Introduction

Twenty-five years ago, a 53–54 kDa cellular protein now referred to as P53 was reported to interact with viral oncoproteins.1, 2, 3, 4, 5 Initial studies suggested that P53 functioned as an oncogenic protein. Interestingly, recent studies that modeled two such mutations in the mouse suggested that they ‘gained function’ by binding to and attenuating the functions of the P53-related proteins P63 and P73.6 This provides one mechanism by which some P53 mutants function as oncogenes. However, structural gene mutations that disable P53 function, or that could engender dominant-negative properties, have been reported in about 50% of human cancers, including those that contribute most to morbidity and mortality (e.g., see http://www-p53.iarc.fr/index.html). Clear loss of both p53 alleles has been reported in a significant fraction of human cancers.7 Together, these data demonstrate that p53 is an important tumor suppressor.8, 9 Importantly, a substantial fraction of human tumors express wild-type p53, but its function is compromised by aberrant expression of proteins that negatively regulate it, such as Hdm2 and HdmX (Hdm2 or HdmX designate the human proteins, and Mdm2 and MdmX the mouse; the corresponding genes are hdm2, hdmx, mdm2, mdmx).10, 11, 12, 13 (Below, for simplicity, these regulators are collectively referred to as Mdm2 or MdmX.) Together, the data show that the P53 pathway is disabled in the majority, if not all, human cancers.

P53 is an unstable transcription factor that can regulate numerous downstream targets to induce permanent or reversible cell cycle arrest, apoptosis, and DNA repair.14, 15 While P53 has also been reported to induce apoptosis by nontranscriptional mechanisms,16, 17, 18 substantial evidence indicates that its transcriptional functions are critical for effective tumor suppression.19

Signals elicited by stresses including DNA damage (e.g., see, see, Huang et al.,20 Kastan et al.,21 and Wahl and Carr22), short or abnormally structured telomeres,23, 24 high-level oncogene signaling,25, 26, 27, 28 hypoxia,29 and glucose availability30 activate P53. Cells that encounter such conditions either arrest cell division or die if they possess a functional P53 pathway. By contrast, if P53 is dysfunctional, cells proliferate when exposed to oncogenic stimuli, unrepaired DNA damage, or metabolic perturbations that induce chromosome instability.31, 32 Thus, by eliminating P53 function, key controls to prevent cell cycle entry under inappropriate, potentially genome destabilizing conditions are lifted, allowing outgrowth of the genetically unstable variants that fuel tumor progression.

Since P53 output has the potential to kill cells, stringent regulatory mechanisms have evolved to prevent its errant activation, as well as to allow for rapid activation when appropriate. However, the regulatory mechanisms that enable the P53 pathway to generate different transcriptional responses to different stimuli in different tissues remain to be defined. There is significant medical importance to having a clear idea of P53 regulatory mechanisms as 3 million human cancers per year are estimated to contain a wild-type P53 protein whose function is attenuated. Understanding the mechanisms underlying such negative regulation affords a substantial opportunity for therapeutic intervention. This possibility has been realized with the development of agents such as cis-imidazolines (e.g., Nutlin 3A) that interfere with Mdm2–P53 interactions to activate P53 in cells with overexpressed Mdm2.33 Below, I review data from cell culture and mouse models that provide a current view of mechanisms that keep P53 off in unstressed cells, allow its activation in response to conditions that produce double-strand breaks, and then turn off P53 upon stress abatement.

Integration of P53 Structure with Regulation

P53 structure suggests many potential levels at which its transcription function could be regulated (Figure 1). Transfection studies suggest that P53 contains one N-terminal transactivation domain (TAD1) comprising the first 40 amino acids,34 and a second (TAD2) consisting of amino acids 43–63 that is revealed upon inactivation of TAD1.35 A potential conformational switch may be contained within a domain comprising proline-X-X-proline motifs between the TAD and the large central DNA-binding domain.36, 37, 38, 39 A C-terminal oligomerization domain is necessary to form tetramers.40 These tetramers not only constitute the most active transcriptional form of P5340, 41, 42, 43, 44, 45, 46, 47 but also conceal the dominant nuclear export signal that lies within the oligomerization domain.48 The lysine-rich C-terminus constitutes a domain implicated in both positive regulation by post-translational modifications involving acetylation,49, 50, 51, 52 methylation,53, 54 and sumoylation,55, 56 and negative regulation by ubiquitylation57 or neddylation.58 The C-terminus has also been proposed to function as an allosteric regulator.59, 60, 61 However, recent data make it more likely that this domain enables P53 to bind nonspecifically to DNA to increase the rate at which specific P53 response elements are detected through linear diffusion.62, 63

Figure 1
figure 1

A competition model for positive and negative regulation. This is a very simplified view of p53, the N-terminal binding region for Mdm2 and MdmX, as well as for p300/CBP. The structure of the p53 alpha-helical region binding within the hydrophobic cleft of Mdm2 is shown. The N-terminal region of MdmX that binds p53 is very similar. The positions of S15, T18, and S20 are shown as they have been implicated in altering the N-terminal structure when phosphorylated to either decrease Mdm2 (MdmX?) binding or increase association with p300/CBP. Also shown in simplified form is the introduction of either positive or negative post-translational modifications to multiple C-terminal lysines by either the HATs or by the E3 Mdm2. The N-terminal residues L22W23 that have been shown to prevent Mdm2 binding and severely reduce transactivation when mutated to Q22 and S23 are shown in the primary structure and in their locations deduced from crystallographic analyses. See the text for more details and references

Full size image

The N-Terminal TAD

Post-translational P53 modifications have been proposed to induce structural alterations to stabilize the protein and to effect the chromatin modifications needed for transcriptional activation. The extreme N-terminal TAD has residues critical for the binding of both the coactivators needed for transactivation, and the E3 ubiquitin ligase MDM2 that contributes most significantly to determining the abundance and short half-life of P53 in unstressed cells.19 Indeed, mutation of just two amino acids within this domain (L22Q, W23S in human, and L25Q, W26S in mouse; referred to as P53QS below) generates a stable but largely inactive P53 as the mutant can neither interact with Mdm2 (and presumably MdmX, which has a similar P53-binding domain) nor with the required coactivators.

One attractive model is that DNA damage activates the ATM kinase to induce a phosphorylation cascade beginning with human ser15 (i.e., h-ser15), equivalent to mouse ser18 (m-ser18), which triggers the subsequent phosphorylation of h-thr18 (m-thr21) culminating in phosphorylation of h-ser20 (m-ser23).64 These residues constitute part of a helical domain that interacts with a hydrophobic pocket in the Mdm2 N-terminus.65 In vitro studies with peptides from this region indicate that thr18 phosphorylation can significantly destabilize this structure, leading to Mdm2 dissociation.66 Transfection studies suggest that h-ser20 phosphorylation significantly destabilizes the Mdm2/P53 interaction,67 while others indicate that h-ser15 phosphorylation increases the affinity for p300/CBP.68, 69 These events together have been suggested to stimulate P53 transactivation function. By contrast, other studies showed that P53 mutants in which all serine residues in the entire protein were changed to alanine displayed wild-type stability and transactivation.19 Of great significance, while mouse mutants generated by homologous recombination to encode m-ser18ala appeared to exhibited tissue-specific defects in p53 target gene activation, its capacity to suppress spontaneous tumor formation remained intact.70, 71, 72 Mice expressing the ser23ala mutation also exhibited modest, tissue-specific deficiencies, but not the substantial destabilization and functional inactivation that would have been predicted were phosphorylation of this residue critical for reducing Mdm2 binding.73, 74 These data are most consistent with N-terminal phosphorylations contributing to modulating tissue-specific responses rather than serving as an on–off switch as occurs in Drosophila.75, 76

The interpretation of the significance of N-terminal phosphorylation on P53 function also depends on whether P53 contains two TADs. If it does, then mutations in m-ser18 and m-thr21 might only inactivate the first one, but preserve the function of the second to allow some level of transactivation. The question of whether P53 contains more than one functional TAD was raised at the Dunedin meeting. The human L22QW23S mutant was found by transfection analysis to retain some residual function, and residues 43–63 were subsequently identified as a second potential TAD (see above). Recent data indicate that the second TAD mediates activation of the proapoptotic IGFBP3 gene, and that the basic C-terminus inhibits this function.77 This interpretation is based on transfection studies, and it is important to emphasize that the function of the second TAD is revealed only in the context of P53 mutants deleted for TAD1, which contains the Mdm2 binding site. Thus, TAD1 mutants should encode a very stable P53 variant capable of binding chromatin constitutively (see below). The idea that the C-terminus is inhibitory to P53 function is inconsistent with other studies indicating that it mediates linear diffusion to augment the ability of P53 to find its response elements (see below). Therefore, it remains to be determined whether the second putative TAD can contribute to P53 functions in the context of a structurally wild-type P53 with compromised function for TAD1 while preserving stability control by Mdm2.

A second approach to define the importance of transactivation for P53-mediated tumor suppression, and to assess the relative contribution of each P53 TAD in vivo, uses homologous recombination to generate mice with inactivating mutations in TAD1. Mice expressing a p53 with the mouse equivalent of the L22QW23S mutations (i.e., L25QW26S, referred to as p53QS) that largely inactivate human TAD1 have been generated by three groups.78, 79, 80 Two previous studies showed that the p53QS mutant appeared to be completely devoid of function.78, 79 However, we now know that the targeting construct we used to generate this mutant contained an additional alanine to valine mutation at codon 135 in the DNA-binding domain (Ala135 to Val135). The Val135 mutation also generates a temperature-sensitive allele.81 Unfortunately, a second study did not attempt to produce mice expressing p53QS, but did generate differentiated ES cells and reconstituted thymus expressing this allele.78 The data from both our studies and those of Chao et al.,78 as well as our more recent comparisons of the p53QS alleles with Ala135 (p53QSA) or Val135 (p53QSV),81 show that the p53QS mutations themselves generate p53 molecules devoid of detectable transcriptional activity, ability to induce cell cycle arrest or apoptosis, or suppress tumor formation under the conditions tested. In stark contrast to this conclusion, data presented at the Dunedin meeting showed that an independently generated p53QS with Ala135 is embryonic lethal in heterozygotes.80 As no homozygous p53QSA embryos were obtained, the source of the lethality remains to be determined. It is possible that the P53QSA protein could stabilize and activate the remaining wild-type P53, that the mutant allele could retain transcription function in the proposed second TAD, or that another mechanism such as that described below could underlie the observed lethality.

Our recent studies provide another way of envisioning lethality generated by a stable, but transcriptionally inert transcriptional regulator.81 We carefully compared the properties of endogenous p53QSV with exogenously expressed p53QSA (introduced into p53-null MEFs by lentiviral infection, and clones expressing the same levels of P53QSV and P53QSA proteins were identified and analyzed). We detected no transactivation activity of either mutant on genes such as p21, mdm2, and Puma, but, consistent with the results from the Attardi lab, we did detect bax transactivation by our p53QSA. However, bax activation by either our p53QSV or QSA MEFs or those from the Attardi lab was only twice that observed in p53-null cells. We have yet to observe induction of cell cycle arrest, apoptosis, or suppression of xenograft formation by p53QSA or p53QSV.81 p53QSA and p53QSV are, as expected, extremely stable proteins (T1/2>8 h). There is, however, one very important difference between P53QSA and P53QSV proteins. While p53QSA binds p21, mdm2, Puma, and Noxa promoters constitutively and at levels equal to that of activated wild-type p53, p53QSV binds these promoters far less efficiently. Interestingly, we found that the p53QSA generated in the Attardi lab is at least five times as abundant in MEFs as p53QSV in MEFs derived from the mice we generated, even though both proteins are equivalently stable and p53 mRNA synthesis is similar for each allele.82

Based on all the available data, we propose the following explanations for the embryonic lethality exhibited by the mice produced by Johnson et al.80 First, the very high level of the p53QSA protein in their mice could exhibit low activity in particular cell types under specific conditions in vivo due to its constitutive chromatin binding and weak transactivation mediated by the second TAD. Another interpretation suggested by our recent data suggests a transactivation-independent mechanism of lethality. That is, although p53QSA lacks measurable transactivation function, we speculate that its accumulation to high levels, combined with its tight and constitutive binding to chromatin, could interfere with cellular transcription, replication, or other DNA-associated molecular events to engender lethality.

The C-Terminus and p53 Regulation

The data summarized above lead us to suggest that there is one dominant TAD in the N-terminus, and that, as deduced from in vitro studies, there are residues within it that are essential for the control of both P53 stability and transactivation. A model that is accepted by most of the field has emerged in which the binding of coactivators competes for binding of the negative regulators Mdm2 and MdmX to the N-terminus. This sets up a competition for competing modifications such as acetylation and ubiquitylation of key highly conserved lysines (six in human, seven in mouse) in the extreme C-terminus.51

As described above, the C-terminus is likely to contribute to P53 function by increasing the efficiency with which it finds its specific response elements. Indeed, elegant studies presented at Dunedin by C Prives showed that the C-terminus enables p53 to engage circular DNA molecules, regardless of whether they have a p53 response element. Deletion of the C-terminus reduces association with random DNA, as found earlier in studies of chromatin association.83 Once P53 engages random DNA, the rate at which it detects specific P53 response elements is increased by linear diffusion.63 However, neither acetylation nor phosphorylation of the C-terminus affected linear diffusion or target gene transactivation.63, 83

These in vitro observations raise the question of whether C-terminal modifications affect P53 function in vivo. Human P53 mutants in which all six highly conserved C-terminal lysine residues were mutated to arginine to prevent post-translational modifications including ubiquitylation and acetylation proved to be stable and more active than wild-type P53.84, 85 However, it is difficult to recreate the normal stoichiometric relationships between P53, Mdm2, and MdmX using cotransfection of P53 with Mdm2 into human cancer cell lines that contain other alterations that could affect p53 function.

I presented data at the Dunedin meeting obtained from mice generated by homologous recombination to encode p53 with arginine substituted for the seven C-terminal lysines that are analogous to the corresponding residues in human p53 (i.e., p537KR;86 also see87 for a study that generated a p53 allele in which six lys were changed to arg but mice were not produced). We developed this model to determine the impact of C-terminal modifications on P53 regulation. While the in vitro data suggested this mutation should be lethal if homozygous, we observed normal Mendelian transmission and no evidence of early mortality, sex bias, or developmental abnormalities.86 Surprisingly, mouse P537KR and wild-type P53 exhibit the same half-life, and like wild-type P53, p537KR is degraded by Mdm2-mediated proteasome-dependent proteolysis. In vitro analyses showed that P537KR can be ubiquitylated by Mdm2 after transfection, but the pattern was not the same as that of wild-type P53. As other studies indicate that lysines other than in the C-terminus may control P53 stability,88 it is formally possible that alternative lysine(s) are targeted for ubiquitylation when those in the C-terminus are not available. However, as P537KR has precisely the same half-life as wild-type P53, it seems unlikely that Mdm2 could engage widely separated lysines with equal efficiency. Furthermore, while Mdm2 can polyubiquitylate itself, it can only polyubiquitylate P53 at multiple lysines. This means that if P53 needs to be polyubiquitylated for degradation, this must require the action of an E4 ubiquitin ligase, such as p300.89 It is difficult to reconcile the identical half-lives of these two proteins with the observation that Mdm2 is sufficient to meditate its own polyubiquitylation, while P53 requires both Mdm2 and an E4. Thus, our data raise the intriguing possibility that P53 degradation may merely require association with Mdm2, and it is the self-ubiquitylation of Mdm2 that drives the degradation of both proteins. However, this model seems simplistic as Mdm2 can also bind to the P53-related proteins P63 and P73, yet it does not mediate their degradation.90, 91 Consequently, there must be other critical steps, such as correct presentation of the substrate to the proteasome, or engagement of linking molecules such as hHR23 to effect substrate degradation.92, 93 Clearly, more research is needed to resolve the precise mechanism by which Mdm2 induces P53 degradation, and to reveal how Mdm2 selectively degrades p53.

We also investigated the activity of P537KR in cell culture and in thymocytes in vivo. P537KR and wild-type P53 proved to be experimentally indistinguishable in terms of their activation kinetics, ability to induce target genes, cell cycle arrest, and apoptosis in MEFs. However, P537KR was activated at lower doses of ionizing radiation in mice when freshly explanted thymocytes were analyzed. Importantly, this apparently increased activity of P537KR was also observed in vitro when MEFs were grown according to a 3T3 passage protocol. We observed identical initial growth rates of MEFs expressing wild type and P537KR, but MEFs encoding P537KR entered a senescent state from which immortal variants did not arise. This contrasts with MEFs encoding wild-type P53 that entered crisis, but emerged as immortal variants. These data suggest that while the conserved C-terminal lysines and associated modifications are not essential for P53 control, they are likely to fine-tune P53 activity, and to ensure that the magnitude of a stress responses is appropriate. This fine-tuning is likely to be critical for optimal lifespan of metazoans to guarantee that cells that undergo numerous divisions, or at risk of stress exposure, do not experience errant P53 activation.

The Proline-Rich Domain (PRD) and Mdm2 Interaction

P53 is exquisitely sensitive to Mdm2 expression levels. For example, mice expressing a hypomorphic mdm2 allele produced 30–50% of the normal Mdm2 protein level.94 This resulted in precocious P53 activation in lymphoid tissue and in some epithelial cells. Recent studies showed that mdm2+/− heterozygous mice were far more resistant to the development of lymphoid tumors induced by expression of the Eu-Myc transgene.95 Finally, earlier onset and increased cancer predisposition has been reported in premenopausal women in whom a single-nucleotide polymorphism changes a T to a G in the mdm2 promoter.96 This base change creates a binding site for the ubiquitous transcription factor Sp1 adjacent to a response element for the estrogen receptor, which apparently together contribute to two- to four-fold elevated Mdm2 levels that attenuate P53 function. Together, these data clearly show that Mdm2 abundance is an important determinant of the output of the P53 stress response pathway.

One way to modulate the impact of Mdm2 on P53 function is to regulate the efficiency with which it binds P53. As stated above, N-terminal phosphorylation may participate in this, but its importance appears more significant in specific cell types. Another region that may affect Mdm2 binding consists of a loosely conserved PRD between the TAD and the DNA-binding domain. This region contains a variable number of PXXP motifs, where P=proline and X=any amino acid. PXXP motifs provide potential interaction sites with proteins containing src homology 3 motifs, and the PXXP domains of P53 have been reported to interact with proteins including Mdm2,38, 39, 97 p300,98 WWOX1,99 and the corepressor mSin3a.100 Early studies showed that deletion of the PRD in human P53 alter P53 transactivation in such a way as to preserve the apoptotic function, while compromising cell cycle arrest.101 Initial studies suggested that PRD deletion reduced the affinity of P53 for apoptotic target genes,102, 103 but analyses performed at lower, more natural expression levels, revealed reduced regulation of a broader spectrum of target genes.104

Recent analyses have provided a potential molecular mechanism for the compromised functionality of P53 lacking the PRD. DNA damage can induce phosphorylation of multiple threonine residues in threonine–proline motifs. This creates sites that can be bound by the prolyl isomerase Pin1, which can then induce cis–trans isomerization of the adjacent proline. Prolyl isomerization has been reported to reduce Mdm2 binding, and to stabilize and activate P53.36, 37, 39 Deleting the prolines, or mutating the threonine to nonphosphorylatable alanine, prevents the required prolyl isomerization, and P53 damage responses appear to be compromised in Pin1-deficient MEFs. A recent twist on this story was added by Ygal Haupt at the Dunedin meeting, when he reported that phosphorylation of thr81 is required to recruit Pin1, and that Pin1 recruitment is required for efficient association with the damage-activated kinase Chk2.39 Chk2 recruitment induces phosphorylation of ser20, which then reduces affinity for Mdm2, allowing P53 activation. Importantly, some humans with the Li–Fraumeni cancer predisposition syndrome have been reported to have chk2 mutations,105, 106 while others encode mutant p53 with proline 82 replaced by leucine (i.e., pro82leu).107 The latter data suggest the importance of pro82 as an important contributor to P53 function in vivo. Consistent with this interpretation, Haupt described transfection studies showing that P53 Pro82Leu accumulated less efficiently after ionizing radiation, and exhibited reduced ability to activate a p21 promoter-driven luciferase reporter.39 These data suggest that the PRD may be important for regulating P53 output by fine-tuning Mdm2 binding. One note of caution: earlier studies by Dumaz and Meek showed that the impact of deleting the PRD on P53 activation was critically dependent on the ratio of Mdm2 to P53.97 Therefore, firm assessment of the contribution of the PRD to P53 control will await the generation and analysis of mice expressing p53 with a PRD deletion, and others in which the prolines and threonines have been mutated to rigorously evaluate the validity of the conformational hypothesis advanced above.

New Thoughts on an Old Dogma

Over 36 000 papers addressing p53 function have been published since its discovery 25 years ago. By contrast, a fraction of this number has dealt with Mdm2 and MdmX, which partly reflects their more recent discoveries.19, 57, 108, 109 However, Mdm2 and MdmX are clearly critical negative regulators of P53, and their functions are nonoverlapping since deletion of either engenders early embryonic lethality.110, 111, 112, 113, 114 This raises the possibility that important aspects of P53 control remain to be elucidated due to the paucity of attention focused on Mdm2 and MdmX and proteins that control their functionality, such as Arf,115, 116 gankyrin,117, 118 and the ubiquitin-specific protease HAUSP (herpes virus-associated ubiquitin-specific protease) (Li et al.,119, 120 and Meulmeester et al.121; and see below). Another puzzle is why this system requires the two similar RING domain proteins Mdm2 and MdmX to each serve as a P53-negative regulator. Below, I present a model I proposed at the Dunedin meeting in which I divided p53 regulation into four phases, and accounts for the kinetic relationships between p53, Mdm2, and MdmX we had observed at that time. I will review recent unpublished and published data relevant to this model (Figure 2).

Figure 2
figure 2

Models for attenuating and destabilizing p53, and for activating and stabilizing it in response to stress. P53 regulation is presented in four phases as described in detail in the text. It is important to note that the relative numbers of molecules of P53, Mdm2, MdmX, and HAUSP are not known. Neither is the extent to which Mdm2 and MdmX exist as homo- and heterodimers, and whether this changes as Mdm2 and MdmX levels increase or decline during a stress response. *Indicate damage-kinase phosphorylations that prevent interaction with HAUSP. See the text for a detailed explanation of the model and additional limitations

Full size image

Phase 1: Homeostasis in unstressed cells

P53 is maintained at low levels and in an inactive state in unstressed cells. This can be achieved in several ways. First, Mdm2 can bind to P53 to mediate proteasomal degradation.122, 123 Second, Mdm2 can bind to the P53 TAD to prevent recruitment of coactivators that access the same site.124 Third, as described by Moshe Oren125 in Dunedin, Mdm2 can monoubiquitinate the histones located in the vicinity of the P53 response element when a P53–Mdm2 complex engages chromatin, and Mdm2 can also monoubiquitinate histones in a P53-independent manner when it is overexpressed. Less appreciated is the contribution of MdmX to attenuating P53 function in unstressed cells. This could be substantial as MdmX is produced constitutively and is very stable (see Marine and Jochemsen108, 109 for reviews). The molecular abundance of Mdm2 relative to MdmX in nonstressed and stressed cells remains to be determined. Thus, P53 is held at bay in unstressed cells by interactions with the two negative regulators, Mdm2 and MdmX.

What then controls the levels of Mdm2 and MdmX? The Mdm2 level in unstressed cells is determined by a number of factors including, but not limited to the following: (1) P53-dependent transactivation of the Mdm2 gene (see Lahav et al.125 for recent kinetic modeling); (2) mitogen-dependent activation of factors such as Erk that also transactivate Mdm2;126 (3) mitogen-dependent post-translational modifications that modulate Mdm2 stability (e.g., see Ashcroft et al.127 and Gottlieb et al.128); and (4) interaction with HAUSP.119, 120, 121, 129 The factors that regulate MdmX abundance have not been widely studied, but one that appears increasingly important involves interaction with HAUSP.121 HAUSP was first identified to participate in P53 control by Gu and co-workers as a P53-associated protein.120 Inactivating HAUSP by siRNA or by homologous recombination in HCT116 cells resulted in Mdm2 and MdmX destabilization.120, 121, 130 The importance of HAUSP in regulating MdmX level is demonstrated by the lack of detectable MdmX in HAUSP-deficient cells.121 As eliminating HAUSP greatly destabilized both Mdm2 and MdmX, P53 became stabilized and activated.130 These observations imply that HAUSP interaction with, and deubiquitination of, Mdm2 and MdmX contributes significantly to their steady-state levels in unstressed cells. As MdmX has a longer half-life than either Mdm2 or P53, it is reasonable to speculate that MdmX may be the preferred substrate for HAUSP-mediated deubiquitination.

To summarize regulation of Phase 1: Mdm2 accumulates as a consequence of transcriptional activation by P53 and mitogen-activated factors, as well as by ‘stabilization’ through HAUSP interaction. While Mdm2 can ubiquitinate MdmX, MdmX is deubiquitinated and stabilized via interaction with HAUSP. The aggregate levels of Mdm2 and MdmX determine both P53 abundance and transcriptional activity. Mdm2-mediated ubiquitination of histones may provide an additional means of controlling P53 transcription functions.

Phase 2: Early activation

DNA damage leads to the activation of ATM and other kinases within minutes, as well as phosphorylation of N-terminal serines that may modulate association with Mdm2/X and coactivators. However, of equal importance, these kinases also rapidly phosphorylate Mdm2 and MdmX.131, 132, 133, 134 We showed that phosphorylation of MDM2 by damage kinases mediates its accelerated degradation early in the damage response.135 Studies from the Yuan and Chen labs showed that DNA damage also leads to MdmX degradation,132, 136 and we now know this requires the RING and acidic domains and ubiquitin ligase activity of Mdm2.131, 137, 138, 139

A picture of what triggers the accelerated degradation of Mdm2 early in the damage response, and what converts MdmX from a stable to an unstable protein is now emerging. Recent work from the Shiloh and Jochemsen labs show that the damage-induced phosphorylations on Mdm2 and MdmX destabilize both proteins by preventing them from interacting with HAUSP.121, 134 Another study provided a somewhat different explanation that phosphorylation increased association between Mdm2 and MdmX.140 Regardless of which mechanism is correct, the data taken together imply that DNA damage triggers accelerated degradation of both negative regulators. However, early in the DNA damage response in normal fibroblasts and epithelial cells, it appears that MdmX levels do not decline immediately, and that P53 target genes are activated weakly. Indeed, full activation takes between 1 and 2 h.135 The next section provides one explanation of this delay.

Phase 3. Full activation

It is clear that full P53 activation occurs only after a lag of about 2 h.135 Damage kinases are active over this time, P53 remains phosphorylated at h-ser15, and presumably Mdm2 and Mdmx are also phosphorylated over this interval. One of the first genes activated by P53 is Mdm2, but early in the damage response, the Mdm2 that is made is unstable, likely because it is phosphorylated and cannot interact with HAUSP. Mdm2 does accumulate over time, however, and this correlates with progressively decreasing MdmX levels. This raises the possibility that P53 activates Mdm2 so that its increased abundance ensures MdmX degradation. This model predicts that conditions that lead to P53 activation in the absence of activation of damage kinases should also lead to MdmX degradation due to the elevated Mdm2 abundance. As shown by Lubo Vassilev at the Dunedin meeting, P53 can be fully activated by Nutlin3a, which prevents Mdm2 from binding to P53.141 Indeed, we found that Nutlin3a results in substantial increases in Mdm2 abundance, and that MdmX levels decline in parallel (J Stommel, M Wade, M Tang, and G Wahl, unpublished data). Importantly, we also showed that proteasome inhibitors prevent P53 activation after DNA damage, even though P53 was phosphorylated in its N-terminus.135 By contrast, Nutlin addition during proteasome inhibitor treatment enabled P53 activation by reducing Mdm2 (and presumably)–MdmX interaction with P53.135 Together, the data indicate that P53 activation requires Mdm2 and Mdmx phosphorylation to decrease HAUSP interaction with both proteins, the net effect of which is to increase Mdm2-mediated self-ubiquitylation and ubiquitylation of MdmX. This decreases the levels of Mdm2 at early time points and as a consequence activates P53, resulting in increased Mdm2 transactivation. The increased levels of Hdm2 can then degrade MdmX, resulting in full activation of P53. In this model, P53 activation requires a positive feedback loop in which increasing Mdm2 abundance titrates the amount of MdmX degradation to assure the fine-tuning of the timing and magnitude of the P53 transcriptional response.

Phase 4: Attenuation

The components of this system also provide the potential to attenuate P53 signaling should a stress dissipate or DNA damage be repaired. I suggest this could occur in the following way. ATM and P53 ser15 phosphorylation return to background levels within 4–6 h of induction of DNA damage with a low dose of the radiomimetic agent neocarzinostatin.135 In normal human fibroblasts, Mdm2 levels are high at 4 h, and start to decline thereafter, along with a parallel decrease in P53 transactivation. In light of the data summarized above, it is reasonable to propose that the damage-phosphorylated Mdm2 pool is replaced with a nonphosphorylated pool as a consequence of new synthesis and diminished abundance of activated ATM. The elevated levels of Mdm2 should be able to interact with and inactivate P53 by the mechanisms described above. Furthermore, in the absence of activated damage kinases, MdmX will be allowed to interact with HAUSP, leading to deubiquitination and stabilization. This provides a second barrier to continued P53 activation.

Perspectives

The integration of recent data described above provides a new way of conceptualizing P53 regulation. Homeostasis is maintained by the well-known negative feedback loop. A new twist to the loop is provided by the idea that the activation of Mdm2 during a stress response actually creates a positive feedback loop in which Mdm2 mediates MdmX degradation to achieve full P53 activation. This elegant system enables gradual and tightly controlled increases in P53 activity rather than an all-or-none response that might be created were the system solely controlled solely by phosphorylation. Finally, the reservoir of Mdm2 created by P53-mediated activation enables attenuation of the P53 response since most of the MdmX is eliminated in the activation phase.

This model accounts for most of the major regulators known to operate in this system, and it is consistent with the kinetics of the response in the normal fibroblasts and epithelial cells we have analyzed. It is clear that the system is very sensitive to small changes in the stoichiometric relationships between HAUSP, Mdm2, MdmX, and P53, and it is likely that proteins that regulate Mdm2 function, such as Arf or Gankyrin, may enable finer tuning of the system, or be required for responses to stressors other than DNA damage. Clearly, an important goal for the near future will be to determine the precise numbers of each of these molecules in normal and neoplastic cells, and to determine how differences in stoichiometry affect the dynamics and magnitude of stress-activated responses. We have begun a cursory analysis of cell lines with wild-type P53, and have found substantial variations in the relative abundance of each molecule. This may well explain the attenuation of the P53 response in the numerous cancers expressing wild-type P53.

An important unresolved question concerns the molecular mechanism that switches Mdm2 target specificity from P53 to MdmX. Clearly, one component of the mechanism is determined by whether HAUSP interacts with and deubiquitinates Mdm2 and MdmX. However, as described above, the binding of Mdm2 to P53 may be all that is needed for P53 to be degraded. As N-terminal phosphorylation of P53 seems unlikely to prevent Mdm2 from binding to P53, we need to determine whether P53 stabilization involves preventing Mdm2 from interacting with P53, or whether additional factors are involved. As one example, perhaps proteins that connect Mdm2 to the proteasome, such as hHR23, comprise part of the switch.92, 93

The excitement in this field has grown with the successful isolation of a variety of compounds such as the Nutlins that can activate wild-type P53 in some tumor lines. What we now need to determine is whether the different expression patterns of the key proteins that control P53 function will impact on the utility of these agents, or if the molecular signatures of cancer cells can be used to define subsets of patients most likely to respond to P53 activator therapies. Of significance, as MdmX has emerged to play an increasingly important role in P53 activation, the optimal drugs for p53 activation clearly need to target both Mdm2 and MdmX to prevent both from interacting with and inhibiting P53.