p14ARF forms meso-scale assemblies upon phase separation with NPM1

NPM1 is an abundant nucleolar chaperone that, in addition to facilitating ribosome biogenesis, contributes to nucleolar stress responses and tumor suppression through its regulation of the p14 Alternative Reading Frame tumor suppressor protein (p14ARF). Oncogenic stress induces p14ARF to inhibit MDM2, stabilize p53 and arrest the cell cycle. Under non-stress conditions, NPM1 stabilizes p14ARF in nucleoli, preventing its degradation and blocking p53 activation. However, the mechanisms underlying the regulation of p14ARF by NPM1 are unclear because the structural features of the p14ARF-NPM1 complex remain elusive. Here we show that NPM1 sequesters p14ARF within phase-separated condensates, facilitating the assembly of p14ARF into a gel-like meso-scale network. This assembly is mediated by intermolecular contacts formed by hydrophobic residues in an α-helix and β-strands within a partially folded N-terminal domain of p14ARF. Those hydrophobic interactions promote phase separation with NPM1, enhance nucleolar partitioning of p14ARF, restrict p14ARF and NPM1 diffusion within condensates and in nucleoli, and reduce cell viability. Our structural model provides novel insights into the multifaceted chaperone function of NPM1 in nucleoli by mechanistically linking the nucleolar localization of p14ARF to its partial folding and meso-scale assembly upon phase separation with NPM1.


Introduction
Arf (Alternative Reading Frame; p14 ARF in human, p19 Arf in mouse) is an intrinsically disordered protein and key tumor suppressor that is lost or silenced in most human cancers.Arf is induced in response to oncogene activation, e.g., Myc and Ras signaling, and binds MDM2, an E3 ubiquitin ligase for p53, leading to MDM2 inhibition, p53 stabilization and cell cycle arrest 1 .In proliferating cells, Arf is maintained at low levels and localizes to the granular component (GC) of the nucleolus through its interaction with Nucleophosmin (NPM1) 2,3 .Tight regulation of nucleolar Arf by NPM1 maintains stable pools of Arf, NPM1, and MDM2 4 .NPM1 regulates Arf stability by binding Arf and sequestering it in the nucleolus, and disruption of the Arf-NPM1 interaction releases Arf from nucleoli and induces proteasomal degradation of Arf in the nucleus 5,6 .In addition, binding of respiratory cytochrome c to NPM1, causes an extended-to-compact conformational change in NPM1, triggering p19 Arf release 7 .
Similarly, binding of p14 ARF by GLTSCR2 blocks the p14 ARF -NPM1 interaction, enhancing p14 ARF nuclear translocation and degradation 8 .Furthermore, p19 Arf mutants which lack conserved Nterminal segments fail to bind NPM1 and are rapidly degraded 5 .Release from NPM1 facilitates Arf targeting of MDM2 9 and occurs in response to various stressors, including DNA damage 7,10 and nucleolar disruption 11,12 .Conversely, Arf overexpression induces NPM1 degradation through the SUMO pathway 13 .Thus, interactions with NPM1 in the nucleolus are critical for regulating Arf stability and function.
Nucleoli are liquid-like membrane-less organelles (MLOs) assembled in part through liquid-liquid phase separation (LLPS) 14,15 .NPM1 forms pentamers and mediates the assembly of the GC in part through multivalent interactions of acidic tracts (A-tracts, A2 and A3) within its central intrinsically disordered region (IDR) with multivalent arginine-rich motifs (R-motifs) in nucleolar proteins, e.g., ribosomal proteins and non-ribosomal proteins such as SURF6 16,17 .
Interaction with NPM1 facilitates the localization of R-motif proteins to nucleoli 16,18 .p14 ARF contains several multivalent R-motifs, which are required for nucleolar localization and are mutated in certain cancers, causing redistribution of p14 ARF throughout the cell 19,20 .Purified NPM1 undergoes phase separation with R-motif proteins in vitro, forming condensates that mimic the liquid-like features of the nucleolus 15 .We previously showed that p14 ARF promotes phase separation when mixed with NPM1 in vitro, and that the presence of p14 ARF attenuates NPM1 mobility within condensates 21 .
To gain insight into the molecular basis of p14 ARF -NPM1 interactions in the nucleolus, here we characterize the structure and dynamics of p14 ARF and NPM1 within condensates using an integrated structural biology approach, encompassing solution-and solid-state nuclear magnetic resonance (NMR) spectroscopy and small-angle neutron scattering (SANS).We found that p14 ARF forms meso-scale assemblies within condensates with NPM1, mediated by intermolecular hydrophobic interactions between p14 ARF residues within a partially folded Nterminal domain.Based on this information, we hypothesized that hydrophobic interactions mediated by p14 ARF cause NPM1 immobilization within condensates in vitro and reduced NPM1 diffusion in nucleoli.We found that substitution mutagenesis to block p14 ARF hydrophobic interactions restored p14 ARF and NPM1 mobility in condensates while reducing the propensity for phase separation.In cells, p14 ARF and NPM1 exhibited reduced diffusion and mobility in nucleoli, consistent with the formation of higher order p14 ARF -NPM1 assemblies.This correlated with p14 ARF levels and was dependent upon hydrophobic residues within the p14 ARF N-terminal domain.These results demonstrate that although the R-motifs are sufficient to induce phase separation of NPM1, the hydrophobicity of p14 ARF potentiates phase separation and is required for the restriction of p14 ARF and NPM1 within the nucleolus.Based on our model, NPM1 promotes sequestration of p14 ARF in nucleoli by facilitating the phase separation and partial folding of p14 ARF .

p14 ARF Exhibits Local and Long-Range Ordering within Condensates with NPM1
Pentameric NPM1 engages its binding partners in part through multivalent electrostatic interactions of its disordered A2 and A3 acidic tracts (Fig. 1A) and R-motifs in partner proteins.p14 ARF contains several multivalent R-motifs (termed R1-3) (Fig. 1B, Supplementary Table 1).p14 ARF also displays three well-conserved N-terminal clusters of hydrophobic residues (termed H1-H3) (Supplementary Fig. 1), two of which are predicted by ZipperBD 22 and PSI-PRED4 23 to form aggregation-prone α-helical and β-sheet secondary structures (Fig. 1B).To gain insight into the structural organization within phase-separated p14 ARF -NPM1 complexes, we applied contrast variation small-angle neutron scattering (CV-SANS).This approach leverages the differences in the neutron scattering length densities of protons and deuterons to isolate the scattering contributions from select biomolecules in complex mixtures through protein perdeuteration (replacement of H-atoms with D-atoms) and adjustment of the H2O/D2O ratio within buffers 24 .Fitting the CV-SANS curve of p14 ARF -NPM1 condensates under p14 ARFmatched conditions (only scattering from NPM1 detected) to a correlation length model (Fig. 1C, green trace; Supplementary Table 2, see Methods for fitting procedure) suggests that the IDRs of pentameric NPM1 16,17 are in extended conformations in condensates.Strikingly, the CV-SANS curves for p14 ARF -NPM1 condensates under full-scattering conditions (scattering from both NPM1 and p14 ARF detected) and NPM1-matched conditions (only scattering from p14 ARF detected) exhibited prominent Bragg peaks (Fig. 1C; grey and blue traces, respectively).The CV-SANS curve from NPM1-matched conditions was fit to a broad peak model, which revealed that p14 ARF molecules also assume extended conformations ( = 0.66) and form a meso-scale (10-100 nm) assembly with a characteristic intermolecular spacing, d ≈ 180Å, within the condensed phase with NPM1 (Fig. 1C).This assembly appears branched at the longest length scales measured ( = 0.35) with inter-chain contacts 25 occurring over a distance of ~160 Å. Meso-scale ordering of this type is common within phase-separated materials, e.g., polymer gels, and can be caused by physical crosslinks 24 .
We next sought to characterize the residue-level structure of p14 ARF within condensates with NPM1 and to identify sites of intra-and intermolecular p14 ARF contacts using solution-state NMR spectroscopy.The two-dimensional transverse relaxation-optimized spectroscopy, heteronuclear single-quantum 1 H-15 N correlation (2D 1 H-15 N TROSY-HSQC) spectrum of [ 13 C, 15 N]-p14 ARF within condensates with unlabeled NPM1 revealed resonances for a subset of residues (Supplementary Fig. 2).Using triple-resonance NMR methods (see Methods), these were assigned to residues in the C-terminal region of p14 ARF , following R-motif R3 (Fig. 1B, Supplementary Table 3), indicating that this region of p14 ARF is disordered in condensates with NPM1.In contrast, N-terminal p14 ARF residues showed extensive resonance broadening and could not be analyzed using solution-state NMR methods.
We reasoned that resonance broadening resulted from limited mobility of p14 ARF within its phase-separated meso-scale assemblies, as indicated by previous fluorescence recovery after photobleaching (FRAP) results 21 .Therefore, we applied cross-polarization magic-angle spinning solid-state NMR (CP-MAS ssNMR) methods, which can detect resonances for both mobile and immobile segments of proteins within condensates 21 (Supplementary Table 4, Fig. 1D, Supplementary Fig. 3).Analysis of multiple two-and three-dimensional ssNMR spectra enabled resonance assignments for residues within the p14 ARF N-terminus (Supplementary Figs. 4, 5; Supplementary Table 5; see Methods).Analysis of secondary 13 C chemical shifts, which report on secondary structure, revealed that the N-terminal domain (NTD) of p14 ARF , which is disordered in isolation 26 , adopts α-helical and β-strand secondary structure in condensates with NPM1 (Fig. 1E).
Consistent with the findings from CV-SANS, we observed only one intra-molecular contact in p14 ARF , between T8 and H26, in 2D 13 C- 13 C dipolar assisted rotational resonance (CC-DARR) spectra at long mixing times (200 ms and above; Supplementary Fig. 6A), suggesting that compact conformations are not highly populated or form only transiently.To probe for inter-molecular p14 ARF -p14 ARF contacts, we recorded NHHC spectra 27 for a p14 ARF -NPM1 condensate containing a 1∶1 mixture of independently 15 N-or 13 C-labeled p14 ARF molecules, to ensure that only inter-molecular 15 N− 13 C correlations were detected 28 .The resulting spectrum showed a high degree of similarity to DARR spectra, demonstrating that structured regions within the p14 ARF N-terminus engage in inter-molecular contacts (Supplementary Fig. 6B).Furthermore, based on the low signal-to-noise ratio observed for NHHC spectra, persistent p14 ARF contacts either constitute a minor state or occur over long distances.

Structural model for p14 ARF within p14 ARF -NPM1 condensed phase
Next, we visualized the structure of p14 ARF within the condensed phase with NPM1 by integrating constraints obtained from analysis of NMR and CV-SANS data into a structural ensemble model (Supplementary Fig. 7, see Methods).First, we used PSI-PRED4 29 to predict residue-level p14 ARF secondary structure.We then used Flexible Meccano 30 to generate large ensembles of conformers (10,000), where the secondary structure propensity of non-structured regions was systematically varied from random coil to β-sheet/poly-proline type II (PPII), in a cooperative or non-cooperative manner.These structural ensembles were processed using Cryson 31 and ShiftX2 32 to calculate polymer scaling factors, and predict chemical shifts for each conformer, respectively.Finally, we applied Bayesian statistics 33 to calculate the probability of each conformer based on experimental SANS and NMR data and selected a refined p14 ARF ensemble containing the highest probability conformers (Supplementary Fig. 8A-C).
The refined p14 ARF ensemble exhibited a mean Cα-Cα distance of 80 ± 31Å (Fig. 2A) and a mean scaling factor   = 0.659 ± 0.001 (Fig. 2B), which are in agreement with the experimental values ( 0 = 85 ± 31Å and  = 0.659 ± 0.296, respectively).Furthermore, the ensemble average predicted chemical shifts showed good agreement with experimental NMR data (Fig. 2C, D, Supplementary Fig. 8D-F).The resulting model shows p14 ARF in extended conformations that expose the hydrophobic surfaces and R-motifs (Fig. 2E).In this way, p14 ARF may engage in inter-molecular interactions with both itself and NPM1 within the condensed phase.
To model p14 ARF within the meso-scale p14 ARF -NPM1 assembly, we used D+ 34 to assemble the refined p14 ARF ensemble into domains of diverse sizes and space groups, with a chi-squared minimization yielding the best model.We obtained the best agreement with experimental data for p14 ARF in a 4 x 3 domain with 2D rectangular symmetry and X, Y lattice point distances of 180 Å and 200 Å, respectively (Fig. 2F, G).Examination of intermolecular Cα-Cα distances within the meso-scale p14 ARF assembly revealed characteristic spacings of ~200 and ~400 Å (Supplementary Fig. 8G).Consistent with the low signal-to-noise ratio observed in the NHHC spectrum, only a small subset of close-range interchain distances were observed (<30 Å).The final model shows an ensemble of p14 ARF molecules assembled in an ordered lattice, which permits conformers at individual lattice points to assume a high degree of conformational disorder (Fig. 2G).Furthermore, the p14 ARF meso-scale pores can accommodate NPM1 pentamers (~60 Å correlation length; Fig. 1C).

NPM1 IDR remains disordered within the condensed phase with p14 ARF
We previously applied CP-MAS ssNMR to show that the N-terminal NPM1 oligomerization domain (OD) retains secondary structure in condensates with p14 ARF and experiences limited mobility 21 .However, we detected no resonances corresponding to residues in the NPM1 central IDR or the C-terminal, nucleic acid binding domain (NBD), suggesting that these structural elements remain dynamic.Here, we applied solution-state NMR to probe the structure and dynamics of the NPM1 IDR within p14 ARF -NPM1 condensates.2D 1 H-15 N TROSY-HSQC spectra for [ 13 C, 15 N]-NPM1 showed resonances for residues in the IDR, although resonance broadening was apparent (Fig. 3A).This stemmed from an enhancement in 15 N R2 relaxation, as detected through measurements of different types of nuclear spin relaxation (Fig. 3B).This was most pronounced for residues closest to the A3 acidic tract (residues 161-188), which mediates interactions with R-motif-containing proteins 18 including Arf 2 .Interestingly, R2 enhancement was due in part to chemical exchange as measured by 15 N Carr-Purcell-Meiboom-Gill ( 15 N-CPMG) relaxation dispersion (Fig. 3C).Fitting to a 2-state exchange model showed that interconversion of NPM1 IDR conformations occurred on the 100s µs timescale (Fig. 3C, Supplementary Fig. 9, Supplementary Table 6), suggesting that the condensate environment restrains conformational dynamics of the NPM1 IDR (Fig. 3D).
p14 ARF hydrophobic residues contribute to p14 ARF meso-scale ordering and to reduced

NPM1 mobility within condensates
We hypothesized that the hydrophobic interfaces in the p14 ARF N-terminal region are involved in interactions that drive phase separation and reduce NPM1 mobility within condensates.To test this, we substituted multiple aliphatic residues (Ile, Leu, and Val) within the p14 ARF N-terminus with Gly and Ser (termed p14 ARF ΔH1-3) (Fig 4A, Supplementary Table 1).
Taken together, these results show that hydrophobic residues within the p14 ARF -NTD act as "stickers" 35 that mediate self-association, enhance multivalent heterotypic interactions to drive phase separation with NPM1, and promote meso-scale assembly of p14 ARF molecules, thus restraining NPM1 translational diffusion.

p14 ARF reduces nucleolar NPM1 diffusion in a concentration-dependent manner
NPM1 sequesters p14 ARF in nucleoli to inhibit it from engaging other binding partners and activating anti-proliferative pathways 4 .Given that NPM1 usually forms dynamic, liquid-like condensates 36 and purified p14 ARF rapidly precipitates from solution 37 , we reasoned that p14 ARF and NPM1 form condensates that block p14 ARF aggregation by capturing it within the gel-like interaction network of the meso-scale assemblies (Fig. 5A).This is akin to NPM1's role as a chaperone for misfolded proteins in the nucleolus during cellular stress 38 .On the other hand, overexpression of p19 Arf promotes NPM1 degradation 13 and assembly of high molecular weight p19 Arf -containing complexes 2 .Based on these observations, we reasoned that an abundance of NPM1 is needed to form p14 ARF -NPM1 complexes in nucleoli, to stabilize p14 ARF and limit its potential for homo-and hetero-oligomerization with other nucleolar biomacromolecules.
Therefore, we next asked whether expression of p14 ARF alters the dynamics of NPM1 in nucleoli.
To assess the dependence of this effect on the level of p14 ARF -iRFP expression, we first used flow cytometry to isolate DLD-1 NPM1-G clones that expressed p14 ARF -iRFP at different levels (Supplementary Fig. 15A).Consistent with our observations with unsorted DLD-1 NPM1-G cells, expression of p14 ARF -iRFP in the isolated DLD-1 NPM1-G clones caused dose-dependent reductions in DApp and mobility for NPM1-GFP, which was correlated with values for p14 ARF -iRFP (Supplementary Fig. 15B, C).We then monitored p14 ARF -iRFP and NPM1-GFP diffusion for two clones (termed G2 and B11) before, and 24 hours and 48 hours after doxycycline induction of p14 ARF -iRFP expression.Both DLD-1 NPM1-G clones showed significant reductions in the DApp value for p14 ARF -iRFP and NPM1-GFP within 24 hours, which persisted after 48 hours of p14 ARF -iRFP expression (Supplementary Fig. 15D, E).Furthermore, NPM1-GFP mobility was reduced in both DLD-1 NPM1-G cell clones at the 48-hour time point (Supplementary Fig. 15F, G).
Consistent with previous reports of p14 ARF expression in p53-null cell lines 6,44 , expression of p14 ARF -iRFP correlated with reduced viability of DLD-1 NPM1-G cells in a dose-and timedependent manner (Supplementary Fig. 15H, I).

Discussion
p14 ARF is a highly basic intrinsically disordered protein that functions as a tumor suppressor through p53-dependent and -independent mechanisms 45 .Here, we probed the structural features of p14 ARF in condensates with NPM1 and within nucleoli in DLD-1 cells lacking functional p53 and endogenous p14 ARF .Strikingly, p14 ARF adopts elements of secondary structure and induces meso-scale ordering upon phase separation to form gel-like condensates with NPM1.In addition to the R-motifs, which mediate electrostatic interactions with multivalent acidic tracts within pentameric NPM1's central IDR 2,16,18 , hydrophobic residues within p14 ARF mediate homotypic interactions that underlie meso-scale ordering.The extended nature of p14 ARF creates voids within the meso-scale assembly that are compatible with the dimensions of pentameric NPM1, the IDR of which remains flexible despite mediating key interactions with p14 ARF .Formation of this meso-scale assembly significantly attenuates the mobility of p14 ARF and NPM1 within condensates in comparison with their dynamic states in condensates formed by NPM1 and the hydrophobic residue-depleted p14 ARF mutant (p14 ARF ΔH1-3).While it is impossible to probe the meso-scale structure of p14 ARF and NPM1 within nucleoli using SANS and ssNMR, we did probe the dynamics of these proteins within cells, which revealed unexpected functional interplay.The mobility of p14 ARF declined as its level within nucleoli increased and this was paralleled by p14 ARF level-dependent declines in NPM1 mobility.Further, the levels of p14 ARF and NPM1 within nucleoli were anti-correlated, suggesting that multicomponent phase separation 46 underlies the functional relationship between these two proteins within nucleoli.This functional interplay was eliminated through mutation of hydrophobic residues of p14 ARF .Cell viability tracked downwards with increased p14 ARF expression levels, suggesting that p14 ARF serves as a viability rheostat through multicomponent phase separation with NPM1 and likely other nucleolar components.Our results provide mechanistic insight into how NPM1 stabilizes Arf within nucleoli 5,6 , consistent with NPM1's role as a nucleolar chaperone upon protein unfolding stress 38 .However, we also show that, as its levels rise, p14 ARF intoxicates cells, consistent with its tumor suppressor activity in response to oncogene activation 47 .
Many intrinsically disordered proteins, or intrinsically disordered protein regions, adopt compact conformations in isolation under physiological conditions but some assume more expanded conformations after a phase transition 35,48 .Conformational expansion exposes socalled sticker residues within polypeptide chains for multivalent interactions that underlie intermolecular network formation and phase separation 35 .Crosslinks may also be mediated by folded segments within stretches of otherwise disordered regions.For example, FG nucleoporin hydrogels are scaffolded by intermolecular β-sheet interactions 49 , and TDP-43 C-terminal domain phase separation requires transient contacts between a conserved α-helix 50 .Here we show that p14 ARF populates an ensemble of extended conformations with elements of β-strand and α-helical secondary structure within meso-scale assemblies with NPM1.Interestingly, similar binding-induced induction of secondary structure, albeit without a phase transition, was previously observed with fragments of both p14 ARF and p19 Arf containing conserved R-motifs that form soluble, -strand-rich structures upon binding to acidic-residue-rich stretches derived from the central IDR of HDM2 [51][52][53][54] .We propose that adoption of secondary structure within Arf is a common mechanism underlying its interactions with acid-tract-containing binding partners.

Cell Lines
The following cell lines were purchased from American Type Culture Collection (ATCC): DLD-1 (male, adult, age not reported, Dukes' type C colon cancer), DLD-1 cells were cultured in RPMI 1640 medium (ThermoFisher) supplemented with 10% fetal bovine serum and 100 U/mL penicillin/streptomycin.The DLD-1 cells harboring doxycycline-inducible p14 ARF -miRFP670, p14 ARF ΔH1-3-miRFP670, miRFP670, were maintained in RPMI 1640 medium supplemented with 10% Tet system approved fetal bovine serum (ThermoFisher), and 250 µg/ml G418.All cell lines were incubated at 37°C in a humidified incubator with 5% CO2.Gene edited cell lines were authenticated by short tandem repeat (STR) profiling.Cells were tested negative for mycoplasma by the e-Myco PLUS Mycoplasma PCR Detection Kit (Bulldog Bio).

Escherichia coli Strains
Escherichia coli BL21(DE3) cells were used to produce recombinant proteins.NEB Stable Competent Escherichia coli cells (New England Biolabs) were used when subcloning genes into lentiviral vectors.All other vectors were transformed to DH5α competent cells (taxid: 668369).The NEB Stable cells and the other E. coli strains were grown at 30 o C and 37 °C, respectively.

Plasmid and Cloning Methods
For E. Coli expression of the recombinant proteins including NPM1 and wild-type p14 ARF , their DNA coding sequences were subcloned to the pET-28a(+) plasmid (EMD Biosciences) as previously described 21,55 .The DNA sequence encoding the p14 ARF ΔH1-3 mutant was de novo synthesized as gBlocks (Integrated DNA Technologies) and subcloned into pET-28a(+) using the BamHI and HindIII sites.The protein sequence of the p14 ARF ΔH1-3 mutant is provided in Supplementary Table 1.To express proteins tagged with the monomeric, near-infrared fluorescent protein, miRFP670 43 , we synthesized the cDNAs of miRFP670, and p14 ARF or p14 ARF ΔH1-3 C-terminally fused with miRFP670 following a (GGS)5 linker.These were subcloned into the NheI and SalI restriction sites of the pCDH-PGK vector, a gift from Kazuhiro Oka (Addgene plasmid # 72268; http://n2t.net/addgene:72268;RRID: Addgene_72268).The protein sequences of these constructs are provided in Table S1.The coding regions were then PCR-amplified with a common pair of primers (forward: 5'-CACCCATTCTGCACGCTTCAAAAG-3'; reverse: 5'-CCACATAGCGTAAAAGGAGCAAC-3').The PCR products were subsequently TOPO cloned into the pENTR vector using the pENTR/SD/D-TOPO Cloning Kit (ThermoFisher).All plasmid constructs were verified with DNA sequencing performed by Hartwell Center DNA Sequencing Core at St. Jude Children's Research Hospital and by Massachusetts General Hospital CCIB DNA core.
Recombinant p14 ARF proteins were prepared as described 21 .Briefly, p14 ARF and p14 ARF ΔH1-3 were expressed in E. coli BL21 cells grown at 37 °C in 30 µg/ml Kanamycin supplemented LB medium.For isotopic labeling to generate [U 13 C, 15 N]-p14 ARF , cells were grown in MOPS-based minimal media containing [U 13 C6]-D-glucose and 15 NH4Cl (Cambridge Isotope Laboratories) 56 .For [U 13 C]-p14 ARF and [ 15 N]-p14 ARF labeled p14 ARF , [U 13 C6]-D-glucose/NH4Cl and D-glucose/ 15 NH4Cl were used, respectively.At OD600nm = 0.8, 0.5 mM IPTG was added, cells were incubated at 37 °C for an additional 3 h and harvested by centrifugation at 3,800 rpm at 4 °C.Cells were resuspended in 50 mM Tris pH 8.0, 500 mM NaCl, 5 mM βmercaptoethanol, and one SIGMAFAST protease inhibitor cocktail tablet (Sigma) and disrupted by sonication.The lysate was cleared by centrifugation at 30,000 rpm at 4 °C and Urea was added to a final concentration of 6 M; this fraction was set aside.In parallel, the cell pellet was resuspended in 6 M Guanidine HCl, 0.1% Triton X-100, 5 mM β-mercaptoethanol and subjected to mechanical disruption followed by sonication.This fraction was cleared by centrifugation at 30,000 rpm at 4 °C and the supernatant was removed, combined with the initial lysate, and purified by Ni-NTA-affinity chromatography on an ÄKTA FPLC (GE) using a linear gradient of 50 mM Tris pH 8.0, 500 mM NaCl, 5 mM β-mercaptoethanol and 500 mM Imidazole and further purified using C4 HPLC (Higgins Analytical, Mountain View, CA, USA) with a H2O/CH3CN/0.1% trifluoroacetic acid solvent system.
To generate calibration curves for mEGFP and miRFP670 fluorescence, recombinant poly-histidine-tagged mIRFP670 and mEGFP in pET28a (+) (Novagen) were expressed in BL21 (DE3) Escherichia coli cells (Millipore Sigma, Burlington, MA, USA).Cells were grown at 37 °C in LB medium supplemented with 30 µg/ml of Kanamycin.At OD600nm = 0.8, 0.5 mM Isopropyl β-D-1 thiogalactopyranoside (IPTG) was added, cells were incubated at 37 °C for 3 h and harvested by centrifugation at 3,800 rpm at 4 °C.Proteins were purified from the soluble lysates fraction using Ni-NTA affinity chromatography.Affinity tags were removed via proteolytic cleavage with tobacco etch virus (TEV) protease and purified using a S75 10/300 (GE) gel filtration column on an ÄKTA FPLC (GE).Biliverdin HCl (Sigma-Aldrich) was dissolved into PBS, added to mIRFP670 at a 2.5-fold molar excess and incubated at 37 o C for 3hrs.Excess biliverdin was removed by buffer exchange using a centrifugal filtration device.

Small-Angle Neutron Scattering
SANS experiments were performed on the extended q-range small-angle neutron scattering (EQ-SANS) beam line at the Spallation Neutron Source (SNS) at the Oak Ridge National Laboratory (ORNL).The detector was set at 4m from the sample position.The choppers ran at 30 Hz in frame-skipping mode to give two wavelength bands: 2.5 Å to 6.1 Å and 9.4 Å to 13.1 Å.This configuration provided a q-range from ~0.004 Å -1 < q < ~0.45 Å -1 .The source aperture was 25mm diameter and the sample aperture was 10mm diameter.
To prepare p14 ARF -NPM1 condensates for CV-SANS analysis recombinant p14 ARF & p14 ARF ΔH1-3 proteins were resuspended from lyophilized powders in 100% deuterated dimethyl sulfoxide (DMSO) and added directly to solutions of NPM1 at room temperature (~23 °C) to induce formation of phase-separated condensates.All samples contained 10 mM sodium phosphate pH 7, 150 mM NaCl, 2 mM TCEP with p14 ARF proteins and NPM1 at 40 µM.Full scatter measurements were performed in buffer containing 100% D2O and using protonated proteins.For contrast variation measurements, the H2O/D2O ratios were adjusted to 84.9% D2O to match 2 H-NPM1, 44.7% for p14 ARF , and 49.6% for p14 ARF ΔH1-3.The match point for NPM1 was determined experimentally 18 and verified independently for the current study (data not shown).Due to the instability of p14 ARF in solution, the match points for p14 ARF and p14 ARF ΔH1-3 were calculated using the MULCh contrast calculator tool 58 .The samples were loaded into 2 mm pathlength circular-shaped quartz cuvettes (Hellma USA, Plainville, NY) and SANS measurements were performed at 25 ˚C while the samples rotated on a tumbler to prevent droplets from settling out of suspension.Data reduction was performed using MantidPlot 59 .The measured scattering intensities were corrected for the detector sensitivity, the scattering contribution from the buffer and empty cells and re-scaled to an absolute scale using a calibrated standard 60 .
For p14 ARF -NPM1 condensates under full scattering conditions, the scattering curve was fit to a broad peak model 16 : where,  0 is the correlation length from the scattering at high-q and Ξ 1 is the correlation length from scattering at low-q.The peak corresponds to the d-spacing ( 0 = 2  0 ), i.e., the characteristic distance between scattering inhomogeneities.The scaling exponent,  0 = 1  0 , and  accounts for the background scattering.For NPM1-matched, p14 ARF -detected conditions, scattering was fit to a broad peak model with a correlation length term 61 : For p14 ARF -matched, NPM1-detected conditions, scattering was fit to a correlation length model: For p14 ARF ΔH1-3-NPM1 condensates, all scattering curves were fit to equation 3.

Condensate Formation for NMR Analysis
To prepare p14 ARF -NPM1 condensates for NMR analysis, recombinant unlabeled and isotopically enriched p14 ARF proteins (including [U 13 C, 15 N]-p14 ARF , [U 13 C]-p14 ARF and [ 15 N]-p14 ARF ) were resuspended from lyophilized powders in 100% deuterated dimethyl sulfoxide (DMSO-d6) and added directly to solutions of NPM1 to induce formation of phase-separated p14 ARF -NPM1 condensates.These condensates were formed at room temperature (~23 °C), such that the final p14 ARF and NPM1 concentrations were 20 µM.For assignment of p14 ARF by solution state NMR, a condensed phase was prepared by mixing 50 µM [ 13 C, 15 N]-p14 Arf and 50 µM NPM-IDR.The final buffer contained 10 mM sodium phosphate pH 7.0, 150 mM NaCl, 2 mM TCEP, 0.015% NaN3, 1.67% DMSO-d6, 7% D2O.Low concentrations of DMSO-d6 have no effect on the structure of NPM1 as confirmed previously by solution state NMR 21 .Following phase separation, samples were incubated for 20 min at room temperature.Samples were then ultracentrifuged at 100,000 rpm (436,000 x g) for 2 hours at 4 o C to pellet the condensates.The light phases were removed prior to NMR analysis.

Solution-State NMR Spectroscopy
Solution state NMR experiments were performed on Bruker AVANCE NEO spectrometers.Measurements of p14 ARF were made on a spectrometer operating at a 1 H Larmor frequency of 600 MHz, equipped with a 5 mm triple-resonance 1H/13C/15N TCI cryoprobe.Measurements of NPM1 were made on a spectrometer operating at a 1 H Larmor frequency of 800 MHz, equipped with a TXO cryoprobe optimized for 13 C. Spectra were processed in Topspin 4.0 or NMRPipe and analyzed in Sparky.
In solution dynamics measurements of NPM1 (Fig. 3) were performed on 65µM [ 2 H, 15 N]-NPM1 in 10 mM sodium phosphate pH=7.0, 150 mM NaCl, 2 mM DTT, 10% D2O at 25 o C. 1 H-15 N NOE values were calculated as the ratio between peak intensities in spectra recorded with and without 1 H saturation.The 15 N relaxation rates, R1 and R2, were determined by fitting crosspeak intensities, measured as a function of variable delay periods, to a single-exponential decay. 15N-CPMG relaxation dispersion was fitted using the protein dynamics toolset in the Bruker Dynamics Center 2.5.6 with the following fitted function alternatives: where  2 is the effective relaxation rate,   is the exchange rate, Δ is the chemical shift difference between states A and B, and  =     Δ 2 .Error estimation was performed using Monte-Carlo simulation.Fitted parameters were calculated with a 95% confidence interval.

Solid-State NMR Spectroscopy
Solid-state NMR experiments were performed on a Bruker Avance NEO spectrometer operating at 14.1 T ( 1 H Larmor frequency of 600 MHz) using a Bruker MAS CryoProbe TM , a cryogenically cooled magic-angle spinning (MAS) triple resonance (HCN) probe head 63 .The samples were packed in specially designed 3.2 mm MAS rotors with Teflon inserts to ensure proper centering of the p14 ARF -NPM1 condensate samples.Detailed description of the acquisition parameters can be found in Supplementary Table 4.In general, all the MAS experiments were performed at MAS speeds between 10-15 kHz.Typical radio-frequency (RF) fields used in the experiments for the 1 H, 13 C and 15 N channels were 80-100 kHz, 60-65 kHz and 40 kHz, respectively.Double cross polarization (CP), dipolar assisted rotational resonance (DARR) and COmbined R2vn-Driven (CORD) mixing requires lower RF fields and are reported in Supplementary Table 4. Contact times for CP and double CP were typically 1 ms with recycle delays of 2 s.The CP-MAS NMR acquisition times varying from 1-2 hours for two-dimensional (2D) NCO 64,65 and NCaCX experiments [66][67][68] 2D experiments) to several hours (7-10 hours) for the 2D CC correlation experiments (with DARR, CORD or insensitive nuclei enhanced by polarization transfer (INEPT) mixing).Three-dimensional (3D) experiments were recorded over 1.5 (3D NCOCX 64,65 ) and 2.5 days (3D NCaCx, through co-addition of two experiments of one day each and another of 10 hours; 34 hours of acquisition in total).The NHHC experiment used to probe contacts between the 15 N-p14 ARF and the 13 C-p14 ARF molecules within condensates with NPM1, based on proton spin diffusion between 15 N-coupled amide protons (in one p14 ARF molecule and 13 C-coupled aliphatic protons in another p14 ARF molecule), required the longest experimental time: two spectra acquired with identical parameters were co-added; these were acquired for 3 days and 9 hours, respectively.All spectra were referenced using adamantane ( 13 C δ = 38.5 ppm).

Structural Ensemble Generation and Refinement
For p14 ARF structural ensemble generation, PSI-PRED4.0 29was first used to predict the positions of α-helical and β-strand segments.Regions with a probability of >5 were used.Flexible Meccano 30 was then used to generate large starting pools of conformers (10,000).The 2 o structure propensity of regions outside of the predicted α-helical and β-strand segments was systematically varied from random coil to β-strand/PPII in a fully cooperative and noncooperative manner.In the former scenario, 0, 12.5, 25, and 37.5% fully structured β-strand conformers were introduced among conformers produced using random coil dihedrals.For the non-cooperative ensembles, dihedrals were sampled randomly in the same sequence about a gaussian distribution centered at a φ, ψ angle of -112.6, 123, and the gaussian dispersion parameter was varied from 115-140.For each pool, Cryson 31 was used to calculate the radius of gyration, which was transformed into a polymer scaling factor () through the relationship: where  0 is an empirical prefactor 69 and  is the number of amino acids.ShiftX2 32 was used to predict the Cα, Cβ, NH, HN, and Cʹ chemical shifts for each conformer.
Bayesian statistics were used to estimate the probability of each conformer based on experimental SANS and NMR data.The posterior probability density of the weights based on the observed experimental data was determined from Bayes' theorem 33 The posterior was then calculated using equation ( 1), with the integral approximated using the trapezium method as implemented in the scipy integration submodule.The posterior was then used as a new prior and equation ( 1) was evaluated for 15,000 iterations to improve the estimate of the weight vector.Convergence was assessed by evaluating the ensemble average scaling factor over the refinement trajectory (Supplementary Fig. 8A, B). 160 of the most probable conformers were selected from each starting pool and a rank sums test, as implemented in the scipy statistical functions submodule, was performed to determine the most probable refined ensemble, and identify degenerate ensembles (Supplementary Fig. 8C).
D+ 34 was used to assemble the conformers from the refined ensemble into configurations of various sizes and space groups and compute their scattering intensities.Initially, p14 ARF conformers were assembled into 2D square and rectangular space groups.For each space group, array sizes ranging from 2x2 to 4x4 were assembled with X, Y interchain distances ranging from 160-200Å in 10Å increments.The reciprocal grid size was then calculated using the "suggest parameters" tool and scattering curves were computed using the classic Monte-Carlo integration method with 1e 6 iterations and a 500ms update interval.The best model was determined based on the smallest chi-squared difference between the ensemble average scattering curves from a given configuration and the experimental CV-SANS curve.

Endogenously-Tagged Cell Line Generation
Endogenously C-terminally mEGFP-tagged NPM1 in DLD-1 cells (DLD-1 NPM1-G ) were generated using CRISPR-Cas9 technology in the Center for Advanced Genome Engineering (St.Jude Children's Research Hospital).The donor homology directed repair (HDR) template containing a (GGS)5 linker DNA coding sequence upstream of the mEGFP sequence flanked by ~800 bases homology arms was synthesized and blunt-end cloned into pUC57 (the plasmid pUC57_NPM1-mEGFP_HDR donor repair template, CAGE117.g1.meGFP donor) by Bio Basic.Briefly, 500,000 DLD1 cells were transiently co-transfected with precomplexed ribonuclear proteins (RNPs) consisting of 100 pmol of chemically modified sgRNA (CAGE117.NPM1.g1,5'-UCCAGGCUAUUCAAGAUCUC-3', Synthego), 33 pmol of Cas9 protein (St.Jude Protein Production Core), 500 ng of plasmid donor.The transfection was performed via nucleofection (Lonza, 4D-Nucleofector™ X-unit) using solution P3 and program CA-137 in a small (20 µl) cuvette according to the manufacturer's recommended protocol.Single cells were sorted based on viability five days post-nucleofection into 96-well plates containing prewarmed media and clonally expanded.Clones were screened and verified for the desired modification using PCRbased assays and confirmed via sequencing.Final clones were authenticated using the PowerPlex Fusion System (Promega) performed at the Hartwell Center (St.Jude).

Lentiviral Transduction and Generation of Cell Lines
Lentiviral vectors were used to make lentiviral particles by the Vector Development and Production Shared Resource at St. Jude Children's Research Hospital.Cells were transduced with virus in the presence of 10 µg/ml polybrene (Sigma).For pINDUCER20 lentivirus transduced cells, the selection by G418 (500 µg/ml) lasted until mock-transfected, control cells were completely eliminated, and the cells were constantly maintained in the culture medium containing G418 at 250 µg/ml.

Single-Cell Cloning
Each population of the virally transduced, G418 resistant cells were sorted one cell/well into three 96-well plates.After growing in G418-containing media for 7-10 days, each viable single colony was further passaged into two corresponding wells in one Nunc 96-well cell culture treated plate (ThermoFisher) and one glass bottom black 96-well plate (Greiner Bio-One, Cat.#655891).The clones in the glass bottom 96-well plates were treated with 1 µg/ml doxycycline to induce the expression of miRFP670-tagged protein in cells, and the expression levels were quantified by measuring miRFP670 fluorescence intensity in the live cells using fluorescence microscopy.Single cell clones in the corresponding wells in the Nunc 96-well plates, which could express miRFP670-tagged protein at high, medium, or low levels, were selected and expanded.As miRFP670 requires the cofactor biliverdin for fluorescence 70 , the protein expression levels in these single-cell clones were further assessed by immunoblotting analysis.

Cell Growth Assays
Aliquots of cell suspensions were seeded in 96-or 24-well plates at 5,000 or 10,000 cells per well, respectively.After culturing for 20-24 h, the cells were counted for the starting time point and/or subjected to treatments as needed, and then cultured for the indicated times.For cell counting, existing culture medium in each well was replaced with fresh culture medium containing 10-fold diluted Cell Counting Kit-8 (CCK-8, APExBIO), and the absorbance at 450 nm was measured after 1-2 h of incubation.Cell growth was calculated as the ratio of A450 at later time points relative to that of the starting time point.The relative cell viability was expressed as the ratio of A450 of the treated versus that of untreated controls cells.Biological replicates were performed separately at different times.

Fluorescence Recovery After Photobleaching
Analysis of fluorescence recovery after photo-bleaching (FRAP) images to determine the apparent diffusion coefficient (DApp) and percent mobility was performed following a modified version of the protocol from 71 , using in-house pipelines written in Python (Supplementary Fig. 10).For FRAP in live cells, all images were corrected (()  ) to account for background fluorescence (()  ) and for photofading and irreversible loss of molecules during the bleach event, using the mean intensity of the cell nucleus (()  ), where: ()  = () − ()  ()  − ()  (12)   Here, the background and mean nuclear intensities were extracted from freehand drawn regions of interest (ROI) using the Slidebook 6.0 (Intelligent Imaging Innovations, Gottingen, Germany).For FRAP of droplets, all images were corrected using an unbleached reference droplet (()  ).

Fig. 2 .
Fig. 2. Structural Model for the p14 ARF Component of the p14 ARF -NPM1 Condensed Phase.A) Intramolecular Cα-Cα distances for the p14 ARF ensemble.B) Comparison of the ensemble and experimental polymer scaling factors.C) Comparison of the ensemble and experimental Cα chemical shifts.D) Comparison of the ensemble and experimental Cβ chemical shifts.E) Representative conformers from the p14 ARF ensemble.F) Comparison of the experimental p14 ARF CV-SANS curve (light blue scatter points) and the p14 ARF ensemble model (blue trace).Points represent the average and standard deviation.G) Ensemble model for the p14 ARF mesoscale assembly.

Fig. 3 .
Fig. 3.The NPM1 IDR Retains Disorder and Experiences Attenuated Backbone Motions within the Condensed Phase with p14 ARF .A) 2D 1 H-15 N TROSY-HSQC spectrum of [ 13 C, 15 N]-NPM1 within the p14 ARF -NPM1 condensed phase, displaying signals from the NPM1 IDR.B) Linear net charge per residue (LNCPR) for the NPM1 IDR. 1 H-15 N heteronuclear NOE, R1 and R2 transverse relaxation profiles for NPM1 in solution (blue) and within the p14 ARF -NPM1 condensed phase (red), which show a restriction of IDR backbone motions on the ps-ns timescale.Exchange broadening rates Rex for condensed NPM1 are shown on the bottom.C) 15 N-CPMG relaxation dispersion profiles for Ala186, A201 and T199 collected at 800 MHz, with fits to a two-state model.D) Upon phase separation with p14 ARF the NPM1 IDR exchanges slowly between multiple conformations on the µs-ms timescale.All error bars represent the standard deviations.

Fig. 5 .
Fig. 5. p14 ARF Reduces Nucleolar NPM1 Diffusion in a Concentration Dependent Manner.A) Schematic constant-temperature and pressure phase diagram for p14 ARF -NPM1.Single phase regions are shown in white; coexistence regions are shown in gray.The curved arrow represents a concentration vector that crosses through the coexistence regions, initially sampling a liquid-like NPM1-rich phase, followed by a gel-like p14 ARF -NPM1 phase, terminating in a solid-like p14 ARF -rich phase.B) Fluorescence microscopy images of live B11 cells before and 48 hours after doxycycline induced p14 ARF -iRFP expression.Scale bars = 2 µm.C) Z-score analysis of NPM1-GFP and p14 ARF -iRFP levels in DLD-1 NPM1-G cells, showing that p14 ARF and NPM1 levels are anti-correlated (two-sided Mann-Whitney U-test, n = 2272, 122, 54), (*) p < 0.05, (****) p < 0.0001.D) Representative single-cell FRAP for two cells selected from the DLD-1 population shown in C. The curves on the left are from a cell expressing a high level of nucleolar NPM1 and low level of p14 ARF .The curves on the right are from a cell expressing a low level of nucleolar NPM1 and a high level of p14 ARF .E) The DApp and F) the mobility for nucleolar NPM1-GFP is reduced as nucleolar p14 ARF -iRFP levels increase (small, transparent markers) and as the duration of p14 ARF -iRFP expression is extended (large, opaque markers).
:   ⃑⃑⃑ ( ⃑⃑ ), represents a priori knowledge about the underlying weights,  ⃑⃑ , and a likelihood function,   ⃑⃑ | ⃑⃑⃑ ( ⃑⃑ | ⃑⃑ ), describes the probability of observing the experimental data,  ⃑⃑ = { 1 , … ,   }, a vector of  experimental measurements.The uniform probability density function from the scipy statistical functions submodule was used to generate the initial prior distribution.The likelihood function, which describes the uncertainty of each chemical shift measurement was:   | ⃑⃑ ] is the chemical shift calculated from the ensemble,   2 is the experimental error and   2 is the chemical shift prediction error.The likelihood function, which describes the uncertainty of the polymer scaling factor was: