RNA self-assembly contributes to stress granuleformation and defining the stressgranule transcriptomeBriana Van Treecka, David S. W. Prottera, Tyler Mathenya, Anthony Khonga,b, Christopher D. Linkc, and Roy Parkera,b,1

aDepartment of Chemistry and Biochemistry, University of Colorado Boulder, Boulder, CO 80309; bHoward Hughes Medical Institute, University of ColoradoBoulder, Boulder, CO 80309; and cDepartment of Integrative Physiology, University of Colorado Boulder, Boulder, CO 80309

Contributed by Roy Parker, February 4, 2018 (sent for review January 2, 2018; reviewed by Paul Anderson and Imed E. Gallouzi)

Stress granules are higher order assemblies of nontranslating mRNAsand proteins that form when translation initiation is inhibited. Stressgranules are thought to form by protein–protein interactions of RNA-binding proteins. We demonstrate RNA homopolymers or purifiedcellular RNA forms assemblies in vitro analogous to stress gran-ules. Remarkably, under conditions representative of an intracel-lular stress response, the mRNAs enriched in assemblies from totalyeast RNA largely recapitulate the stress granule transcriptome.We suggest stress granules are formed by a summation of pro-tein–protein and RNA–RNA interactions, with RNA self-assemblylikely to contribute to other RNP assemblies wherever there is a highlocal concentration of RNA. RNA assembly in vitro is also increased byGR and PR dipeptide repeats, which are known to increase stressgranule formation in cells. Since GR and PR dipeptides are involvedin neurodegenerative diseases, this suggests that perturbations in-creasing RNA–RNA assembly in cells could lead to disease.

RNP granules | dipeptides | RNA self-assembly | stress granules

R ibonucleoprotein (RNP) granules are eukaryotic non–membrane-bound organelles composed of RNA and protein.

RNP granules are located in the cytoplasm or the nucleus andinclude P-bodies, germ granules, neuronal granules, paraspeckles,Cajal bodies, and stress granules (1–6). Stress granules are cyto-plasmic assemblies of nontranslating mRNAs and RNA-bindingproteins that form during stress responses where bulk translationinitiation is inhibited.Stress granules are of interest for four reasons. First, stress gran-

ules are thought to play a role in the stress response and generegulation (7, 8). Second, related RNP granules exist in neuronsand can affect synaptic plasticity (9, 10). Third, mutations in RNA-binding proteins, or stress granule-remodeling complexes, whichlead to constitutive or increased stress granule formation, arecausative in amyotrophic lateral sclerosis, frontotemporal lobardegeneration, inclusion body myopathy, and other degenerativedisorders (11–13). Fourth, stress granule formation can influenceboth tumor progression and viral infection (14–17).Stress granules, as well as other RNP granules, are thought to

form by RNA providing a scaffold for RNA-binding proteins thatform protein–protein interactions in trans to drive assembly. Forexample, mammalian stress granule assembly is promoted by inter-actions of the TIA1, G3BP1, and ATXN2 proteins (18–20). Simi-larly, assembly of yeast P-bodies is promoted by the dimerization ofthe RNA-binding protein Edc3, as well as an intrinsically disorderedregion (IDR) on Lsm4p (21). Consistent with interactions betweenRNA-binding proteins promoting RNP granule assembly, severalpurified RNA-binding proteins, in isolation or with RNA, can formself-assemblies in vitro as coacervates, fibers, or hydrogels, althoughthe specific relationship of these in vitro assemblies to RNP granulesremains unclear (22–27). The role of trans-RNA–RNA interactionsin stress granule formation remains relatively unexplored.Herein, we illustrate that RNA–RNA interactions contribute

to the assembly of stress granules and the targeting of RNAs tostress granules. We show that trans-RNA–RNA interactions playa role in the recruitment of given RNAs to stress granules. We

find that various RNAs are capable of self-interactions, even in theabsence of Watson–Crick base pairing. More importantly, weshow that mixtures of cellular RNAs partition into self-assembliesin vitro under physiologically relevant conditions and that theRNAs enriched in these assemblies reflect the RNAs enriched instress granules in vivo. This argues that RNA–RNA interactionscontribute to both stress granule assembly and to the determination ofwhich RNAs are enriched in stress granules. Strikingly, we also ob-served that RNA self-assembly is promoted by the pathogenic di-peptide repeats GR and PR. This work highlights that RNPs exchangebetween monomeric and multimeric states within the cell and that thecell utilizes a variety of mechanisms to regulate this equilibrium.

ResultsFour points led us to hypothesize that RNA–RNA interactions intrans might contribute to stress granule formation. First, stressgranules form when there is ribosome run-off, exposing the pre-viously ribosome-occupied coding regions that would be expectedto form RNA–RNA interactions both in cis and in trans. Second,long mRNAs partition highly into stress granules but only show amodest increase in the binding sites of stress granule proteins,suggesting length might contribute to the partitioning of mRNAsinto stress granules through trans-RNA–RNA interactions (28).Third, stress granule cores in lysates are resistant to high salt or

Significance

Stress granules, which are ubiquitous, non–membrane-boundassemblies of protein and RNA, form when translation initia-tion is inhibited, contribute to the regulation of gene expression,and are implicated in the pathologies of cancer and neurode-generative disease. Understanding the mechanisms of stressgranule assembly is crucial to gaining greater insight into theirbiological function and pathological misregulation. We provideevidence that RNA–RNA interactions contribute to the assemblyof stress granules. Furthermore, we show that pathogenic di-peptides increase the propensity of RNA to assemble. Together,this argues that RNAs are assembly prone and must be carefullyregulated. A summative model of stress granule assembly, whichincludes trans-RNA–RNA interactions, can be extended to otherribonucleoprotein granules in the cell.

Author contributions: B.V.T., D.S.W.P., A.K., and R.P. designed research; B.V.T., D.S.W.P.,and C.D.L. performed research; B.V.T., D.S.W.P., T.M., and A.K. analyzed data; and B.V.T.and R.P. wrote the paper.

Reviewers: P.A., Brigham and Women’s Hospital; and I.E.G., McGill University.

The authors declare no conflict of interest.

Published under the PNAS license.

Data deposition: The data reported in this paper have been deposited in the Gene Ex-pression Omnibus (GEO) database, https://www.ncbi.nlm.nih.gov/geo (accession no.GSE99170).1To whom correspondence should be addressed. Email: roy.parker@colorado.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1800038115/-/DCSupplemental.

Published online February 26, 2018.

2734–2739 | PNAS | March 13, 2018 | vol. 115 | no. 11 www.pnas.org/cgi/doi/10.1073/pnas.1800038115

Dow

nloa

ded

by g

uest

on

Mar

ch 3

, 202

1

aliphatic alcohols, which are known to disrupt many protein–protein,but not RNA–RNA, interactions (29). Stress granule cores havebeen suggested to be independent of RNA based on their resistanceto RNase treatment (29), but a reexamination of this observationshowed that extensive RNase treatment fails to degrade the RNAwithin stress granule cores (Fig. S1). Fourth, trans-RNA–RNA in-teractions contribute to the assembly of multimeric RNPs in Dro-sophila embryos, as well as RNA foci containing transcripts withrepeat expansions (30, 31).Additional evidence that RNA–RNA interactions could con-

tribute to stress granule assembly came from the partitioning ofnoncoding RNAs (ncRNAs) into stress granules in mammaliancells (28). By data-mining the mammalian stress granule tran-scriptome (28), we observed that ncRNAs shorter than 3,000 baseswere generally depleted from stress granules, as predicted by length(Fig. 1A). However, antisense ncRNAs shorter than 3,000 basesshowed a bimodal distribution, with one population enriched instress granules and one depleted (Fig. 1B). Of the 14 antisensencRNAs significantly enriched greater than twofold in stress gran-ules and with fragments per kilobase of transcript per millionmapped reads (FPKM) values above 1 in total RNA samples,10 were antisense to long mRNAs that were enriched in stressgranules (Fig. 1D), five of which are illustrated (Fig. 1F). This isstriking, since only 14.5% of mRNAs in the cell are enriched instress granules (Fig. 1C). In contrast, of the 23 antisense ncRNAssignificantly depleted greater than twofold from stress granules andwith FPKM values above 1 in total RNA samples, only one had anantisense partner that was enriched in stress granules (Fig. 1E). Thus,there is a correlation between the localization of antisense ncRNAsand their ability to base-pair to longer mRNAs, implying that RNA–RNA interactions may partition specific RNAs into stress granules.To determine if RNA–RNA interactions could contribute to

stress granule formation, we calculated the approximate concentra-tion of the coding region of mRNAs exposed during polysome col-lapse. For yeast, we estimated the concentration of exposed codingregions to be between 170 and 800 μg/mL (Materials and Methods),while in the mammalian U-2 OS cell line, the exposed ORFs wereestimated at ∼180 μg/mL (Materials and Methods). If RNA–RNAinteractions contribute to stress granule formation, then we pre-dicted that RNA at these concentrations would spontaneously as-semble under conditions mimicking the intracellular milieu.To test this prediction, we assessed whether purified protein-

free total RNA from Saccharomyces cerevisiae at 150 μg/mL wouldform assemblies under conditions where we varied the salt andused PEG to mimic molecular crowding. We observed that totalyeast RNA readily self-assembled and formed two types of as-semblies (Fig. 2A). At higher PEG and lower salt, the RNA wasobserved to form small droplets (Fig. 2 A, a). With increasing salt,we observed the formation of more amorphous assemblies thatcontained a larger percentage of the RNA, which, due to theirmorphology, we refer to as RNA tangles (Fig. 2 A, b and c and Fig.S2 A and B). Based on fluorescence recovery after photobleaching(FRAP) of spiked-in fluorescent RNAs or dilution into lower ionicstrength, droplets were more dynamic and less stable than RNAtangles (Fig. S2 C and D). Specifically, 97% of the RNA in tanglesformed at high salt is immobile (Fig. S2C), and dilution takesnearly an hour to disrupt tangles, whereas droplets disperse on theorder of seconds (Fig. S2D). Importantly, both droplets and tan-gles were enriched for the RNA-specific dye SYTO RNASelect(Fig. S2E) and depleted for PEG (Fig. S2F). This is consistent withPEG functioning primarily as a crowding agent, which is furthersupported by Ficoll-promoting RNA self-assembly (Fig. S2G).Moreover, in the absence of crowding agents, RNA also self-assembled under physiological concentrations of salt (150 mM)and the polyamines spermine (223 μM) and spermidine (1,339 μM),which are known to stabilize RNA–RNA interactions (32) (Fig. 2B).Thus, under in vitro conditions analogous to the cytosol during astress response, purified RNA undergoes self-assembly.Stress granules recruit a diversity of RNA-binding proteins in

addition to RNA. Similarly, we observed that RNA self-assembliesrecruited RNA-binding proteins (Fig. S3). Specifically, when fused

to GFP, the IDRs of yeast Lsm4 and eIF4GII, both of which bindRNA (24), are recruited to RNA droplets more than the IDR ofFUS or GFP alone (Fig. S3 A and B). Similarly, RNA dropletsrecruit the RNA-binding protein hnRNPA1 (Fig. S3C). Thus,assemblies formed by RNA interactions in cells would be expectedto recruit RNA-binding proteins.RNA–RNA interactions contributing to RNA self-assemblies

could be helical stacking (33), specific base-pairing (31), orpromiscuous interactions between mRNA involving both tradi-tional Watson–Crick interactions and additional interactions(34). Evidence that non-Watson–Crick interactions can promoteRNA self-assembly is that all four homopolymers self-assemble,with polyU forming rapidly relaxing droplets [as previously ob-served (32)], polyC forming slower relaxing droplets, polyAforming asymmetrical assemblies with very slow relaxation rates,and polyG forming an aggregate that is presumably based onG-quadruplexes (Fig. 2C and Fig. S4). Thus, RNA sequencesmay impart biophysical characteristics to their assemblies, but,

A B

Fre

quen

cy(c

ount

s)

Fre

quen

cy(c

ount

s)

Log (Fold Change) Log (Fold Change)2 2

-10 -5 0 5 10 -10 -5 0 5 10

300

200

100

0 0

10

20

30

Antisense ncRNAsncRNAs (antisense excluded)

DC E

F

Enriched (SG)

Neither (SG)

Depleted (SG)

10*p<0.0001

p=0.0004510

12

1

4

7789

1780 1626

Enriched asRNAbinding partner localization

All mRNAs Depleted asRNAbinding partner localization

*1 has a secondary bindingpartner that is depleted

PEG3

MDM2

IRF2BP2RP4-781K5.2

DYRK2RP11-335O4.3

PLCD4RP11-548H3.1

RP11-611O2.5

5.5 83757.7

2.94.5

4.13.6

2.14.6

3.76.4

1314

12633785

4663693

8912777

4350759

SGEnrichment

TranscriptLength(bases)

PEG3-AS1

Fig. 1. RNA–RNA base-pairing influences RNA localization. All ncRNAs(excluding antisense) (A) and antisense ncRNAs (B) with lengths <3,000 ntand significant sequencing reads were plotted in a histogram showing theirlog2(fold change) enrichment in stress granule/total RNA. A red brackethighlights the second peak for RNAs enriched in stress granules. (C) Pie chartillustrating the proportion of all mammalian mRNAs enriched (red), de-pleted (blue), or neither (gray) in stress granules. Stress granule localizationof binding partners to enriched (D) and depleted (E) antisense ncRNAs isshown. *One enriched antisense ncRNA transcript has a secondary bindingpartner to a depleted mRNA. (F) Examples of enriched antisense ncRNAs andtheir binding partners. SG, stress granule.

Van Treeck et al. PNAS | March 13, 2018 | vol. 115 | no. 11 | 2735

BIOCH

EMISTR

Y

Dow

nloa

ded

by g

uest

on

Mar

ch 3

, 202

1

more importantly, RNA self-assembly is a general property ofdiverse RNAs and is not restricted to Watson–Crick interactions.If the self-assembly of RNA in vitro is relevant to stress granule

assembly, we predicted that similar mRNAs would assemble intoRNA droplets in vitro and stress granules in vivo. To test thisprediction, we purified RNA droplets formed under physiologicalsalt (150 mM NaCl) and PEG from total yeast RNA in triplicate,sequenced the assembled RNA, and compared that RNA pop-ulation with the mRNAs that partition into stress granules (28).This experiment revealed several important observations.First, triplicates of in vitro assemblies and total RNA agreed

within themselves but showed clear differences (Fig. S5A), allowingthe identification of both enriched and depleted RNAs in the invitro assemblies (Fig. 3A). RNAs significantly (P < 0.01) and two-fold enriched and depleted under these conditions will be referredto throughout the remainder of this paper as in vitro-enrichedand -depleted, respectively. On average, the in vitro-enriched RNAswere significantly longer, while the in vitro-depleted RNAs wereshorter (Fig. 3B). This closely mirrored the length effect seen in theanalysis of the yeast stress granule transcriptome (28) (Fig. 3C).A second key observation was a strong overlap between in vitro-

and stress granule-enriched mRNAs (Fig. 3D and Fig. S5B).Specifically, of the 916 mRNAs enriched in yeast stress granules(28), 634 are enriched in vitro. Conversely, of the 1,111 mRNAsdepleted from yeast stress granules (28), only 56 are enrichedin vitro. This correlation extends to RNAs depleted from in vitroassemblies and stress granules in vivo (Fig. 3E and Fig. S5C). Ofthe 1,456 mRNAs depleted from in vitro assemblies, only 10 areenriched in stress granules, and of the 1,111 mRNAs depletedfrom stress granules, 573 are also depleted from the RNA self-assembly in vitro. This indicates that the biophysical properties ofRNA that drive RNA self-assembly in vitro correlate with a criticalmRNA feature for partitioning mRNAs into stress granules in vivo.Plotting the relative enrichment of RNAs in in vitro assemblies

vs. in stress granules shows a correlation in the degree of enrich-ment in both cases, with a Pearson’s correlation of 0.53 (Fig. 3F).In general, RNAs more enriched in stress granules are also moreenriched in assemblies in vitro, and vice versa. The RNAs thatshow less correlation between in vivo and in vitro recruitment canbe explained by the effects of translation efficiency on mRNA re-cruitment into stress granules, where efficient translation has beenshown to correlate with depletion from stress granules (28). Re-moval of transcripts with fractions of optimal codons below 0.4 orabove 0.6 increases the Pearson’s correlation coefficient to 0.61(Fig. S5E). This makes sense, as translation would be expected to

affect mRNAs partitioning into stress granules in vivo but not toaffect RNA self-assembly in vitro, where there is no translationalapparatus. Taken together, these observations suggest that thepartitioning of mRNAs into stress granules is modulated, in part,by the self-assembly capabilities of RNA.We suggest a working model where stress granules form when

the summation of protein–protein, protein–RNA, and RNA–RNAinteractions increase over a threshold for assembly. This model canbe illustrated in a phase diagram, with transitions between “phases”explaining stress granule formation (Fig. 4). For example, increasingthe concentration of exposed RNA, either by inhibiting translationinitiation leading to ribosome run-off or by transfection or injectionof RNA into cells (35–37), could cause a transition in the cell thatleads toward stress granule assembly via increased RNA–RNA in-teractions (Fig. 4, yellow arrow). Similarly, deletion of G3BP1 andG3BP2, which are abundant proteins that contribute to stressgranule assembly (19), prevents assembly under most stresses andmoves components back into the nonassembled phase in our model(Fig. 4, red arrow). Conversely, overexpression of TIA1, G3BP1, orFMR1, all of which can contribute to stress granule assembly (18,38–40), may drive the cell into a regime of assembled stress granulesby increasing the protein–protein interactions (Fig. 4, black arrow).Since RNA–RNA interactions contribute to stress granule as-

sembly, we hypothesized that any small molecule or peptide thatstabilizes RNA–RNA interactions would promote stress granuleassembly. Strikingly, prior work has shown that arginine-containingdipeptides GR and PR produced by RAN translation of hex-anucleotide (G4C2) repeat expansions of C9orf72 (41, 42) are bothtoxic to cells and trigger stress granule formation (43, 44). Al-though these dipeptides perturb many cellular processes (45–47),we predicted they may directly induce stress granules by stabilizingRNA–RNA interactions. Indeed, we observed that (PR)10 and(GR)10 robustly stimulated RNA self-assembly in vitro, while (GP)10had little effect (Fig. 5 and Fig. S6). Assemblies with (PR)10 or(GR)10 were enriched for both RNA and dipeptides (Fig. 5A), andneither RNA nor protein alone was sufficient for robust assemblyunder these conditions (Fig. 5A and Fig. S6A). Since the most toxicdipeptides are the same dipeptides that stimulate the assembly ofRNA in vitro, we suggest that dipeptides may exert some of theirtoxic effects by promoting RNA–RNA assemblies in the cell, such asthrough the formation of stress granules.

DiscussionWe present several lines of evidence that stress granules form, inpart, by RNA–RNA interactions and that those interactions can

PE

G(%

)

b

c

d

In between

Droplets

Nothing

Tangles

NaCl (mM)

Total Yeast RNA Phase DiagramA B

10 µm 5 µm

Yeast total RNA

a b

c d

polyGpolyApolyCpolyU

10µm

C

12

a10

8

6

4

2

00 200 400 600 800

-po

lyam

ines

+po

lyam

ines

Fig. 2. Various RNAs self-assemble in vitro. (A)Phase diagram of RNA assembly morphology undervarying PEG and NaCl concentrations. Images corre-spond to labeled positions in the phase diagram:Droplets formed at 0 mM NaCl, 10% PEG (a); droplet/tangles formed at 300 mM NaCl, 7.5% PEG (b); tan-gles formed at 750 mM NaCl, 10% PEG (c); and noassemblies at 300 mM NaCl, 2.5% PEG (d). All con-ditions contain 1 mM MgCl2 and 150 μg/mL yeasttotal RNA. (B) Total yeast RNA in 150 mM NaCl and1 mM MgCl2 with and without physiologically rele-vant conditions of spermine (223 μM) and spermidine(1,339 μM). (C) polyU, polyC, polyA, and polyG self-partition in vitro at 500 μg/mL (respective) homo-polymers, 10% PEG, and 750 mM NaCl.

2736 | www.pnas.org/cgi/doi/10.1073/pnas.1800038115 Van Treeck et al.

Dow

nloa

ded

by g

uest

on

Mar

ch 3

, 202

1

influence the partitioning of mRNA. First, we show that a por-tion of total yeast RNA effectively self-assembles in vitro underconditions mimicking intracellular stress conditions (Fig. 2). Im-portantly, this self-assembly of yeast RNA largely reproduces thestress granule transcriptome (Fig. 3). Moreover, the enrichment ofshort antisense ncRNAs in mammalian stress granules correlatedwith their ability to base-pair to a longer mRNA enriched in stressgranules (Fig. 1). Finally, dipeptide repeats, which are known toinduce stress granules in cells, strongly increase RNA self-assemblyin vitro (Fig. 5). In combination with genetic experiments showingproteins can enhance stress granule assembly, we conclude thatstress granules assemble by a summation of protein–RNA, protein–protein, and RNA–RNA interactions (Fig. 4). The precise set ofinteractions that drive stress granule formation can vary under dif-ferent conditions as long as the total summation is above the criticalthreshold for assembly.There is growing evidence that RNA–RNA interactions can

help drive the assembly of stress granules. A role for RNA–RNAinteractions could explain why purified stress granule cores arestable against many insults (29). In addition, a role for RNA–RNAinteractions could suggest why the ATPase activity of Ded1p isrequired to disassemble stress granules (48). Our results also high-light the importance of charge shielding in promoting RNA self-assembly, as increasing salt concentrations or adding positivelycharged molecules, such as polyamines or arginine-containingdipeptides, greatly increases assembly. This is consistent with ob-servations in the literature suggesting that ionic strength, whichhas a strong impact on RNA self-assembly in vitro (Fig. 2 and Fig.S2), could be an important regulator of stress granule formation.Stress granule assembly can be triggered by hypertonic shock and,conversely, can fail to form in conditions of low intracellular os-molality (49). Moreover, sorbitol, another osmotic stressor, canpartially rescue formation of stress granules in ΔΔG3BP1/2 cells(19), perhaps because the sorbitol increases the intracellular os-motic strength, thereby stabilizing RNA–RNA interactions. Inanother line of evidence, the addition of G4C2 RNA to U-2 OScell lysates condenses an assembly containing many stress granulecomponents and cellular RNAs, perhaps by nucleating interac-tions between various mRNAs in the lysates (37). Finally, theprevalence of RNA–RNA interactions helps explain why injectionor transfection of concentrated RNA into cells leads to the for-mation of large, higher order assemblies (35–37). This is analo-gous to the huge influx of exposed RNAs during a stress response,

Log (fold change) in stress granules

Log

(fol

dch

ange

)in

vitr

oas

sem

blie

s

510

510

410

410

310

310

210

210

110

110

010

010

A B C

D

E

F

In vitro assembly enriched (150 mM NaCl)

2

2

854

Enrichedn=1488

Enriched (SG)n= 916

634 28256

Enrichedn=1488

Depleted (SG)n=1111

1432 1055

In vitro assembly depleted (150 mM NaCl)

Enriched (SG)n= 916

Depletedn= 1456

1446 90610

Depleted (SG)n=1111

Depletedn= 1456

883 573 538

Overlap: p<1x10 Overlap: n.s.Lack of overlap: n.s. Lack of overlap: p<1x10

-15

Overlap: p<1x10Lack of overlap: n.s.

-15

-15

Overlap: n.s.Lack of overlap: p<1x10-15

Total RNA (reads) Enr

iche

dN

eith

er

Nei

ther

Dep

lete

d

Dep

lete

d

Enr

iche

dInvi

tro

asse

mbl

ies

(rea

ds)

RN

Ale

ngth

(kb)

RN

Ale

ngth

(kb)

Assembliesin vitro

Yeaststress granules

0 0

2 2

4 44 4

10 1016 16

*** ***(Khong et al., 2017)

EnrichedNeitherDepleted

CodonOptimality

0.30 0.80

4

3

3 4 5 6

2

2

1

1

0

0

-1

-1

-2

-2

-3

-3

-4

-4-5-6

R=0.53

Fig. 3. RNAs in self-assemblies in vitro largely recapitulate the stress gran-ule transcriptome. (A) RNAs from in vitro assemblies formed at 150 mM NaClidentified as significantly (P < 0.01) and twofold enriched (1,488 RNAs, reddots) and depleted (1,456 RNAs, blue dots) compared to total yeast RNA. Thebox plots show the correlation between transcript length and RNA locali-zation to in vitro assemblies (B) and yeast stress granules (C) (28). ***P <0.001 between any three box plots. (D) In vitro-enriched mRNAs significantlyoverlap with mRNAs identified in the yeast stress granule transcriptome andexhibit significant lack of overlap with RNAs depleted from stress granules.n.s., not significant. (E) In vitro-depleted RNAs significantly overlap withRNAs depleted from stress granules and exhibit a significant lack of overlapwith RNAs identified in the yeast stress granule transcriptome. (F) Degree ofenrichment and depletion between in vitro assemblies and stress granulescorrelates (Pearson’s correlation, R = 0.53). RNAs that are more enrichedin vitro than in vivo (Upper Left) tend to have greater proportions of optimalcodons (orange).

prot

ein-

prot

ein

inte

ract

ions

RNA-RNA interactions

Proteindominatedassembly

Assembly driven by protein-proteinand RNA-RNA interactions

No stressgranules

Stressgranules

Stressgranules

No stressgranules

RNA dominatedassemblyNo assembly

Overexpressmultivalent RBPs

free RNA

Depletion ofkey RBPs

Fig. 4. Four-phase model of stress granule assembly. Model illustrating howassemblies may be RNA-dominated (Bottom Right) or protein-dominated(Top Left); however, a combination of interactions is often responsible forassembly within cells (Top Right). Arrows denote examples from the litera-ture that lead to stress granule formation or dissolution. The yellow arrowshows the formation of stress granules through a large influx of non-translating RNAs. The red arrow signifies a lack of stress granules when keyproteins are deleted. The black arrow denotes the formation of stressgranules through overexpression of certain RNA-binding proteins.

Van Treeck et al. PNAS | March 13, 2018 | vol. 115 | no. 11 | 2737

BIOCH

EMISTR

Y

Dow

nloa

ded

by g

uest

on

Mar

ch 3

, 202

1

which may allow for the emergence of interactions that are nor-mally outcompeted by more specific, high-affinity interactions.The role of RNA–RNA interactions in stress granule assembly

has two broader implications. First, one anticipates that RNA–

RNA interactions will contribute to other RNP granules. Indeed,given the thermodynamic strength of RNA–RNA interactions, theyshould be expected to be in a stable state in cells and will form intrans whenever there is a sufficiently high local concentration ofRNA. For example, we suggest that very efficient transcription oflong RNAs would be expected to drive the formation of RNA–RNA interactions between nascent transcripts. One possible ex-ample of this process would be the assembly of paraspeckles, whichform at sites of transcription of NEAT1 RNA and contain multipleNEAT1 copies (50).A second implication is that RNA–RNA interactions can be

promiscuous and form between any two RNAs with single-strandedregions. For example, by chance, the average mammalian mRNA instress granules (7.5 kb) (28) should have over 300 possible sites ofsix consecutive base pairs with another 7.5 kb of mRNA. Even ifthe vast majority of these sites are lost to intramolecular RNAfolding or RNA-binding proteins, numerous possible sites for RNA–RNA interaction will remain. Thus, one anticipates that RNA–RNAinteractions can arise between many different RNAs. In somecases, evolution will have created definitive interaction sites to give

specificity to assemblies, as has been seen with Oscar and BicoidmRNAs in Drosophila embryos (30).Although promiscuous RNA–RNA interactions are capable of

forming between any two RNAs, it is important to note that therewill be a gradient of any given RNA’s propensity to assemble. Weshow that longer RNAs are more enriched in assemblies in vitroand in stress granules (Fig. 3). In addition, one anticipates that theability of RNAs to form base-pairing interactions will increase theirself-assembly properties. Given these two inputs, short-structuredRNAs, such as tRNAs (originally “soluble”RNAs), will be excludedfrom RNA assemblies. Alternatively, longer RNAs, particularlythose with repeat sequences capable of self–base-pairing, will behighly efficient at self-assembly. Recent reports have illustrated thatlonger repeat RNAs more effectively form intracellular and in vitroRNA assemblies (31) and that transfection of G-quadruplex–capableG4C2 RNA elicits robust stress granule assembly, whereas its an-tisense counterpart C4G2 does not (37). In this light, it is notablethat many repeat-containing RNAs, including both pathogenictoxic RNAs and satellite RNAs, form specific nuclear foci (43, 51–53). This suggests that formation of hyperstable RNA assemblies incells is toxic, and this provides a possible explanation for why re-peat RNAs with a strong tendency to base-pair cause disease oncethey are expanded beyond a certain length.This work suggests that RNPs exist within cells at an equilibrium

between monomeric RNPs and multimeric RNP granules that isinfluenced by many parameters. For example, RNA helicases, ri-bosome association, and monovalent RNA-binding proteins areexpected to play a role in maintaining RNPs in the monomericstate. Consistent with this view, it is known that depletion of theabundant Tdp-43 ortholog in Caenorhabditis elegans leads to theaccumulation of dsRNA foci in the nucleus (54). In contrast, longerRNA lengths, high local or transient RNA concentrations, and thepropensity of RNA to interact with itself will promote RNA–RNAassociation. Taken together, we suggest that RNA–RNA assembliesin cells may be a default state and that cells prevent this RNAaggregation by active means, ribosomes, and RNA-binding proteins.

Materials and MethodsMaterials. Homopolymer RNAs were purchased from Amersham Pharmacia Bio-tech, Inc. (polyA, polyC, polyU, andpolyG) and Sigma (polyU). Fluorescently labeledRNA containing consensus sequences for polypyrimidine tract-bindingprotein (PTBRNA) (24) and short homopolymer oligos were purchased from IDT (Dataset S1).Dipeptides were ordered through New England Peptide (Dataset S1).

Antisense RNA Analysis. Briefly, ncRNAs from the mammalian stress granuletranscriptome (28) were split into those defined as antisense and all otherncRNAs (55). Binding partners were found on the UCSC Genome Browser,and their enrichment in stress granule was graphed.

Preparation of Stress Granule-Enriched Fraction and RNase Treatment. Stress gran-ules were enriched from BY4741 cells transformed with Ded1Δ141–150 as de-scribed previously (29). RNase mixture (AM2286; Ambion) or RNase III (M0245L;New England Biolabs) was added to the granule-enriched fraction as per therelevant manufacturer’s directions and placed at 37 °C for 2 h before imaging ona microscope. Additional details are provided in Supporting Information.

In Vitro Assembly and Dynamics. Specific information regarding in vitro as-sembly formation (total RNA, homopolymer, RNA-binding protein re-cruitment, and dipeptide/RNA assemblies) as well as dynamics (relaxationtimes, FRAP, and dilution) can be found in Supporting Information.

Microscopy of RNA Assemblies.Mixtures were placed in 96-well glass-bottomedplateswith high-performance no. 1.5 cover glass (Fisher Scientific). Imageswereacquired on a DeltaVision epifluorescence microscope with a 100× objective(Applied Biosystems) equipped with a sCMOS camera. Images for FRAP analysiswere acquired using a Nikon A1R laser scanning confocal microscope.

Total RNA Pelleting for Tape Station and Sequencing Analysis. Total yeast RNAwas mixed with 150 mM NaCl and 10% PEG, and then pelleted for se-quencing. RNA pellets and total RNA were treated with a DNA-free DNAremoval kit (Thermo Fisher Scientific) and sent to the University of ColoradoBioFrontiers Institute Next-Gen Sequencing Core Facility for Ribo-Zero

B

A(G

P)1

0N

odi

pept

ides

Bright FieldBright Field Dipeptides SYTO 17

(GR

)10

(PR

)10

5

No assembly

Sparse assembly

Moderate assembly*

Robust assembly

Protein only

Protein and RNA

0

20

40

60

80

100

120

140

160

0 10 20 30 40 50

Phase Diagram (GR)10 and RNA

[Dipeptides] (

[RN

A]

Legend

Fig. 5. Pathogenic dipeptides increase RNA assembly. (A) Dipeptides (GR)10and (PR)10 promote assembly, while (GP)10 does not. Fluorescent dipeptides(green) and RNA (SYTO 17, red) are both enriched in assemblies. (B) Phasediagram illustrating the assembly of (GR)10 and RNA. The squares signify lackof assembly. +, sparse and small assembly; ++, moderate assemblies (consti-tuting either frequent but smaller assemblies or larger assemblies that weremore sparse); +++, robust assembly. Green indicates protein-only assembly.

2738 | www.pnas.org/cgi/doi/10.1073/pnas.1800038115 Van Treeck et al.

Dow

nloa

ded

by g

uest

on

Mar

ch 3

, 202

1

treatment, library construction, and NextSeq run. Read quality was assessedusing fastqc. Illumina adaptors were trimmed with Trimmomatic 0.32 (56).An index genome was built with Bowtie 0.12.7 using the S288C referencegenome (57). Reads were aligned using Bowtie 0.12.7, and mapped readswere counted using HTSeq (58). Normalization and differential expres-sion was performed with DESeq 1.22.1 (59). Sequences can be viewed inthe National Center for Biotechnology Information Gene ExpressionOmnibus database (accession no. GSE99170). More details are provided inSupporting Information.

Length and Optimal Codon Analysis. Sequencing data for in vitro RNA as-semblies were analyzed for length and codon optimality based on previousreports (60, 61). Specific details are provided in Supporting Information.

Numerical Calculations.All numericalmanipulations, includingestimationofexposedRNA during stress, and statistical tests can be found in Supporting Information.

ACKNOWLEDGMENTS. We thank all members of the R.P. laboratory and OlkeUhlenbeck for valuable conversations. We thank Yuan Lin for purifying andsending all MBP-GFP-IDR-HIS protein constructs. We thank the University of Col-orado BioFrontiers Institute Next-Gen Sequencing Core Facility, which performedthe Illumina sequencing and library construction. Some of the imaging in thiswork was performed at the BioFrontiers Institute Advanced Light MicroscopyCore. Laser scanning confocal microscopy was performed on a Nikon A1R micro-scope supported by National Institute of Standards and Technology-University ofColorado Cooperative Agreement Award 70NANB15H226. National ScienceFoundation SCR Training Grant T32GM08759 (to B.V.T.) and the Howard HughesMedical Institute (R.P.) funded this work.

1. Buchan JR (2014) mRNP granules. Assembly, function, and connections with disease.RNA Biol 11:1019–1030.

2. Voronina E, Paix A, Seydoux G (2012) The P granule component PGL-1 promotes thelocalization and silencing activity of the PUF protein FBF-2 in germline stem cells.Development 139:3732–3740.

3. Kiebler MA, Bassell GJ (2006) Neuronal RNA granules: Movers and makers. Neuron 51:685–690.

4. Anderson P, Kedersha N (2006) RNA granules. J Cell Biol 172:803–808.5. Fox AH, et al. (2002) Paraspeckles: A novel nuclear domain. Curr Biol 12:13–25.6. Gall JG (2000) Cajal bodies: The first 100 years. Annu Rev Cell Dev Biol 16:273–300.7. Buchan JR, Parker R (2009) Eukaryotic stress granules: The ins and outs of translation.

Mol Cell 36:932–941.8. Anderson P, Kedersha N (2009) RNA granules: Post-transcriptional and epigenetic

modulators of gene expression. Nat Rev Mol Cell Biol 10:430–436.9. Barbee SA, et al. (2006) Staufen- and FMRP-containing neuronal RNPs are structurally

and functionally related to somatic P bodies. Neuron 52:997–1009.10. McCann C, et al. (2011) The Ataxin-2 protein is required for microRNA function and

synapse-specific long-term olfactory habituation. Proc Natl Acad Sci USA 108:E655–E662.

11. Kim HJ, et al. (2013) Mutations in prion-like domains in hnRNPA2B1 andhnRNPA1 cause multisystem proteinopathy and ALS. Nature 495:467–473.

12. Buchan JR, Kolaitis R-M, Taylor JP, Parker R (2013) Eukaryotic stress granules arecleared by autophagy and Cdc48/VCP function. Cell 153:1461–1474.

13. Mackenzie IR, et al. (2017) TIA1 mutations in amyotrophic lateral sclerosis andfrontotemporal dementia promote phase separation and alter stress granule dynamics.Neuron 95:808–816.e9.

14. Anderson P, Kedersha N, Ivanov P (2015) Stress granules, P-bodies and cancer. BiochimBiophys Acta 1849:861–870.

15. Somasekharan SP, et al. (2015) YB-1 regulates stress granule formation and tumorprogression by translationally activating G3BP1. J Cell Biol 208:913–929.

16. Reineke LC, Lloyd RE (2013) Diversion of stress granules and P-bodies during viralinfection. Virology 436:255–267.

17. Beckham CJ, Parker R (2008) P bodies, stress granules, and viral life cycles. Cell HostMicrobe 3:206–212.

18. Gilks N, et al. (2004) Stress granule assembly is mediated by prion-like aggregation ofTIA-1. Mol Biol Cell 15:5383–5398.

19. Kedersha N, et al. (2016) G3BP-Caprin1-USP10 complexes mediate stress granulecondensation and associate with 40S subunits. J Cell Biol 212:845–860.

20. Nonhoff U, et al. (2007) Ataxin-2 interacts with the DEAD/H-box RNA helicaseDDX6 and interferes with P-bodies and stress granules. Mol Biol Cell 18:1385–1396.

21. Decker CJ, Teixeira D, Parker R (2007) Edc3p and a glutamine/asparagine-rich domainof Lsm4p function in processing body assembly in Saccharomyces cerevisiae. J Cell Biol179:437–449.

22. Hennig S, et al. (2015) Prion-like domains in RNA binding proteins are essential forbuilding subnuclear paraspeckles. J Cell Biol 210:529–539.

23. Kato M, et al. (2012) Cell-free formation of RNA granules: Low complexity sequencedomains form dynamic fibers within hydrogels. Cell 149:753–767.

24. Lin Y, Protter DSW, Rosen MK, Parker R (2015) Formation and maturation of phase-separated liquid droplets by RNA-binding proteins. Mol Cell 60:208–219.

25. Nott TJ, et al. (2015) Phase transition of a disordered nuage protein generates en-vironmentally responsive membraneless organelles. Mol Cell 57:936–947.

26. Elbaum-Garfinkle S, et al. (2015) The disordered P granule protein LAF-1 drives phaseseparation into droplets with tunable viscosity and dynamics. Proc Natl Acad Sci USA112:7189–7194.

27. Zhang H, et al. (2015) RNA controls PolyQ protein phase transitions. Mol Cell 60:220–230.

28. Khong A, et al. (2017) The stress granule transcriptome reveals principles of mRNAaccumulation in stress granules. Mol Cell 68:808–820.e5.

29. Jain S, et al. (2016) ATPase-modulated stress granules contain a diverse proteome andsubstructure. Cell 164:487–498.

30. Ferrandon D, Koch I, Westhof E, Nüsslein-Volhard C (1997) RNA-RNA interaction isrequired for the formation of specific bicoid mRNA 3′ UTR-STAUFEN ribonucleopro-tein particles. EMBO J 16:1751–1758.

31. Jain A, Vale RD (2017) RNA phase transitions in repeat expansion disorders. Nature546:243–247.

32. Aumiller WM, Jr, Pir Cakmak F, Davis BW, Keating CD (2016) RNA-based coacervatesas a model for membraneless organelles: Formation, properties, and interfacial li-posome assembly. Langmuir 32:10042–10053.

33. Zanchetta G, Bellini T, Nakata M, Clark NA (2008) Physical polymerization and liquid

crystallization of RNA oligomers. J Am Chem Soc 130:12864–12865.34. Leontis NB, Stombaugh J, Westhof E (2002) The non-Watson-Crick base pairs and their

associated isostericity matrices. Nucleic Acids Res 30:3497–3531.35. Bounedjah O, et al. (2014) Free mRNA in excess upon polysome dissociation is a

scaffold for protein multimerization to form stress granules. Nucleic Acids Res 42:

8678–8691.36. Mahadevan K, et al. (2013) RanBP2/Nup358 potentiates the translation of a subset of

mRNAs encoding secretory proteins. PLoS Biol 11:e1001545.37. Fay MM, Anderson PJ, Ivanov P (2017) ALS/FTD-associated C9ORF72 repeat RNA

promotes phase transitions in vitro and in cells. Cell Rep 21:3573–3584.38. Tourrière H, et al. (2003) The RasGAP-associated endoribonuclease G3BP assembles

stress granules. J Cell Biol 160:823–831.39. Anderson P, Kedersha N (2008) Stress granules: The Tao of RNA triage. Trends

Biochem Sci 33:141–150.40. Mazroui R, et al. (2002) Trapping of messenger RNA by fragile X mental retardation

protein into cytoplasmic granules induces translation repression. Hum Mol Genet 11:

3007–3017.41. Ash PEA, et al. (2013) Unconventional translation of C9ORF72 GGGGCC expansion

generates insoluble polypeptides specific to c9FTD/ALS. Neuron 77:639–646.42. Zu T, et al. (2013) RAN proteins and RNA foci from antisense transcripts in C9ORF72

ALS and frontotemporal dementia. Proc Natl Acad Sci USA 110:E4968–E4977.43. Lee Y-B, et al. (2013) Hexanucleotide repeats in ALS/FTD form length-dependent RNA

foci, sequester RNA binding proteins, and are neurotoxic. Cell Rep 5:1178–1186.44. Boeynaems S, et al. (2017) Phase separation of C9orf72 dipeptide repeats perturbs

stress granule dynamics. Mol Cell 65:1044–1055.e5.45. Kwon I, et al. (2014) Poly-dipeptides encoded by the C9orf72 repeats bind nucleoli,

impede RNA biogenesis, and kill cells. Science 345:1139–1145.46. Lin Y, et al. (2016) Toxic PR poly-dipeptides encoded by the C9orf72 repeat expansion

target LC domain polymers. Cell 167:789–802.e12.47. Shi KY, et al. (2017) Toxic PRn poly-dipeptides encoded by the C9orf72 repeat ex-

pansion block nuclear import and export. Proc Natl Acad Sci USA 114:E1111–E1117.48. Hilliker A, Gao Z, Jankowsky E, Parker R (2011) The DEAD-box protein Ded1 modulates

translation by the formation and resolution of an eIF4F-mRNA complex. Mol Cell 43:

962–972.49. Bounedjah O, et al. (2012) Macromolecular crowding regulates assembly of mRNA

stress granules after osmotic stress: New role for compatible osmolytes. J Biol Chem

287:2446–2458.50. Clemson CM, et al. (2009) An architectural role for a nuclear noncoding RNA: NEAT1

RNA is essential for the structure of paraspeckles. Mol Cell 33:717–726.51. de Mezer M, Wojciechowska M, Napierala M, Sobczak K, Krzyzosiak WJ (2011) Mu-

tant CAG repeats of Huntingtin transcript fold into hairpins, form nuclear foci and are

targets for RNA interference. Nucleic Acids Res 39:3852–3863.52. Mankodi A, Lin X, Blaxall BC, Swanson MS, Thornton CA (2005) Nuclear RNA foci in

the heart in myotonic dystrophy. Circ Res 97:1152–1155.53. Hall LL, et al. (2017) Demethylated HSATII DNA and HSATII RNA foci sequester

PRC1 and MeCP2 into cancer-specific nuclear bodies. Cell Rep 18:2943–2956.54. Saldi TK, et al. (2014) TDP-1, the Caenorhabditis elegans ortholog of TDP-43, limits the

accumulation of double-stranded RNA. EMBO J 33:2947–2966.55. Kinsella RJ, et al. (2011) Ensembl BioMarts: A hub for data retrieval across taxonomic

space. Database (Oxford) 2011:bar030.56. Bolger AM, Lohse M, Usadel B (2014) Trimmomatic: A flexible trimmer for Illumina

sequence data. Bioinformatics 30:2114–2120.57. Langmead B, Trapnell C, Pop M, Salzberg SL (2009) Ultrafast and memory-efficient

alignment of short DNA sequences to the human genome. Genome Biol 10:R25.58. Anders S, Pyl PT, Huber W (2015) HTSeq–A Python framework to work with high-

throughput sequencing data. Bioinformatics 31:166–169.59. Anders S, Huber W (2010) Differential expression analysis for sequence count data.

Genome Biol 11:R106.60. Nagalakshmi U, et al. (2008) The transcriptional landscape of the yeast genome de-

fined by RNA sequencing. Science 320:1344–1349.61. Presnyak V, et al. (2015) Codon optimality is a major determinant of mRNA stability.

Cell 160:1111–1124.62. Schindelin J, et al. (2012) Fiji: An open-source platform for biological-image analysis.

Nat Methods 9:676–682.

Van Treeck et al. PNAS | March 13, 2018 | vol. 115 | no. 11 | 2739

BIOCH

EMISTR

Y

Dow

nloa

ded

by g

uest

on

Mar

ch 3

, 202

1