Activation of the p53 pathway by small-molecule-induced MDM2 and MDMX dimerizationBradford Graves1, Thelma Thompson, Mingxuan Xia, Cheryl Janson, Christine Lukacs, Dayanand Deo, Paola Di Lello,David Fry, Colin Garvie, Kuo-Sen Huang, Lin Gao, Christian Tovar, Allen Lovey, Jutta Wanner, and Lyubomir T. Vassilev1

Roche Research Center, Hoffmann-La Roche Inc., Nutley, NJ 07110

Edited by Alan R. Fersht, Medical Research Council Laboratory of Molecular Biology, Cambridge, United Kingdom, and approved May 30, 2012 (received forreview March 3, 2012)

Activation of p53 tumor suppressor by antagonizing its negativeregulator murine double minute (MDM)2 has been considered anattractive strategy for cancer therapy and several classes of p53-MDM2 binding inhibitors have been developed. However, thesecompounds do not inhibit the p53-MDMX interaction, and theireffectiveness can be compromised in tumors overexpressingMDMX. Here, we identify small molecules that potently blockp53 binding with both MDM2 and MDMX by inhibitor-drivenhomo- and/or heterodimerization of MDM2 and MDMX proteins.Structural studies revealed that the inhibitors bind into and oc-clude the p53 pockets of MDM2 and MDMX by inducing the for-mation of dimeric protein complexes kept together by a dimericsmall-molecule core. This mode of action effectively stabilized p53and activated p53 signaling in cancer cells, leading to cell cyclearrest and apoptosis. Dual MDM2/MDMX antagonists restoredp53 apoptotic activity in the presence of high levels of MDMXand may offer a more effective therapeutic modality for MDMX-overexpressing cancers.

The tumor suppressor p53 is a powerful growth-suppressiveand proapoptotic protein tightly controlled by its negative

regulators: murine double minute (MDM)2 and MDMX (1, 2).These proteins bind p53 with their structurally similar N-termi-nal domains and effectively inhibit p53 transcriptional activity (1,3). They both possess a RING (really interesting new gene)domain in their C termini, but it is only functional in MDM2,which serves as a specific E3 ligase and main regulator of p53stability (4, 5). Despite its RING domain, MDMX does not havean intrinsic ligase activity and does not affect directly p53 sta-bility (6). However, MDMX can enhance ligase activity ofMDM2 toward p53 by forming MDM2/MDMX heterodimers (7,8). It has been reported that the MDM2/MDMX complex isresponsible for polyubiquitination of p53, whereas MDM2 aloneprimarily induces monoubiquitination (9). Targeted disruptionof MDM2/MDMX heterocomplexes is embryonic-lethal in mice,suggesting that complex formation is essential for p53 regulationin vivo (10). On the other hand, MDM2 can also ubiquitinateMDMX and is, therefore, responsible for its stability as well (11,12). MDM2 is a transcriptional target of p53, and both proteinsform an autoregulatory feedback loop by which they mutuallycontrol their cellular levels (13).The functional relationship between MDM2 and MDMX is

still being refined at the molecular level, but it is well establishedthat these two negative regulators play a critical role in con-trolling p53 tumor-suppressor function in normal cells (2, 14).This is why they are frequently overproduced through geneamplification and/or overexpression in tumors that retain wild-type p53 (14). Therefore, antagonizing the binding of MDM2and MDMX to p53 is expected to restore p53 function and mayoffer a strategy for cancer therapy (15). Recently identifiedsmall-molecule inhibitors of the p53-MDM2 interaction havevalidated this approach, and the first pharmacological MDM2antagonists are now undergoing clinical evaluation (16, 17).MDM2 inhibitors have shown effective p53 activation followedby cell cycle arrest, induction of apoptosis, and tumor regressionin cancer cells with mdm2 gene amplification (18, 19). However,their apoptotic activity has been found to be moderate to

marginal in many tumor cell lines expressing normal levels ofMDM2, suggesting that cancer uses other mechanisms to at-tenuate or disable p53 signaling (20), such as the overexpressionof the other negative p53 regulator, MDMX. High levels ofMDMX protein can make MDM2 antagonists, which haveshown very low activity against p53-MDMX binding, ineffectivein killing cancer cells (21–23). Thus, simultaneous inhibition ofMDM2 and MDMX is needed to release the full activity ofstabilized p53 (15, 17). Therefore, recent efforts have been fo-cused on identification of dual MDM2/MDMX antagonists.Because of distinct structural differences between MDM2 and

MDMX in their p53-binding pockets (24–26), small moleculesoptimized for MDM2 have shown very low affinity for MDMX(27). For example, the first potent and selective small-moleculeMDM2 antagonist, nutlin-3a, has ∼400-fold lower potencyagainst MDMX than MDM2 (28). This trend has been followedby other MDM2 inhibitors (19). Efforts to identify MDMX-specific inhibitors have recently yielded a class of small moleculeswith in vitro binding activity in the high nanomolar range butrelatively poor cellular potency and uncertain mechanism ofcellular activity (29). Nearly equipotent MDM2/MDMX peptideinhibitors have been identified and characterized structurally buttheir activity has been detected only in cell-free systems (30).Recently, a cell-penetrating “stapled” peptide with good MDMXbinding affinity has been identified and evaluated in cancer cells(31). Although cellular potency against p53-MDMX interactionhas been found adequate, this peptide was unable to disrupteffectively p53-MDM2 binding, and it has been combined withthe MDM2 antagonist, nutlin-3, to assess the antitumor potentialof this emerging therapeutic modality.Here, we identify a class of small molecules that can potently

inhibit p53 interactions with both MDM2 and MDMX by in-duced protein dimerization and effectively restore p53 activity inMDMX-overexpressing cancer cells. We show that antagonizingboth negative p53 regulators significantly improves the apoptoticoutcome in cancer cells overproducing MDMX.

ResultsIdentification of Indolyl Hydantoins as MDM2/MDMX Antagonists. Adiverse library of small molecules was screened for suppressionof p53-MDMX binding (Table S1). The hits were then tested foractivity against the p53-MDM2 interaction. One series of indolylhydantoin compounds emerged as potent, dual MDM2/MDMX

Author contributions: B.G., T.T., P.D.L., D.F., C.G., K.-S.H., L.G., C.T., J.W., and L.T.V. de-signed research; B.G., T.T., M.X., C.J., C.L., D.D., P.D.L., C.G., L.G., C.T., and J.W. performedresearch; A.L. and J.W. contributed new reagents/analytic tools; B.G., T.T., M.X., C.L., D.D.,P.D.L., D.F., C.G., K.-S.H., L.G., C.T., and L.T.V. analyzed data; and B.G. and L.T.V. wrotethe paper.

Conflict of interest statement: The authors are employees of Hoffmann-La Roche Inc.

This article is a PNAS Direct Submission.

Data deposition: The atomic coordinates and structure factors have been deposited in theProtein Data Bank, www.pdb.org (PDB ID codes 3U15 and 3VBG).1To whom correspondence may be addressed. E-mail: bradford.graves@roche.com orlyubomir.vassilev@roche.com.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1203789109/-/DCSupplemental.

11788–11793 | PNAS | July 17, 2012 | vol. 109 | no. 29 www.pnas.org/cgi/doi/10.1073/pnas.1203789109

Dow

nloa

ded

by g

uest

on

Apr

il 23

, 202

0

antagonists. For example, RO-2443 (Fig. 1A) showed a re-markably similar inhibitory activity against both MDM2 (IC50 =33 nM) and MDMX (IC50 = 41 nM) binding to p53. For its size,RO-2443 appeared highly potent (ligand efficiency, defined asbinding energy per heavy atom, is 0.36) and likely to bind into, atmost, two of the three subpockets on the surface of MDMX orMDM2. These pockets were defined by the original structure ofa p53 peptide bound to MDM2 (32) which showed that therewere three key residues from the p53 peptide, Phe19, Trp23, andLeu26. Throughout this report, these binding pockets on thesurface of MDM2 and MDMX are referred to as the Phe, Trp,and Leu pockets.To investigate the mechanism of action of RO-2443, we first

used NMR spectroscopy. The [1H-15N]HSQC spectrum of the

MDMX N-terminal fragment (Fig. 1B) in the presence and ab-sence of the small molecule indicated that: (i) the compound wasbinding to the p53 pocket; and (ii) there was a substantial up-field shift for Y63, which is consistent with shielding by an aro-matic group. A similar shift had been observed upon binding ofa p53 peptide to MDMX attributable to shielding by Phe19. Inaddition, there was an overall resonance broadening in thepresence of the compound, which was manifested in the HSQCspectrum as a decrease in cross-peak intensities, suggesting for-mation of a higher-molecular-mass species. To determine theeffect of RO-2443 on the state of MDMX in solution, size-ex-clusion chromatography with static light scattering (SEC-SLS)was performed (Fig. 1C). In the absence of RO-2443, the proteingave a SLS-calculated mass of 12.8 kDa, which agrees with thetheoretical mass of a monomer (12.3 kDa). Addition of RO-2443resulted in the protein eluting at an earlier elution volume, in-dicating that the shape and/or mass of the protein had changed.The SLS-calculated mass of the complex was 24.1 kDa, whichsuggests that MDMX forms a dimer when bound to RO-2443.Kinetic analysis of the binding of RO-2443 to MDMX (Fig. 1D)showed that the binding was in line with a two-molecule bindingmodel. Isothermal calorimetry (ITC) confirmed a 1:1 ratio be-tween compound and protein but could not distinguish between1:1, 2:2, or higher-order species (Fig. 1E). It revealed a bindingconstant (Kd = 78 nM) that is in good agreement with thebinding assay. The ITC also shows that the binding is completelydominated by the entropy component, consistent with bindinginteractions involving primarily hydrophobic effects. Thus,a consistent picture emerged that RO-2443 induces some sort ofdimer formation of MDMX and MDM2.

Crystal Structures Reveal Tight MDMX Dimer Formation. Crystals ofMDMX bound to RO-2443 were grown that diffracted to rela-tively high resolution (1.8 Å) and molecular replacement withthe structure of MDMX bound to a p53 peptide (33) was suc-cessful (Table S2). The structure is comprised of four monomersin the asymmetric unit arranged as a pair of dimers (Fig. 2A).These dimers show at their core two molecules of the inhibitor,each of which has binding interactions to both protein mono-mers. For each inhibitor molecule, the indolyl-hydantoin moietyoccupies the Phe pocket of one protein monomer, whereas thedi-fluoro-phenyl group reaches into the Trp pocket of the other.From a different viewpoint, Fig. 2B shows an overlay of the in-hibitor dimer structure with that of the p53 peptide (33). Clearly,among the key interactions that the compounds form is an ex-tensive aromatic stacking interaction between the two indolyl-hydantoin groups with interplanar distances ranging from 3.3 to3.7 Å. The stacking interaction extends to include a tyrosineresidue (Y63) from each of the protein monomers, resulting ina four-level π-sandwich. The small molecule–protein interactionswere even slightly shorter, ranging from 3.1 to 3.5 Å. Anotherpositive interaction between the small molecules is the σ-hole ofthe chlorine attached to the indole ring pointing at the di-fluoro-phenyl ring of the other molecule. There are multiple contacts at4 Å between the chlorine atom and the phenyl ring, which arewithin the range described by Bissantz et al. (34).Modeling studies indicated that MDM2 should be able to form

the exact same homodimer and that MDMX and MDM2 couldform heterodimers. Indeed, crystal structures of MDM2 boundto RO-2443 and other analogs confirm the expected MDM2homodimer structure (Fig. S1A). Using both structures, a modelof the potential heterodimer can be assembled showing no se-rious conflicts (Fig. S1B). Fig. 2C provides a conceptually clearerpicture of the nature of the MDMX/RO-2443 dimer. As sug-gested by the model (also see Fig. S2A), the p53 binding pocketson MDMX (or MDM2) are nearly completely occluded. Themost exposed part of the inhibitor is the methylene bridgebetween the hydantoin and phenyl groups, which providesa means for extending the inhibitor to reach into the Leu pocket(Fig. S2B). Partly for this reason, but also to impact the

0

50

100

150

200

250

1.2 1.4 1.6 1.8 2 2.2 2.4

MDMX(15-106)MDMX(15-106) + RO5457707MDMX(15-106) + RO5472443

0.0 0.5 1.0 1.5 2.0 2.5 3.0

-1.2

-1.0

-0.8

-0.6

-0.4

-0.2

0.0

0.2-0.08

-0.06

-0.04

-0.02

0.00

0.020 10 20 30 40

Time (min)

µcal

/sec

Molar Ratio

kcal

/mol

e of

inje

ctan

t

0

20

40

60

80

100

1.E-04 1.E-02 1.E+00

MDM2MDMX

-10.0

-5.0

0.0

5.0

kcal

/mol

e

DGDH -TDS

RO-2443

A

C

B

IC50p53-MDM2: 33 nMp53-MDMX: 41 nM

D

E

1H (ppm)

15N

(ppm

)

Y63

Y63

G54

G54

15N-MDMX15N-MDMX/RO-2443

Elution volume (ml)

Abs

orbt

ion

(mA

U)

MDMXMDMX + RO-2443

% o

f con

trol

Concentration (uM)

Kd = 0.078 μMΔH = -1.13 kcal/mole-TΔS = -8.6 kcal/moleΔG = -9.69 kcal/mole

NH

N

NH

O

O

Cl

F

F

0 50 100 1500

2000

4000

RO-2443 (uM)

Δ F

I 1-molecular model

2-molecular model

Fig. 1. RO-2443 inhibits the interaction of p53 with MDM2 and MDMX. (A)Chemical structure and in vitro activity of RO-2443. (B) [1H-15N]HSQC NMRspectra of hzMDMX (His6-15-106, L45V, V95L) in the absence (black contours)and presence (red contours) of RO-2443. The decrease in cross-peak in-tensities observed upon binding of RO-2443 is evidenced by fewer contoursin the red spectrum. (C) Binding of RO-2443 to MDMX in solution changes itsmobility consistent with protein dimerization. SLS detection of N-terminalzebrafish MDMX fragment eluting from a size-exclusion column shows theshift of the protein peak to a position consistent with that of a dimer. (D)Kinetic analysis of the change in Trp fluorescence intensity over a wide rangeof RO-2443 concentrations. Curves based on a binding model with themonomer or a dimer show that the data are consistent with the dimer model(Kdhigh = 0.4 μM and Kdlow = 57.2 μM). (E) Isothermal calorimetry shows theimpact of binding RO-2443 to hMDMX and that the energy of the in-teraction is dominated by the entropy component.

Graves et al. PNAS | July 17, 2012 | vol. 109 | no. 29 | 11789

MED

ICALSC

IENCE

S

Dow

nloa

ded

by g

uest

on

Apr

il 23

, 202

0

physicochemical properties of the compounds, this site was tar-geted for modification, resulting in analogs such as RO-5963.

RO-5963 Inhibited p53 Binding to MDM2 and MDMX in Cancer Cells.RO-2443 showed potent MDM2/MDMX inhibitory activityin vitro, but poor water solubility did not allow for a meaningfulassessment of its cellular activity. Further chemical optimizationof RO-2443 yielded RO-5963 (Fig. 3A), a close analog withslightly increased potency but substantially improved solubility.In the same binding assay, RO-5963 showed p53-MDM2 in-hibitory activity (IC50, ∼17 nM) similar to that of nutlin-3a (IC50,∼19 nM) (Fig. 3A). Its p53-MDMX inhibitory activity (IC50, ∼24nM) was nearly equivalent to MDM2 activity but ∼400-foldbetter than the MDMX potency of nutlin-3a (IC50, ∼9 μM). RO-5963 penetrated MDMX-overexpressing breast cancer cells(MCF7), stabilized p53, and elevated protein levels of its

transcription targets, p21 and MDM2, in a dose-dependentmanner (Fig. 3B). The increase in p21 and MDM2 protein levelswas attributable to induction of their transcription as revealed bythe dose-dependent increase of their mRNA and of two otherp53 transcriptional targets, macrophage inhibitory cytokine-1(MIC-1) and bcl-2 associated X protein (BAX), but not MDMX,which is not under p53 control (Fig. 3C). Stabilization and ac-tivation of p53 was induced by disrupting its interaction withMDM2 and MDMX in MCF7 cells as demonstrated by immu-noprecipitation of either p53 or MDMX proteins from cells ex-posed to RO-5963 followed by Western analysis (Fig. 3D). At 20μM, RO-5963 was equivalent to 10 μM nutlin in inhibiting p53-MDM2 binding and also effectively blocked p53-MDMX bindingat both 10 and 20 μM concentration. As expected, the MDM2-specific inhibitor, nutlin-3a, showed no effect on the p53-MDMXinteraction. A notable increase in MDMX and MDM2 proteinspulled down by immunoprecipitated MDMX (Fig. 3D, Right)was consistent with predicted RO-5963-induced formation ofMDMX/MDMX and MDMX/MDM2 dimers in MCF7 cells.

RO-5963 Activates p53 Pathway in Cancer Cells Expressing Wild-Typep53. Inhibition of p53-MDM2 binding should induce p53 sig-naling only if the cells express wild-type but not mutant p53,which generally loses its transcriptional activity. Indeed, the EC50of RO-5963 was ∼10-fold lower in cancer cells expressing wild-type p53 compared with mutant p53 cells (Fig. 4A) and onlyactivated the p53 transcriptional targets p21 and MDM2 in p53wild-type but not p53 mutant cancer cells (Fig. S3). RO-5963affected the viability of HCT116 cells but not their nutlin-re-sistant clone HCT116R1, which has lost ability to induce p53response (Fig. S4). Treatment of four cancer cell lines with RO-5963 did not increase the levels of p53Ser15 phosphorylation,suggesting that p53 activation is not caused by genotoxic stressinduced by the compound (Fig. 4B).

+RO-2443

High Kd

+Low Kd

Step 1:

Step 2:

XMDMXMDM

MDMX

MDMX

MDMXMDMX

A

C

B

Fig. 2. RO-2443 binds to the p53 pocket of MDMX and induces proteindimerization. (A) Crystal structure of RO-2443 bound to MDMX. Close-upview of the p53 binding regions of two MDMX molecules forming a dimer.One MDMX molecule is shown as a surface rendition (carbon, white; oxygen,red; nitrogen, blue; and sulfur, yellow). The other MDMX molecule is shownas a stick figure with the same color scheme, but the carbon atoms arecolored cyan. The two RO-2443 molecules are shown as stick figures. Themolecule with cyan-colored carbon atoms is binding with the indole-hydantoin moiety in the Phe pocket of MDMX, shown as a stick figure andthe di-fluoro-phenyl group in the Trp pocket of MDMX shown as a surface.The molecule with green-colored carbon atoms binds in the reverse mode.(B) View of the two inhibitor molecules (∼90° rotation) showing how theyrelate to the binding of a p53 peptide with carbon atoms colored magenta(32). The indolyl group of one inhibitor (green) overlays with Phe19 of thepeptide, whereas the di-fluoro-phenyl group of the other inhibitor (cyan)overlays with Trp23. It is worth noting that the chlorine atom of the 6-chloro-tryptophan of the peptide is nearly coincident with the parafluoro atom ofthe inhibitor. (C) Dimer model for binding of RO-2443 to MDMX. This ren-dition of the possible steps in the formation of the dimer indicates thatmonomeric interactions likely form first, followed by a pairing of monomersto form the dimer. This is based on the lack of any detection of dimerinteractions by the small molecules alone. A binding model in which theformation of the monomer is followed by the addition of a second smallmolecule and subsequent addition of a vacant copy of MDMX cannot beruled out.

F

N

NHO

F

NH

O

O

NH

Cl

OH OHA

B C

0

20

40

6080

100

120

1.E-04

1.E-03

1.E-02

1.E-01

1.E+00

1.E+01

1.E+02

MDM2MDM X

Nutlin-3a: IC50MDM2: 18.7 nMMDMX: 9300 nM

RO-5963: IC50MDM2: 17.3 nMMDMX: 24.7 nM

% o

f con

trol

0

10

20

30

40

1 2 3 4

p21MDM2MDMXMIC-1BAX

Concentration (µM)1.25 2.5 5 10

mRN

A (fo

ld c

hang

e)

020406080

100120

1.E-04

1.E-03

1.E-02

1.E-01

1.E+00

1.E+01

1.E+02

MDM2MDMX

Log concentration (µM)

D

Cont

rol

Nut

lin

5963

-10

5963

-20

Cont

rol

Nut

lin

5963

-10

5963

-20

IP: p53 IP: MDMX

p53 MDMXMDM2

Actin

Cont

rol

Nut

lin

5963

-10

5963

-20

p53p21MDM2MDMXActin

Cont

rol

5963

-2.5

5963

-559

63-1

059

63-2

0N

utlin

-10

Fig. 3. RO-5963 stabilizes p53 and activates the p53 pathway in cancer cells.(A) Chemical structure and in vitro inhibitory activity of RO-5963 and theMDM2 antagonist, nutlin-3a. (B) RO-5963 stabilizes p53 and elevates proteinlevels of p53 targets, p21 and MDM2. Log-phase MCF7 cells were incubatedwith RO-5953 for 24 h, and cell lysates were analyzed by Western blotting.(C) Dose-dependent induction of p53 target genes in MCF7 cells 24 h postRO-5963 addition. (D) RO-5963 inhibits p53-MDM2 and p53-MDMX bindingin cancer cells. MCF7 cells were incubated with 10 μM nutlin-3a and 10 or 20μM RO-5963 for 4 h, and the levels of p53, MDM2, and MDMX were de-termined in protein complexes immunoprecipitated with anti-MDMX oranti-p53 antibodies by Western blotting.

11790 | www.pnas.org/cgi/doi/10.1073/pnas.1203789109 Graves et al.

Dow

nloa

ded

by g

uest

on

Apr

il 23

, 202

0

By disrupting p53-MDM2 binding and the autoregulatoryfeedback loop, small-molecule antagonists (e.g., nutlins) can ele-vate MDM2 protein levels and, thus, facilitate MDMX ubiquiti-nation and degradation. As a result, MDMX protein levels werefound reduced in many cancer cell lines in the presence of nutlin-3(28). If RO-5963 induces homo- or heterodimerization of MDM2and MDMX inside living cells, that might interfere with MDM2ligase activity and/or its ability to ubiquitinate MDMX. To in-vestigate this possibility, we incubated G401 cancer cells with in-creasing concentrations of RO-5963 in the presence or absence of10 μM nutlin-3a. As expected, nutlin substantially reducedMDMX protein levels (Fig. 4C). RO-5963 dose-dependently in-creased p53, MDM2, and p21 levels but only slightly reducedMDMX. However, it protected MDMX from nutlin-induceddegradation by MDM2 in both G401 and H460 cells (Fig. 4C).These results suggest that RO-5963-induced dimerization ofMDM2 and MDMX is the likely cause for protection of MDMXfrom MDM2-mediated degradation. However, despite increasedMDMX levels in cells exposed to RO-5963, compared with nutlin,MDMX was prevented from binding to and inhibiting p53 in thepresence of the dual inhibitor.Next, we tested RO-5963 in a wider panel of 11 solid tumor

cell lines expressing wild-type p53 and representing diverse tu-mor types: breast (MCF7), prostate (LNCaP, 22Rv1), colon(HCT116, RKO), lung (H460, A549), kidney (A498), osteosar-coma (U2OS), melanoma (LOX). The dual inhibitor was able to

effectively activate p53 and elevate p21 and MDM2 levels, sug-gesting that it penetrates well cultured cells and can be used indiverse cellular context (Fig. 4D).

RO-5963 Effectively Activates Main Functions of the p53 Pathway inCancer Cells. One of the main functions of activated p53 is in-duction of cell cycle arrest. As previously demonstrated by thespecific MDM2 antagonist, nutlin-3 (20), RO-5963 potentlyarrested cell cycle progression in exponentially growing cancercells in G1 and G2 phase, effectively depleting the S phasecompartment (Fig. S5A). Induction of apoptosis is another majorp53 function that is frequently altered in cancers expressing wild-type p53 (20). Similar to MDM2, MDMX overexpression hasbeen shown to effectively disable this function by inhibiting p53transcriptional activity (6, 14). Therefore, MDMX could bea barrier to p53 apoptotic activity even in the presence of MDM2antagonists (21–23). The SJSA1 osteosarcoma line, whichexpresses very high levels of MDM2 protein, is presumed to befree of other defects in the p53 pathway and has shown a robustapoptotic response to the specific MDM2 antagonists (20).However, its engineered clone, SJSA-X, is nearly completelyresistant to nutlin because of the high levels of exogenouslyexpressed MDMX from a CMV promoter (23). Therefore, theSJSA-X clone offers an excellent mechanistic model for assess-ing the cellular activity of p53-MDMX inhibitors. The high levelsof both MDM2 and MDMX proteins in SJSA-X cells representa fairly high hurdle to dual inhibitors. When RO-5963 was testedfor apoptotic activity on the parental cell line, SJSA-V, it showedslightly lower but still strong Annexin V signal compared withnutlin (Fig. S5B). As expected, similar apoptotic activity wasmeasured also in the SJSA-X clone in which nutlin was practi-cally inactive. The apoptotic activity of RO-5963 was dose-de-pendent and was enhanced when both molecules were combined.Western blotting of SJSA-X cell lysates revealed that 20 μMRO-5963 induced p53 accumulation comparable to 10 μM nutlin-3abecause p53 stabilization is caused by disruption of the p53-MDM2 interaction (Fig. 5C). However, p53 transcriptional ac-tivity, indicated by p21 and MDM2 levels, was higher in thepresence of RO-5963, presumably because of blocking not onlyp53-MDM2 but also p53-MDMX binding and liberating the el-evated p53 protein from both inhibitors. Protein levels of p21and MDM2 were increased further by combining RO-5963 withnutlin, reflecting the increased transcriptional activity of p53(Fig. S5C) that explains enhanced apoptotic activity of thecombination (Fig. S5B). These results indicate that RO-5963 caneffectively inhibit both p53-MDM2 and p53-MDMX binding andcan induce apoptosis in a cancer cell model overexpressing highlevels of both negative p53 regulators.

RO-5963 Overcomes the Resistance of MDMX-Overexpressing CancerCells to MDM2 Antagonists. Cancer cells overexpressing MDMXhave substantially reduced apoptotic response to small-moleculeMDM2 antagonists (e.g., nutlins) (21–23). This resistance may bepartly attributable to the inability of nutlin to inhibit p53-MDMXbinding, leading to incomplete restoration of p53 activity. MCF7breast cancer cell line represents such a MDMX-dependentmodel. These cells have high levels of MDMX protein and arefairly insensitive to nutlin-induced apoptosis. Therefore, we firsttested the apoptotic activity of RO-5963 in MCF7 and two otherbreast cancer cell lines with low (ZR75-1) and intermediate(ZR75-30) MDMX levels (35). As an MDM2-only inhibitor (20,28), nutlin-3a had relatively low apoptotic activity in MCF7 cellsand practically no activity in ZR75-30 cells (Fig. 5A). ZR75-30cells express moderate levels of MDMX but very low levels ofMDM2, suggesting that MDMX plays an important role in p53regulation in these cells. RO-5963 (20 μM) showed much higherapoptotic activity than nutlin in both MCF7 and ZR75-30 celllines. These results suggest that RO-5963 is capable of releasingp53 from MDMX inhibition and restoring its activity. Althoughthis led to efficient apoptosis and cell death in the MCF7 line (Fig.5 A and B), ZR75-30 cells had an enhanced but incomplete

D

C

- + - + - + - + - - 5963

MC

F-7

H46

0

RKO

HC

T116

HC

T116

RKO

p53

p-p53Ser15

p21

MDM2

Actin

Dox

- 1.25 2.5 5 10 - 1.25 2.5 5 10 - 1.25 2.5 5 10 µM RO-5963

p53p21MDM2MDMXActin

- - - - - + + + + + + + + + + Nutlin

BA

G401 H460

% o

f con

trol

Concentration (µM)

0

20

40

60

80

100

0 1 10

MCF-7HCT-116RKOSW480MDA-MB435

WT-p53

Mut-p53

LnC

aP

U2O

S

RKO

A498

22R

V1

HC

T116

H46

0

LOX

MC

F7

A549

G40

1

RO-5963 - + - + - + - + - + - + - + - + - + - + - +

p53p21

MDM2Actin

Fig. 4. RO-5963 activates p53 signaling in diverse cellular context bya nongenotoxic mechanism. (A) Antitumor activity of RO-5963 depends onthe p53 status. Viability of three wild-type p53 (MCF7, HCT116, RKO) andtwo mutant p53 (SW480, MDA-MB-435) cancer cell lines was determined bythe CellTiter-Glo assay after 5 d of incubation with RO-5963 and expressed aspercentage of controls ± SD. (B) RO-5963 does not induce genotoxic re-sponse in cancer cells. Cells were incubated with 10 μM RO-5963 or 1 μMdoxorubicin for 24 h, and the levels of total and Ser15-phosphorylated p53were determined by Western blotting. (C) Binding of RO-5963 preventsMDM2-mediated degradation of MDMX. G401 and H460 cells were in-cubated with the indicated concentrations of RO-5963 with or without10 μM nutlin-3a, and relative levels of p53, p21, MDM2, and MDMX weredetermined by Western blotting. Blots representative of three independentexperiments are shown. (D) RO-5963 activates p53 signaling in multiplecancer cell lines with wild-type p53. Exponentially growing cancer cell wereincubated with 10 μM RO-5963 for 24 h, and the relative levels of p53, p21,and MDM2 were determined by Western blotting.

Graves et al. PNAS | July 17, 2012 | vol. 109 | no. 29 | 11791

MED

ICALSC

IENCE

S

Dow

nloa

ded

by g

uest

on

Apr

il 23

, 202

0

apoptotic response. This is likely attributable to other abnormal-ities in cancer signaling that can attenuate p53 apoptotic pathwaysas seen in many epithelial cancer cell lines with normal MDM2and relatively low MDMX levels (20, 28). Interestingly, the dualantagonist was less active than nutlin in the ZR75-1 cell lineexpressing barely detectable MDMX levels. This may result fromthe fact that nutlin is slightly more effective than RO-5963 atinhibiting p53-MDM2 binding in cells.Next, we tested the apoptotic activity of RO-5963 and nutlin

against a randomly selected panel of nine cancer cell linesexpressing wild-type p53 and two with mutant p53. As expected,no apoptotic activity was detected in the mutant p53 lines, andvariable levels of apoptosis were measured in the wild-type lines(Fig. 5D). RO-5963 showed better apoptotic activity than nutlinin 4/9 (H460, RKO, LS174T, AGS) and similar or slightly lowerin five of nine wild-type p53 cell lines. Three out of the four lineswith enhanced apoptosis had relatively high levels of MDMX.Overall, MDMX protein levels correlated with the enhancedapoptotic response to the dual MDM2/MDMX antagonists (Fig.5 C and D). These data suggest that inhibition of p53-MDM2

interaction is critical for stabilization of p53 and that simulta-neous inhibition of p53-MDMX binding can enhance p53 apo-ptotic activity in cancer cells with MDMX overexpression.However, inhibition of both MDM2 and MDMX does not sub-stantially enhance p53-dependent apoptotic response in cancercells with normal or low levels of MDMX that is likely affectedby other factors than MDMX.

DiscussionThe role of MDMX in the fine regulation of p53 is still emerging,but it is an established fact that MDMX overexpression canblock p53 function and render cancer cells resistant to MDM2antagonists (6, 15). Even normal levels of MDMX could partiallysilence activated p53 because all known inhibitors of the p53-MDM2 interaction are unable to liberate p53 from the remain-ing MDMX, suggesting that dual MDM2/MDMX inhibitors maysubstantially improve the outcome of this p53 activation strategy.However, recent efforts to develop inhibitors of p53 binding withboth MDM2 and MDMX have been hindered by the structuraldifferences in the p53 pockets of the proteins (26). Using high-throughput screening and a diverse small-molecule library, weidentified a class of indolyl-hydantoin compounds that areroughly equipotent. The series demonstrated clear SAR and thebetter analogs showed remarkable potency given their small sizeand the ability to occupy only two of the three surface subpocketsin the p53 binding region. The presence of the 6-chloro-indolegroup provided the expectation that the binding mode could bepredicted because this same group is used in some of the knownMDM2 antagonists (36).The predicted binding mode of the dual MDM2/MDMX

antagonists placed the chloro-indole moiety in the Trp pocketwith the remainder of the compound extending into the Phepocket. This was, in general, consistent with the available in-formation. First, the fact that the compounds were equipotentagainst MDM2 and MDMX made it less likely that they boundin the Leu pocket, where the two protein structures diverge themost. Secondly, the large up-field shift in the NMR spectrum forY63 meant that this residue was likely interacting with an aro-matic group from the inhibitor, the di-fluoro-phenyl group, inthis model. However, the actual structure turned these pre-dictions completely upside down (Fig. 2A). The indole andhydantoin moieties remain essentially coplanar and occupy anextended Phe pocket. In this orientation, it is the indolyl-hydantoin that is providing the shielding for the Y63 residue. Italso forms a stacking interaction with the phenolic ring of Y63(Fig. S1A). Most surprisingly, the di-fluoro-phenyl group of onemolecule bound to one protein monomer reaches over to bindinto the Trp pocket of the other protein monomer (Fig. 2B).Although unusual, the structure nicely explained the preferencefor fluorine in the paraposition for binding to MDMX. It alsoprovided an explanation for the preference for chlorine at the 6-position of the indole.Although MDM2 and MDMX naturally form homodimers

and heterodimers, these are driven by the C-terminal RINGdomains (6, 37). Thus, the dimerization of the N-terminal p53-binding domains in this fashion has not been observed pre-viously. It represents a significant advantage in that small mol-ecules are able to achieve potent binding and inhibitory effectsand to do it in a way that avoids addressing the Leu subpocket,where the two protein structures are the most different. Asa result, these compounds are equipotent against MDM2 andMDMX. Nevertheless, it is possible to modify these compoundsto enable them to address the Leu pocket and RO-5963 (Fig. 3A)represents one such approach.RO-5963 penetrated cancer cells and inhibited the binding of

p53 to MDM2 and MDMX. Although nearly equipotent in vitro,RO-5963 was more effective in disrupting the p53-MDMX in-teraction than p53-MDM2 interaction in cultured cancer cells.One possible reason for this is the increasing cellular levels ofMDM2 in the presence of MDM2 inhibitors. Whereas MDMXlevels are generally stable, MDM2 protein levels rise as a result

0

10

20

30

40

50

60

70

80

ZR- 75- 1 ZR- 75- 30 MCF- 7

Series1Series2Series3Series4

ZR75-1 ZR75-30 MCF7

DMSONutlin (10 μM)RO-5963 (10 μM)RO-5963 (20 μM)

Ann

exin

V+

cells

(%)

0

1020

30

40

5060

70

80

A54

9H46

0

HCT116

RKO

LS17

4T A

GS G

401U2O

S

Skmel-

5

MDA-MB-43

5

SW480

Series1Series2Series3Series4

DMSONutlin (10 μM)RO-5963 (10 μM)RO-5963 (20 μM)

Ann

exin

V+

cells

(%)

A

MDM2

MDMX

Actin

C

D

RO-5963

Control

B

Fig. 5. Apoptotic activity of RO-5963 and MDMX status. (A) Breast cancercells respond to RO-5963 depending on the levels of MDM2 and MDMX.Three cell lines with variable ratios of MDMX/MDM2 proteins were exposedto nutlin-3a (10 μM) and RO-5963 (10 or 20 μM) for 48 h, and the percentageof apoptotic cells (± SD) was determined by the Annexin V assay. (B) Cyto-toxicity of RO-5963 on MCF7 cells. Phase-contrast images were taken 48 hafter addition of 20 μM RO5963. (C) Relative protein levels of MDM2 andMDMX in a panel of solid tumor cell lines. Cells lines at subconfluent stage ofgrowth were analyzed for protein levels by Western blotting. (D) Apoptoticresponse to RO-5963 and nutlin in a panel of cancer cell lines with wild-typeand mutant p53. Log-phase cells were incubated with nutlin-3a and RO-5963for 48 h and analyzed for apoptosis as in A. Two mutant p53 cell lines, MDA-MB-435 and SW480, were included as controls.

11792 | www.pnas.org/cgi/doi/10.1073/pnas.1203789109 Graves et al.

Dow

nloa

ded

by g

uest

on

Apr

il 23

, 202

0

of inducing its expression by the accumulating p53. IncreasingMDM2 protein may change the balance between homo- andheterodimers induced by RO-5963 and make MDM2 inhibitoryactivity a limiting factor in the course of exposure to the dualinhibitor. This is a likely reason why addition of nutlin furtherenhanced the apoptotic outcome of RO-5963 treatment espe-cially in the MDM2 amplified SJSA-X line (Fig. S5B).Forced or natural overexpression of MDMX can significantly

impede the activity of MDM2 antagonists (19, 21–23). Theseagents have been optimized for binding to the p53 pocket ofMDM2 and are practically inactive against the p53-MDMX in-teraction (18, 19, 28). Unobstructed binding of MDMX to p53could silence its transcriptional activity and may alter otherproperties, leading to inadequate activation of p53 functions.Therefore, the effectiveness of MDM2 antagonists in tumorswith high levels of MDMX will be diminished. Our experimentalresults with the dual MDM2/MDMX inhibitor are in fullagreement with this prediction. RO-5963 restored p53 tran-scriptional activity and overcame the apoptotic resistance ofMDMX-overexpressing cell line, SJSA-X, to nutlin-3 (Fig. 5C).The same is true in the “naturally” MDMX-overexpressingbreast cancer cell lines MCF7 and ZR75-30 (Fig. 5). Both celllines exhibit higher MDMX to MDM2 ratios, suggesting thatMDMX may play a role in their resistance to nutlin-inducedapoptosis. The relatively incomplete apoptotic response inZR75-30 line is likely attributable to factors other than MDMXattenuating p53-dependent apoptosis signaling. In MCF7 cells,used routinely as a cellular model of MDMX-overexpressingbreast cancer, RO-5963 showed massive cell death with nearly allcells losing their viability within 72 h of treatment (Fig. 5B).These results are in agreement with previously published data

using RNA interference to address the role of MDMX in theresponse to nutlin (23, 28, 35).Nearly one-fifth of breast, colon, and lung cancers overexpress

MDMX because of gene amplification (38), and that could bethe only abnormality in their p53 signaling as proposed forMDM2-amplified tumors (20). Our data suggest that patientswith this signature may greatly benefit from treatment with dualMDM2/MDMX antagonists. The identification of the indolylhydantoin class of dual inhibitors is the first step toward de-velopment of MDM2/MDMX-targeted cancer therapy. Furtheroptimization of the potency and pharmacological properties ofthis chemical class should allow the extension of our observationsto in vivo cancer models and eventually the clinic.

Materials and MethodsMaterials and detailed methods are available in SI Materials and Methods.

Cell proliferation/viability was evaluated by the methyl-thiazolyl-tetra-zolium (MTT) and CellTiter-Glo (Promega) assays. Cell cycle analysis, AnnexinV assays, and Western blotting were performed as described previously (20).MDM2-p53 and MDMX-p53 binding was assessed by time-resolved (TR)-FRETbinding and fluorescence quenching assays. NMR spectra were collectedusing 15N-labeled humanized-zebrafish (hz)MDMX. For cocrystallization withRO-2443, a 1.3:1 stoichiometric excess of compound was added to HDMXprotein. The protein construct (14-111,C17S) was as described in ref. 33. Crys-tallization conditions, data collection, molecular replacement, and refinementwere performed as detailed in SI Materials and Methods. Statistics for therefined model are in Table S2.

ACKNOWLEDGMENTS. We thank Ann Petersen and Honju Li for chemicalsynthesis; Charles Belunis for protein purification; Sonal Sojitra, John Koss,and Stephen Wasserman for X-ray data collection; Geoffrey Wahl for the giftof SJSA-X cells; and Nader Fotouhi for his suggestions and critical reading ofthe manuscript.

1. Harris SL, Levine AJ (2005) The p53 pathway: Positive and negative feedback loops.Oncogene 24:2899–2908.

2. Wade M, Wang YV, Wahl GM (2010) The p53 orchestra: Mdm2 and Mdmx set thetone. Trends Cell Biol 20:299–309.

3. Toledo F, Wahl GM (2006) Regulating the p53 pathway: In vitro hypotheses, in vivoveritas. Nat Rev Cancer 6:909–923.

4. Marine JC, Lozano G (2010) Mdm2-mediated ubiquitylation: p53 and beyond. CellDeath Differ 17:93–102.

5. Lenos K, Jochemsen AG (2011) Functions of MDMX in the modulation of the p53-response. J Biomed Biotechnol 2011:876173.

6. Marine JC, Jochemsen AG (2005) Mdmx as an essential regulator of p53 activity. Bi-ochem Biophys Res Commun 331:750–760.

7. de Graaf P, et al. (2003) Hdmx protein stability is regulated by the ubiquitin ligaseactivity of Mdm2. J Biol Chem 278:38315–38324.

8. Kawai H, Lopez-Pajares V, Kim MM, Wiederschain D, Yuan ZM (2007) RING domain-mediated interaction is a requirement for MDM2’s E3 ligase activity. Cancer Res 67:6026–6030.

9. Wang X, Wang J, Jiang X (2011) MdmX protein is essential for Mdm2 protein-medi-ated p53 polyubiquitination. J Biol Chem 286:23725–23734.

10. Huang L, et al. (2011) The p53 inhibitors MDM2/MDMX complex is required forcontrol of p53 activity in vivo. Proc Natl Acad Sci USA 108:12001–12006.

11. Pan Y, Chen J (2003) MDM2 promotes ubiquitination and degradation of MDMX.MolCell Biol 23:5113–5121.

12. Kawai H, et al. (2003) DNA damage-induced MDMX degradation is mediated byMDM2. J Biol Chem 278:45946–45953.

13. Momand J, Zambetti GP, Olson DC, George D, Levine AJ (1992) The mdm-2 oncogeneproduct forms a complex with the p53 protein and inhibits p53-mediated trans-activation. Cell 69:1237–1245.

14. Marine JC, et al. (2006) Keeping p53 in check: Essential and synergistic functions ofMdm2 and Mdm4. Cell Death Differ 13:927–934.

15. Wade M, Wahl GM (2009) Targeting Mdm2 and Mdmx in cancer therapy: Better livingthrough medicinal chemistry? Mol Cancer Res 7:1–11.

16. Vassilev LT (2007) MDM2 inhibitors for cancer therapy. Trends Mol Med 13:23–31.17. Brown CJ, Lain S, Verma CS, Fersht AR, Lane DP (2009) Awakening guardian angels:

Drugging the p53 pathway. Nat Rev Cancer 9:862–873.18. Vassilev LT, et al. (2004) In vivo activation of the p53 pathway by small-molecule

antagonists of MDM2. Science 303:844–848.19. Shangary S, et al. (2008) Temporal activation of p53 by a specific MDM2 inhibitor is

selectively toxic to tumors and leads to complete tumor growth inhibition. Proc NatlAcad Sci USA 105:3933–3938.

20. Tovar C, et al. (2006) Small-molecule MDM2 antagonists reveal aberrant p53 signalingin cancer: Implications for therapy. Proc Natl Acad Sci USA 103:1888–1893.

21. Koblish HK, et al. (2006) Benzodiazepinedione inhibitors of the Hdm2:p53 complex

suppress human tumor cell proliferation in vitro and sensitize tumors to doxorubicin

in vivo. Mol Cancer Ther 5:160–169.22. Patton JT, et al. (2006) Levels of HdmX expression dictate the sensitivity of normal and

transformed cells to Nutlin-3. Cancer Res 66:3169–3176.23. Wade M, Wong ET, Tang M, Stommel JM, Wahl GM (2006) Hdmx modulates the

outcome of p53 activation in human tumor cells. J Biol Chem 281:33036–33044.24. Popowicz GM, et al. (2007) Molecular basis for the inhibition of p53 by Mdmx. Cell

Cycle 6:2386–2392.25. Popowicz GM, Czarna A, Holak TA (2008) Structure of the human Mdmx protein

bound to the p53 tumor suppressor transactivation domain. Cell Cycle 7:2441–2443.26. Riedinger C, McDonnell JM (2009) Inhibitors of MDM2 and MDMX: A structural

perspective. Future Med Chem 1:1075–1094.27. Vassilev LT (2005) p53 Activation by small molecules: Application in oncology. J Med

Chem 48:4491–4499.28. Xia M, et al. (2008) Elevated MDM2 boosts the apoptotic activity of p53-MDM2

binding inhibitors by facilitating MDMX degradation. Cell Cycle 7:1604–1612.29. Reed D, et al. (2010) Identification and characterization of the first small molecule

inhibitor of MDMX. J Biol Chem 285:10786–10796.30. Pazgier M, et al. (2009) Structural basis for high-affinity peptide inhibition of p53

interactions with MDM2 and MDMX. Proc Natl Acad Sci USA 106:4665–4670.31. Bernal F, et al. (2010) A stapled p53 helix overcomes HDMX-mediated suppression of

p53. Cancer Cell 18:411–422.32. Kussie PH, et al. (1996) Structure of the MDM2 oncoprotein bound to the p53 tumor

suppressor transactivation domain. Science 274:948–953.33. Kallen J, et al. (2009) Crystal Structures of HumanMdmX (HdmX) in Complex with p53

Peptide Analogues Reveal Surprising Conformational Changes. J Biol Chem 284:

8812–8821.34. Bissantz C, Kuhn B, Stahl M (2010) A medicinal chemist’s guide to molecular inter-

actions. J Med Chem 53:5061–5084.35. Lam S, et al. (2010) Role of Mdm4 in drug sensitivity of breast cancer cells. Oncogene

29:2415–2426.36. Yu S, et al. (2009) Potent and orally active small-molecule inhibitors of the MDM2-p53

interaction. J Med Chem 52:7970–7973.37. Linke K, et al. (2008) Structure of the MDM2/MDMX RING domain heterodimer re-

veals dimerization is required for their ubiquitylation in trans. Cell Death Differ 15:

841–848.38. Danovi D, et al. (2004) Amplification of Mdmx (or Mdm4) directly contributes to tu-

mor formation by inhibiting p53 tumor suppressor activity. Mol Cell Biol 24:

5835–5843.

Graves et al. PNAS | July 17, 2012 | vol. 109 | no. 29 | 11793

MED

ICALSC

IENCE

S

Dow

nloa

ded

by g

uest

on

Apr

il 23

, 202

0