RESEARCH ARTICLE SUMMARY◥

MOLECULAR BIOLOGY

Imaging dynamic and selectivelow-complexity domain interactionsthat control gene transcriptionShasha Chong, Claire Dugast-Darzacq, Zhe Liu, Peng Dong, Gina M. Dailey,Claudia Cattoglio, Alec Heckert, Sambashiva Banala, Luke Lavis,Xavier Darzacq, Robert Tjian*

INTRODUCTION: DNA binding transcriptionfactors (TFs) are quintessential regulators ofeukaryotic gene expression. Early studies ofTFs revealed their well-structuredDNA bindingdomains (DBDs) and identified functionally crit-ical activation domains (ADs) required for tran-scription. It later became evident thatmanyADscontain intrinsically disordered low-complexitysequence domains (LCDs), but how LCDs activatetranscription has remained unclear. Althoughit is known that transcriptional activation byLCDs requires selective interaction with bind-ing partners, it has been challenging to direct-ly measure selective LCD-LCD recognition invivo and unravel its mechanism of action.

RATIONALE: Traditional biochemical recon-stitution and genetics studies have identifiedmost of the molecular players central to tran-scription regulation. However, the mechanismbywhichweak, dynamic protein-protein interac-tions drive gene activation in living cells hasremained unknown. Advances in live-cell single-molecule imaging have opened a new frontierfor studying transcription in vivo. In this study,we used synthetic LacO (Lac operator) arraysas well as endogenous GGAA microsatelliteloci to study LCD-LCD interactions of TFssuch as EWS/FLI1, TAF15, and Sp1 in livecells. To probe the dynamic behavior of TFLCDs at target genomic loci, we have combined

CRISPR-Cas9 genome editing, mutagenesis,gene activation, cell transformation assays,and various high-resolution imaging approachesincluding fluorescence correlation spectroscopy,fluorescence recovery after photobleaching, lat-tice light-sheet microscopy, three-dimensionalDNA fluorescence in situ hybridization, andlive-cell single-particle tracking.

RESULTS: Live-cell single-molecule imagingrevealed that TF LCDs interact to form localhigh-concentration hubs at both syntheticDNAarrays and endogenous genomic loci. TF LCDhubs stabilize DNA binding, recruit RNA poly-merase II (RNA Pol II), and activate transcrip-tion. LCD-LCD interactions within hubs arehighly dynamic (seconds tominutes), selective forbinding partners, and differentially sensitive to

disruption by hexanediols.These findings suggest thatunder physiological condi-tions, rapid, reversible, andselective multivalent LCD-LCD interactions occur be-tween TFs and the RNA

Pol II machinery to activate transcription. Weobserved formation of functional TF LCDhubsat a wide range of intranuclear TF concentra-tions. Although we detected apparent liquid-liquidphase separationwithgross overexpressionof LCDs, transcriptionally competent TF LCDhubswere observed at physiological TF levels atendogenous chromosomal loci in the absence ofdetectable phase separation. In addition, muta-genesis, gene expression, and cell transforma-tion assays in Ewing’s sarcoma cells revealeda functional link between LCD-LCD inter-actions, transactivation capacity, and oncogenicpotential.

CONCLUSION: The use of various imagingmethods in live cells powerfully complementsin vitro studies and provides new insights intothe nature of LCD interactions and their rolein gene regulation. We propose that trans-activation domains function by forming localhigh-concentration hubs of TFs via dynamic,multivalent, and specific LCD-LCD interactions.It also seems likely that weak, dynamic, andtransient contacts between TFs play a rolein disease-causing dysregulation of gene ex-pression (i.e., EWS/FLI1 in Ewing’s sarcoma),suggesting that LCD-LCD interactions mayrepresent a new class of viable drug targets.Although we examined a small subset of TFLCDs, the principles uncovered regardingthe dynamics and mechanisms driving LCD-LCD interactions may be applicable to otherclasses of proteins and biomolecular inter-actions occurring in many cell types.▪

RESEARCH

Chong et al., Science 361, 378 (2018) 27 July 2018 1 of 1

The list of author affiliations is available in the full article online.*Corresponding author. Email: jmlim@berkeley.eduCite this article as S. Chong et al., Science 361, eaar2555(2018). DOI: 10.1126/science.aar2555

From hubs to phase separation: Activation occurs in a wide range of TF concentrations. Invivo LCD-dependent transactivation occurs in hubs formed over a broad range of TFconcentrations(100 nM to 100 mM) and time scales (<1 s tominutes). At endogenous concentrations,TF LCDs formtransactivation hubs at native genomic loci without undergoing evident phase separation. Upon TFLCD overexpression, phase separation is observed at syntheticTF binding site arrays.

ON OUR WEBSITE◥

Read the full articleat http://dx.doi.org/10.1126/science.aar2555..................................................

on February 5, 2020

http://science.sciencem

ag.org/D

ownloaded from

RESEARCH ARTICLE◥

MOLECULAR BIOLOGY

Imaging dynamic and selectivelow-complexity domain interactionsthat control gene transcriptionShasha Chong1,2, Claire Dugast-Darzacq1,3, Zhe Liu4, Peng Dong4, Gina M. Dailey1,Claudia Cattoglio1,2, Alec Heckert1, Sambashiva Banala4, Luke Lavis4,Xavier Darzacq1,3, Robert Tjian1,2,3*

Many eukaryotic transcription factors (TFs) contain intrinsically disordered low-complexitysequence domains (LCDs), but how these LCDs drive transactivation remains unclear.We usedlive-cell single-molecule imaging to reveal that TF LCDs form local high-concentrationinteraction hubs at synthetic and endogenous genomic loci.TF LCD hubs stabilize DNA binding,recruit RNA polymerase II (RNA Pol II), and activate transcription. LCD-LCD interactionswithin hubs are highly dynamic, display selectivity with binding partners, and are differentiallysensitive to disruption by hexanediols. Under physiological conditions, rapid and reversibleLCD-LCD interactions occur between TFs and the RNA Pol II machinery without detectablephase separation. Our findings reveal fundamental mechanisms underpinning transcriptionalcontrol and suggest a framework for developing single-molecule imaging screens for drugstargeting gene regulatory interactions implicated in disease.

Sequence-specific DNA binding transcrip-tion factors (TFs) are preeminent playersin eukaryotic gene regulation. From theearliest studies of human TFs, it was rec-ognized that regulatory proteins such as

Sp1 contain well-structured DNA binding do-mains (DBDs) and functionally critical trans-activation domains that participate in specificTF-TF interactions to direct gene transcription(1–3). Numerous atomic structures of DBDs haveprovided a concrete understanding of TF-DNAinteractions. In contrast, many transactivationdomains contain low-complexity sequence do-mains (LCDs) that persist in an intrinsicallydisordered conformation not amenable to con-ventional structural determination. Mutationsin TF LCDs not only disrupt transcription butalso have been implicated in cancer and neuro-degenerative disorders (4, 5). However, themechanism by which TF LCDs execute specifictransactivation functions has remained an enig-ma. Elucidation of how LCDs operate in vivo,given the dynamic nature of TF-TF interactionsrequired for gene regulation, has been an equallychallenging problem.Several in vitro studies have suggested that

purified LCDs from the FET protein family (FUS,EWS, andTAF15) canundergo reversible hydrogel

formation or liquid-liquid phase separation athigh concentrations and low temperatures (6–8).Moreover, the C-terminal domain of RNA poly-merase II (RNA Pol II) is itself an LCD and canundergo phase separation (9) and be incorporatedinto FET LCD hydrogels in a phosphorylation-regulated manner (10). FET LCDs were also re-ported to undergo phase separation in live cellsupon overexpression (7, 11).However, stark differences exist between

in vivo physiological conditions and those usedfor in vitro or overexpression studies. Temper-ature, protein concentration, purity, and micro-environment may all substantially affect thebehavior of LCDs. There is also a vigorous debateas to whether LCDs undergo cross-b polymer-ization or remain in a disordered conformationwhen interacting with partners (6–8, 11–17).From the perspective of elucidating how TFswork in vivo, an equally pressing unresolvedmechanistic question concerns the dynamicsand time scales governing LCD-LCD interac-tions that would allow TFs to function in rapidcellular processes. Selectivity of cognate LCD-LCD interactions is another important yet poorlyunderstood feature that is required for properTF function in vivo. Thus far, selective LCD-LCDrecognition has not been directly demonstratedin vivo, let alone understood at a mechanisticlevel. In this study, we combined a variety of high-resolution imaging strategies—including fluo-rescence correlation spectroscopy (FCS) (18, 19),fluorescence recovery after photobleaching (FRAP)(20), lattice light-sheet microscopy (21), three-dimensional (3D)DNA fluorescence in situ hybrid-ization (FISH) (22), and live-cell single-particle

tracking (SPT) (23, 24)—to probe the dynamicbehavior of TF LCDs at target genomic loci underphysiological conditions.

Synthetic LacO arrays mediateformation of LCD interaction hubs

We first established proof-of-concept experi-ments by using a synthetic Lac operator (LacO)array (~50,000 LacO repeats) integrated into thegenome of human U2OS cells (25) that expressvarious enhanced yellow fluorescent protein(EYFP)–tagged TF LCDs fused to the Lac repressor(LacI) (Fig. 1A). To probe potential sequencespecific LCD-LCD interactions, we examined twodistinct classes of LCDs: QGYS-rich LCDs fromthe FET family and a QGTS-rich LCD from Sp1that is low in tyrosine (table S1) (Q, Gln; G, Gly; Y,Tyr; S, Ser; T, Thr).As expected, the LacO array recruits a large

number of EYFP-LCD-LacImolecules via targetedDNA binding, forming a concentrated local inter-action hub in the nucleus. LacO-associated hubsformed by LCD-LacI but not LacI are visible bybright-field microscopy (Fig. 1B), suggesting thatthe refractive index and mass density of LCD-LacI hubs differ considerably from the surround-ing nuclear environment.We found that both TAF15-LacI and Sp1-LacI

give rise to much brighter and larger LacO-associated LCD hubs than LacI alone. To quan-tify this effect, we used two orthogonal methodstomeasure the protein concentrations in live cells.Specifically, we performed fluorescence intensityand FCS measurements of intracellular proteinand compared the results with standard concen-tration curves of a purified fluorescent tag (EYFPor mCherry) calibrated by the same method (fig.S1). Next, we estimated the absolute protein copynumber in each hub by using the average in-hubprotein concentration (fig. S2A) and hub dimen-sions measured in single-cell images (fig. S2B).The two independent concentration measure-ments consistently showed that the LCD-LacIcopy number in the LacO-associated hub increaseswith the TF nuclear concentration much fasterthan with LacI alone and reaches levels up toorders of magnitude greater than the numberof LacO repeats available for direct LacO-LacIbinding (Fig. 1C and fig. S2, C and D), suggest-ing likely cooperative multivalent interactionsat the LacO array that is contributed by extensiveLCD self-interactions. Similarly, smaller LacO ar-rays containing substantially fewer (~15,000) LacOrepeats similarly nucleate LCD self-interactions(Fig. 1D and fig. S2, E and F).Moreover, LCD-LacI but not LacI alone can

form hundreds of smaller puncta throughoutthe nucleus once its intranuclear concentrationreaches a certain threshold (Figs. 1B and 2A, andfig. S2, G and H). LCDs can form intranuclearpuncta in some cases evenwithout being fused toa DBD such as LacI (Fig. 3A, bottom). These re-sults suggest that LCD-LCD interactions canpromote self-assembly of LCD hubs upon over-expression without assistance from DNA (7, 11).In addition, FRAP dynamics of LCD-LacI at

the LacO array was also significantly different

RESEARCH

Chong et al., Science 361, eaar2555 (2018) 27 July 2018 1 of 9

1Department of Molecular and Cell Biology, University ofCalifornia, Berkeley, CA, USA. 2Howard Hughes MedicalInstitute, University of California, Berkeley, CA, USA. 3CIRMCenter of Excellence, University of California, Berkeley, CA,USA. 4Janelia Research Campus, Howard Hughes MedicalInstitute, Ashburn, VA, USA.*Corresponding author. Email: jmlim@berkeley.edu

on February 5, 2020

http://science.sciencem

ag.org/D

ownloaded from

from that of LacI (Fig. 1E). Because diffusioncontributes negligibly to the FRAPdynamics (fig.S3, A and B), such differences can be attributedto changes in dissociation rates. Specifically, whenwe fit the FRAP curves with a reaction-dominantmodel (26), we found that fusing TAF15 or FUSLCD to LacI leads to more than 60% reduction inthe dissociation rate constant of LacI (fig. S3, Cto F). This result suggests that at increased localTF concentrations, TF LCD hubs driven by LCD-LCD interactions stabilize TF binding to itscognate genomic site via multivalent contactsthat could include both DBDs for tethering tochromatin and one or more LCDs within TFproteins that can form multiple transient inter-actions with different partner proteins.

LCD hubs interact with RNA Pol IIHaving demonstrated homotypic LCD self-interactions, we next investigated the potentialrole of heterotypic LCD-LCD interactions in hubformation. First, we tested whether TF LCD hubscan interact with RNA Pol II in vivo by using aLacO-containing U2OS line in which we replacedthe endogenous RPB1 (major and catalytic sub-unit of RNA Pol II) with an a-amanitin–resistantHalo-tagged RPB1 (27). We subsequently labeledthe cells with a fluorescent HaloTag ligand andvisualized RNA Pol II distribution in vivo. Wefound that mCherry-FET-LacI expression mediatessignificant enrichment of RNA Pol II in LacO-associated hubs compared with background levelsrecorded using LacI alone (Fig. 1, F and G, and

fig. S4, A to C). Moreover, self-assembled LCD-LacI hubs that are unaffiliated with a LacO arrayalso enrich RNA Pol II (fig. S4D), suggesting thatLCD hubs interact with RNA Pol II potentiallywithout assistance from DNA. While recapitulat-ing the in vitro incorporation of RNA Pol II intoLCD hydrogels (10), these experiments go one stepfurther and suggest that LCD hub formation canfacilitate the recruitment of the general transcriptionmachinery in vivo—a key step toward transactivation.

LCD-LCD interactions are sequencespecific and differentially sensitive tohexanediol disruption

Next, we probed the sequence specificity of inter-actions between various classes of LCD. To this

Chong et al., Science 361, eaar2555 (2018) 27 July 2018 2 of 9

Fig. 1. A LacO array can mediate the formationof an LCD hub in live cells, which involvesextensive LCD self-interaction and recruits RNAPol II. (A) Schematic for a LacO array (n ≈ 50,000repeats for array 1, n ≈ 15,000 repeats for array 2)in the U2OS genome nucleating an LCD hubwhen EYFP-LCD-LacI is transiently expressed.Alternatively, EYFP-LacI is expressed as a control.NLS, nuclear localization signal. (B) Confocalfluorescence and bright-field images ofLacO-containing U2OS cells where LacO array 1(highlighted by circles) is bound by EYFP-labeledLCD-LacI or LacI. LCD-LacI–bound, but notLacI-bound, LacO arrays are visible in bright-fieldimages. (C and D) Copy number of EYFP-labeled(C) or mCherry-labeled (D) TAF15 LCD-LacI(red) or LacI (blue) molecules bound to LacOarray 1 (C) or 2 (D) as a function of mean nuclearconcentration of the TF. Concentrations weremeasured by fluorescence intensity comparison.Each dot represents one cell. (E) Averaged FRAPcurves at LacO array 1 bound by mCherry-labeledTAF15 LCD-LacI (red) or LacI (blue). Errorbars represent SD. a.u., arbitrary units;N, number of cells analyzed. (F) (Top) Schematic ofthe proteins expressed in the LacO-containingU2OS line. (Bottom) Confocal fluorescenceimages show that Halo-RPB1 (labeled with200 nM Halo ligand JF500, green) is enriched atLacO array 2 bound by mCherry-EWS LCD-LacI(red). (G) (Left) Averaged Halo-RPB1 imagesat LacO array 2 bound by mCherry-labeledLacI, EWS LCD-LacI, FUS LCD-LacI, or TAF15LCD-LacI (N = 55, 69, 81, or 143). (Right)Fluorescence intensity of Halo-RPB1 at the LacOarray center in the average images aftersubtraction of nuclear Halo-RPB1 background(see supplementary methods). ** denotes astatistically significant increase compared withthe LacI condition (P < 0.01, two-sample t test).Error bars represent bootstrapped SD (45).

RESEARCH | RESEARCH ARTICLEon F

ebruary 5, 2020

http://science.sciencemag.org/

Dow

nloaded from

end, we coexpressed both EYFP-LCD-LacI andmCherry-LCD in LacO-containing cells. mCherry-LCD lacking aDBDbecomes enriched at the arrayonly when it can interact with the coexpressedLCD that is fused to LacI (Fig. 2A and fig. S5A).Notably, the array can enrich mCherry-LCD overawide range of expression levels. The EWS-LacI–bound LacO array also enriches endogenousEWSR1, as detected by immunofluorescence (fig.S5B). Therefore, mCherry-LCD enrichment at

the array is most likely due to specific LCD-LCDinteractions rather than potential nonspecificoverexpression artifacts. Using this two-colorimaging assay, we confirmed homotypic self-interactions of all tested LCDs (from the FETfamily and Sp1). Intriguingly, although all threeFET LCDs interacted among themselves, none ofthem interacted with the Sp1 LCD (Fig. 2, A andB), suggesting that LCD interactions exhibitstrong sequence specificity that is likely an es-

sential feature underlying combinatorial TF reg-ulation of gene expression.To better understand the nature of LCD-LCD

interactions, we treated cells with 1,6-hexanediol(1,6-HD), an aliphatic alcohol known to dissolvevarious intracellularmembraneless compartmentsand FUS hydrogels in vitro through disruption ofhydrophobic interactions (28–30). We observedthat both FUS and Sp1 LCD hubs rapidly dis-assemble within 30 s when exposed to 10% 1,6-HD.TheLacO-associatedLCD–inducedhubshrankto a size comparable to that of the array boundby LacI alone, whereas all nuclear puncta notassociated with LacO disappeared (Fig. 2C andMovie 1). We also found that 2,5-hexanediol (2,5-HD), a less hydrophobic derivative of 1,6-HD thatbarelymelts FUS hydrogels in vitro (28), disruptsLCD hubs less efficiently in live cells (fig. S5, Cand D). This correlation between hydrophobicityof hexanediols and LCD hub melting suggeststhat these aliphatic alcohols may directly in-fluence LCD-LCD interactions by disrupting keyhydrophobic contacts. These in vivo results alsomirror in vitro hydrogel studies using these samedisrupting agents (28).Sp1 LCD hubs were disrupted significantly

faster and more extensively than FUS LCD hubswith 2 or 5% 1,6-HD (Fig. 2D). Thus, although acombination of intermolecular forces may con-tribute to LCD hub formation, our results in-dicate that hydrophobic interactions might bemore sensitive to disruption and play a moredominant role in Sp1 LCD self-interactions thanFUS LCD, consistent with the Sp1 LCD contain-ing hydrophobic residues sparsely interspersedamong Q repeats (31). The differential sequencedependence of LCD-LCD interactions revealedby 1,6-HD treatment may be correlated to theselectivity of homo- and heterotypic LCD inter-actions observed above.

LCD-LCD interactions arehighly dynamic

To study the dynamics of protein-protein inter-actions between LCD pairs, we coexpressed

Chong et al., Science 361, eaar2555 (2018) 27 July 2018 3 of 9

Fig. 2. LCD hub formation involves selective protein-protein interactions, which can bedisrupted by 1,6-HD with sequence-dependent sensitivity. (A) Confocal fluorescenceimages of U2OS cells containing LacO array 1 that coexpress various combinations ofmCherry-LCD and EYFP-LCD-LacI. The region surrounding the LacO array is zoomed in.(B) Quantification of the enrichment of mCherry-LCD (red) at the LacO array 1 bound by variousEYFP-labeled LCD-LacI fusion proteins (green), calculated as the peak mCherry fluorescenceintensity at the array divided by the average intensity immediately surrounding the array(fig. S5A). Null, mCherry not fused to any LCD. An mCherry enrichment at the array above1 suggests LCD-LCD interactions. * denotes a statistically significant difference above 1 (P < 0.05,one-sample t test). NS, nonsignificant difference above 1. Error bars represent SE. (C) Fluorescenceimages of FUS and Sp1 LCD hubs before (0 s) and after (29 s) addition of 10% 1,6-HD.(D) Number of nuclear puncta formed by FUS or Sp1 LCD surviving over time upon additionof 1,6-HD at different concentrations. Error bars represent SE.

Movie 1. LacO array–associated and self-aggregated non-array–affiliated FUS LCDhubs in live cells are quickly disrupted by10% 1,6-hexanediol (related to Fig. 2C).

RESEARCH | RESEARCH ARTICLEon F

ebruary 5, 2020

http://science.sciencemag.org/

Dow

nloaded from

EYFP-LCD-LacI and Halo-LCD in the LacO-containing U2OS cells and performed SPT ofHalo-LCD to measure residence times (RTs) ofLCD-LCD interactionswithin the LacO-associatedhub (Fig. 3A, top). For all LCDs tested, RTs re-sulting from self-interactions fell in the range of11 to 33 s (Fig. 3B). When EYFP-LCD and Halo-LCD from the FET family were coexpressed athigh levels, they spontaneously formed hubs thatare unaffiliated with the array and resemble in-tranuclear puncta (Fig. 3A, bottom). These non-arrayhubs bindHalo-LCDviahomo- or heterotypicinteractions with even shorter RTs (7 to 10 s). Asexpected, the Sp1 LCD that failed to interact withthe FUS LCD had an RT of <1 s at the non-arrayFUS LCD hubs (Fig. 3B). The fact that RTs ofmany LCDs in self-aggregated hubs unaffiliatedwith genomic DNA are substantially shorter thanin hubs formed at the LacO array suggests thatTF-DNA interactions that maintain a high localconcentration of TFLCDs contribute to stabilizingLCD-LCD interactions and vice versa. Together,these findings reveal the rapid, reversible, andinterdependent nature of LCD-LCD and TF-DNAinteractions as well as their propensity to formlocal high-concentration hubs that likely stabilizemulticomponent complexes—e.g., transcription pre-initiation complex, aprerequisite for transactivation.

EWS/FLI1 forms hubs at endogenousGGAA microsatellites

Having unmasked the sequence specificity anddynamic nature of LCD-LCD interactions by usingsynthetic LacO arrays in living cells, we nexttested LCD behavior at native GGAA micro-satellites (>20 GGAA repeats) in the Ewing’ssarcoma cell line A673 (32–35). These cancer-derived cells carry a chromosomal translocationt(11;22)(q24;q12) producing a fusion oncogene,EWS/FLI1, that encodes a potent TF consistingof a trans-activating LCD from EWSR1 and theDBD from FLI1 that targets GGAA sequences(Fig. 4A).To visualize the behavior of endogenously

expressed EWS/FLI1, we fused a HaloTag to itsDBD using CRISPR-Cas9–mediated genome edit-ing of A673 cells (Fig. 4, A and B, and fig. S6A)(36). This knock-in strategy allowed us to imagefluorescently tagged endogenous EWS/FLI1 atits normal expression levels (Fig. 4B), which wasessential because LCDs tend to self-aggregateand behave aberrantly upon overexpression. Toensure that Halo-tagging does not disrupt trans-activation functions of EWS/FLI1, we confirmedthat EWS/FLI1-Halo activates a luciferase reporterconstruct containing a GGAA microsatellite–driven promoter (33) as efficiently as wild-type(WT) EWS/FLI1 (fig. S6C). More importantly,using the gold standard neoplastic transforma-tion assay (37), we confirmed that the EWS/FLI1-Halo knock-in A673 cells form colonies insoft agar much like the WT A673 cells—albeit,less efficiently (fig. S6, D and E).We next performed high-resolution lattice

light-sheet microscopy and found that EWS/FLI1 forms many small interaction hubs (>1000per nucleus) in the nucleus (Fig. 4C andMovie 2).

The detected number of intranuclear hubs hasthe same order ofmagnitude as the total numberof EWS/FLI1-bound GGAAmicrosatellites acrossthe human genome (~6000) estimated by chro-matin immunoprecipitation sequencing and bio-informatics analyses (38). To examine the spatialrelationship between EWS/FLI1 hubs and GGAAmicrosatellites, we performed simultaneous con-focal imaging of EWS/FLI1-Halo and 3D DNAFISH targeting genes adjacent to GGAA micro-satellites that are regulated by EWS/FLI1 (Fig.4D), including CAV1, FCGRT,ABHD6,KDSR, andKIAA1797 (33, 39). Although EWS/FLI1 enrich-ment is detected atmany single loci of these genes,the crowded distribution of intranuclear EWS/FLI1 hubs makes it difficult to clearly visualizeEWS/FLI1 enrichment at single target loci. Byrecording images of ~1000 loci for each gene,the signal-to-noise ratio is markedly improvedto reveal specific EWS/FLI1 enrichment at GGAArepeats, whereas no enrichmentwas seen at non-GGAA gene loci (Fig. 4E). These results indicate

that EWS/FLI1 forms hubs at endogenous GGAAmicrosatellite DNA elements.

Dynamic EWS LCD-LCD interactionsmediate formation of EWS/FLI1 hubs

We previously showed that because of LCD-LCDinteractions, the LCD-LacI copy number in theLacO-associated hub increases much faster withthe TF concentration than that of LacI alone,reaching levels that are orders of magnitudehigher than the number of available TF bind-ing sites at the LacO array (Fig. 1, C and D, andfig. S2, C to F). Using the methods establishedearlier, we estimated the intranuclear concentra-tion of endogenously Halo-tagged EWS/FLI1 inA673 cells to be ~200 nM and the median copynumber of EWS/FLI1 per hub at GGAA micro-satellites to be 24 (fig. S7, A and B). It was re-ported thatmost EWS/FLI1-boundmicrosatellitescontain 11 to 19 GGAA motifs, with a medianaround 15 (38). Because the DBD of FLI1 occupiestwo consecutive GGAA repeats (33), the median

Chong et al., Science 361, eaar2555 (2018) 27 July 2018 4 of 9

Fig. 3. LCD-LCD interactions involved in hub formation are highly dynamic. (A) Snapshotsof a two-color SPTmovie simultaneously imaging EYFP-labeled (green) EWS LCD-LacI (top,forming a LacO-associated LCD hub) or EWS LCD (bottom, forming self-aggregated LCD hubsnot affiliated with the LacO array) and Halo-tagged EWS LCD (2 nM PA-JF646 labeled, red) inU2OS cells containing LacO array 1. A white dashed contour outlines the cell nucleus.We imaged thehubs in the EYFP channel (green) and tracked individual Halo-EWS LCD molecules with anacquisition time of 500 ms in the PA-JF646 channel (red). (B) Residence times of LCD (red) boundat the LacO-array–associated LCD hub or at self-aggregated LCD hubs not affiliated with thearray (green). *P < 0.05, two-sample t test. Error bars represent SE.

RESEARCH | RESEARCH ARTICLEon F

ebruary 5, 2020

http://science.sciencemag.org/

Dow

nloaded from

number of EWS/FLI1 molecules recruited to amicrosatellite via direct DNA-protein interactionis estimated to be 8. The fact that there are asignificantly greater number of EWS/FLI1 mole-cules in a GGAA-affiliated hub than what directbinding to GGAA repeats could accommodatesuggests that, like the synthetic LacO arrays, thesenative genomic elements can also efficientlynucleate EWS LCD-LCD interactions to formlocal high-concentration TF hubs.Moreover, whentransiently expressing EWS/FLI1-Halo or a Halo-tagged LCD deletion mutant (FLI1 DBD) in U2OScells, we again observe that the total number ofEWS/FLI1 molecules in the GGAA-affiliated hubsincreasesmuch faster than FLI1DBDas a functionof TF concentration (fig. S7, C and D), providing

strong evidence that LCD-LCD interactions occurat GGAA-affiliated EWS/FLI1 hubs.We previously observed that the formation of

anLCD interaction hub slows downdissociation ofLCD-LacI from the LacO array (Fig. 1E and fig.S3F). If LCD-LCD interactions are also involvedin EWS/FLI1 hub formation at GGAA micro-satellites, we expect the RT of EWS/FLI1 withinGGAA-affiliated hubs to be longer than that ofEWS/FLI1 outside hubs. We stained the EWS/FLI1-Halo knock-in A673 cells with two fluores-cent ligands (40, 41): High-concentration JF549staining allows visualization of EWS/FLI1 hubsin the cell nucleus, whereas low-concentrationPA-JF646 staining allows real-time tracking ofindividual EWS/FLI1 molecules (Fig. 5A and

Movie 3). SPT revealed the average RTs of EWS/FLI1 in and outside the GGAA-affiliated hubsto be 90 and 16 s, respectively (Fig. 5B and fig. S8,A and B). The fact that EWS/FLI1 binds to GGAArepeats for a significantly longer time againsuggests that EWS LCD-LCD interactions areinvolved in the formation of the GGAA-affiliatedhubs. It is very likely that both LCD-LCD inter-actions and DBD binding to GGAA repeats worktogether to stabilize hub formation, much as weobserved for LCD-LacI at the LacO array.To confirm that hub formation in this native

setting is dependent on the EWS LCD, we de-termined howmutations in the LCDmight affectRTs of EWS/FLI1. We started by replacing dif-ferent numbers (m = 3, 7, 10, 17, or 29) of tyrosines

Chong et al., Science 361, eaar2555 (2018) 27 July 2018 5 of 9

Fig. 4. Combined DNA FISH and EWS/FLI1-Halo imaging show thatendogenous EWS/FLI1 forms hubsat GGAA microsatellites. (A) Schematicfor GGAA microsatellites in theA673 genome nucleating hubs ofendogenously Halo-tagged EWS/FLI1.(B) Western blot of EWS/FLI1 and b-actin(normalization control) from clonalEWS/FLI1-Halo knock-in (KI), WT andclonal EWS/FLI1 knockout (KO) A673lines. (C) z-projected 3D image ofendogenous EWS/FLI1-Halo in an A673cell nucleus (stained with 200 nM Haloligand JF549) taken on the latticelight-sheet microscope. (D) Confocalfluorescence images of 3D DNA FISHtargeting GGAA microsatellite–adjacentCAV1 gene (enhanced Cy5 labeled,red) and endogenous EWS/FLI1-Halo(JF549 labeled, green). The zoomed-inviews depict the region surroundingone particular CAV1 locus. EWS/FLI1-Halo enrichment at the locusis visible but buried in high background.(E) Averaged two-color imagesof five GGAA microsatellite–adjacentgene loci (CAV1, FCGRT, ABHD6, KDSR,and KIAA1797) and two gene loci notcontaining a GGAA microsatellite(Non-GGAA locus 1 targeting ADGRA3and locus 2 targeting REEP5). Theright column shows average surfaceplots of EWS/FLI1-Halo.

RESEARCH | RESEARCH ARTICLEon F

ebruary 5, 2020

http://science.sciencemag.org/

Dow

nloaded from

(Y) in the EWSLCD (residues 47 to 266 of EWSR1)with serines (S) and then testing the self-interaction capability of mutant LCDs [EWS(YSm)] using the LacO array assay establishedearlier. As previously shown, when we coex-

pressed EYFP-EWS-LacI and mCherry-EWS,the mCherry signal became enriched at theLacO array owing to EWS LCD self-interaction.Notably, when we replaced WT EWS in bothfusion proteins with EWS(YSm), mCherry enrich-

ment at the array progressively decreased withan increasing number of Y-to-S mutations (fig.S9A) and vanished for EWS(YS29), in which alltyrosines are replaced (Fig. 5C). Similarly, wefound that EWS(YS29) does not interact with

Chong et al., Science 361, eaar2555 (2018) 27 July 2018 6 of 9

Fig. 5. Dynamic LCD-LCD interactions occur at GGAAmicrosatellites,which stabilize EWS/FLI1 binding and drive its transactivation function.(A) Snapshots of an SPTmovie imaging endogenous EWS/FLI1-Halo labeled withtwo Halo ligands, JF549 (200 nM) and PA-JF646 (20 nM).We imaged theEWS/FLI1-Halo hubs in the JF549 channel (green) and tracked individual EWS/FLI1-Halo molecules in and outside the hubs in the PA-JF646 channel (red).(B) Residence times of EWS/FLI1 bound in hubs are longer than its residencetimes outside hubs, as determined by SPT (**P < 0.01, two-sample t test). Errorbars represent SE. (C) EWS LCD is enriched at LacO array 1 bound by EWSLCD-LacI, but EWS(YS29) LCD is not recruited to the array by EWS(YS29) LCD-LacI. However, EWS(YF29) LCD is recruited to the array by EWS(YF29) LCD-LacI.(D) (Top) Schematic of proteins transiently expressed in EWS/FLI1 KOA673 cells:Halo-tagged EWS/FLI1, EWS(YS)/FLI1, or FLI1 DBD. (Bottom) Residence times ofEWS/FLI1 and its variantsbinding in andoutside their hubs, asdeterminedbySPT.*P < 0.05, two-sample t test. Error bars represent SE. (E) Snapshots of an SPT

movie simultaneously imaging SNAPf-tagged EWS/FLI1 (200 nM JF549 labeled,green) and Halo-tagged EWS or EWS(YS) LCD (20 nM PA-JF646 labeled, red) inEWS/FLI1 KOA673 cells. Individual LCD-Halo molecules were tracked with thestrategy described in (A). (F) Residence times of EWS bound at EWS/FLI1 hubsare longer than for EWS(YS) LCD, as determined by SPT (*P < 0.05, two-samplet test). Error bars represent SE. (G) Luciferase assay shows that EWS/FLI1 butnot EWS(YS)/FLI1 or FLI1 DBD transactivates a GGAAmicrosatellite–drivenreporter (**P < 0.01, two-sample t test). Error bars represent SE. (H) RT-qPCRshows down-regulation of GGAAmicrosatellite–associated EWS/FLI1 targetgenes in A673 cells upon EWS/FLI1 KO. Stable expression of exogeneous (Exo)EWS/FLI1, but not of the mutant EWS(YS)/FLI1, rescues the expression defect inEWS/FLI1 KOA673 cells. For each target gene, the mRNA level was normalizedusing five different invariant genes (fig. S10A) and graphed as a fold changerelative to the mRNA level present in the WTA673 line (set to 1). *P < 0.05,two-sample t test. NS, not statistically significant. Error bars represent SD.

RESEARCH | RESEARCH ARTICLEon F

ebruary 5, 2020

http://science.sciencemag.org/

Dow

nloaded from

WTEWS (fig. S9, B and C). By contrast, a mutantreplacing all 29 tyrosines with phenylalanine (F)[EWS(YF29)] retains hub formation activity withitself and with WT EWS (Fig. 5C and fig. S9D),suggesting that aromatic amino acids and hydro-phobic contacts represent major drivers of EWSLCD-LCD interactions. The conformational dis-order, proline residues, and acidic amino acids ofthe EWS LCD might play a role in keeping thesekey hydrophobic residues exposed to solventand potential binding partners without beingsequestered in a collapsed state (42).Next, we probed the effects of mutations that

disrupt LCD hub formation on RTs of EWS/FLI1.To examine behaviors of EWS/FLI1 variants inA673 cells without interference of endogenousEWS/FLI1, we generated an EWS/FLI1 knockoutA673 line using CRISPR-Cas9–mediated genomeediting (Fig. 4B and fig. S6B) and verified thattransiently and moderately reexpressed EWS/FLI1-Halo in the knockout line exhibited bindingdynamics comparable to that of endogenousEWS/FLI1-Halo (Fig. 5, B and D, and fig. S8C).We then transiently expressed similar levelsof FLI1 DBD-Halo or a Halo-tagged 37-residueY-to-S mutant [EWS(YS)/FLI1]. Both mutantsstill displayed some hubs in the nucleus, but thehubs are considerably diminished. SPT revealedthat the in-hub RTs of the mutants becomesignificantly reduced (by 51 to 65%) relative tothose of WT EWS/FLI1, whereas their outside-hub RTs remain largely unchanged (Fig. 5D).Together, these results confirm that LCD-LCDinteractions drive the formation of EWS hubs atGGAA microsatellites.To measure the dynamics of just the protein-

protein interactions occurring within the EWSLCDhubs,we transiently expressed SNAPf-taggedEWS/FLI1 andHalo-taggedEWS in the EWS/FLI1knockout line and labeled both fusion proteinsusing fluorescent ligands with distinct emissionspectra (40, 41).Whereas EWS/FLI1-SNAPf formshubs at GGAA microsatellites via protein-DNAbinding, EWS LCD-Halo, which does not interactwith DNA, binds to EWS/FLI1 hubs only viaprotein-protein interactions. We visualized EWS/

FLI1 hubs and simultaneously tracked individualEWS LCDmolecules that bind to the hubs (Fig.5E). SPT revealed the average RT of EWS LCD inEWS/FLI1 hubs to be 16 s, which suggests thatLCD-LCD interactions are highly dynamic (Fig.5F). As expected, the mutant EWS(YS) LCD hasa significantly shorter RT (~7 s) at EWS/FLI1hubs, consistent with its diminished interactionwith the EWS LCD.

EWS LCD-LCD interactions are essentialfor transcription and transformationfunctions of EWS/FLI1

Finally, we testedwhether LCD-LCD interactionsinfluence EWS/FLI1 functions. We found thatwhereas EWS/FLI1 efficiently induces gene ac-tivation at a GGAA microsatellite in a luciferaseassay, the mutant EWS(YS)/FLI1 and FLI1 DBDdo not (Fig. 5G). We further engineered the EWS/FLI1 knockout A673 line to stably express EWS/FLI1 or EWS(YS)/FLI1 and performed reversetranscription quantitative polymerase chain re-action (RT-qPCR) tomeasure the expression levelsof GGAA microsatellite–associated EWS/FLI1target genes. As expected, we found that expres-sion of EWS/FLI1, but not EWS(YS)/FLI1, specif-ically rescues the gene expression defect in theknockout line, indicating that EWS LCD-LCDinteractions are required for transactivation (Fig.5H and fig. S10A). Moreover, the knockout linestably expressing EWS/FLI1, but not EWS(YS)/FLI1, forms colonies in soft agar like WT A673cells (fig. S10B). This demonstrates that EWSLCD-LCD interactions are required for oncogenictransformation. Taken together with previouslypublished RT-qPCR and RNA sequencing datain mesenchymal stem cells showing the impor-

tant role of EWS LCD in inducing expression ofGGAA microsatellite–associated genes (43), ourresults suggest that the formation of EWS LCD-dependent hubs is essential for EWS/FLI1 toactivate transcription of these target genes anddrive oncogenic gene expression programs inEwing’s sarcoma.

Discussion

DNA binding TFs are key regulators of eukaryoticgene expression. Early studies of TFs revealedtheir well-structured DBDs and identified func-tionally critical activation domains (ADs) requiredfor transcription. It later became evident thatmany transactivation sequences contain intrin-sically disordered LCDs, but how they mediatetransactivation remained unclear. Despite thecomposition of LCDs and their generally unstruc-tured nature, ADs must interact with specificbinding partners to activate transcription. How-ever, it has been challenging if not impossibleto directly measure selective LCD-mediated tar-get recognition by ADs in vivo. Two other keyaspects of transcriptional control mechanismshave also remained largely uncharted: What isthe nature of the protein-protein transactionsthat drive gene activation, and how stable or dy-namic are these critical interactions in living cells?We addressed these problems by employing

single-molecule imaging to visualize LCD-LCDinteractions in live cells. Our findings indicatethat TF-TF interactions are extremely transient,with RTs of 5 to 20 s and rarely longer than1 min. These studies lead us to propose thattransactivation domains function by formingtransient local regions of high TF concentra-tions or “hubs” (sometimes also referred to asclusters) via dynamic, multivalent, and sequence-specific LCD-LCD interactions. A key future en-deavor will be to unlock the biochemical andstructural basis for selective LCD-LCD inter-actions. Such selectivity is likely essential toimplement the complex combinatorial logic oftranscriptional regulation.Recent advances in live-cell single-molecule

imaging have not only opened a new frontierfor studying transcription in vivo but have alsoprovided an opportunity to gain greater insightinto the nature and behavior of LCDs and theirpropensity to remain largely as intrinsically dis-ordered peptide sequences. By deploying high-resolution single-molecule imaging of TF LCDinteractions in live cells, our studies offer a pow-erful complement to pioneering in vitro studiesthat provided the first clues about LCD inter-actions (10). Importantly, to the extent that onecan make comparisons between hydrogels andintracellular LCD hub formation, many aspects ofFET-LCD function uncovered in vitro are borneout within live cells. For example, the behaviorof LCDs upon mutations, disruption by hexane-diols, and interaction with RNA Pol II observedin vivo are generally consistent with the role ofLCDs proposed on the basis of in vitro assays.In addition, single-molecule live-cell imaging re-vealed several new aspects of LCD-driven inter-actions. Most notable are the fast dynamics and

Chong et al., Science 361, eaar2555 (2018) 27 July 2018 7 of 9

Movie 2. 360° view of a 3D image ofendogenous EWS/FLI1-Halo in an A673 cellnucleus (stained with 200 nM Haloligand JF549) taken on the latticelight-sheet microscope (related to Fig. 4C).

Movie 3. Two-color movie imagingendogenous EWS/FLI1-Halo in a knock-inA673 cell nucleus stained with twoHaloTag ligands, JF549 (200 nM)and PA-JF646 (20 nM). The image acquisitiontime was 500 ms. We took time-lapseimages with a 10-s interval in the JF549channel to visualize EWS/FLI1 hubs andimaged continuously in the PA-JF646 channelfor SPT of individual EWS/FLI1-Halomolecules (related to Fig. 5A).

RESEARCH | RESEARCH ARTICLEon F

ebruary 5, 2020

http://science.sciencemag.org/

Dow

nloaded from

sequence specificity of LCD interactions as theyform transient local high-concentration hubs thatdrive transactivation (Fig. 6). We were also in-trigued by the formation of LCD-dependent hubsthroughout the nucleoplasm that are not asso-ciated with cognate genomic DNA. These LCDinteraction-driven puncta, which display a ca-pacity to interact with RNA Pol II, sensitivity to1,6-HD, and fast dynamics (recovery times of 7to 10 s), may provide an opportunity in futurestudies to further probe mechanisms govern-ing LCD-LCD–mediated hub formation andtransactivation.To analyze the behavior of LCD-LCD inter-

actions in live cells, we exploited the advantagesof various imaging platforms and developed twodistinct in vivo cell-based assays: one involvinglarge synthetic TF binding site (LacO) arrays andthe other targeting endogenous GGAA micro-satellite regulatory elements in the human ge-nome.When used in combination, the two assaysafford powerful and complementary platforms forprobing the properties of TF LCD interactions inthe live-cell context. The highly reiterated LacObinding sites can serve as a useful cell-basedassay system capable of nucleating local high-concentration hubs in vivo for any protein domainor LCD of interest through LacO-LacI mediatedbinding. These arrays can be readily detected byfluorescence imaging with a high signal-to-noiseratio and allow flexibility in probing various inter-action properties of LCDs while offering a con-venient alternative to various in vitro assays suchas hydrogel polymerization and droplet forma-tion for studying gelation and phase separationof intrinsically disordered proteins, two processes

that are likely coupled with each other undercertain physiological settings (44).Our studies were not designed to address

the structure and nature of LCD-driven phaseseparated compartments, but under certain over-expression conditions we detected what appearsto be liquid-liquid phase separation (i.e., sphericalshape and local changes in refractive index). Al-thoughwe can detect apparent liquid-liquid phaseseparation with gross overexpression of LCDs,we did not obtain evidence for phase separationof TF hubs (i.e., EWS/FLI1) formed at endoge-nous expression levels. However, transactivationby TF LCD hubs is observed at endogenous TFlevels at native chromosomal loci in the absenceof detectable phase separation. Given the tran-sience of LCD-LCD interactions and our directmeasurements of TF concentrations in thenucleusand within hubs, we surmise that LCD-dependenttransactivation can occur in hubs formed overa broad range of TF concentrations (100 nM to100 mM) and time scales—from extremely briefspecific and nonspecific LCD-LCD and TF-DNAbinding events (0.1 to 1 s) to assembly of rel-atively stable hubs (minutes) driven by specific,multivalent interactions. Both the compositionand diversity of LCDs in hubs and their inter-action specificity could influence the range oftheir operational concentrations and their po-tential for phase separation and/or polymerformation. New insights regarding the rapidbinding dynamics and functional importanceof TF LCDs (i.e., LCD-dependent oncogenic po-tential of EWS/FLI1) suggest that understand-ing these mechanisms may also inform ourability to develop strategies to modulate gene

expression in the context of disease. Given thetransient interactions exhibited by TFs and thecritical role of gene dysregulation in disease(i.e., EWS/FLI1 in Ewing’s sarcoma), our findingsoffer the potential to develop single-moleculeimaging platforms to screen drugs targeting generegulatory pathways. In particular, we imaginethat moderate- to high-throughput screens forgene expression inhibitors or activators based onhigh-resolution single-cell and single-moleculeimaging could provide a strategy to probe largesectors of the proteome that have resisted tradi-tional in vitro and cell-based screens for smallmolecules.With new classes of chemicals, naturalproducts, or peptidomimetic libraries, it mayeven be possible to eventually target LCDs thatare the key drivers of regulatory protein-proteininteractions and hub formation involved in geneactivation andpotentiallymany other biomolecularinteractions implicated in disease. Finally, al-though we examined only a small subset of TFLCDs, the fundamental principles that we haveuncovered about the rapid dynamics and hubmechanisms driving LCD-LCD interactionsmaybe applicable to other classes of regulatory pro-teins and biomolecular interactions occurring ina variety of cell types.

Materials and methods summary

The number of LacO repeats in the LacO arrayswas determined by RT-qPCR. FCS and fluores-cence intensitymeasurementwere performed onfluorescently labeled TFs in live cells. By com-paring the results with standard concentrationcurves of a purified fluorescent tag, the TF nuclearconcentration and its copy number at hubs were

Chong et al., Science 361, eaar2555 (2018) 27 July 2018 8 of 9

Fig. 6. A model forfunctional LCD-LCDinteractions in vivo:From hubs to phaseseparation. (A) Dynamicand sequence-specificLCD-LCD interactionsdrive hub formation inlive cells. (B) LCD-dependent transactiva-tion occurs in hubsformed over a broadrange of TF concentra-tions. At endogenousconcentrations, TF LCDsform transactivation hubsat native genomic lociwithout undergoingevident phase separation.Upon TF LCDoverexpression,phase separation isobserved at synthetic TFbinding site arrays.

RESEARCH | RESEARCH ARTICLEon F

ebruary 5, 2020

http://science.sciencemag.org/

Dow

nloaded from

determined. FRAP and SPT were performed tomeasure interaction dynamics between variousTFs and their target genomic loci and to examinehow LCD-LCD interactions affect TF-DNA inter-action dynamics. SPT was also used to determineLCD-LCD interaction dynamics. Two-color con-focal fluorescence imaging was used to examineinteractions between different classes of LCDsand between LCD hubs and RNA Pol II.CRISPR-Cas9–mediated genome editing was

performed to label the endogenous EWS/FLI1 inA673 cells with a HaloTag or to knock out theprotein, allowing fluorescence imaging or func-tional studies of EWS/FLI1. Luciferase and softagar colony formation assays were used to verifythe functions of EWS/FLI1-Halo. Lattice light-sheet microscopy was used to visualize intra-nuclear hubs of the endogenous EWS/FLI1-Halo.Simultaneous confocal imaging of EWS/FLI1-Halo and 3D DNA FISH were performed to ex-amine the spatial relationship between hubs ofendogenous EWS/FLI1 andGGAAmicrosatellites.Luciferase, RT-qPCR, and soft agar colony forma-tion assays were used to examine the effects ofY-to-S mutations on the transactivation andtransformation functions of EWS/FLI1. Detaileddescriptions for all materials and methods areprovided in the supplementary materials.

REFERENCES AND NOTES

1. J. T. Kadonaga, A. J. Courey, J. Ladika, R. Tjian, Distinctregions of Sp1 modulate DNA binding and transcriptionalactivation. Science 242, 1566–1570 (1988). doi: 10.1126/science.3059495; pmid: 3059495

2. A. J. Courey, R. Tjian, Analysis of Sp1 in vivo reveals multipletranscriptional domains, including a novel glutamine-richactivation motif. Cell 55, 887–898 (1988). doi: 10.1016/0092-8674(88)90144-4; pmid: 3142690

3. W. Su, S. Jackson, R. Tjian, H. Echols, DNA looping betweensites for transcriptional activation: Self-association ofDNA-bound Sp1. Genes Dev. 5, 820–826 (1991). doi: 10.1101/gad.5.5.820; pmid: 1851121

4. M. J. Friedman et al., Polyglutamine domain modulates theTBP-TFIIB interaction: Implications for its normal function andneurodegeneration. Nat. Neurosci. 10, 1519–1528 (2007).doi: 10.1038/nn2011; pmid: 17994014

5. H. Kovar, Dr. Jekyll and Mr. Hyde: The Two Faces of theFUS/EWS/TAF15 Protein Family. Sarcoma 2011, 837474(2011). doi: 10.1155/2011/837474; pmid: 21197473

6. M. Kato et al., Cell-free formation of RNA granules: Lowcomplexity sequence domains form dynamic fibers withinhydrogels. Cell 149, 753–767 (2012). doi: 10.1016/j.cell.2012.04.017; pmid: 22579281

7. A. Patel et al., A Liquid-to-Solid Phase Transition of the ALSProtein FUS Accelerated by Disease Mutation. Cell 162,1066–1077 (2015). doi: 10.1016/j.cell.2015.07.047; pmid: 26317470

8. K. A. Burke, A. M. Janke, C. L. Rhine, N. L. Fawzi, Residue-by-Residue View of In Vitro FUS Granules that Bind the C-TerminalDomain of RNA Polymerase II. Mol. Cell 60, 231–241 (2015).doi: 10.1016/j.molcel.2015.09.006; pmid: 26455390

9. M. Boehning et al., RNA polymerase II clustering throughCTD phase separation. bioRxiv 316372 [Preprint]. 7 May 2018.doi: 10.1101/316372

10. I. Kwon et al., Phosphorylation-regulated binding of RNApolymerase II to fibrous polymers of low-complexity domains.Cell 155, 1049–1060 (2013). doi: 10.1016/j.cell.2013.10.033;pmid: 24267890

11. M. Altmeyer et al., Liquid demixing of intrinsically disorderedproteins is seeded by poly(ADP-ribose). Nat. Commun. 6, 8088(2015). doi: 10.1038/ncomms9088; pmid: 26286827

12. S. Xiang et al., The LC Domain of hnRNPA2 Adopts SimilarConformations in Hydrogel Polymers, Liquid-like Droplets, andNuclei. Cell 163, 829–839 (2015). doi: 10.1016/j.cell.2015.10.040; pmid: 26544936

13. D. T. Murray et al., Structure of FUS Protein Fibrils andIts Relevance to Self-Assembly and Phase Separation

of Low-Complexity Domains. Cell 171, 615–627 (2017).doi: 10.1016/j.cell.2017.08.048; pmid: 28942918

14. Y. Lin, D. S. Protter, M. K. Rosen, R. Parker, Formationand Maturation of Phase-Separated Liquid Dropletsby RNA-Binding Proteins. Mol. Cell 60, 208–219 (2015).doi: 10.1016/j.molcel.2015.08.018; pmid: 26412307

15. S. Elbaum-Garfinkle et al., The disordered P granule proteinLAF-1 drives phase separation into droplets with tunable viscosityand dynamics. Proc. Natl. Acad. Sci. U.S.A. 112, 7189–7194(2015). doi: 10.1073/pnas.1504822112; pmid: 26015579

16. T. J. Nott et al., Phase transition of a disordered nuage proteingenerates environmentally responsive membranelessorganelles. Mol. Cell 57, 936–947 (2015). doi: 10.1016/j.molcel.2015.01.013; pmid: 25747659

17. A. Molliex et al., Phase separation by low complexity domainspromotes stress granule assembly and drives pathologicalfibrillization. Cell 163, 123–133 (2015). doi: 10.1016/j.cell.2015.09.015; pmid: 26406374

18. E. L. Elson, D. Magde, Fluorescence correlation spectroscopy.I. Conceptual basis and theory. Biopolymers 13, 1–27 (1974).doi: 10.1002/bip.1974.360130102

19. R. T. Youker, H. Teng, Measuring protein dynamics in live cells:Protocols and practical considerations for fluorescencefluctuation microscopy. J. Biomed. Opt. 19, 090801 (2014).doi: 10.1117/1.JBO.19.9.090801; pmid: 25260867

20. B. L. Sprague, J. G. McNally, FRAP analysis of binding: Properand fitting. Trends Cell Biol. 15, 84–91 (2005). doi: 10.1016/j.tcb.2004.12.001; pmid: 15695095

21. B. C. Chen et al., Lattice light-sheet microscopy: Imagingmolecules to embryos at high spatiotemporal resolution.Science 346, 1257998 (2014). doi: 10.1126/science.1257998;pmid: 25342811

22. I. Solovei, M. Cremer, 3D-FISH on cultured cells combinedwith immunostaining. Methods Mol. Biol. 659, 117–126 (2010).doi: 10.1007/978-1-60761-789-1_8; pmid: 20809307

23. J. Chen et al., Single-molecule dynamics of enhanceosomeassembly in embryonic stem cells. Cell 156, 1274–1285 (2014).doi: 10.1016/j.cell.2014.01.062; pmid: 24630727

24. A. S. Hansen, I. Pustova, C. Cattoglio, R. Tjian, X. Darzacq,CTCF and cohesin regulate chromatin loop stability withdistinct dynamics. eLife 6, e25776 (2017). doi: 10.7554/eLife.25776; pmid: 28467304

25. S. M. Janicki et al., From silencing to gene expression:Real-time analysis in single cells. Cell 116, 683–698 (2004).doi: 10.1016/S0092-8674(04)00171-0; pmid: 15006351

26. B. L. Sprague, R. L. Pego, D. A. Stavreva, J. G. McNally,Analysis of binding reactions by fluorescence recovery afterphotobleaching. Biophys. J. 86, 3473–3495 (2004).doi: 10.1529/biophysj.103.026765; pmid: 15189848

27. X. Darzacq et al., In vivo dynamics of RNA polymerase IItranscription. Nat. Struct. Mol. Biol. 14, 796–806 (2007).doi: 10.1038/nsmb1280; pmid: 17676063

28. Y. Lin et al., Toxic PR Poly-Dipeptides Encoded by the C9orf72Repeat Expansion Target LC Domain Polymers. Cell 167,789–802 (2016). doi: 10.1016/j.cell.2016.10.003;pmid: 27768897

29. A. R. Strom et al., Phase separation drives heterochromatindomain formation. Nature 547, 241–245 (2017). doi: 10.1038/nature22989; pmid: 28636597

30. O. Rog, S. Köhler, A. F. Dernburg, The synaptonemal complexhas liquid crystalline properties and spatially regulates meioticrecombination factors. eLife 6, e21455 (2017). doi: 10.7554/eLife.21455; pmid: 28045371

31. G. Gill, E. Pascal, Z. H. Tseng, R. Tjian, A glutamine-richhydrophobic patch in transcription factor Sp1 contactsthe dTAFII110 component of the Drosophila TFIID complexand mediates transcriptional activation. Proc. Natl. Acad.Sci. U.S.A. 91, 192–196 (1994). doi: 10.1073/pnas.91.1.192;pmid: 8278363

32. A. Martínez-Ramírez et al., Characterization of the A673 cellline (Ewing tumor) by molecular cytogenetic techniques.Cancer Genet. Cytogenet. 141, 138–142 (2003). doi: 10.1016/S0165-4608(02)00670-2; pmid: 12606131

33. K. Gangwal et al., Microsatellites as EWS/FLI responseelements in Ewing’s sarcoma. Proc. Natl. Acad. Sci. U.S.A.105, 10149–10154 (2008). doi: 10.1073/pnas.0801073105;pmid: 18626011

34. S. L. Lessnick, M. Ladanyi, Molecular pathogenesis of Ewingsarcoma: New therapeutic and transcriptional targets.Annu. Rev. Pathol. 7, 145–159 (2012). doi: 10.1146/annurev-pathol-011110-130237; pmid: 21942527

35. K. M. Johnson et al., Role for the EWS domain of EWS/FLI in bindingGGAA-microsatellites required for Ewing sarcoma anchorage

independent growth. Proc. Natl. Acad. Sci. U.S.A. 114, 9870–9875(2017). doi: 10.1073/pnas.1701872114; pmid: 28847958

36. F. A. Ran et al., Genome engineering using the CRISPR-Cas9system. Nat. Protoc. 8, 2281–2308 (2013). doi: 10.1038/nprot.2013.143; pmid: 24157548

37. S. L. Lessnick, C. S. Dacwag, T. R. Golub, The Ewing’s sarcomaoncoprotein EWS/FLI induces a p53-dependent growth arrest inprimary human fibroblasts. Cancer Cell 1, 393–401 (2002).doi: 10.1016/S1535-6108(02)00056-9; pmid: 12086853

38. K. M. Johnson, C. Taslim, R. S. Saund, S. L. Lessnick,Identification of two types of GGAA-microsatellites and theirroles in EWS/FLI binding and gene regulation in Ewingsarcoma. PLOS ONE 12, e0186275 (2017). doi: 10.1371/journal.pone.0186275; pmid: 29091716

39. N. Guillon et al., The oncogenic EWS-FLI1 protein binds in vivoGGAA microsatellite sequences with potential transcriptionalactivation function. PLOS ONE 4, e4932 (2009). doi: 10.1371/journal.pone.0004932; pmid: 19305498

40. J. B. Grimm et al., A general method to improve fluorophores forlive-cell and single-molecule microscopy. Nat. Methods 12,244–250 (2015). doi: 10.1038/nmeth.3256; pmid: 25599551

41. J. B. Grimm et al., Bright photoactivatable fluorophores forsingle-molecule imaging. Nat. Methods 13, 985–988 (2016).doi: 10.1038/nmeth.4034; pmid: 27776112

42. M. V. Staller et al., A High-Throughput Mutational Scan of anIntrinsically Disordered Acidic Transcriptional ActivationDomain. Cell Syst. 6, 444–455 (2018). doi: 10.1016/j.cels.2018.01.015; pmid: 29525204

43. G. Boulay et al., Cancer-Specific Retargeting of BAF Complexesby a Prion-like Domain. Cell 171, 163–178 (2017). doi: 10.1016/j.cell.2017.07.036; pmid: 28844694

44. T. S. Harmon, A. S. Holehouse, M. K. Rosen, R. V. Pappu,Intrinsically disordered linkers determine the interplay betweenphase separation and gelation in multivalent proteins. eLife 6,e30294 (2017). doi: 10.7554/eLife.30294; pmid: 29091028

45. B. Efron, R. Tibshirani, An Introduction to the Bootstrap(Chapman & Hall, CRC Press, 1993).

ACKNOWLEDGMENTS

We thank S. Lessnick, S. McKnight, and M. Kato for discussions andproviding reagents; Q. Gan and A. Hansen for providing codes toanalyze imaging data; J. Bosco and P. Sharma for help withmolecular cloning; K. Heydari and the CRL Flow Cytometry Facility forassistance with flow cytometry; the CRL Molecular Imaging Centerfor providing access to confocal fluorescence microscopes; andJ. Goodrich, G. S. Martin, and members of Tjian and Darzacq labs forcritical reading of the manuscript. Funding: This work was supportedby California Institute of Regenerative Medicine grant LA1-08013(to X.D.), NIH grants UO1-EB021236 and U54-DK107980 (to X.D.), andthe Howard Hughes Medical Institute (to Z.L., L.L., and R.T.). Authorcontributions: R.T. conceived of and supervised the project. S.C., R.T.,and X.D. designed experiments and interpreted the data. S.C.performed genome editing, soft colony formation assays, live-cellimaging in the presence of hexanediols, confocal imaging, DNA FISH,immunofluorescence, FRAP and SPT measurements, and all of theimaging data analyses. C.D.-D. established and characterizedHalo-RPB1 cell lines, characterized LacO-containing cell lines, andperformed all of the RT-qPCR assays and related data analyses. S.C.and Z.L. developed codes for SPT data analyses. P.D. performed latticelight-sheet imaging. G.M.D. and S.C. designed and generatedconstructs. S.C. and C.C. performed luciferase assays. A.H. performedFCS measurements. S.B. and L.L. synthesized and characterizedfluorescent HaloTag and SNAPf-tag ligands. S.C., R.T., C.D.-D., C.C.,Z.L., X.D., P.D., and A.H. prepared the manuscript and supplementarymaterials. Competing interests: L.L. has filed a patent and patentapplications (e.g., U.S. Patent 9,933,417) covering azetidine-containingfluorophores such as JF549. Data and materials availability: Allcodes for imaging data analyses have been submitted to GitHub andare publicly available (https://github.com/Shasha-Chong/CodeFor2018SciencePaper).

SUPPLEMENTARY MATERIALS

www.sciencemag.org/content/361/6400/eaar2555/suppl/DC1Materials and MethodsSupplementary TextFigs. S1 to S10Tables S1 to S5References (46–55)

23 October 2017; resubmitted 12 March 2018Accepted 13 June 2018Published online 21 June 201810.1126/science.aar2555

Chong et al., Science 361, eaar2555 (2018) 27 July 2018 9 of 9

RESEARCH | RESEARCH ARTICLEon F

ebruary 5, 2020

http://science.sciencemag.org/

Dow

nloaded from

transcriptionImaging dynamic and selective low-complexity domain interactions that control gene

Banala, Luke Lavis, Xavier Darzacq and Robert TjianShasha Chong, Claire Dugast-Darzacq, Zhe Liu, Peng Dong, Gina M. Dailey, Claudia Cattoglio, Alec Heckert, Sambashiva

originally published online June 21, 2018DOI: 10.1126/science.aar2555 (6400), eaar2555.361Science 

, this issue p. eaar2555, p. eaar3958, p. 412; see also p. 329Scienceinhibitors and how their dynamic interactions might initiate transcription elongation.

further revealed the differential sensitivity of Mediator and RNA polymerase II condensates to selective transcriptional.etcompartmentalize and concentrate the transcription apparatus to maintain expression of key cell-identity genes. Cho

showed that at super-enhancers, BRD4 and Mediator form liquid-like condensates thatet al.Indeed, Sabari sequence-specific protein-protein interaction. These hubs have the potential to phase-separate at higher concentrations.domains of transcription factors form concentrated hubs via functionally relevant dynamic, multivalent, and

found that low-complexityet al.transcription regulation is emerging (see the Perspective by Plys and Kingston). Chong domains. Now a conceptual framework connecting the nature and behavior of their interactions to their functions in

contain intrinsically disordered low-complexity−−BRD4, subunits of the Mediator complex, and RNA polymerase II such as transcription factors and cofactors including−−Many components of eukaryotic transcription machinery

Phase separation and gene control

ARTICLE TOOLS http://science.sciencemag.org/content/361/6400/eaar2555

MATERIALSSUPPLEMENTARY http://science.sciencemag.org/content/suppl/2018/06/20/science.aar2555.DC1

CONTENTRELATED

http://stke.sciencemag.org/content/sigtrans/11/548/eaao1917.fullhttp://science.sciencemag.org/content/sci/361/6400/412.fullhttp://science.sciencemag.org/content/sci/361/6400/eaar3958.fullhttp://science.sciencemag.org/content/sci/361/6400/329.full

REFERENCES

http://science.sciencemag.org/content/361/6400/eaar2555#BIBLThis article cites 54 articles, 9 of which you can access for free

PERMISSIONS http://www.sciencemag.org/help/reprints-and-permissions

Terms of ServiceUse of this article is subject to the

is a registered trademark of AAAS.ScienceScience, 1200 New York Avenue NW, Washington, DC 20005. The title (print ISSN 0036-8075; online ISSN 1095-9203) is published by the American Association for the Advancement ofScience

Science. No claim to original U.S. Government WorksCopyright © 2018 The Authors, some rights reserved; exclusive licensee American Association for the Advancement of

on February 5, 2020

http://science.sciencem

ag.org/D

ownloaded from