-PK can thus be recruited to ssDNA regions formed transiently at replication forks. DNA-PK may then be available for interaction with a double-stranded DNA (dsDNA) end, which could appear in the vicinity as a consequence of replication fork encounters with open topoisomerase I DNA complexes trapped by drugs such as camptothecin (Shao et al., 1999). It is noteworthy that DNA-PK does not always require Ku for activation, because it can be activated by tethering to DNA via other molecules, such as chromatin constituents of the HMG family (B10).

A further capability of RPA could arise from its ability to bind p53 (S6) and from the abrogation of this binding by phosphorylation of the RPA2 subunit by DNA-PK (S11), ATM (S12), or CycA:Cdk2 (S10).

We thus begin to see some of the intricate mechanism of the DNA repair machinery. This example shows how a molecular interaction map can represent DNA-targeted processes and the transitions between multimolecular assemblies.

	DISCUSSION

TOP ABSTRACT INTRODUCTION MAP CONVENTIONS MOLECULAR INTERACTION MAP DISCUSSION APPENDIX REFERNCES

Biological functions must eventually be understood as arising from molecular interactions. It is therefore not surprising that the molecular interaction information so far accumulated forms a highly complex network whose functional behavior may be difficult to comprehend. To provide a foundation for eventual understanding of function, this information must be organized in a manner that allows integrated behavior to be discerned. Molecular interaction maps such as those described here could contribute to this goal. The primary objective here was to suggest how complex molecular interaction networks can be usefully displayed. The map was constructed from evidence relating to molecular interactions with the view that the interaction patterns would suggest biological functions. The complexity of the map, however, demands that great care be taken in the formulation of specific functional hypotheses, which may have to be investigated with the aid of computer simulations (Kohn, 1998).

The molecular constituents were tentatively grouped into subsystems, demarcated on the map by dashed boxes. Some of the numerous connections between the boxes can be arranged in tracts, perhaps analogous to nerve tracts in the brain or communication buses in computers. A task for the future is to find an objective way to make these groupings, or indeed to determine to what extent nature has designed subsystems within the control network.

The map includes 26 individual p53 modification or binding interactions for which evidence of functionality has been presented. Considering the number of modifications and binding combinations that are possible a priori, the number of different possible states of p53 could be so large as to raise the question of whether, at any given time in a cell, any two p53 molecules would likely be in the same state. It seems plausible, however, that certain combinations of states would be strongly favored under particular circumstances.

p53 appears to be a central focus for the concurrence of signals from many pathways, presumably serving as an integrated information processor. It might, for example, function to test the validity of signaling patterns and, depending on the outcome of the test, to initiate cell cycle arrest or apoptosis. The test however must first be activated by allowing p53 (which normally is rapidly degraded and thus nearly absent) to accumulate in response to inputs involving Mdm2 and/or p19ARF. It appears to be a logic unit that monitors the state of the cell based on the pattern of a large number of inputs.

The molecular interaction map helps one discern possible multiprotein assemblies, alternative arrangements of which may operate under different circumstances in a cell or under particular states of cell differentiation. A recent example and possible paradigm is the large multiprotein binder and acetyl transferase p300, which may function as a platform for the assembly of high-order complexes (Grossman et al., 1998). Multiprotein interactions may be competitive, cooperative, or independent of each other, possibly depending on the modification states of the component molecules. Moreover, susceptibility to modification may depend on the multimolecular configuration. A given protein may have more than one binding site for a second protein: p300, for example, has two separated sites for binding of p53 (Grossman et al., 1998). Multimolecular complexes thus may have alternative, functionally different configurations, and covalent modifications could induce configurational switches. The details of these interdependencies may be critically important, but only fragmentary information is as yet available.

Promoter regulation can be very complicated, because transcription factors can have positive or negative effects, depending on their environments and mutual interactions. Genes controlled in part by the same transcription factors sometimes exhibit different regulation patterns. Moreover, the interactions among the different transcription factors are only beginning to be elucidated. It was therefore not feasible, in the present diagrams, to show the full regulation pattern of each individual gene. The regulation of the E2F-dependent S-phase genes and of the p53-dependent genes highlight the difficulties. The major known regulatory actions on these genes are shown (Figure 6, A and B), and some of the details regarding effects on individual genes are mentioned in the annotation list.

A diagram convention such as that described here will be needed for the representation of functional models. These "heuristic" diagrams do not fully define the contingencies among the interactions. For simulation of models, however, "explicit" or fully defined diagrams are needed. Suitable explicit diagrams can be prepared using a subset of the present symbols (Kohn, 1998) or using the conventions of electronic circuitry (McAdams and Shapiro, 1995).

This exercise has suggested how a large body of molecular interaction data can be organized in a map with associated annotations. Many of the known interactions relating to cell cycle control and DNA repair events in the nucleus were included in the map. Some important areas remain to be added: in particular, the control of nuclear-cytoplasmic shuttling of key regulatory molecules and the signaling pathways from growth factor receptors on the cell surface.

The molecular interaction map will require frequent updating as new information accrues. For this purpose, we plan to put updated versions the map on the Internet (http://discover.nci.nih.gov)