Multivesicular bodies associate with components of ...

LETTERS

Multivesicular bodies associate with components of miRNA effector complexes and modulate miRNA activity Derrick J. Gibbings1, Constance Ciaudo1, Mathieu Erhardt1 and Olivier Voinnet1,2 In animals, P-bodies or GW-bodies appear to cause the congregation of proteins involved in microRNA (miRNA)-mediated post-transcriptional silencing. The localization of P‑bodies does not overlap with that of known organelles and are thus considered independent of lipid bilayers. Nonetheless, an miRNA effector protein, argonaute 2 (AGO2), was initially identified as membrane-associated, and some miRNAs have been found in secreted vesicles (exosomes) that derive from endo-lysosomal compartments called multivesicular bodies (MVBs). Proteins can be sorted in a ubiquitin-dependent manner into MVBs by three heteromeric subcomplexes, collectively termed ESCRT (endosomal sorting complex required for transport), to be further secreted in exosomes and/or degraded by the lysosome. Here we show that GW‑bodies containing GW182 and AGO2, two main components of the RNA-induced silencing complex (RISC), are distinct from P‑bodies due to their congregation with endosomes and MVBs. Moreover, miRNAs and miRNA-repressible mRNAs are enriched at these cellular membranes, suggesting that endosomes and/or MVBs are sites of miRNA-loaded RISC (miRISC) accumulation and, possibly, action. We further show that purified exosomelike vesicles secreted by MVBs are considerably enriched in GW182, but not P‑body components, AGO2 or miRNA-repressible mRNA. Moreover, cells depleted of some ESCRT components show compromised miRNA-mediated gene silencing and overaccumulate GW182, which associates with ubiquitylated proteins. Therefore, GW182, possibly in association with a fraction of miRNA-loaded AGO2, is sorted into MVBs for secretion and/or lysosomal degradation. We propose that this process promotes continuous assembly or disassembly of membrane-associated miRISCs, which is possibly required for miRNA loading or target recognition and subsequent silencing. RNA extracted from secreted vesicles that resemble exosomes1 (50–100 nm in diameter) contains miRNAs2. To test whether such vesicles also contain proteins required for miRNA activity, we used cultured monocytes known to secrete exosomes. We purified morphologically uniform vesicles, form‑ ing a population homogenous in size, and consistent with characteristics 1 2

of exosomes (Fig. 1a–b) and highly enriched in CD63, a known exosome marker (Fig. 1a). Moreover, RNAi against BIG2 mRNA (brefeldin‑A-inhib‑ ited guanine nucleotide-exchange protein), required for exosome release3, reduced vesicle yield by almost 50% (Fig. 1c). The purified, exosome-like material contained some AGO2, albeit less than in whole-cell lysates, and was dramatically enriched in GW182 (Fig. 1d), required for miRNA func‑ tion through its binding to AGO2 (ref. 4). Immunogold labelling and elec‑ tron microscopy of permeabilized, purified vesicles further confirmed this GW182 enrichment (Fig.1e). In contrast, the components of the P‑body decapping complex, DCP1A and GE‑1 (ref. 5), were barely detectable in secreted vesicles (Fig.1d). Identical results were obtained with monocytes cultured in serum-free medium (these excluded the contribution of vesi‑ cles from bovine serum, Supplementary Information, Fig. S1a), and with exosomes purified independently by a third party (S. Amigorena and col‑ leagues, Institut Curie, Paris) from HeLa (Supplementary Information, Fig. S1b) and ex vivo-derived dendritic cells (data not shown). We assayed the exosome-like vesicles for the presence of mature miRNAs (as opposed to miRNA passenger strands or miRNA deg‑ radation products), which were not always differentiated in previous quantitative reverse-transcription-PCR (qRT-PCR) or DNA chip anal‑ yses2,6. The 19–33 nucleotide RNA fraction isolated from monocytic exosome-like vesicles was subjected to sequencing. Among the 6,986 genome-matching sequences, 17% were known miRNAs (Fig. 1f). Consistent with previous qRT-PCR studies, the cloning frequency 2,7 and length of mature miRNAs isolated from vesicles and whole cells were similar (Fig. 1g–h and data not shown). Similarly, miRNA pas‑ senger strands (0.793% versus 0.558%) and stem-loops (3.48% versus 1.85%) were cloned at comparable, albeit much lower, frequencies in exosome-like and cellular fractions. Analysis of vesicle preparations indicated that the abundant miRNAs (let-7a) and mRNAs (RRM2) are protected against RNase treatments by membranes (Supplementary Information, Fig. S1c). Purified exosome-like vesicles thus contain single-stranded, mature miRNAs in addition to high levels of GW182 and low levels of AGO2 (Fig.1d, e). As exosomes are secreted by MVBs, we tested whether GW182 and AGO2 associate with MVBs. Monocyte post-nuclear supernatants were fractionated on continuous iodixanol (Optiprep) density gradients (Fig. 2).

IBMP-CNRS, UPR2357 Université de Strasbourg, 12 rue du Général Zimmer, 67084 Strasbourg Cedex, France. Correspondence should be addressed to O.V. (e-mail: [email protected]).

Received 16 March 2009; accepted 28 May 2009; published online 16 August 2009; DOI: 10.1038/ncb1929

nature cell biology VOLUME 11 | NUMBER 9 | SEPTEMBER 2009 © 2009 Macmillan Publishers Limited. All rights reserved.

1143

20

10

1

500nm

e

Ago2

f

Mr(k) -95

GW182 -170 -130

Dcp1a

Ge-1

60 20

100

20 Control

2,000 1,000 15 20 25 Sequence length (nucleotides)

Isotype mAb

h

-72

miRNA (17.08%) rRNA (8.75%)

-55

sc-snoRNA (1.53%) scRNA (0.84%) tRNA (4.84%) 100 nm

Big2

3,000

100 nm

-170 -130

*

60

g

GW182 mAb

os Ex

Ce lls

MonoMac6

1,000

Number of sequences

d

10 100 Size (diameter, nm)

100

Repbase (1.53%) Genome (64.89%)

16 12

Cells Exosomes

8 4

miRNA

16 142-5p let-7f 92a 27a 142-3p let-7a 191 17 19b 21 24 19a 15a

55-

Protein content of exosome Number of cells pellet (percentage) (percentage)

CD63

c

30

Percentage of total miRNA cloned

s

b

Percentage of total objects in solution

Mr(k) 72-

Ex o

a

Ce lls

LETTERS

Figure 1 Purified exosomes contain mature miRNA and are enriched in GW182, but not DCP1A or Ge‑1. (a) Equal amounts of proteins from exosomes (Exos) and cells were analysed by western blotting for exosomeenriched protein, CD63. Electron microscopy was used to image resuspended exosome pellets. (b) Dynamic light scattering of exosomes demonstrates a single population of appropriate size (20–90 nm). Error bars indicate mean ± s.e.m. from five measurements. (c) BIG2 RNAi reduces the recovery of proteins in exosome preparations. Error bars indicate mean ± s.e.m.,

n = 4, *P = 0.0194 (paired Student’s t‑test) α = 0.05. (d) Equal amounts of proteins from exosomes and whole cells were analysed by western blotting. (e) Exosomes were labelled with anti-GW182 mAB 4B6 and observed by electron microscopy. (f) Pie-chart representing the identity of cloned small RNA sequences in purified exosomes, as a percentage of sequences matching genomic human DNA. (g) Size distribution of all miRNA cloned from exosomes shows a population with a median at 21–22 nucleotides. (h) Exosomes contain miRNA profiles similar to those of whole cells.

In three independent experiments (Fig. 2), these gradients could sepa‑ rate early endosomes (Hrs, fraction 1) from late endosomes/MVBs (Hrs, fraction 1; Alix, PRP and LAMP2, fraction 2), lysosomes (LAMP2, frac‑ tions 2–5) and endoplasmic reticulum (group 78, fractions 5 and 8–9) in a fractionation pattern similar to that described previously 8. Signals for GW182 and AGO2 were consistently detected in endosomal-MVB frac‑ tions (fractions 1–4). Some GW182 was also detected in lysosomes (frac‑ tion 5). Moreover, mature miRNAs, including miR‑16 and let-7a, were also enriched in endosomal-MVB fractions 1–2 (Fig. 2). These results suggest that key components of miRISC congregate at endosomes and MVBs. To confirm that GW182 and AGO2 localize to endosomes and MVBs, we compared the cellular distribution of GFP (green fluorescent protein)tagged versions of GW182 and AGO2 with that of exogenously deliv‑ ered N‑rhodamine-labelled phosphatidylethanolamine (NRhPE), which is sorted to and retained within MVBs9. Interestingly, auto-antibodies against PE stain the same foci as GW182 auto-antibodies10. Accurate MVB labelling in monocytes was confirmed by co-localization of NRhPE

with the MVB-associated tetraspanin CD82 (Fig. 3a; Supplementary Information, Fig. S1d). GFP-tagged GW182 (GFP–GW182) and AGO2 (GFP–AGO2) formed punctuate structures, most of which co-local‑ ized with NRhPE (in 93% and 83% of monocytes tested, respectively, n = 83; Fig. 3b, d). Consistent with partitioning of AGO2 and GW182 with endosomes and MVBs, but not by endoplasmic reticulum (Fig. 2), NRhPE did not co-localize with the endoplasmic reticulum marker Sec61β (Supplementary Information, Fig. S1d). Moreover, GFP–AGO2 co-localized with the MVB-targeted HIV‑1 Gag protein9 (Supplementary Information, Fig.1d). In contrast, only 6% of cells showed co-localiza‑ tion of NRhPE and GFP-tagged DCP1A foci, a P‑body-specific marker (Fig. 3e, f). Likewise, most GFP–GW182 and RFP (red fluorescent protein)-tagged DCP1A (RFP–DCP1A) co-localized in only 3% of cells (Fig. 3g, h). Similar, though less dramatic, differences were also observed between the respective localizations of GFP–GW182, GFP–AGO2 and GFP–DCP1A in NRhPE-labelled HeLa cells (Supplementary Information, Fig. S1e–f). Consistent with previous demonstrations showing that stress

1144


LETTERS Density g ml–1

MVBs Endosomes Lysosomes

1.20 1.15 1.10 1.05 1.00 0.95 Fraction

1

2

3

4

5

Cytoplasm ribosomes

ER

6

7

8

9

10

11

a NRhPE

YFP–CD82

Merge

b NRhPE

GFP–GW182

Merge

c NRhPE

GFP–GW182

Merge

d NRhPE

GFP–AGO2 GFP–Ag GFP –Ag Ago2 o2 GFP–Ago2

Merge

e NRhPE

GFP–DCP1A

Merge

f NRhPE

GFP–DCP1A

Merge

g RFP–DCP1A

GFP–GW182

Merge

h RFP–DCP1A

GFP–GW182

Merge

12

Mr(K) Hrs

-95

Alix

-95 -37

PRP

-25 -130

LAMP2

-95

grp78

-72

rRNA AGO2

-95

Relative quantity (×10–9)

GW182

-170

14 12 10 8 6 4 2

miR-16

Fraction 1 10 8 6 4 2

2

Fraction 1

2

3

4

5

6

7

8

9

10

11

12

7

8

9

10

11

12

Let7a

3

4

5

6

Figure 2 GW182, AGO2 and miRNA partition in continuous density gradients with endosomes and multivesicular bodies. A 5–30% Optiprep gradient was collected in 12 fractions and analysed by western blotting , ethidium bromide-stained 1% agarose gel (rRNA) or qRT-PCR (miR‑16 and let-7a). The refractive index was used to calculate density (top) among three gradients (error bars indicate mean ± s.e.m.). ER, endoplasmic reticulum. Typical results for one such gradient are depicted.

granules are GW182-free11, neither induction nor suppression of stress granules altered co-localization of GFP–GW182 with MVBs in mono‑ cytes (Supplementary Information, Fig. S2a). The results suggest that the co-localization of AGO2 and GW182 foci with MVBs defines subcellular structures that are distinct from P‑bodies, as proposed previously 12. To rule out possible ambiguities arising from the use of ectopically expressed tagged proteins, we used antibodies against GW182. We first used the reference GW182 antiserum 18033, which also contains antibod‑ ies recognizing the DCP1A-interacting protein Ge‑1 and should there‑ fore partially label P‑bodies13. Of foci labelled with antiserum 18033, 26% (63/244) were not labelled with anti-DCP1A (a P-body marker; Fig. 4a)11,14, confirming that a proportion of GW182-positive structures is independ‑ ent of decapping P‑bodies. Distinct structures were also labelled by antiDCP1A and the monoclonal antibody 4B6 that recognizes GW182A (albeit minimally; Fig.4b). We then compared labelling with anti-DCP1A or GW182 antiserum to that with anti-CD63 (enriched at MVBs)15 or anti-Hrs (enriched at endosomes; Fig. 2) monoclonal antibodies. A signifi‑ cant proportion of structures labelled with antiserum 18033 gave signals coinciding with CD63 (49%, 87/171) and Hrs (42%, 82/196; Fig. 4c, d). Less co-labelling was observed between CD63 (34%, 44/128) or Hrs (34%, 62/185) and the 1C6 antiserum (Fig. 4e, f), which is predominated by

Figure 3 GW182, but not DCP1A, colocalizes with MVBs. (a–f) Cells were loaded with NRhPE after transfection with plasmids expressing YFP (yellow fluorescent protein)–CD82, a MVB-enriched protein (a), GFP–GW182 (b, c), GFP–AGO2 (d) or GFP–DCP1A (e, f). (g, h) Cells were transfected with GFP–GW182 and RFP–DCP1A. RFP or NRhPE is shown in red (left), GFP or YFP-tagged proteins in green (middle) and colocalization is shown in yellow in merged panels (right). Scale bars, 2 μm

Ge‑1 antibodies but may also contain low levels of anti-GW182 antibod‑ ies13. Similarly, only 33% (39/117) of CD63 signals and 24% (33/140) of Hrs signals coincided with DCP1A staining (Fig. 4g, h), although, strik‑ ingly, DCP1A-positive foci were often adjacent to endosomes or MVBs (see histogram profiles, Fig. 4g, h). Similar results were obtained with a second MVB marker (Supplementary Information, Fig. 2b). We conclude


1145

LETTERS a

b

c

d

18033 (GW182 + Ge-1)

DCP1A

Merge

4B6 (GW182A)

DCP1A

Merge

18033 (GW182 + Ge-1)

CD63

18033 (GW182 + Ge-1)

Hrs

Merge Me Mer erge er ge Merge

2

Merge

1

2

1

e

1C6 (Ge-1>GW182)

CD63

Merge

f

1C6 (Ge-1>GW182)

Hrs

Merge

g

DCP1A

CD63

Merge

h

DCP1A

Hrs

Merge

Figure 4 Endogenous GW182-positive/DCP1A-negative foci coincide with endosomes and MVBs more frequently than DCP1A-positive foci. (a) Approximately 1/4 foci labelled with anti-human GW182 anti-serum 18033 in Mono-Mac6 are independent of DCP1A-positive foci. Note that as anti-GW182 anti-serum recognizes Ge‑1 (ref. 13) in addition to GW182, it should label both P‑bodies and GW‑bodies. (b) Cells (293T cells) were co-labelled with anti-GW182 mAb 4B6 and anti-DCP1A as described previously12. (c–h) Foci labelled with antiserum containing predominantly anti-GW182 antibodies13 (18033; c, d), anti-Ge‑1 antibodies (1C6; e, f) or a rabbit polyclonal anti-DCP1A antibody (g, h) were examined for their localization with endosomes (Hrs) or late endosomes–MVB (CD63). Fluorescence intensity distribution along inscribed white lines is shown in histograms, (green fluorescence, black line; red fluorescence, grey line).

1146

that most 18033-positive/DCP1A-negative structures localize with endo‑ somes or MVBs, consistent with the results obtained using GFP–GW182. Nonetheless, a minor proportion of DCP1A-positive structures coincided with endosomes or MVBs, consistent with the results obtained using GFP– DCP1A (Supplementary Information, Fig. 1e, f). Collectively, Figs 3 and 4 strongly suggest that at least two cellular aggregations of AGO2-associated RNA-silencing components exist. One pool, defined as GW-bodies, is GW182-rich, DCP1A-poor and often associated with MVBs. The sec‑ ond pool is non-membranous, DCP1A-rich and GW182-poor; therefore it probably comprises P‑bodies. The relative co-localization of GW182, AGO2 and DCP1A with MVBs parallels their presence or absence in membrane fractions and exosome-like vesicles. We therefore considered the possibility that MVBs constitute sites of miRISC assembly and/or function. We transfected a plasmid expressing Renilla mRNA with a 3UTR containing two let7a target sites16, or a negative control containing two seed mismatches (Fig. 5a). Remarkably, let-7a-repressible mRNAs were significantly enriched, compared with control mRNAs, in GW182-rich, DCP1A-poor crude membrane pellets (Supplementary Information, Fig. S3a), but not in supernatants (Fig. 5b). Furthermore, in Optiprep density gradients, miRNA-repressible mRNAs were more highly concentrated in fractions corresponding to late endosomes, MVBs and lysosomes than control mRNAs (Supplementary Information, Fig. 3b). Having established that AGO2, GW182 and let-7a-repressible mRNAs congregate onto similar membranes (Figs 2 and 5a; Supplementary Information, Fig. S3b), we tested the possible targeting of the latter to exosomes. Strikingly, how‑ ever, let-7a-repressible mRNAs were markedly under-represented in exo‑ some-like vesicles (Fig. 5b). Comparing the whole-cell versus exosomal mRNA content of glioblastoma2 similarly uncovered that known miRNA target transcripts are under-represented in exosomes compared with all detected mRNAs (Fig. 5b; Supplementary Information, Table S1). In con‑ trast, housekeeping gene mRNAs, which may be less subject to miRNAmediated repression17, are enriched in exosomes (Fig. 5b, Supplementary Information, Table S1). Thus, whereas miRNA-repressible transcripts are enriched in GW182- and AGO2-associated membrane fractions, they seem selectively excluded from exosome-like vesicles. To reconcile the data obtained thus far, we envisaged that a pool of GW182 selectively dissociates from membrane-bound, AGO–miRNA–mRNA complexes to be sorted into MVBs and subsequently secreted or degraded through the exosome-lysosome pathway. A major MVB channelling mechanism relies on the recognition of ubiquitylated proteins by the ESCRT complex. Immunoprecipitates of GW182, but not of DCP1A, were indeed found to contain ubiq‑ uitylated proteins (Supplementary Information, Fig. S3c). This is also consistent with observations that AGO2, which binds GW182, is puri‑ fied with ubiquitylated proteins18, and that ubiquitin is found in some AGO complexes19. Ubiquitin is integral to many cellular processes, so to investigate more directly whether GW182 may be subject to ESCRTdependent sorting into MVBs, short interefering RNAs (siRNAs) were used to knockdown ESCRT components (Supplementary Information, Fig. S4), including Vps36, Tsg101, Alix (which is found in exosomes and participates in invagination of intraluminal vesicles) and Hrs (necessary for intraluminal vesicle accumulation within MVBs)20. Indeed, Hrs- and Alix-knockdown increased cellular GW182, but not DCP1A, levels, indi‑ cating a possible role for these two proteins in exosomal secretion and/ or lysosomal degradation of GW182 (Fig. 5c). Notably, tsg101, vps36,


LETTERS

hrs

alix

Mr(K)

GW182

-170

DCP1A

-72 -55 -43

Tubulin

Actin + RRM2

+

+

Sup’t

Actin +

+ RRM2

Pellet

+

+

+

30 20 10

Cells Exosomes

alix

vps36 hrs

tsg101

Control

2 18

alix

hrs

s3

6

1

vp

ol

10

GW

+ miR-206

10

2 18 GW

hrs

6

U6

alix

alix siRNA

– 20

s3

Merge

Transfected miR-206

30

vp

GFP–GW182

U6

20

tsg

Merge

let-7a

40

ntr

GFP–GW182

60

Co

Merge

ol

GFP–GW182

vps36 siRNA NRhPE

20

40

e

control siRNA NRhPE

40

ntr

NRhPE

60

Co

d

80

80 60 40 20

House keeping

vps36

Glioblastoma exosomes

100

100

Percentage inhibition of miR206 target (normalized to FLuc)

Cont

tsg101

**

miRNA repressed

Non-repressible mRNA

Repressible mRNA

c

**

Let-7a

All

Renilla mRNA

Let-7a

Percentage of miRNA enriched in exosomes

Renilla mRNA

Percentage normalized Renilla Percentage miRNA acitivity mRNA intact vs mut miRNA site Let7a vs mut Let7a (normalized to RLuc)

b

a

Figure 5 Knockdown of ESCRT complex components compromises miRNA activities. (a) Model of reporter constructs showing let-7a interaction with target 3´UTR fused to Renilla luciferase (adapted from ref. 16). Asterisks indicate ‘seed mismatches’ in the reporter with a mutated miRNA target site. (b) Left, miRNA-repressed mRNA is lacking in cytoplasmic, but not membrane, fractions (16000g for 45 min; actin, n = 3, P = 0.0390; RRM2, n = 3, P = 0.0225). Renilla mRNA was normalized to actin or RRM2 mRNAs where indicated. Sup’t, supernatant; mut, mutated. Middle, miRNA-repressible mRNA is excluded from exosomes compared with whole cells (actin, n = 4, P = 0.0019; RRM2, n = 3, P = 0.006). Renilla mRNA was normalized to actin or RRM2 where indicated. Right, post-hoc bioinformatic analysis of data from a study by Skog et al.2 demonstrates that validated mRNA targets of miRNA are under-represented in exosomes (7.89%, 3/35 probes, χ2 = 6.636, P = 0.01), and housekeeping genes undergoing less miRNA targeting are over-represented in exosomes (35.89%, 566/1577 probes, χ2 = 16.641, P