NORAD-induced Pumilio phase separation is required for genome stability

Abstract

Liquid-liquid phase separation is a major mechanism of subcellular compartmentalization^1,2. Although the segregation of RNA into phase-separated condensates broadly affects RNA metabolism^3,4, whether and how specific RNAs use phase separation to regulate interacting factors such as RNA-binding proteins (RBPs), and the phenotypic consequences of such regulatory interactions, are poorly understood. Here we show that RNA-driven phase separation is a key mechanism through which a long noncoding RNA (lncRNA) controls the activity of RBPs and maintains genomic stability in mammalian cells. The lncRNA NORAD prevents aberrant mitosis by inhibiting Pumilio (PUM) proteins^5-8. We show that NORAD can out-compete thousands of other PUM-binding transcripts to inhibit PUM by nucleating the formation of phase-separated PUM condensates, termed NP bodies. Dual mechanisms of PUM recruitment, involving multivalent PUM-NORAD and PUM-PUM interactions, enable NORAD to competitively sequester a super-stoichiometric amount of PUM in NP bodies. Disruption of NORAD-driven PUM phase separation leads to PUM hyperactivity and genome instability that is rescued by synthetic RNAs that induce the formation of PUM condensates. These results reveal a mechanism by which RNA-driven phase separation can regulate RBP activity and identify an essential role for this process in genome maintenance. The repetitive sequence architecture of NORAD and other lncRNAs^9-11 suggests that phase separation may be a widely used mechanism of lncRNA-mediated regulation.

Extended Data Fig. 1 |. NORAD and PUM co-localize in cytoplasmic condensates.

a, (Left) Confocal images of NORAD and PUM1 co-stained HCT116 cells with or without treatment with camptothecin (200 nM) for 24 hours. (Right) Box plots of the mean fluorescence intensity and area of NORAD and PUM1 foci. NORAD, n= 149 foci; PUM1, n = 58 foci; NORAD + camptothecin, n = 97 foci; and PUM1 + camptothecin, n = 67 foci from at least two different fields analyzed. Boxes extend from 25^th to 75^th percentiles; middle line represents median; whiskers extend from minima to maxima. P values calculated using two-tailed t-test. b, Western blot of PUM1 or PUM2 in HCT116 cells of the indicated genotypes. Molecular weight in kDa shown on right of blots. For gel source data, see Supplementary Figure 1. c, Confocal images of PUM1 or PUM2 co-stained with P-body marker XRN1. d, Confocal images of PUM2 and stress granule marker G3BP1 with or without treatment with sodium arsenite (0.5 mM, 60 min incubation). e, Confocal images of NORAD and PUM localization in the indicated cell lines and genotypes. BJ-5ta cells are TERT-immortalized human fibroblasts. f, Quantification of NORAD and PUM1 co-localization (n = 20 cells for each cell line). Mean co-localization shown above box plots. Boxes extend from 25^th to 75^th percentiles; the line in the middle of the box represents the median; whiskers extend from minima to maxima.

Extended Data Fig. 2 |. NP bodies are liquid-like condensates distinct from other known cytoplasmic granules.

a, PUM1 western blot in HCT116 clones with GFP knock-in at the endogenous PUM1 locus. Three genotype-confirmed wild-type and NORAD^−/− clones shown alongside parental wild-type and NORAD^−/− cells. Molecular weight in kDa shown on right of blots. For gel source data, see Supplementary Figure 1. b, Time-lapse images showing fusion of endogenous PUM1-GFP condensates in HCT116 cells. Scale bar, 5 μm. c, Representative 3D-reconstructed live cell confocal images used to estimate the volume of NP bodies, P-bodies, and stress granules. Endogenously-tagged PUM1-GFP HCT116 cells were used for NP body or stress granule measurements (the latter after treatment with 0.5 mM sodium arsenite for 1 hour). P-body marker DCP2-GFP was used to estimate the volume of P-bodies in HCT116 cells. d, Quantification of condensate volumes (n = 1072 NP bodies from 20 cells, 51 P-bodies from 11 cells, and 62 stress granules from 10 cells). Average volume of each granule shown above plots.

Extended Data Fig. 3 |. Liquid-like properties of PUM droplets in vitro.

a, Predicted disordered regions of PUM1 and PUM2, scored by PONDR VSL2. HD, homology domain. b, Schematic (top) and coomassie stain (bottom) of purified MBP-SNAP-PUM-His proteins used for in vitro experiments. For gel source data, see Supplementary Figure 1. c, PUM liquid droplet formation at 5 μM upon TEV-mediated cleavage of the solubilizing MBP tag. Proteins were fluorescently labeled with SNAP₄₈₈ and visualized by DIC or fluorescence microscopy. Scale bar, 10 μm. d, PUM1 liquid droplet formation at 5 μM in the presence or absence of 10% PEG₃₃₅₀ after 1 hour incubation. e, Confocal images (left) and quantification (right) of PUM1 droplet FRAP (5 μM PUM1). Fluorescence intensities plotted relative to pre-bleach time point (t = −5 s). Data shown as mean ± SD (n = 3 droplets). f, Time-lapse confocal images showing fusion of PUM1 droplets (5 μM PUM1). g, DIC and fluorescence microscopy images of Cy3-labeled NORAD RNA (2.7 nM, red) after addition to preformed PUM1 or PUM2 droplets (5 μM, green). h, Fluorescence microscopy images of PUM1 droplets (5 μM, green) formed in the presence or absence of NORAD or PREmut RNA (2.7 nM, red). PREmut contains UGU to ACA mutations in all 18 NORAD PREs, which abolish PUM binding. i, Images and analysis of PUM1-NORAD droplet FRAP. Droplets were formed with Cy3-labeled NORAD RNA (2 nM) and SNAP₄₈₈-labeled PUM1 (5 μM). Data shown as mean ± SD (n = 3 droplets).

Extended Data Fig. 4 |. NORAD-induced PUM phase separation at physiologic concentrations in vitro.

a, Representative 3D-reconstructed confocal image of a PUM1-stained NORAD^−/− HCT116 cell used to estimate cytoplasmic volume. The length (l), width (w), and depth (d) of the entire cell and its nucleus were measured and used to calculate the total cellular and nuclear volumes using the ellipsoid volume formula. Cytoplasmic volume for each cell was determined by subtracting nuclear volume from total volume. b, Box-and-whisker plot of measured cytoplasmic volumes (n = 20 cells). Mean cytoplasmic volume shown on the right. Boxes extend from 25^th to 75^th percentiles; middle line represents median; whiskers extend from minima to maxima. c, Schematic of wild-type and mutant NORAD transcripts. Location of repeated NORAD domains (ND1-ND5) indicated with gray boxes and mammalian sequence conservation shown in green (UCSC Genome Browser hg38 PhastCons track). Location of PREs indicated with red and yellow arrowheads. PREmut transcript contains 18 UGU to ACA mutations in PREs (gray arrowheads). ND4 represents the most conserved segment of NORAD and contains 4 PREs. Figure modified from Elguindy et al.. d, DIC and fluorescence microscopy images of PUM1 (150 nM, green) droplets in the presence or absence of NORAD (2 nM) or PEG₃₃₅₀.

Extended Data Fig. 5 |. PUM target transcripts do not induce PUM1 droplet formation in vitro at physiologic concentrations.

a, Dot plot of PRE-containing mRNA copy numbers in HCT116 cells (estimated by RNA-seq; see Methods) versus the number of PREs in their 3´ UTRs. Transcripts selected for in vitro assays labeled in blue with the number of PREs in their 3´ UTRs in parentheses. Note that the selected mRNAs were also identified as PUM CLIP targets that were downregulated upon PUM overexpression^,. b, Plot of transcript PRE number multiplied by estimated copy number, highlighting the unique combination of NORAD abundance and PRE valency compared to other PRE-containing RNAs. c, qRT-PCR validation of copy numbers of the indicated transcripts in HCT116 cells. Mean copy number is shown above each bar. Data shown as mean ± SD. n = 3 biological replicates. d, Approximate cytoplasmic concentration of each PUM target transcript in HCT116 cells. e, Confocal images of PUM1 droplets, formed with 150 nM PUM1 plus 2 nM NORAD or increasing concentrations of the indicated in vitro transcribed 3´ UTR of each PUM target. Red boxes highlight assays performed at the estimated physiologic concentration of each transcript in HCT116 cells. The same PUM1 protein preparation was used for droplet formation assays with NORAD and 3´ UTRs.

Extended Data Fig. 6 |. PUM target mRNAs weakly co-localize with PUM1 foci.

a, Confocal images of HCT116 cells co-stained for the indicated RNA and PUM1 by RNA FISH and immunofluorescence, respectively. b, Quantification of the indicated PUM target mRNA and PUM1 co-localization (n = 20 cells for each RNA). Mean co-localization shown above box plots. Boxes extend from 25^th to 75^th percentiles; middle line represents median. Whiskers extend from minimum to maximum co-localization for each target mRNA.

Extended Data Fig. 7 |. Competitive, super-stoichiometric recruitment of PUM into RNA-induced droplets.

a, Quantification of the number of PUM1 molecules per PRE in droplets nucleated by NORAD (2 nM) or PRE₈ oligonucleotide (10 nM) and the indicated concentration of PUM1 in vitro. Black line represents the mean (n = 101, 115, 113, 107, 114, or 115 droplets for each condition from left to right). b, Quantification of PUM1 partition coefficients formed with 150 nM PUM1 and 10 nM PRE₈ RNA normalized to partition coefficient at 0 μM competitor RNA (PRE₁ RNA). IC₅₀ represents concentration of PRE₁ RNA needed to reduce PUM1 partitioning by 50%. Data shown as mean ± SD. n = 44 or more droplets analyzed for each data point. Each PRE in the PRE₈ RNA is ~9 times more efficient at PUM1 recruitment than a monovalent PRE.

Extended Data Fig. 8 |. Purification and characterization of PUM1^HDmut and IDR-deletion proteins.

a, Structure of the human PUM1-HD domain (PDB: 1M8X) in complex with PRE-RNA, showing mutated residues in PUM1^HDmut (pink). b, Electrophoretic mobility shift assay (EMSA) demonstrating loss of PUM1^HDmut RNA binding. c, Coomassie stain of purified MBP-SNAP-PUM1^HDmut-His protein used for in vitro experiments. d, Droplet formation by the indicated PUM proteins in the presence or absence of PRE₈ RNA oligonucleotide. e, Coomassie stain of purified MBP-SNAP-PUM1^HD-WTΔIDR-His and MBP-SNAP-PUM1^HDmutΔIDR-His proteins used for in vitro experiments. Gels were cropped to remove irrelevant lanes where indicated with vertical lines. For gel source data, see Supplementary Figure 1.

Extended Data Fig. 9 |. Recruitment of PUM proteins into pre-formed NORAD-PUM condensates independently of RNA binding.

a, Western blot analysis of PUM1 (left) or PUM2 (right) in HCT116 cells of the indicated genotypes transduced with PUM^WT-GFP or PUM^HDmut-GFP expressing lentiviruses. Molecular weight in kDa shown on right of blots. For gel source data, see Supplementary Figure 1. b, Time-lapse live-cell confocal images showing fusion of PUM1^WT-GFP or PUM1^HDmut-GFP condensates in wild-type HCT116 cells. Scale bar, 5 μm. c, d, (Top) Images of PUM1^WT-GFP and PUM1^HDmut-GFP FRAP (c) or PUM2^WT-GFP and PUM2^HDmut-GFP FRAP (d) in wild-type HCT116 cells. Puncta undergoing photobleaching shown in dashed boxes. (Bottom) FRAP quantification with fluorescence intensities plotted relative to pre-bleach time point (t = −5 s). Data shown as mean ± SD (panel c; n = 3 puncta) or mean (panel d; n = 2 puncta). Scale bar, 5 μm. e, Confocal images of PUM2^WT-GFP (top) and PUM2^HDmut-GFP (bottom) in HCT116 cells of the indicated genotypes. f, (Left) Confocal images of GFP-tagged full-length or IDR-deleted PUM1^WT or PUM1^HDmut expressed in wild-type HCT116 cells. (Right) Quantification of PUM1 partition coefficients, defined as the intensity of PUM1-GFP inside condensates relative to the surrounding cytoplasm. Partition coefficients were calculated for n = 175, 116, 36, or 121 condensates from 5 different cells for each protein from left to right. Black bar depicts the mean partition coefficient. P values calculated using two-tailed t-test comparing each PUM1 mutant to PUM1^WT. ****, p < 1 × 10⁻¹⁵; n.s., not significant.

Extended Data Fig. 10 |. NORAD expression and PUM localization in circPRE-expressing cell lines.

a, Schematic of circPRE-producing constructs, which encode the Broccoli aptamer and 0–8 PREs. b, Copy number analysis of circPRE-transcripts in HCT116 CRISPRi cells expressing control or NORAD-targeting sgRNAs. Mean copy number in sgNORAD cell lines is shown above each bar. circPRE₄-low and circPRE₄-mid represent distinct cell populations sorted for different circPRE copy numbers. Data shown as mean ± SD. n = 3 biological replicates. c, qRT-PCR analysis of NORAD expression in the indicated circPRE HCT116 CRISPRi cell lines expressing control or NORAD-targeting sgRNAs. Data shown as mean ± SD. n = 3 technical replicates. d, e, (Top) Confocal images of PUM1 (d) or PUM2 (e) immunofluorescence in the indicated cell lines. (Bottom) Quantification of the number of PUM1 (d) or PUM2 (e) foci per cell in the indicated sgControl or sgNORAD-infected cell lines. (n = 20 cells for each cell line). Mean number of foci shown above each box plot. Boxes extend from 25^th to 75^th percentiles; middle line represents median; whiskers extend from minima to maxima.

Fig. 1 |. NORAD and PUM co-localize in cytoplasmic foci.

a, Confocal images of NORAD RNA FISH in HCT116 cells. b, PUM immunofluorescence in HCT116 cells of the indicated genotypes, imaged with structured illumination microscopy (SIM). c, SIM images of NORAD and PUM co-stained cells. d, Quantification of NORAD and PUM co-localization (n = 20 cells for each PUM protein). Mean co-localization shown below box plots. Boxes extend from 25^th to 75^th percentiles; middle line represents median; whiskers extend from minima to maxima. e, Live cell confocal images of endogenous PUM1-GFP. Dashed lines indicate nuclear boundaries. f, Images (left) and quantification (right) of PUM1-GFP FRAP. Fluorescence intensities plotted relative to pre-bleach time point (t = −5 s). Data shown as mean ± SD (n = 3 droplets).

Fig. 2 |. NORAD induces PUM phase separation through multivalent RNA binding.

a-e, Confocal images (a, c) or turbidity measurements (absorbance at 600 nm, A₆₀₀) (b, e) of 150 nM PUM1 solutions in the presence of increasing concentrations of the indicated RNAs or FAM-labeled PRE-containing RNA oligonucleotides (d). Turbidity normalized to A₆₀₀ at 0 nM RNA (3 replicates per RNA concentration). Red box, consensus PRE; blue box, mutant PRE.

Fig. 3 |. Competitive, IDR-driven recruitment of PUM into NORAD-PUM condensates.

a, (Top) Confocal images of PUM1 droplets (150 nM) formed with NORAD (2 nM) and the indicated concentrations of competitor RNA. (Bottom) Quantification of PUM1 partition coefficients, normalized to partition coefficient at 0 μM competitor RNA. IC₅₀, concentration of PRE₁ RNA that reduced PUM1 partitioning by 50%. Data shown as mean ± SD. n = 49 or more droplets per data point. Scale bar, 5 μm. b, 3D-reconstructed live cell confocal images of endogenously-tagged PUM1-GFP HCT116 cells. c, Quantification of PUM1 molecules in NP bodies or cytoplasm per cell (n = 20 cells). Mean number in each compartment shown above box plots. Boxes extend from 25^th to 75^th percentiles; middle line represents median; whiskers extend from minima to maxima. d, e, Time-lapse confocal images showing recruitment of the indicated wild-type or mutant PUM proteins (red) into preformed PUM1^WT-RNA droplets (green, PRE₈ RNA oligo). f, Confocal images of PUM1^WT-GFP or PUM1^HDmut-GFP localization in HCT116 cells of the indicated genotypes.

Fig. 4 |. PUM condensate-inducing RNAs limit PUM activity and rescue genomic instability in NORAD-deficient cells.

a, Confocal images of circPRE RNA FISH (red) and PUM1 immunofluorescence (green). b, Quantification of circPRE and PUM1 co-localization in sgNORAD cell lines (n = 20 cells for each circPRE cell line). Mean co-localization shown above box plots. Boxes extend from 25^th to 75^th percentiles; middle line represents median; whiskers extend from minima to maxima. c, Cumulative distribution plot of mRNA expression fold-changes in the indicated sgNORAD-infected cell lines compared to parental sgControl-infected cell line. PUM target genes indicated with gray or colored lines (n = 242 genes). Non-target genes in parental cells indicated with black line (n = 1415 genes). P values calculated using one-sided Kolmogorov-Smirnov (K-S) test comparing PUM targets in parental (gray), circPRE₂ (yellow), or circPRE₈ (red) to non-PUM targets (black). d, (Top) Representative images of DAPI-stained anaphase HCT116 CRISPRi cells. (Bottom) Frequency of mitotic cells exhibiting chromosome segregation defects in circPRE cell populations (n = 100 anaphase cells assayed per sample). Dotted line denotes highest percentage of segregation defects observed in sgControl-infected cells. P values calculated using two-tailed chi-square test comparing sgNORAD to sgControl for each circPRE population.