Bivalent binding of p14ARF to MDM2 RING and acidic domains inhibits E3 ligase function

This study shows that MDM2 acidic and zinc finger regions weakly interact with the RING domain and a fragment of p14ARF strengthens this interaction to inhibit MDM2 ubiquitin ligase activity.


Introduction
The activity of p53 tumor suppressor protein is tightly regulated by MDM2 ubiquitin ligase (E3). Under unstressed, homeostatic conditions, levels of p53 are low because of the E3 activity of MDM2, where MDM2 catalyzes p53 ubiquitination leading to its proteasomal degradation (Kubbutat et al, 1997;Pant & Lozano, 2014). In response to cellular stresses, numerous factors function to uncouple p53 from MDM2 to enable p53 stabilization and activation of its transcriptional activity to restore cellular homeostasis (Hu et al, 2012). In a variety of wild-type p53 cancers, including soft tissues sarcomas, osteosarcomas, and glioblastomas, amplification of the MDM2 gene has been frequently reported; this drastically reduces p53 protein levels, which in turn promotes cancer growth (Momand et al, 1998;Stefanou et al, 1998;Burton et al, 2002;Karni-Schmidt et al, 2016). In contrast, deficiency in MDM2 is lethal because of uncontrolled p53 activation as demonstrated by embryonic lethality in mice lacking the Mdm2 gene (Jones et al, 1995;Montes de Oca Luna et al, 1995). This fine balance between p53 activity and cell fate signifies the importance of precise regulation of MDM2.
MDM2 inhibits p53's transcriptional activity by directly binding p53 and promoting p53 ubiquitination and subsequent degradation by the proteasome (Momand et al, 1992;Oliner et al, 1993;Haupt et al, 1997;Honda et al, 1997;Kubbutat et al, 1997;Shi & Gu, 2012). The importance of MDM2 E3 activity in p53 regulation is underscored by mouse studies, where expression of catalytically inactive Mdm2 mutants results in embryonic lethality that is rescued by Trp53 deletion (Itahana et al, 2007;Humpton et al, 2021). The core of MDM2 E3 activity resides in its C-terminal RING domain. The RING domain functions by recruiting E2 thioesterified ubiquitin (E2~Ub;~indicates thioester bond) and priming the E2~Ub thioester bond for nucleophilic attack by the lysine side chain of substrate (Dou et al, 2012;Plechanovova et al, 2012). MDM2's RING domain activity is regulated by dimerization, where MDM2 RING domain homodimerization or heterodimerization with the MDMX RING domain is essential for binding E2~Ub in the active conformation to confer E3 activity (Nomura et al, 2017;Magnussen et al, 2020). Furthermore, phosphorylation of Ser429 after DNA damage directly contributes to stabilization of E2~Ub to enhance MDM2 E3 activity (Magnussen et al, 2020). In addition to the RING domain, MDM2 consists of three other structured domains connected through regions predicted to be unstructured (Fig 1A). This "beads-on-a-string" architecture enables protein-protein interactions and posttranslational modifications (PTMs) to modulate MDM2 E3 activity (Wade et al, 2013;Fåhraeus & Olivares-Illana, 2014). For example, binding of ribosomal proteins by the zinc finger (ZnF) region inhibits MDM2's activity toward p53 (Lohrum et al, 2003;Zhang et al, 2003;Jin et al, 2004), and phosphorylation of MDM2 by Chk2, ATM, and CKI promotes MDM2 ubiquitination and degradation (Chen et al, 2005;Pereg et al, 2005;Inuzuka et al, 2010).
The ARF tumor suppressor is one of the most extensively studied negative regulators of MDM2 (Sherr, 2006). Human p14ARF and its close homolog, mouse p19ARF, are reported to interact with MDM2's  (p53BD), a region containing nuclear localization sequence and nuclear export sequence (NLS/NES), an acidic and zinc finger (AD-ZnF) region, and a C-terminal RING domain. MDM2 constructs used in this study are indicated. (B) Schematic showing the domain structure of human p14ARF. Sequence alignment of the N-terminal 29 residues in p14ARF and p19ARF with conserved residues highlighted in red and ARF motifs (RxFxV) are indicated below. (C) SDS-PAGE showing pull-downs of co-expressed GST-MDM2 RING and His-GFP-MDM2 220-334 . Lane 1 contains GST-MDM2 RING expressed on its own and purified by glutathione affinity chromatography. Co-expression of GST-MDM2 RING and His-GFP-MDM2 220-334 was first p14ARF and MDM2 AD inhibit RING activity Kowalczyk et al.
https://doi.org/10.26508/lsa.202201472 vol 5 | no 12 | e202201472 2 of 13 acidic domain (AD), leading to stabilization and activation of p53 (Kamijo et al, 1998;Pomerantz et al, 1998;Stott et al, 1998;Zhang et al, 1998). The N-terminal 60 residues share the highest sequence similarity in p14ARF and p19ARF. Studies showed that the N-terminal 29 residues of p14ARF are sufficient for MDM2 binding and p53 activation ( Fig 1B) (Kamijo et al, 1998;Midgley et al, 2000;Lohrum et al, 2000a). This sequence is disordered in solution but forms two extended β-strands upon binding to MDM2 AD DiGiammarino et al, 2001). Importantly, the predicted placement of the β-strands corresponds to the double repetition of the so-called ARF motif, which is a conserved RxFxV sequence present in both homologs of ARF. The mechanism by which the formation of the ARF/MDM2 complex influences p53 has been actively studied over the past two decades. Collectively, studies showed that ARF can localize MDM2 within the nucleolus, thereby separating it from p53, prevent MDM2-mediated p53 nuclear export, and inhibit the E3 activity of MDM2 resulting in p53 stabilization (Honda & Yasuda, 1999;Tao & Levine, 1999;Weber et al, 1999;Zhang & Xiong, 1999;Midgley et al, 2000;Lohrum et al, 2000b;Llanos et al, 2001). In vitro characterization of ARF proteins remains challenging as the N-terminal regions encompassing the first 29 residues of ARF proteins are extremely hydrophobic and basic, whereas also reported to undergo oligomerization and precipitation in aqueous buffers Sivakolundu et al, 2008). Furthermore, protein aggregation prediction (http://old.protein.bio.unipd.it/pasta2/) suggests that the N-terminal 29 residues of ARF proteins have a propensity to form aggregates (Walsh et al, 2014). Thus, it remains difficult to characterize how ARF binding to MDM2 AD could influence MDM2 E3 activity.
A recent study showed that MDM2 could form intramolecular interactions between its AD and RING domains and that this interaction appeared to be enhanced by p14ARF in a pull-down experiment (Cheng et al, 2014). To better understand how p14ARF binding to MDM2 AD could modulate the activity of the RING domain, we focused on the 32 N-terminal residues of p14ARF (N32p14ARF) that is sufficient for MDM2 binding and performed nuclear magnetic resonance (NMR) and biochemical analyses to investigate the interactions between MDM2 AD, MDM2 RING domain, and N32p14ARF. We developed a stable fusion construct containing the N32p14ARF and MDM2 AD-ZnF domains to overcome the challenging behavior of isolated ARF protein. In this study, we show that MDM2 AD-ZnF forms weak interactions with the E2~Ub binding surface of the RING domain to reduce E2~Ub binding affinity and activity. N32p14ARF binding to MDM2 AD-ZnF greatly enhances the interaction with the RING domain and outcompetes E2~Ub to inhibit E3 activity. Moreover, we demonstrate that the N-terminal RxFxV motifs of N32p14ARF are essential for inhibition of MDM2 RING activity. These findings provide insights into how the ARF tumor suppressor protein inhibits the E3 activity of MDM2 and have implications for therapies directed at MDM2 and how the regulation of MDM2 and p53 contribute to oncogenesis.

MDM2 AD-ZnF contacts the E2~Ub binding surface of MDM2 RING domain
A prior study showed that direct fusion of AD (residues 230-260) to the RING domain (residues 410-491) stimulated the catalytic activity of the RING domain (Cheng et al, 2014). Because this fusion construct lacks the region encompassing residues 261-409, how this intramolecular interaction modulates MDM2 activity in the context of native MDM2 sequence remains unclear. To assess interactions between the AD and RING domains, we generated constructs encoding His-GFP-tagged MDM2 220-334 containing AD and ZnF region (GFP-AD-ZnF) and GST-tagged MDM2 419-C containing the RING domain (MDM2 RING ), co-expressed both constructs in Escherichia coli, and performed Ni-NTA affinity pull-down followed by glutathione-sepharose affinity pull-down. The double pull-down experiment revealed the formation of a complex between MDM2 AD-ZnF and RING domains ( Fig 1C). To confirm this interaction and to map the AD-ZnF-binding site on the RING domain, we purified 15 N-labeled MDM2 RING and acquired 1 H, 15 N-HSQC spectra while titrating MDM2 220-334 (AD-ZnF). Signals in the 15 N-MDM2 RING spectra were assigned based on a previous study (Kostic et al, 2006) ( Fig  S1A). Titration of 15 N-MDM2 RING with a molar excess of AD-ZnF produced moderate chemical-shift perturbations (CSPs) with several attenuations, suggesting a weak interaction between AD-ZnF and the RING domains (Fig 1D-F). The NMR data allowed us to map the AD-ZnF binding surface on the RING domain ( Fig 1G). CSPs for many residues near the RING/E2~Ub interface were above average, including Val 439 , Ile 440 , Gln 442 , Met 459 , and Ala 460 , which are involved in binding the hydrophobic region of UBE2D2, and His 457 , Val 477 , Cys 478 , and Gln 480 , which are involved in stabilizing the donor Ub (Nomura et al, 2017;Magnussen et al, 2020) (Fig 1G and H). In addition, several signals from the Lys 466 /Ile 485 region distal from the E2~Ub binding site were perturbed. To verify that E2~Ub induces CSPs in 15 N-MDM2 RING , we titrated 15 N-MDM2 RING with substoichiometric ratios of UBE2D2-Ub and showed that addition of UBE2D2-Ub led to attenuation of many cross-peaks including those of Ile 440 and His 457 along the E2~Ub binding surface, but not Lys 466 or Ile 485 (Fig S1B), consistent with the published structure of the MDM2 RING domain bound to UBE2D2-Ub (Magnussen et al, 2020). Together, this demonstrates that AD-ZnF binds to the RING domain and shares a binding surface with E2~Ub.

AD-ZnF inhibits E3 ligase activity
Given that AD-ZnF and E2~Ub contact the same surface on the RING domain, we set out to determine how AD-ZnF impacts the E3 activity of MDM2. For this purpose, we compared the E3 activity and E2~Ub binding affinity of MDM2 RING and MDM2 230-C (containing AD-ZnF and the RING domain). We purified GST-MDM2 RING and GST-MDM2 230-C for these analyses (Fig 2A). We performed single-turnover lysine discharge assay by monitoring the rate of UBE2D2~Ub discharge to free lysine in the presence of MDM2 variants. By using saturated lysine as the acceptor, UBE2D2~Ub preferentially transfers Ub to free lysine and eliminates contributions from autoubiquitination arising from differences in lysine residue accessibility in the two MDM2 constructs. We found that MDM2 230-C displayed reduced activity compared with MDM2 RING (Fig 2B and C). We also performed surface plasmon resonance (SPR) analysis to measure the UBE2D2-Ub binding affinity. MDM2 RING exhibited higher affinity for UBE2D2-Ub (K d = 14.5 ± 1.4 μM) compared with that of MDM2 230-C (K d = 57.5 ± 1.6 μM) ( Fig 2D). These results support the hypothesis that AD-ZnF restricts E2~Ub binding and thereby reduces the E3 activity of MDM2.
To investigate the roles of AD and ZnF in MDM2 RING binding, we generated GFP-MDM2 220-288 (GFP-AD) and GFP-MDM2 289-334 (GFP-ZnF), where GFP was used to improve the stability of AD and ZnF. We titrated 15 N-MDM2 RING with molar excess of GFP-AD-ZnF, GFP-AD, or GFP-ZnF. At 1:2 molar ratio of 15 N-MDM2 RING :GFP-AD-ZnF, several residues exhibited moderate CSPs with several attenuations ( Fig  S2A), similar to the titration of 15 N-MDM2 RING with molar excess of AD-ZnF lacking GFP ( Fig 1D). In contrast, at the same molar ratio, GFP-AD produced minimal CSPs, whereas GFP-ZnF had no effect ( Fig  S2B and C). Next, we assessed the effects of GFP-AD-ZnF, GFP-AD, and GFP-ZnF on MDM2 RING -mediated UBE2D2~Ub discharge. Addition of GFP-AD-ZnF in trans reduced MDM2 RING activity compared with GFP alone, whereas GFP-AD and GFP-ZnF had no effect (Figs 2E and F and S3). Collectively, these results showed that the combination of AD and ZnF regions are required for binding to the MDM2 RING domain to exert the inhibitory effect.

The N-terminus of p14ARF interacts with MDM2 AD
The N-terminal region of p14ARF has been shown to bind to the MDM2 AD and reduce MDM2 in vitro autoubiquitination (Midgley et al, 2000). Given that AD-ZnF interacts with the RING domain, we next assessed how p14ARF binding to the AD could modulate the RING domain activity. We verified the interaction between the N-terminal 32 residues of p14ARF (N32p14ARF) and MDM2 by co-expressing His-GFP-tagged N32p14ARF and GST-tagged MDM2 230-C in E. coli and then performing Ni-NTA affinity pull-down followed by glutathionesepharose affinity pull-down. A complex of His-GFP-N32p14ARF and GST-MDM2 230-C was detectable after double-affinity pull-down ( Fig  3A). We then compared the E3 activity of the His-GFP-N32p14ARF/ GST-MDM2 230-C complex and GST-MDM2 230-C using a lysine discharge assay and found that N32p14ARF had an inhibitory effect on the E3 activity (Fig 3B and C). To assess how p14ARF interacts with MDM2, it would be ideal to work with N32p14ARF; however, this construct is not expressed in the soluble form in E. coli, and the synthetic peptide does not dissolve in aqueous buffers. Our double-pull-down result suggests that when N32p14ARF is coexpressed with its binding partner, there is a stabilizing effect. With this in mind, we designed a multidomain fusion protein with GFPtagged N32p14ARF fused to MDM2 220-334 at the C-terminus (Fig 3D). The resulting GFP-N32p14ARF-[GGSG] 6 -MDM2 220-334 fusion protein (hereafter referred to as GFP-N32p14ARF-AD-ZnF) was eluted from the gel filtration column at a volume consistent with a monomer (Fig 3E). Furthermore, the [GGSG] 6 linker was flanked by two thrombin cleavage sites, allowing GFP-N32p14ARF to be freed from AD-ZnF to verify that the two proteins were still bound when no longer fused (Fig 3D-F). With GFP-AD-ZnF and the GFP-N32p14ARF-AD-ZnF fusion, we proceeded to compare how p14ARF-binding to the AD of MDM2 modulates the AD-ZnF inhibition of the RING domain.

MDM2 AD-ZnF/RING interaction is strengthened by p14ARF
Having shown that AD-ZnF binds MDM2 RING domain and N32p14ARF, we set out to address how E3 activity was affected in the presence of all three components. We titrated the GFP-N32p14ARF-AD-ZnF construct into 15 N-MDM2 RING and monitored CSPs by 1 H, 15 N-HSQC. At the sub-stoichiometric 1:0.15 molar ratio of 15 N-MDM2 RING :GFP-N32p14ARF-AD-ZnF, several residues had detectable CSPs, whereas some residues were attenuated (Figs 4A and S4A). There were widespread signal attenuations upon further titration to a stoichiometric molar ratio, indicative of the formation of a larger protein complex leading to enhanced relaxation and line broadening. We mapped the residues with CSPs and attenuated signals obtained at a 1:0.15 molar ratio of 15 N-MDM2 RING :GFP-N32p14ARF-AD-ZnF onto the structure of the MDM2 RING domain homodimer (Fig 4B). Similar to AD-ZnF, the GFP-N32p14ARF-AD-ZnF fusion affected the E2~Ub binding surface of MDM2 RING domain including aboveaverage CSPs present for Ile 440 , His 457 , Lys 473 , Cys 478 , and Gln 480 ( Fig  4A-C). In addition, Asn 433 and Glu 436 from the 3 10 -helical turn preceding the RING domain involved in donor Ub binding (Magnussen et al, 2020) displayed above-average CSPs. As with AD-ZnF, few CSPs were observed in the Phe 462 /Ile 485 patch distal from the E2~Ubbinding site (Fig 4B). Overlay of 15 N-MDM2 RING spectra in the presence of GFP-N32p14ARF-AD-ZnF or GFP-AD-ZnF demonstrates incorporation of N32p14ARF greatly enhances attenuation of the signals indicative of enhanced RING/AD-ZnF interaction ( Fig 4D). As the NMR titration experiments were performed with GFP fusions to stabilize N32p14ARF, we verified that GFP alone did not affect signals of 15 N-MDM2 RING in 1 H, 15 N-HSQC spectra (Fig S4B). To assess whether MDM2 RING was still active in this ternary complex and with N32p14ARF, we carried out lysine discharge assays. In this assay, discharge of UBE2D2~Ub by MDM2 RING was monitored across increasing concentrations of GFP-AD-ZnF or GFP-N32p14ARF-AD-ZnF. Addition of GFP-AD-ZnF reduced MDM2 RING activity compared with GFP alone. In contrast, addition of GFP-N32p14ARF-AD-ZnF further hindered MDM2 RING activity compared with GFP-AD-ZnF (Figs 4E and S5).
p14ARF directly binds MDM2 RING to inhibit E3 activity Thus far, we have shown that N32p14ARF strengthens the MDM2 AD-ZnF-driven inhibition of the RING domain. However, whether this results from direct contact between p14ARF and the RING domain or whether p14ARF-bound AD indirectly modulates the AD-ZnF/RING interaction remains ambiguous. To investigate whether p14ARF interacts directly with MDM2 RING domain, we co-expressed His-GFP-tagged N32p14ARF and GST-tagged MDM2 350-C in E. coli and performed Ni-NTA affinity pull-down followed by glutathionesepharose affinity pull-down. GST-MDM2 230-C was pulled down together with His-GFP-N32p14ARF after double-affinity pull-down (Figs 3A and 5A). In contrast, His-GFP-N32p14ARF pulled down GST-MDM2 350-C on the first Ni-NTA affinity pull-down, but in the second glutathione-sepharose affinity pull-down, only GST-MDM2 350-C was present (Fig 5A). We noticed that the His-GFP-N32p14ARF pull-down product was considerably less abundant after the first Ni-NTA affinity pull-down when co-expressed with GST-MDM2 350-C , suggesting that MDM2 lacking the AD might not be able to make a stable complex with N32p14ARF. Nonetheless, the results suggested that His-GFP-N32p14ARF exhibits weak affinity for MDM2 350-C . Although the synthetic N32p14ARF peptide is soluble in DMSO, addition of N32p14ARF peptide to MDM2 RING in the aqueous buffer caused precipitation. Thus, to further validate whether p14ARF directly contacts MDM2 RING , we generated a fusion construct, N32p14ARF-MDM2 350-C , in which N32p14ARF was linked to a variant of the MDM2 RING domain that included~80 residues of the native sequence before the RING domain to allow for flexibility and solubility of N32p14ARF (Fig 5B). Discharge of UBE2D2~Ub was reduced with the N32p14ARF-MDM2 350-C fusion, suggesting that N32p14ARF has an inhibitory effect on the MDM2 RING domain (Fig 5C). In addition, binding of UBE2D2-Ub to the N32p14ARF-MDM2 350-C fusion was abolished ( Fig 5D).
Next, we focused on the role of the "ARF motifs" in N32p14ARF in the inhibition of the MDM2 RING domain. Both p14ARF and mouse p19ARF contain two conserved RxFxV motifs within the first and second N-terminal β-strands (Fig 1B). We mutated six key residues in the RxFxV motifs to alanine (R3A,F5A,V7A,R21A,F23A,V25A; 6Ala; Fig  5B) in the N32p14ARF-MDM2 350-C fusion and showed that this N32p14ARF 6Ala -MDM2 350-C fusion had enhanced activity in the UBE2D2~Ub discharge assay and improved UBE2D2-Ub binding affinity compared with the N32p14ARF-MDM2 350-C fusion (Fig 5C and  D). The N32p14ARF 6Ala -MDM2 350-C fusion was slightly less active and bound UBE2D2-Ub marginally weaker compared with that of MDM2 350-C , suggesting that other residues in N32p14ARF likely contribute to binding of the RING domain. These data demonstrate that N32p14ARF directly binds and inhibits the E3 activity of the MDM2 RING domain.

Discussion
Uncovering how MDM2's E3 activity is modulated is critical for understanding how MDM2 regulates p53 levels. The combined NMR and biochemical assays presented in this study demonstrate that the RING domain of MDM2 forms an intramolecular interaction with the AD-ZnF region and recognizes E2~Ub through a conserved surface. Our data support the hypothesis that the AD-ZnF region of MDM2 directly binds the RING domain and restricts the access of E2~Ub, leading to a reduction in E3 activity. NMR and activity assays revealed that the observed AD-ZnF/RING interaction is weaker than the RING/E2~Ub interaction. This suggests that under normal conditions, MDM2 RING domain could still access E2~Ub to exert its E3 activity despite having a slightly reduced E2~Ub binding affinity. Cheng et al (2014) previously reported that the AD could stimulate RING domain activity (Cheng et al, 2014). These differences in the observed effects on RING domain activity may arise from directly fusing the AD to the RING domain in their study, whereas our construct contains the native sequence (MDM2 230-C ) that encompasses the AD-ZnF region connected to the RING domain by a flexible linker. Indeed, our data showed that the combination of AD and ZnF regions are required for optimal AD-ZnF/RING interactions to occlude E2~Ub binding.
Stress signals, such as DNA damage, have been shown to affect MDM2 PTMs, as well as its interactions with binding partners. Therefore, either PTMs or binding partners could potentially modulate the intramolecular interaction between MDM2 AD-ZnF and the RING domain, thereby increasing or abolishing inhibition of MDM2 E3 activity. p14ARF has been shown to bind MDM2 AD, but how it inhibits MDM2 E3 activity remains elusive. We showed that N32p14ARF binding to the AD of MDM2 hindered the activity of the C-terminal RING domain. Furthermore, the N32p14ARF/MDM2 AD-ZnF fusion complex greatly reduced MDM2 E3 activity by blocking the E2~Ub binding surface of the RING domain. The extent of CSPs and signal attenuations in 15 N-MDM2 RING spectra and additional CSPs in the 3 10 -helical region in the presence of GFP-N32p14ARF-AD-ZnF compared with GFP-AD-ZnF hinted that N32p14ARF not only strengthens the AD-ZnF and RING domain interaction but also directly binds to the RING domain. Indeed, when N32p14ARF was fused to the N-terminus of MDM2 RING domain, the conserved ARF motifs were essential for inhibiting MDM2 E3 activity. Our data suggest a model whereby the weak AD-ZnF interaction with the RING domain perturbs E2~Ub binding and allows MDM2 to be "primed" for inhibition (Fig 6). The binding of N32p14ARF to MDM2 AD enhances MDM2 AD-ZnF interaction with the RING domain and enables N32p14ARF to directly engage the RING domain. MDM2 AD-ZnF and the RING domains are separated a long unstructured linker. Therefore, it is likely that N32p14ARF binding would induce a global conformational change in the AD-ZnF and RING domains, restraining the MDM2 RING domain in an autoinhibited conformation and thereby restricting MDM2-mediated ubiquitination.
As documented in over three decades of studies, numerous effectors have been reported to bind to MDM2 (Fåhraeus & Olivares-Illana, 2014). For example, the AD-ZnF region is required for MDM2 interaction with ribosomal proteins, which in turn leads to activation of p53 (Liu et al, 2016). It remains unknown how other effectors of MDM2 exert their inhibitory effect, but it is plausible that they could take advantage of MDM2's "primed" state and inhibit MDM2 through a similar mechanism as ARF. In contrast, when MDM2 binds to p53, MDM2 AD is reported to contact p53 to facilitate MDM2mediated p53 ubiquitination (Kawai et al, 2003;Meulmeester et al, 2003;Yu et al, 2006). How MDM2 AD facilitates p53 ubiquitination is not clear. It seems likely that MDM2 AD could bind to and orient p53 for optimal ubiquitination by E2~Ub bound to the MDM2 RING domain. At the same time, MDM2 AD binding to p53 would perturb MDM2 AD-ZnF/RING interaction and relieve the inhibition of MDM2 E3 activity to allow MDM2-mediated p53 ubiquitination. p53 has been shown to form a ternary complex with MDM2 and ARF (Kamijo et al, 1998;Stott et al, 1998;Zhang et al, 1998). Although a detailed binding mechanism of the ternary complex is unknown, it seems plausible that ARF would bind the well-established MDM2 AD and thereby uncouples MDM2 AD-p53 contact to perturb MDM2mediated p53 ubiquitination. Furthermore, based on our findings, we posit that the ARF could engage in bivalent interaction with MDM2 AD and RING domains to restrain MDM2 E3 activity in the ternary complex, resulting in p53 stabilization. Future structural characterization of MDM2/p53/ARF complexes will be required to unveil the detailed binding mechanism.
p14ARF and p19ARF proteins remain challenging to study in vitro because of the amyloid-like feature of their N-terminal sequences . The methodology employed to generate stable and monomeric fusions of N32p14ARF and MDM2-AD-ZnF could be useful for future studies in characterizing this complex alone and in the complex with the RING domain at atomic resolution. Structural insight into the binding mode could pave the way for the development of MDM2 RING domain inhibitors that will have paramount clinical importance. Most of the therapeutic strategies focus on the MDM2/p53 interaction by targeting MDM2's N-terminal p53 binding domain with nutlin derivatives or synthetic peptides (Skalniak et al, 2019). Our related work has shown that targeting the RING domain might be a suitable alternative to reactivate p53 (Nomura et al, 2017;Humpton et al, 2021). Therefore, this study will guide future work on MDM2 RING and expands on a promising mechanism to inhibit the E3 activity of MDM2 through the RING domain.

Generation of constructs
All DNA constructs were generated using standard PCR techniques with the Q5 High-Fidelity kit (NEB) and verified by automated sequencing. GST-tagged constructs were cloned into a modified pGEX-4T-1 vector (Cytiva), which provides an N-terminal GST sequence followed by a TEV cleavage site. His-tagged constructs were cloned into modified pRSFDuet-1 (Merck Millipore), which provides an N-terminal 6xHis or 12xHis tag, followed by TEV cleavage sequence. MDM2 419-C (MDM2 RING ) and MDM2 230-C used in the activity assays and MDM2 220-334 used in the NMR studies were expressed with a GST tag. 15 N-labeled MDM2 RING was expressed with a 6xHis tag using a stable variant of MDM2 RING harboring G443T . All constructs containing GFP were expressed with a 12xHis tag. The N32p14ARF 6Ala -MDM2 350-C construct was generated by site-directed mutagenesis using PfuUltra High-Fidelity DNA Polymerase (Agilent) and verified by DNA sequencing. The GFP-N32p14ARF-AD-ZnF construct was generated by linking N32p14ARF sequence to GFP via a short GGS linker and further fusing MDM2 220-334 to the C-terminus of GFP-N32p14ARF via a GLVPRGSGGGSGGGSGGGSG GGSGGGSGGGSGLVPRGS sequence, where LVPRGS is a thrombin recognition and cleavage site.

Recombinant proteins expression
All recombinant proteins were expressed in E. coli BL21-DE3 (Thermo Fisher Scientific). Cells were grown at 37°C in Luria Bertani medium, until reaching an OD 600 of 0.6-0.7 and then induced with 0.2 mM isopropyl β-D-1-thiogalactopyranoside (IPTG) at 20°C overnight. 15 N-MDM2 RING was obtained from M9 medium based on a previously published protocol (Studier, 2005). In summary, 300 ml of Luria Bertani medium was inoculated from a single colony and grown overnight at 37°C. Cells were pelleted, washed in M9 medium, and resuspended to a final volume of 60 ml. Cells were then transferred to 6 liters of M9 medium containing 15 NH 4 Cl (1 g/l) and grown at 37°C until reaching an OD 600 of 0.6-0.7 and induced with 0.2 mM IPTG at 16°C for 20 h.

SPR assays
All experiments were done at 25°C using a Biacore T200, equilibrated in running buffer containing 25 mM Tris-HCl (pH 7.6), 300 mM NaCl, 0.5 mM TCEP, and 0.005% (vol/vol) Tween-20. CM-5 chips with anti-GST-nanobodies (GST VHH; Chromotech) were used to couple GST-MDM2 variants to a level of~500 response units, with GST as a control. UBE2D2 S22R C85K -Ub (UBE2D2-Ub), used as an analyte, was purified as described previously (Buetow et al, 2015) and serially diluted in running buffer. Binding between GST-MDM2 variants and UBE2D2-Ub was measured in duplicate, across seven concentrations of UBE2D2-Ub, starting at 100 μM and decreasing in a twofold manner. Data shown are the difference between SPR signal recorded for GST-MDM2 variants and GST alone. The data were analyzed by steady-state affinity using Biacore T200 evaluation software (GE Healthcare) and plotted in Prism8 (GraphPad).

Single-turnover lysine discharge assays
Lysine discharge assays were performed as described previously (Buetow et al, 2015). Arabidopsis thaliana UBA1, UBE2D2 S22R , and IRDye 800CW maleimide-labeled GGSC-Ub were prepared as described previously (Magnussen et al, 2020). Briefly, 2 μM of UBE2D2 S22R was charged with 4 μM of maleimide-labeled Ub in the presence of 0.2 μM A. thaliana UBA1 in buffer containing 50 mM Tris-HCl (pH 7.6), 50 mM NaCl, 5 mM MgCl 2 , and 20 mM ATP for 20 min at room temperature and stopped by addition of 25 mM EDTA for 5 min. Lysine (final concentration of 250 mM for Fig 2B; 50 mM for Figs 2F, 3B, 4E, and 5C) and the indicated MDM2 variant or MDM2/ p14ARF complex (final concentration: 100 nM for Fig 2B; 200 nM for Figs 2F, 3B, 4E, and 5C) were then added to initiate the reaction. Because of the nanomolar concentration of MDM2 variants used in the assay, it was not possible to visualize MDM2 variant bands by Coomassie staining. To standardize the MDM2 variant concentration used in the assay, we determined the MDM2 variants protein concentration and then loaded equimolar quantities on SDS-PAGE to ensure similar loading of MDM2 variants (Fig S6) before dilution for use in the assay. For Fig 4E, indicated concentrations of 12xHis-GFP-MDM2 220-334 or 12xHis-GFP-N32p14ARF-AD-ZnF were incubated with MDM2 RING for 30 min at room temperature before addition to the charged UBE2D2~Ub to initiate the reaction. Reactions were quenched with SDS loading buffer at the indicated times, analyzed by SDS-PAGE and visualized using a LI-COR Odyssey CLx scanner. Images were prepared using Image Studio software (LI-COR Biosciences). The reactions in Figs 2B and 3B were performed in duplicate and triplicate, respectively. Image Studio software (LI-COR Biosciences) and ImageJ were used to analyze the scanned gels and quantify the intensities of the bands. The plots were generated using Prism8 (GraphPad).

Size exclusion analysis
Size exclusion analyses in Fig 2A were done using the 10/300 Superdex 200 chromatography column. 500 μl of protein (~200-500 μg) was applied onto the column pre-equilibrated in buffer containing 25 mM Tris-HCl (pH 7.6), 400 mM NaCl, and 1 mM TCEP. Data were presented using wavelength at 215 nm and normalized for comparison. To demonstrate that N32p14ARF interacts with MDM2 220-334 in the GFP-N32p14ARF-AD-ZnF (Fig 3E), 300 μg of GFP-N32p14ARF-AD-ZnF was incubated in either buffer alone or in the presence of 3 μg of thrombin at 4°C overnight before loading onto the 10/300 Superdex 200 column.

Solution NMR experiments
All NMR experiments were acquired using a Bruker Avance III 600-MHz spectrometer equipped with a cryogenic triple resonance inverse probe. Before titration, each sample was exchanged into NMR buffer (20 mM sodium phosphate, 300 mM NaCl, 0.5 mM TCEP, 5.0% [vol/vol] D 2 O, pH 7.0). Experiments were carried out at 298 K, and 1 H-15 N HSQC spectra were recorded with 16 scans using 64 complex points with a sweep width of 36 parts per million (ppm) in the 15 N dimension. All spectra were processed with 256 points in the indirect dimension using Bruker TopSpin version 3.5 patch level 7 and analyzed using CARA (http://cara.nmr.ch/) and CCPnmr (Skinner et al, 2016).
CSPs were calculated following:

CSP =
h ðδ HA − δ HB Þ 2 + ððδ ΝA − δ ΝΒ Þ/5Þ 2 i 1/2 For a given residue in state A or B, δ H and δ N are the differences in proton and nitrogen chemical shifts, respectively.

Data Availability
The data supporting the findings of this study are within the paper and its supplemental information.