Structural analysis of MDM2 RING separates degradation from regulation of p53 transcription activity

Abstract

MDM2-MDMX complexes bind the p53 tumor suppressor protein, inhibiting p53’s transcriptional activity and targeting p53 for proteasomal degradation. Inhibitors that disrupt binding between p53 and MDM2 efficiently activate a p53 response although their use in the treatment of cancers that retain wild type p53 may be limited by on-target toxicities due to p53 activation in normal tissue. Guided by a novel crystal structure of the MDM2-MDMX-E2(UbcH5B)–ubiquitin complex, we designed MDM2 mutants that prevent E2–ubiquitin binding without altering the RING domain structure. These mutants lose MDM2’s E3 activity, but retain the ability to limit p53's transcriptional activity and allow cell proliferation. Cells expressing these mutants responded more quickly to cellular stress than cells expressing wild type MDM2. Targeting the MDM2 E3 ligase activity could widen the therapeutic window of p53 activation in tumors, since rapid p53 induction can be achieved while basal p53 control by MDM2 is maintained.

Introduction

The p53 tumor suppressor protein is induced in response to various types of oncogenic stress, functioning to arrest cell proliferation and induce cell death. Numerous mechanisms to control p53 have evolved, with a key role for the related RING domain proteins MDM2 and MDMX. In mice, deletion of either MDM2 or MDMX results in early embryonic lethality due to uncontrolled p53 activation^1–7.

In healthy unstressed cells, MDM2 is primarily localized in the nucleus where it binds and inhibits p53, partly by ubiquitinating p53 and targeting it for proteasomal degradation⁸. The MDM2 RING domain is necessary for ubiquitin ligase (E3) activity and dimerization with itself or the RING domain of MDMX. Although MDMX can homodimerize at micro-molar concentration in vitro⁹, in cells, MDMX does not homodimerize and possesses no intrinsic E3 activity^10–13. However, MDM2 and MDMX form an active E3 heterodimer, and some studies suggest this complex is the physiologically relevant ligase. MDM2’s E3 activity also contributes to transcriptional repression by ubiquitinating histones within promoters¹⁴. Furthermore, interactions between MDM2 and p53 can also directly inhibit the transcription factor function of p53 by altering interactions with other components of the transcriptional machinery^15–17. In addition to a requirement to bind p53, other functions of MDM2 are important for the control of p53 activity.

Mutations that prevent MDM2 RING domain homo- or heterodimerization lead to the loss of E3 activity and a failure to restrain p53 function^{6, 7, 18}. The C-terminal tail of MDM2 is also important for E3 activity, although this function can be provided by MDMX in vitro^{19, 20}. Interestingly, in vivo an MDM2 tail mutant controlled p53 activity sufficiently to allow for normal development, despite accumulation of p53 above basal levels in some tissues²¹. These results suggest that MDM2 can control p53 through degradation-independent mechanisms.

The induction of p53 in response to stress is accompanied by the inhibition of MDM2, leading to the stabilization of transcriptionally active p53. Defects in the ability to inactivate MDM2 are frequently seen in cancers that retain wild type p53, and in which small molecule inhibitors of MDM2 may have a therapeutic benefit²². Several drugs that target the binding of p53 to MDM2 have been developed^23–25 and show excellent specificity in stabilizing and activating p53. However, on-target toxicities including thrombocytopenia and neutropenia appear to limit the efficacy and utility of these compounds^{26, 27}.

Previous studies have shown that the RING domain of MDM2 functions to recruit E2 ubiquitin-conjugating enzyme thioesterified with ubiquitin (E2~Ub; ~ indicates thioester bond) and facilitate ubiquitination^{28, 29}. We have solved the crystal structure of MDM2-MDMX RING domain heterodimer bound to E2(UbcH5B) covalently linked to ubiquitin (UbcH5B–Ub; – indicates covalent bond) and generated MDM2 mutants that retain RING domain integrity but do not interact with UbcH5B–Ub. While these mutants are unable to ubiquitinate and degrade p53, they retain the ability to limit p53’s activity.

Results

Structure of MDM2-MDMX RING domain bound to UbcH5B–Ub

To understand how MDM2 RING domain binds and activates E2~Ub for catalysis, we assembled both MDM2 RING homodimer (MDM2^RH; MDM2 428-C) and MDM2-MDMX RING heterodimer (MDM2^R-MDMX^R; MDM2 428-C and MDMX 427-C) in complex with UbcH5B–Ub for crystallization. UbcH5B–Ub complex was obtained by mutating UbcH5B’s C85 to a lysine to form an isopeptide bond with the C-terminal glycine of Ub^{30, 31}. Only MDM2^R-MDMX^R-UbcH5B–Ub complex yielded diffracting crystals and the structure was determined to 2.16 Å (Fig. 1a-c and Table 1).

Figure 1. Structure of MDM2^R-MDMX^R-UbcH5B–Ub.

(a) Cartoon representation of the complex. MDM2^R is colored green, MDMX^R orange, UbcH5B cyan and Ub yellow. UbcH5B–Ub linkage colored in blue, UbcH5B’s α1, L1 and L2 loops, MDM2 and MDMX’s C-terminus are indicated. Zinc atoms are depicted in gray sphere. (b) Surface representation of the complex, colored and oriented as in a. MDMX’s C-terminus is indicated with an arrow. (c) Schematic drawing of the complex, colored as in a. The C-terminal tail of Ub is shown as a black line. Last three residues of MDM2 and MDMX’s C-terminal tails (Y489-F490-P491 in MDM2 and F488-I489-A490 in MDMX) are shown as thick dark green and brown lines, respectively. (d) Close-up view of MDM2^R-UbcH5B interactions. (e) Close-up view of Ub-UbcH5B interactions. (f) Close-up view of MDM2^R-Ub interactions. Left and right panels are related by 90º rotation about the y axis. For d, e and f, key residues are shown as sticks and colored as in a. Nitrogen and oxygen atoms are in blue and red, respectively. (g,h) Nonreduced SDS-PAGE showing the discharge of UbcH5B~Ub variants to l-lysine over time catalyzed by MDM2^R-MDMX^R. Ub’s D58 is not involved in the binding interface and D58A substitution had no effect on UbcH5B~Ub discharge. (i) Nonreduced SDS-PAGE showing the discharge of UbcH5B~Ub to l-lysine over time catalyzed by MDM2^R-MDMX^R variants. Uncropped gel images are shown in Supplementary Data Set 1.

Table 1. Data collection and refinement statistics.

	MDM2R-MDMXR-UbcH5B–Ub (PDB: 5MNJ)
Data collection
Space group	P1
Cell dimensions
a, b, c (Å)	54.24, 62.76, 66.35
α, β, γ (°)	69.83. 69.22, 78.21
Resolution (Å)	50.5–2.16(2.22–2.16)a
Rmerge	0.075(0.402)
I/σ(I)	6.8(1.4)
CC1/2	0.992(0.73)
Completeness (%)	95.3(93.3)
Redundancy	3.2(2.9)
Refinement
Resolution (Å)	50.5–2.16
No. reflections	37881
Rwork / Rfree	0.190/0.231
No. atoms
Protein	5318
Ligand/ion (Zn and SO4)	10
Water	77
B factors
Protein	55.1
Ligand/ion (Zn and SO4)	51.7
Water	52.3
R.m.s. deviations
Bond lengths (Å)	0.008
Bond angles (°)	1.254

^aValues in parentheses are for highest-resolution shell.

There are two copies of MDM2^R-MDMX^R-UbcH5B–Ub complex in the asymmetric unit (r.m.s deviation of 0.29 Å for Cα atoms). In both copies, MDM2^R binds both UbcH5B and Ub whereas MDMX^R only contacts Ub via its C-terminal tail (Fig. 1a-c). This UbcH5B–Ub binding arrangement is similar to that observed in the structures of other homodimeric RING E3-E2–Ub complexes^31–33. However, unlike homodimeric RING E3s where two molecules of E2–Ub bind, MDM2^R-MDMX^R only binds one molecule of UbcH5B–Ub because MDMX does not bind UbcH5B–Ub.

Interactions between MDM2^R-MDMX^R and UbcH5B–Ub

The structure of UbcH5B–Ub adopts the closed, “folded back” configuration similar to those observed in other RING E3-E2–Ub complexes^30–37 (Fig. 1a-c) suggesting that MDM2^R-MDMX^R primes Ub for transfer. The closed UbcH5B–Ub conformation is stabilized by numerous interactions involving MDM2^R-UbcH5B, Ub-UbcH5B, MDM2^R-Ub and Ub-MDMX tail interfaces. The MDM2^R-UbcH5B interaction involves UbcH5B’s N-terminal helix, L1 and L2 loops and the hydrophobic core of MDM2^R centering on I440 (Fig. 1d). Ub-UbcH5B interactions include UbcH5B’s α2 contacting Ub’s I44 surface and UbcH5B’s active site residues clamping Ub’s C-terminal tail (Fig. 1e). MDM2^R-Ub interactions involve MDM2’s V477 and Zn²⁺-ligand H457 and Ub’s L8/I36/L71 surface. Additionally, MDM2’s R479, known as the “linchpin Arg”³⁸, forms hydrogen bonds with the carbonyl oxygens of UbcH5B’s Q92 and Ub’s R72 and the sidechain of Ub’s Q40 (Fig. 1f). These interactions are similar to those observed in other RING E3-E2–Ub complexes^30–37. MDM2 also contains a short helix that precedes the RING domain, where S429 and N433 from this helix contact Ub’s T9/K11 surface (Fig. 1f).

We mutated a selection of key residues involved in Ub-UbcH5B (N77A, D87A, S108R, N114A and D117A in UbcH5B, and I44A and L73D in Ub), MDM2^R-Ub (N433R, V477A and R479A in MDM2, and T9A, I36A and Q40R in Ub) and MDM2^R-UbcH5B (I440A in MDM2) interfaces and performed single-turnover Ub transfer to lysine. All mutants were defective in discharging UbcH5B~Ub suggesting that the observed interactions are important for MDM2-MDMX-catalyzed Ub transfer (Fig. 1g-i and Supplementary Fig. 1a-f).

The importance of the MDMX C-terminal tail in MDM2-MDMX heterodimer

E3 defective mutations in the C-terminal tail of MDM2 can be reactivated by MDMX^{19, 20, 39}. Our structure shows that the last three residues of MDMX (F488-I489-A490) play a crucial role in stabilizing the closed UbcH5B–Ub conformation. F488 packs against Ub’s G35 surface and A490 is completely buried within the MDM2^R-Ub interface to support Ub binding (Fig. 2a). This binding mode suggests that MDMX’s C-terminus cannot accommodate extra residues. Furthermore, the structure shows that MDM2’s C-terminal tail does not contact UbcH5B–Ub.

Figure 2. Importance of MDMX’s C-terminal tail.

(a) Close-up view of MDMX’s C-terminal tail. Transparent surface representation was shown. Last three C-terminal residues (F488-I489-A490) are shown as sticks and colored as in Fig. 1. (b) Schematic drawing of the complex showing disruption in MDMX’s C-terminus (left panel) would hinder the activity whereas disruption in MDM2’s C-terminus (right panel) would not have any effect. (c, d) Nonreduced SDS-PAGE showing the discharge of UbcH5B~Ub to l-lysine over time catalyzed by MDM2^R-MDMX^R variants. Uncropped gel images are shown in Supplementary Data Set 1.

To assess the importance of MDMX’s C-terminus, we mutated F488 to histidine and alanine and found both mutants were defective in discharging UbcH5B~Ub (Fig. 2b,c and Supplementary Fig. 1g,h). Addition of a hexahistidine tag to MDMX’s C-terminus also hindered discharge of UbcH5B~Ub (Fig. 2d and Supplementary Fig. 1i,j). However, substitution of the corresponding C-terminal residue in MDM2, Y489, to histidine or alanine or addition of a hexahistidine tag to MDM2’s C-terminus had a minimal effect (Fig. 2b-d and Supplementary Fig. 1g-j), consistent with our structure. Thus, the length of MDMX’s C-terminus impacts E2~Ub recruitment by MDM2-MDMX heterodimer. MDMX’s F488-Ub interaction is conserved in homodimeric RING E3s such as RNF4 and BIRC7^31–33 but in these examples the C-terminus is only partially buried and introducing exogenous C-terminal residues minimally affected activity⁴⁰.

A model for MDM2 homodimer bound to E2~Ub

Given that MDMX only contributes to the tail-Ub interaction, we reasoned that MDM2’s C-terminus could fulfill this role in the MDM2 homodimer. Indeed, substituting MDMX’s C-terminal F488-I489-A490 with MDM2’s C-terminal Y489-F490-P491 marginally improved MDM2-MDMX’s activity (Fig. 2c and Supplementary Fig. 1g,h) suggesting that the MDM2 homodimer likely uses a similar mechanism to activate E2~Ub. Since MDM2 and MDMX RING domain structures are similar (r.m.s. deviation of 0.44 Å for Cα atoms) and the NMR structure of MDM2^{RH 29} is similar to MDM2^R-MDMX^R (r.m.s. deviation of 1.47 Å for Cα atoms), we generated a model of MDM2^RH bound to UbcH5B–Ub by superimposition of MDM2^R-UbcH5B–Ub from our structure onto MDMX^R in the MDM2^R-MDMX^R-UbcH5B–Ub structure (Fig. 3a,b). In this model, each MDM2 RING domain binds one molecule of UbcH5B–Ub and their C-terminal tails are buried within the adjacent MDM2^R-Ub interface to support Ub binding.

Figure 3. Mechanism of E2~Ub recruitment and activation by MDM2^RH.

(a) A model of MDM2^RH bound to UbcH5B–Ub generated by superimposition of MDM2^R-UbcH5B–Ub portion of the structure onto MDMX^R in the MDM2^R-MDMX^R-UbcH5B–Ub structure. All colorings are as in Fig. 1. (b) Schematic drawing of the MDM2^RH-UbcH5B–Ub model in a. MDM2^RH’s C-terminal tails are shown as thick dark green lines. (c) Nonreduced SDS-PAGE showing the discharge of UbcH5B~Ub variants to l-lysine over time catalyzed by GST-MDM2 variants. (d) Nonreduced SDS-PAGE showing the discharge of fluorescently-labeled UbcH5B~Ub to l-lysine over time catalyzed by MDM2^RH and MDM2^R-MDMX^R visualized by InstantBlue (top panel) and LI-COR Odyssey scanner (bottom panel). (e) A plot showing the rate of the discharge of fluorescently-labeled UbcH5B~Ub to l-lysine in d. The reactions were performed in duplicate. Uncropped gel images are shown in Supplementary Data Set 1.

To validate this model, we mutated a selection of key residues in the Ub-UbcH5B (UbcH5B S108R and Ub I44A), MDM2^R-Ub (MDM2 R479A) and Ub-MDM2 tail (MDM2 Y489A) binding interface and performed UbcH5B~Ub discharge assay catalyzed by MDM2 homodimer. All mutants were defective in discharging UbcH5B~Ub. Moreover, addition of a C-terminal hexahistidine tag reduced the activity (Fig. 3c and Supplementary Fig. 1k,l). Thus, the MDM2 homodimer uses a similar mechanism as the MDM2-MDMX heterodimer to activate E2~Ub for catalysis.

MDM2 homodimer exhibits higher intrinsic ligase activity

The structure of the MDM2^R-MDMX^R-UbcH5B–Ub complex shows that the MDM2-MDMX heterodimer can only recruit one molecule of UbcH5B–Ub at a time while the model of the MDM2^RH-UbcH5B–Ub complex suggests that the MDM2 homodimer can bind two molecules of UbcH5B–Ub and perhaps enable transfer of two Ub molecules simultaneously. To investigate MDM2^RH and MDM2^R-MDMX^R Ub transfer mechanisms, we compared their rate of UbcH5B~Ub discharge. Because the MDM2 RING domain is known to form large oligomers and dimers during expression and purification from E. coli²⁰, we developed a protocol to purify the dimeric fraction of MDM2 RING domain and then performed UbcH5B~Ub discharge assay by using fluorescently-labeled Ub for quantification. At the same E3 concentration, MDM2^RH discharged UbcH5B~Ub at a faster rate as compared to MDM2^R-MDMX^R. When MDM2^RH concentration is halved, it displayed a similar rate of UbcH5B~Ub discharge as MDM2^R-MDMX^R (Fig. 3d,e). Thus, MDM2^RH displays two-fold enhancement in rate as compared to MDM2^R-MDMX^R.

Engineering MDM2 RING domain lacking ligase activity

So far, most studies on MDM2’s ligase-independent regulation of p53 are based on MDM2 C464A (or C462A in mouse), a mutant that disrupts the Zn²⁺-coordination^{18, 41–43}. This mutation has severe consequences on the RING domain fold and hence the ability to homo- or heterodimerize. We sought to design an MDM2 RING mutant that abrogates the ligase activity while maintaining the RING domain fold to study the ligase independent function of MDM2. Our structure and model suggest that disrupting the MDM2^R-UbcH5B, MDM2^R-Ub or MDM2 tail-Ub interaction could hamper the ligase activity. Given that MDM2 tail mutants can be reactivated by MDMX^{20, 44}, we focused on the MDM2^R-UbcH5B and MDM2^R-Ub interfaces. MDM2 I440 and R479 were chosen because I440 plays the central role in E2-binding and R479 helps stabilize the closed E2~Ub conformation. The corresponding residue to I440 in BRCA1, I26, has been mutated to alanine and was found to be defective in activity in vitro^{45, 46}.

We mutated MDM2 I440 and R479 to several different amino acid residues (I440E, I440A, I440D, I440Y, I440R, I440K, R479A, R479P, R479I, R479G, R479F and R479K). The basic and acidic residues are expected to severely impact MDM2^R-UbcH5B hydrophobic interactions whereas residues with no or short sidechains should not support interactions initiated by the linchpin Arg. In vitro UbcH5B~Ub discharge and UbcH5B–Ub binding assays showed that all MDM2 mutants were defective in discharging UbcH5B~Ub compared to the wild type MDM2 (Supplementary Fig. 2a). Most I440 variants had no measurable UbcH5B–Ub binding affinity except I440Y, which displayed ~60-fold weaker affinity than the wild type MDM2 (Supplementary Table 1 and Supplementary Fig. 3a). All R479 variants bound UbcH5B–Ub with weaker affinities and only R479P mutant displayed no measurable binding affinity (Supplementary Table 1 and Supplementary Fig. 3a). Consistent with UbcH5B–Ub binding properties, MDM2 mutants that exhibited no measurable UbcH5B–Ub binding affinity were severely compromised in their p53 ubiquitination activity in cells, with MDM2 I440K, I440E and R479P mutants showing basal p53 ubiquitination levels similar to the empty vector (EV) while MDM2 I440A, I440R and I440D displaying trace p53 ubiquitination activity (Fig. 4a, b).

Figure 4. Redesigning ligase-dead MDM2 mutants.

(a, b) Ubiquitination assay showing that MDM2 I440 or R479 mutants abolish or reduce p53 ubiquitination. (c) Co-immunoprecipitation assay showing that, unlike the MDM2 C464A mutant, MDM2 I440K, I440E and R479P can interact with MDMX. (d) MDMX is a cytoplasmic protein and can enter nucleus when dimerized with MDM2. GFP-tagged MDMX can be found in the nucleus with MDM2 wild type and I440 mutants, demomstrating dimerization between the two proteins. This function is lost in the C464A mutant. Scale bar indicates 10 μm. (e) Co-immunoprecipitation assay showing all the MDM2 mutants can interact with p53 in solution. An MDM2 protein lacking the p53 binding domain (Δp53BD) was used as a negative control. (f) Ubiquitination assay showing that, unlike the C-terminal mutant Y489D, I440 mutants cannot be reactivated by MDMX. Uncropped blot images are shown in Supplementary Data Set 1.

Based on these results, MDM2 I440K, I440E and R479P were selected for subsequent analyses. To verify these mutations do not have an impact on MDM2 homo- or heterodimerization, we performed pull-down experiments in vitro. All GST-tagged MDM2 variants were able to pull-down their corresponding His-tagged MDM2 and MDMX variants, suggesting that these mutants are competent in forming MDM2 homodimers and MDM2-MDMX heterodimers (Supplementary Fig. 2b,c).

Immunoprecipitation results show that all the I440 and R479 mutants retained the ability to interact with MDMX in cells (Fig. 4c), while the negative control MDM2 C464A did not show any interaction. These results were confirmed by showing that wild type MDM2, I440K and I440E mutants can interact with and import MDMX into the nucleus (Fig. 4d), while C464A did not relocalize MDMX. As expected, all mutants, including C464A mutant, interacted with p53 (Fig. 4e), showing that neither an intact RING domain nor the ability of MDM2 to interact with the E2~Ub complex is required for the N-terminus of MDM2 to bind to p53. Furthermore, addition of MDMX did not complement the loss of activity in MDM2 I440K or I440E mutants whereas the defective MDM2 Y489D tail-Ub interaction can be restored with MDMX’s tail in the heterodimer (Fig. 4f and Supplementary Fig. 2d). Together, these results showed that MDM2 I440K, I440E and R479P retain the RING domain fold to form homodimers and heterodimers with MDMX, but cannot interact with E2~Ub to catalyze ubiquitination.

MDMX RING domain is not optimized for E2~Ub activation

To understand the basis for the lack of activity in MDMX we superimposed the structure of MDMX^R onto MDM2^R in our MDM2^R-MDMX^R-UbcH5B–Ub structure to generate a model of MDMX^R-UbcH5B–Ub complex. The model showed that MDMX^R harbors residues that are less optimal in UbcH5B and Ub binding compared to MDM2^R (Fig. 5a,b). Strikingly, MDMX^R lacks the linchpin Arg, and instead contains K478, which cannot initiate the same hydrogen bond network as MDM2’s R479. Furthermore, the MDMX^R-MDMX^R dimerization interface is suboptimal: the two MDMX^R N448 sidechains at this interface potentially clash or create an electrostatic repulsion (Fig. 5c). MDM2 contains C449 at this position that does not clash with C449 in the MDM2 homodimer model or MDMX’s N448 in MDM2-MDMX heterodimer (Fig. 5d,e).

Figure 5. Reactivation of MDMX’s ligase activity.

(a) Structural comparison of MDM2^R-UbcH5B (left panel) and MDMX^R-UbcH5B (right panel) interface. (b) Structural comparison of MDM2^R-Ub (left panel) and MDMX^R-Ub (right panel) interface. MDMX^R-UbcH5B–Ub model was generated by superimposition of MDMX^R onto MDM2^R in the structure of MDM2^R-MDMX^R-UbcH5B–Ub complex. Residues are shown as sticks and all colorings are as in Fig. 1. (c) A model of MDMX^R-MDMX^R homodimer generated as in a showing the presence of N448 at the dimerization interface separated by 2.5 Å. (d) A model of MDM2^RH generated as in Fig. 3a showing the presence of C449 at the dimerization interface separated by 5.0 Å. (e) The dimerization interface of MDM2^R-MDMX^R showing MDM2’s C449 and MDMX’s N448 separated by 3.7 Å. (f) Nonreduced SDS-PAGE showing the discharge of UbcH5B~Ub to l-lysine over time catalyzed by MDM2^R-MDMX^R and MDMX^R variants at 250 nM (top panel) and 2.5 μM (bottom panel). (g) Ubiquitination assay showing the double mutation of N448C and K478R in the MDM2-MDMX chimera (1-421: MDM2, 422-490: MDMX, “MDM2_XRING”) is sufficient to reactivate MDMX's E3 ligase activity. (h) MDM2 I440 mutants are reactivated by MDMX K478R but not by WT MDMX, reflecting the ability of MDMX K478R to interact with UbcH5B–Ub complex and ubiquitinate p53. (i) Models illustrating that, unlike the MDM2 C-terminus mutant (Y489D), MDM2 I440 mutants cannot be reactivated by wild type MDMX (Fig. 4f) because neither MDM2 I440 mutants nor wild type MDMX recruit the UbcH5B–Ub complex (Supplementary Fig. 3). However, these mutants can be reactivated by MDMX K478R that binds the UbcH5B–Ub complex (Fig. 6h and Supplementary Fig. 3). Red cross indicates the point of mutation. Uncropped gel and blot images are shown in Supplementary Data Set 1.

We wondered whether substituting N448C and K478R could reactivate MDMX’s ligase activity. A similar MDMX mutant harboring N448C with residues 465-480 replaced with MDM2 sequence was previously shown to promote p53 ubiquitination and degradation in cells¹³. MDMX displayed no activity under the reaction conditions of our UbcH5B~Ub discharge assay, MDMX N448C showed marginal activity, while MDMX K478R massively improved the activity. The MDMX N448C K478R mutant further enhanced the activity (Fig. 5f). To assess whether these MDMX variants are active in cells, we performed p53 ubiquitination assays by transfecting p53, full-length MDMX variants and His-Ub into U2OS cells. MDMX N448C and MDMX K478R exhibited no appreciable activity while MDMX N448C K478R displayed modest p53 ubiquitination activity, compared to full-length MDM2. Remarkably, when we generated an MDM2-MDMX chimera by replacing MDM2’s RING domain with MDMX’s RING domain (1-421: MDM2, 422-490: MDMX, “MDM2_XRING”) to avoid the complications of the lack of NLS and NES and a shorter acidic domain in MDMX, the N448C K478R mutant exhibited p53 ubiquitination activity similar to wild type MDM2 (Fig. 5g). Based on the model of active E3 dimers, the MDM2 I440 and R479 mutants, which have lost the ability to bind E2~Ub, may regain E3 activity by forming a dimer with a binding partner - like MDMX or other MDM2 proteins - that retain the ability to bind to the E2~Ub complex. To test this hypothesis, MDMX K478R (not chimera) was co-expressed with MDM2 I440 mutants. All mutants were reactivated by MDMX K478R but not by wild type MDMX (Fig. 5h) and binding analysis showed that MDMX K478R has acquired the ability to interact with the UbcH5B–Ub complex (Supplementary Fig. 3b). The ability of MDMX K478R to bind E2~Ub and dimerize with MDM2 I440 mutants allows it to ubiquitinate p53 (Fig. 5i). Together, these results showed that MDMX RING domain sequence is not optimized for E2~Ub activation and highlighted the importance of the linchpin Arg in regulating the ligase activity. Moreover, these results are consistent with earlier findings that demonstrated the requirement of MDM2 acidic domain in promoting p53 ubiquitination and degradation in cells^47–49.

MDM2 I440 or R479 mutants limit p53 activity

To assess the effect of the MDM2 mutants on endogenous p53, we generated U2OS cells that were conditionally depleted p53 through use of a doxycycline inducible shRNA, then removed endogenous MDM2 from these cells by CRISPR/Cas9 followed by stable re-expression of various MDM2 mutants (Supplementary Fig. 4a-c). Removal of doxycycline then allowed for the re-expression of endogenous p53. Expression of wild type MDM2 in this system resulted in a clear degradation of p53 compared to EV transfected cells, while – as expected - the C464A, I440K, I440E and R479P mutants failed to degrade p53 (Fig. 6a,b). Interestingly, however, despite failing to target p53 for degradation, the expression of the I440 or R479 mutants resulted in a significantly lower activation of the p53 target genes including p21, PIG3 and TIGAR, compared to that seen in cells expressing the MDM2 C464A mutant (Fig. 6a,b, Fig. 7a, and Supplementary Fig. 5).

Figure 6. MDM2 I440 and R479 mutants can limit p53 activity.

(a, b) Western blot analysis showing that although MDM2 I440K, I440E and R479P do not degrade p53, induction of p21 and other p53 target genes is attenuated by these mutants. n = 3 independent experiments. p values: v.s. EV (vertical), v.s. C464A (horizontal). n.s. indicates non-significant v.s. EV. (c) ChIP for p53 and MDM2 with quantitative PCR for a p53 response element in the p21 promoter (-2350 bp) and a non-specific binding region (+50 bp). n = 3 independent experiments (each with 2 technical replicates). p values: v.s. EV (ChIP: p53), v.s. C464A (ChIP: MDM2). (d) Ratio of MDM2/p53 on the p21 promoter based on the ChIP (Fig. 6c, p21 -2350 No Doxy). MDM2 wild type, MDM2 I440 and R479 mutants were found at the p21 promoter with p53. p values: v.s. wild type (vertical), v.s. C464A (horizontal). (e) The inhibition of cell growth seen following restoration of p53 by removal of Doxycycline (Doxy)) in MDM2 knock-out cells was rescued by expressing MDM2 wild type, I440 and R479 mutants. n = 3 independent experiments. p values: v.s. C464A. F-statistics and degrees of freedom of ANOVAs are reported in Supplementary Table 3. Uncropped blot images are shown in Supplementary Data Set 1.

Figure 7. MDM2 I440 and R479 mutants can quickly respond to the stress.

(a) qPCR showing that expression of the indicated p53 target genes is attenuated by MDM2 I440K, I440E and R479P, and induced more strongly than in cells expressing wild type MDM2 following 18 hours treatment with Actinomycin D (10 nM). n = 3 independent experiments. p values: v.s. EV No Doxy (vertical), v.s. wild type No Doxy ActD (horizontal). n.s. indicates non-significant v.s.EV No Doxy. (b, c) Examples of immunoblots and quantification of replicate blots showing that p21 and PIG3 are more rapidly induced in cells expressing MDM2 I440 and R479 mutants upon Actinomycin D (10 nM) treatment compared to cells expressing wild type MDM2 n = 3 independent experiments. Immunoblots of all samples are available in Supplementary Fig. 6a. (d) The indicated cells were treated with vehicle or 10nM Actinomycin D for varying times as shown then were fixed and stained 120 hours after plating. Cells expressing MDM2 I440 or R479 mutants more rapidly attenuate cell growth in comparison to MDM2 wild type. (e) Quantification of cell growth experiments. SRB intensity was quantified using ImageJ software. p values: v.s. each wild type. n.s. indicates non-significant. n = 5 independent experiments. F-statistics and degrees of freedom of ANOVAs are reported in Supplementary Table 3. Uncropped blot images are shown in Supplementary Data Set 1.

Chromatin immunoprecipitation (ChIP) assays to look for MDM2 at the p21 promoter showed that although wild type MDM2 expression reduced the overall amount of p53 protein, the ratio of p53 to MDM2 at the p21 promoter suggested that low levels of both proteins were present. By contrast, although MDM2 C464A retained the ability to interact with p53 when co-precipitated in solution (Fig. 4e), this interaction was not maintained at the p21 promoter, explaining why MDM2 C464A cannot efficiently limit p53 activity. Interestingly, the I440 and R479 mutants were all found at the p21 promoter with p53 (Fig. 6c,d), suggesting that their ability to limit p53’s transcriptional activity reflects a direct perturbation of the formation of an active transcriptional complex at the promoter. In terms of proliferation, in the absence of p53 (with doxycycline) all of the cell lines grew at the same rate (Fig. 6e). However, following reintroduction of endogenous p53, control cells (lacking MDM2) stopped growing while cells expressing wild type MDM2 continued proliferating (Fig. 6e). Importantly, despite not targeting p53 for degradation, the ability of the MDM2 I440 or R479 mutants to restrain p53 activity was sufficient to allow proliferation of these cells (Fig. 6e).

Next, we monitored the response of cells expressing the I440 or R479 MDM2 mutants to stress signals that lead to the dissociation of the p53-MDM2 interaction. While low levels of Actinomycin D inhibit MDM2 by disrupting nucleoli and releasing ribosomal proteins that bind an inhibit MDM2^{50, 51}, we used 10 nM Actinomycin D, which leads to the phosphorylation of N-terminal residues of p53 and loss of MDM2 binding^{52, 53} (Supplementary Fig. 4d). Although the expression of the MDM2 I440 or R479 mutants restrained the activation of p53 target genes similarly to the expression of wild type MDM2, cells expressing MDM2 I440 or R479 mutants responded more robustly to a 18 hour treatment of Actinomycin D (Fig. 7a-c and Supplementary Fig. 6), leading to a higher expression of p53 target genes than that seen in cells expressing wild type MDM2 at this time. As expected, Actinomycin D increased p53 levels in cells expressing wild type MDM2, but did not affect the higher p53 levels in cells expressing the E3 defective MDM2 I440, R479 or C464A mutants (Fig. 7b,c). As previously reported^{54, 55}, stabilization of p53 was accompanied by a decrease in MDMX levels – a response that was only evident in cells expressing wild type MDM2, but not the E3 defective mutant. A time course of the response showed that the lack of requirement to stabilize p53 in cells expressing MDM2 I440 or R479 correlated with a more rapid increase in expression of p21 and PIG3 after Actinomycin D treatment compared to a more gradual increased seen in cells expressing MDM2 wild type (Fig. 7b,c). Similar results were obtained in cells treated with two other inhibitors of the p53-MDM2 interaction, Nutlin (Supplementary Fig. 6b) and Doxorubicin (Supplementary Fig. 6c). Consistently, cells expressing MDM2 I440 or R479 mutants showed a more rapid attenuation of cell growth in response to Actinomycin D than cells with wild type MDM2 (Fig. 7d,e). Therefore, although these MDM2 mutants are unable to control p53 levels, they can still repress p53 activity Moreover, the high levels of p53 in cells expressing these MDM2 mutants can be more rapidly activated in response to stress than p53 controlled by wild type MDM2 (which degrades p53), where the p53 protein must still accumulate.

Discussion

The ability of MDMX to interact with MDM2 through the RING domain has been described to both inhibit or enhance MDM2 E3 ligase activity^{10, 56, 57}. Our present work provides the structural and mechanistic insights into how MDM2 homodimer and MDM2-MDMX heterodimer bind and activate E2~Ub for transfer. Both the MDM2 homodimer and MDM2-MDMX heterodimer utilize a similar mechanism as other RING E3s reported previously^{30–38, 58}, where MDM2 RING-E2, MDM2-Ub and Ub-E2 interactions restrain E2~Ub into a closed conformation that optimally positions the E2~Ub thioester bond for transfer. Complementing previous studies^{19, 20, 39}, we show that the C-terminal tail of MDM2 and MDMX in the MDM2 homodimer and MDM2-MDMX heterodimer, respectively, and their lengths play a critical role in supporting Ub binding. In vivo, a similar C-terminal tail mutant of MDM2 was able to control p53 to allow normal development through a mechanism that required MDMX heterodimerization²¹. Surprisingly, although this complex retains E3 activity, p53 protein accumulated in some tissues and cells from these mice, suggesting that degradation was compromised. While the mechanism of how this C-terminal mutant controls p53 under these conditions remains unclear, these data are consistent with our observations that MDM2 shows E3 independent functions to regulate p53.

Our data indicate that despite their apparent structural similarity, MDM2 homodimer and MDM2-MDMX heterodimer have different ability to transfer Ub due to their ability to interact with E2~Ub complexes, and that MDM2 homodimer exhibits higher intrinsic ligase activity than MDM2-MDMX heterodimer. This can be attributed to the inability of MDMX to recruit E2~Ub therefore halving the amount of E2~Ub recruited by a heterodimer (compared to an MDM2 homodimer) or the MDM2 RING domain exhibiting a higher turnover rate in the context of the MDM2 homodimer. However, the experiments performed here are based on the RING domain and the regulation of these complexes by post-translational modification⁵⁹, other domains^{9, 60} or other protein partners which could favor the activity of the MDM2-MDMX heterodimer.

Consistent with an earlier study showing that replacement of MDMX’s N448 and residues 465-480 with MDM2’s sequences was able to stimulate MDMX’s ligase activity¹³, we found that substituting MDMX’s N448 and K478 to cysteine and arginine, respectively, were sufficient to unmask its ligase activity. The N448C mutation enables RING domain homodimerization, thereby allowing MDMX tail-Ub interaction, whereas K478R presumably stabilizes the closed E2~Ub conformation as observed in our structure. These results further underscore the importance of C-terminal tail-Ub interaction and the linchpin Arg in modulating the ligase activity.

Prior studies have used MDM2 C464A mutant to assess ligase-independent functions in regulating p53, but C464A also abolishes the RING domain fold. It is therefore not clear whether failure of this mutant (or the mouse equivalent) to control p53¹⁸ reflects a loss of E3 function, a loss of dimerization or a failure to bind to p53 at promoters. We show that MDM2 I440K, I440E and R479P abrogate the interaction with E2~Ub complexes and abolish MDM2’s E3 activity, while retaining the ability to interact with p53, and to homodimerize or heterodimerize (with MDMX), suggesting they retain an intact RING domain fold. These mutants can be localized on p21 promoter (via p53) and efficiently reduce p53’s transcriptional activity, allowing proliferation but rendering cells hypersensitive to stress. Interestingly, MDM2 C464A also retains some effect on cell growth and modestly limits the induction of p53 target genes including p21 (Fig. 6a,b), potentially reflecting the ability of this mutant tobind p53 in solution (a summary of the activity of these MDM2 mutants is available in Supplementary Table 2).

Finally, it is possible that targeting the catalytic activity of MDM2 to reactivate wild type p53 in tumors (which have elevated levels of stress) could open a therapeutic window by avoiding the deleterious on-target side effects of completely disrupting the MDM2-p53 interaction in normal tissue. Previous work has described small molecule inhibitors of MDM2 that bind to the RING domain, inhibiting the E3 activity and stabilizing p53⁶¹, suggesting that such an approach is feasible.

Online Methods

Protein preparation

All constructs were generated by standard PCR-ligation techniques. All proteins are human unless otherwise specified and were expressed in BL21 (DE3) cells or BL21 (DE3) RIL (Stratagene). To generate MDM2^R-MDMX^R (MDM2 428-C and MDMX 427-C) for crystallization and assays, MDM2^R and MDMX^R were cloned into the first and second multiple cloning site of a bicistronic vector, RSF_Duet (Novagen), respectively, where MDM2^R contains an N-terminal 6xHis-tag followed by a TEV protease cleavage site. 6xHis-MDM2^R-MDMX^R was purified by Ni-NTA affinity, followed by cation exchange and size exclusion chromatography. All MDM2^R-MDMX^R variants were purified in the same manner (Supplementary Fig. 7a). To generate C-terminal 6xHis-tagged MDM2^R-MDMX^R variants, MDM2^R and MDMX^R variants were cloned into pGEX4T1 (GE Healthcare) containing an N-terminal GST-tag and RSF_Duet containing an N-terminal His-tag. MDM2^R-MDMX^R variants were formed by co-expressing GST-MDM2^R with His-MDMX^R variants or GST-MDMX^R with His-MDM2^R variants and subsequently purified by Ni-NTA affinity chromatography, followed by glutathione affinity chromatography (Supplementary Fig. 7b). For MDMX reactivation and binding assays, MDMX^R variants were cloned into pGEX4T1 containing an N-terminal GST-tag followed by a TEV protease cleavage site. MDMX^R variants were purified by glutathione affinity chromatography for binding analyses or released from the beads by incubation with TEV overnight, followed by cation exchange chromatography for reactivation assays (Supplementary Fig. 7c). For binding analyses and assays, MDM2 398-C variants were cloned into pGEX4T1 and purified by glutathione affinity chromatography (Supplementary Fig. 7d,e). MDMX^R variants (10 μM; 100 μl) were loaded onto superdex 75 10/300 to assess their dimerization state (Supplementary Fig. 7f). To obtain pure MDM2^RH for assay, MDM2^R was cloned into pGEX4T1 containing an N-terminal 6xHis-GST-tag followed by a TEV protease cleavage site, expressed in 36 L of LB broth, purified by glutathione affinity chromatography and subsequently buffer exchanged into 50 mM Tris-HCl, pH 7.6, 0.4 M NaCl, 5 %(v/v) glycerol and 1 mM DTT. 6xHis-GST tag was removed by incubation with TEV followed by Ni-NTA pass-back. The cleaved MDM2^R was further purified by HiLoad 26/600 Superdex 200 chromatography (Supplementary Fig. 7c and Supplementary Fig. 7g). For ubiquitination assays, Arabidopsis thaliana Uba1, UbcH5B variants and Ub variants were prepared as described previously³⁴. UbcH5B^S22R–Ub and UbcH5B–Ub used for crystallization and binding analyses, respectively, were prepared as described previously³⁴. To generate fluorescently-labeled Ub, a MHHHHHHCG sequence was inserted at the N-terminus of Ub and purified by Ni-NTA affinity and size exclusion chromatography in PBS buffer, pH 6.6. The protein was labeled by incubation with IRDye® 800CW Maleimide (LI-COR) at a molar ratio of 4:1 (protein:dye) for 2 h at 22 °C in the dark and subsequently buffer exchanged into 25 mM HEPES, pH 7.0 and 150 mM NaCl using a Zeba spin desalting column (Thermo Fisher) to remove excess dye. Protein concentrations were determined by Bio-RAD protein assay using BSA as a standard and Ub concentration was determined as described previously³⁴. All MDM2, MDMX and MDM2-MDMX variants were stored in 25 mM Tris-HCl, pH 7.6, 0.4 M NaCl and 1 mM DTT and other proteins were stored in 25 mM Tris-HCl, pH 7.6, 0.15 M NaCl and 1 mM DTT. All plasmids generated for protein expression were described in Supplementary Table 4.

Crystallization

MDM2^R-MDMX^R-UbcH5B–Ub complex was obtained by mixing 6xHis-MDM2^R-MDMX^R (6.6 mg/ml) and UbcH5B^S22R–Ub (20 mg/ml) at a 1:1 molar ratio. Crystals were grown at 19 °C by microseeding using the hanging-drop vapor diffusion method by mixing the complex with an equal volume of reservoir solution containing 0.1 M Tris-HCl, pH 8.5, 0.175 M Li₂SO₄ and 16-20 %(v/v) PEG 3350. The crystals were flash-frozen in 0.1 M Tris-HCl, pH 8.5, 0.175 M Li₂SO₄, 22 %(v/v) PEG 3350 and 20% (v/v) ethylene glycol. Data were collected at beamline I24 at Diamond Light Source (DLS).

Structural determination

The data were integrated with automated XDS⁶² and scaled using the CCP4 program suite⁶³. Initial phases were obtained by molecular replacement with PHASER⁶⁴ using UbcH5B and Ub from PDB 3ZNI³⁰ and MDM2^R-MDMX^R from PDB 3VJF²⁸ as the search models. Two copies of MDM2^R-MDMX^R-UbcH5B–Ub complex were found in the asymmetric unit. The model was built in COOT⁶⁵ and refined using PHENIX⁶⁶. The complex was refined to a resolution of 2.16 Å and the final model contained two copies of UbcH5B (Chains A and E residues 2-147), Ub (Chains B and F residues 1–76), MDM2^R (Chains C and G residues 428-491) and MDMX^R (Chains D and H residues 427-490). Residues with no observable side chain electron density were built as alanine stub. Details of the refinement statistics are shown in Table 1. All figure models were generated using PYMOL (Schrödinger).

Surface Plasmon resonance (SPR) analyses

All SPR experiments were performed at 25 ºC on a Biacore T200 with a CM-5 chip (GE Healthcare). GST-MDM2 398-C and GST-MDMX^R variants were coupled to CM-5 chips as described previously³⁴. For GST-MDM2 398-C variants, UbcH5B–Ub variants were serially diluted in running buffer containing 25 mM Tris-HCl, pH 7.6, 150 mM NaCl, 0.1 mg/ml BSA, 1 mM DTT, 0.005% (v/v) Tween-20 and 300 μM UbΔGG. UbΔGG was included to saturate UbcH5B’s backside for optimal RING E3-UbcH5B–Ub interaction as described previously³⁴. For GST-MDMX^R variants, UbcH5B–Ub variants were serially diluted in running buffer containing 25 mM Tris-HCl, pH 7.6, 150 mM NaCl, 0.1 mg/ml BSA, 1 mM DTT and 0.005% (v/v) Tween-20. Binding was measured at indicated concentration ranges in Supplementary Fig. 3 in duplicate. Data reported are the difference in SPR signal between GST-MDM2 398-C or GST-MDMX^R variants and GST alone. The data were analyzed by steady-state affinity analysis using Biacore T200 BIAevaluation software (GE Healthcare) and Scrubber2 (BioLogic Software).

Single-turnover lysine discharge assays

UbcH5B variants (7-10 μM) were charged with Ub variants (20-50 μM) in buffer containing 50 mM Tris-HCl, pH 7.6, 50 mM NaCl, 0.5 μM Arabidopsis thaliana UBA1 and 1 mg/ml BSA for 20 min at 23 °C and charging was stopped by incubating the reaction with 0.01 U/ml apyrase (Sigma) and 30 mM EDTA for 2-3 min at 23 °C as described previously³⁴. The lysine discharge reactions were then initiated by adding a mixture containing 50 mM Tris-HCl, pH 7.6, 50 mM NaCl, 1 mg/ml BSA, L-lysine (150 mM or 50 mM for Supplementary Fig. 1) and E3 variants (as indicated or 0.3 μM in Figs. 1g-i, 2c,d, 3c and Supplementary Fig. 2d; 0.8 μM in Supplementary Fig. 1a,c,g,i,k; 0.6 μM in Supplementary Fig. 1e; and 0.25 μM in Supplementary Fig. 2a). Final concentrations are in parenthesis. Reactions were quenched with SDS loading buffer at the indicated times, resolved by SDS-PAGE and visualized by staining with InstantBlue (Expedeon). For quantification in Supplementary Fig. 1, the reactions were performed in triplicate. LI-COR Odyssey scanner was used to analyze the stained gels and bands were quantified by using Image Studio Lite software (LI-COR Biosciences). The fraction of UbcH5B~Ub left was plotted on a bar graph. In Fig. 3d, Fluorescently-labeled Ub was used instead of Ub and visualization was performed using a LI-COR Odyssey scanner prior to staining with InstantBlue. The fluorescently-labeled UbcH5B~Ub band was quantified by using Image Studio Lite software (LI-COR Biosciences).

In vitro pull-down experiments

GST-MDM2 398-C variants were co-expressed with the respective His-MDM2 398-C variants or His-MDMX 383-C as shown in Supplementary 2b,c. Cells were harvested, lysed and centrifuge. Lysates were applied onto 400 μl of Ni-NTA beads, washed with 25 mM Tris-HCl, pH 7.6, 0.2 M NaCl, 20 mM imidazole and 5 mM β-mercaptoethnaol and eluted with 25 mM Tris-HCl, pH 7.6, 0.2 M NaCl, 200 mM imidazole and 5 mM β-mercaptoethnaol. The eluted product was then applied onto 400 μl of glutathione sepharose beads, washed with 50 mM Tris-HCl, pH 7.6, 0.2 M NaCl, 5 mM DTT and eluted with 50 mM Tris-HCl, pH 7.6, 0.2 M NaCl, 5 mM DTT and 10 mM glutathione. 3 μg of eluted product was then resolved on the SDS-PAGE. For control experiments, GST-MDM2 398-C, His-MDM2 398-C and His-MDMX 383-C were expressed individually. The lysates were divided in half, applied onto 400 μl of Ni-NTA or glutathione sepharose beads, washed and eluted with buffer as above. 3 μg of eluted product or equal volume (for proteins that do not bind the beads, but eluted with the same volume) was then resolved on the SDS-PAGE and visualized by staining with InstantBlue.

Cell culture and generation of tetracycline inducible p53 knockdown-MDM2 knock-out cells

Human osteosarcoma U2OS or human embryonic kidney HEK293 cells were obtained from ATCC and were cultured in Dulbecco's Modified Eagle Medium (DMEM) supplemented with 10 % FBS and 2 mM L-Glutamine. All cell lines were mycoplasma negative and were authenticated by STR multiplex assay.

In the experiments showed in Fig. 6, Fig. 7 and Supplementary Fig. 4-6, tetracycline inducible p53 knockdown-MDM2 knock-out cells were used. U2OS cells were infected with pLKO tet-on shRNA targeting p53 and selected by puromycin for 5 days. Knock-down of p53 was confirmed by Western blotting. Doxycycline treatment induced shRNA expression and p53 depletion within 48 hours and this effect was reversed within 5 days after discontinuing doxycycline treatment (Supplementary Fig. 4a). CRISPR/Cas9 vectors targeting the MDM2 p53 binding domain was then transiently transfected, clones isolated and CRISPR disruptions assessed by Western blotting, then confirmed genomic PCR followed by DNA sequencing (Supplementary Fig. 4b,c).

Doxycycline was discontinued for 5 days before plating for "No Doxy" condition. Plasmids used and target sequences were described in Supplementary Tables 4 and 5.

Mutagenesis PCR

Plasmids used are described in Supplementary Table 4. All mutations within the RING domain of MDM2 and MDMX were generated by site-directed mutagenesis using KOD hot start master mix (Cat. No. 71842, Novagen) and verified by DNA sequencing.

Immunoblotting

Cells were washed with PBS and proteins were extracted with lysis buffer (1 % Triton X-100, 150 mM NaCl and 50 mM Tris-HCl pH8.0), which were then centrifuged at 16,100 r.c.f. at 4 °C for 10 minutes. Where appropriate, protein concentration in cellular lysate was determined using the BCA protein assay, with concentrations normalized to the lowest-concentration sample. Samples were heated to 99 °C for 10 minutes with sample buffer (Final concentration: 10 % glycerol, 2 % SDS, 62.5 mM Tris-HCl pH6.8 and 5 % 2-mercaptoethanol) before separation by SDS-PAGE in a 10 % or 4-12 % bis-tris polyacrylamide gel (Life Technologies) and transfer to 0.2 µm nitrocellulose membrane (GE Healthcare). After the transfer, the membrane was blocked by TTBS supplemented with 5 % skimmed milk or Odyssey blocking buffer (LI-COR) for 1 hour, and was then incubated with primary antibodies (Supplementary Table 6) overnight at 4 °C.

After this incubation time, the membrane was washed with TTBS and incubated with appropriate secondary antibodies (Supplementary Table 6) for 1 hour, then washed with TTBS and TBS, and membrane-bound secondary antibodies were detected using Odyssey scanner (LI-COR). Images were analyzed by Image Studio (LI-COR).

Data were normalized to actin expression, as appropriate.

Immunoprecipitation

Proteins were extracted and normalized as described in immunoblotting section. Cellular lysates containing 250 μg proteins were incubated with 2 μg of appropriate primary antibody (Supplementary Table 6) or control IgG for 1 hour at 4 °C, and then incubated with Dynabeads protein G magnetic beads overnight at 4 °C on the rotator. The immunoprecipitates were washed 3 times with PBS and heated to 99 °C for 10 minutes with sample buffer. After removing magnetic beads, SDS-PAGE and membrane transfer were performed as described in immunoblotting section.

Ubiquitination assay

U2OS cells grown in 100 mm dishes were transiently transfected with various plasmids (Supplementary Table 4) using Genejuice transfection reagent (Cat. No. 70967, Novagen) and cultivated for further 24 hours. Cells were treated for 5 hours with proteasome inhibitor MG132 (10 mM) and lysed in UBA buffer (6 M Guanidine HCl, 300 mM NaCl and 50 mM phosphate pH 8 supplemented with 100 µg/ml N-Ethylmaleimide (NEM)). Lysates were boiled at 99 °C for 20 minutes with 1,000 RPM shake. His-tag pull-down was performed using 50 µl of Dynabeads His-Tag Isolation & Pull down Beads (Cat. No. 10104D, Novex) overnight and beads were washed with UBA, UBB (3 M Guanidine HCl, 300 mM NaCl and 50 mM phosphate pH 8), UBC (300 mM NaCl and 50 mM phosphate pH 8) then PBS, followed by SDS–PAGE and Western blotting.

Ubiquitinated and input proteins were detected by immunoblotting.

Immunofluorescence

HEK293 cells were grown on 13 mm glass-coverslips (VWR) placed into 6 well plates. Cells were transfected with MDM2 wild type or mutants and GFP-tagged MDMX overnight then washed and fixed with 4 % paraformaldehyde (PFA) (Sigma) in PBS for 20 minutes at room temperature and PFA was quenched with 30 mM glycine in PBS. Cells were washed and permeabilized with 0.2 % Triton X-100 in PBS for 10 minutes at room temperature. Cells were then washed with PBS and blocked with 0.5 % BSA in PBS for 3 hours at room temperature and were then incubated with MDM2 antibody (Supplementary Table 6) overnight at 4 °C. After this incubation time, cells were washed with TTBS and incubated with Alexa568-conjugated secondary antibody (Supplementary Table 6) for 2 hours, then washed with TPBS and then PBS. Coverslips were washed with double-distilled water to remove any salt residues and then mounted with VECTASHIELD Mounting Medium with DAPI (Vector laboratories). Images were taken using confocal microscope Fluoview FV1000 (Olympus).

RNA Extraction and Quantitative Real-Time PCR

Cells were grown in 100 mm dishes and RNA was extracted using RNeasy Mini Kit (Qiagen) RNA concentration was determined using a Nanodrop 2000c (Thermo Scientific). cDNA was synthesized through reverse transcription polymerase chain reaction (RT-PCR) from 1 μg RNA using High-capacity RNA-to-cDNA Kit (Life Technologies). The quantitative real-time PCR (qPCR) reaction was performed with 10 ng of cDNA using Fast SYBR green master mix. Primers used are described in Supplementary Table 5. The amount of fluorescent PCR product accumulating during the PCR programme (20 sec at 95 °C hot start, 40 cycles of 3 sec denaturing at 95 °C, 30 sec annealing-elongation at 60 °C) was detected by the ABI 7500 Fast (Thermo scientific). Gene expression was quantified relative to the housekeeping gene β-2 microglobulin (B2M) according to the comparative ΔΔCt method.

Chromatin Immunoprecipitation (ChIP)

Approximately 5 x 10⁷ cells (after 5 days of doxycycline discontinuation, where appropriate) were seeded on 150 mm dish. After 2 days, cells were washed and fixed in 1 % formaldehyde in serum-free DMEM for 10 min. Cross-linking reaction was stopped by adding glycine (final concentration 125 mM) for 5 min. Media was completely aspirated and cells were washed 4 times with ice cold PBS. Cells were then harvested in 5 ml of ChIP cell lysis buffer (10 mM Tris-HCl pH 8.0, 10 mM NaCl, 0.20 % Igepal CA-630, 10 mM NaBu, 50 μg/ml PMSF and 1 μg/ml Leupeptin) and incubated for 10 min on ice. Cells were then centrifuged at 2,500 rpm for 5 min at 4 °C and nuclei was resuspended and incubated in 1.2 ml of ChIP nuclei lysis buffer (NLB: 50 mM Tris-HCl pH 8.0, 10 mM EDTA, 1 % SDS, 10 mM NaBu, 50 μg/ml PMSF and 1 μg/ml Leupeptin) for 10 min on ice. 0.72 ml of IP dilution buffer (IPDB: 20 mM Tris-HCl, 150 mM NaCl, 2 mM EDTA, 1 % Triton X-100, 0.01% SDS, 10 mM NaBu, 50 μg/ml PMSF and 1 μg/ml Leupeptin) and samples were sonicated using the Bioruptor sonicator (high setting, 30s on-30s off) for 30 min (replaced ice every 10 min). 4.1 ml of IPDB was added and chromatin was precleared by 20 μl of normal mouse IgG (Santa Cruz Biotechnology) and 50 μl of Dynabeads protein G (Life technologies) for 4 hours on a rotating wheel at 4 °C. Modified IPDB (IPDBmod) was made by combining 1:4 NLB:IPDM. Supernatant was transferred to new Eppendorf tubes and set up in necessary conditions for ChIP with p53 and MDM2. Samples were incubated on a rotating wheel overnight at 4 °C.

ChIP Target conditions

p53: 0.675 ml chromatin + 0.675 ml IPDBmod + 7.5 μg p53 antibody (DO-1) MDM2: 0.675 ml chromatin + 0.675 ml IPDBmod + 7.5 μg MDM2 antibody (Ab-1) Negative control: 0.675 ml chromatin + 0.675 ml IPDBmod + 7.5 μg IgG

50 μl of Dynabeads protein G was added to the samples and they were incubated for 3 hours on a rotating wheel overnight at 4 °C. Beads were washed twice with IP wash buffer 1 (20 mM Tris-HCl pH8.0, 50 mM NaCl, 2 mM EDTA, 1 % Triton X-100 and 0.001 % SDS), once with IP wash buffer 2 (10 mM Tris-HCl pH8.0, 250 mM LiCl, 1 mM EDTA, 1 % Igepal CA-630 and 1% Deoxycholate), then twice with TE buffer pH8.0. ChIP materials were eluted in 300 μl of ChIP elution buffer (100 mM NaHCO3, 1% SDS, 20 μg RNase and 40 μg Proteinase K) on a Thermomixer heat block (Eppendorf) at 900 rpm overnight for 65 °C. Input samples were similarly processed.

Following this, 200 μl of samples were transferred to fresh Eppendorf tubes and 200 μl of TE buffer. 400 μl of phenol-chloroform-isoamylalcohol (PCI) was added to the samples and they were transferred to the phase-lock tubes (5 Prime) for DNA isolation. Samples were spun at 14,000 r.c.f. for 5 minutes at RT and upper phase was transferred to Eppendorf tubes. 16 μl of 5 M NaCl and 8 μl of glycogen (5 mg/ml) was added and vortexed. 1 ml of ice cold 100 % ethanol was then added to the samples, voetexed and stored at -80 °C for 30 min. Following this, samples were spun at 16,100 r.c.f. for 15 min at 4 °C, and the supernatant was discarded. Pellets were washed with 1 ml ice cold 80 % ethanol and spun at 16,100 r.c.f. for 10 min at 4 °C. The supernatant was discarded and pellets were dried at RT then DNA was eluted in 50 μl 10 mM Tris-HCl and heated on a Thermomixer heat block (Eppendorf) for 10 min at 50 °C in order to dissolve the pellet. Samples were then analyzed via qPCR as described with a modified PCR programme (20 sec at 95 °C hot start, 45 cycles of 3 sec denaturing at 95 °C, 30 sec annealing-elongation at 60 °C). Primers used are provided in Supplementary Table 5. Values were calculated as % input.

Cell growth assay and cell staining

Cells were plated in 24-well plates (after 5 days of doxycycline discontinuation where appropriate). After trypsinization, the number of viable cells was counted using CASY cell counter.

For cell staining, cells were fixed with ice-cold 100 % methanol and stained with 0.04 % SRB (Sulforhodamine B) in 1 % acetic acid. SRB intensity was quantified by ImageJ software.

Statistical analysis and reproducibility

Experimental data are expressed as mean ± S.D. unless otherwise indicted. Statistical differences were analyzed by one- or two-way analysis of variance (ANOVA) followed by Tukey's post-hoc test (two-tailed), as appropriate. F statistics and degrees of freedom for all ANOVAs were reported in Supplementary Table 3. All statistical analyses were performed with Prism (GraphPad Software). Statistical significance was defined as p<0.05. Each experiment was repeated independently at least three times and sample sizes and number of repeats are defined in each figure legend.