REPORTS
Nanotubular Highways for
Intercellular Organelle Transport
Amin Rustom,1 Rainer Saffrich,2 Ivanka Markovic,3
Paul Walther,4 Hans-Hermann Gerdes1,5*
Cell-to-cell communication is a crucial prerequisite for the development and
maintenance of multicellular organisms. To date, diverse mechanisms of intercellular exchange of information have been documented, including chemical
synapses, gap junctions, and plasmodesmata. Here, we describe highly sensitive
nanotubular structures formed de novo between cells that create complex
networks. These structures facilitate the selective transfer of membrane vesicles and organelles but seem to impede the flow of small molecules. Accordingly, we propose a novel biological principle of cell-to-cell interaction based
on membrane continuity and intercellular transfer of organelles.
Performing three-dimensional (3D) live-cell
microscopy (1) in the presence of fluorescently labeled lectin wheat germ agglutinin
(1), we observed ultrafine intercellular structures of cultured rat pheochromocytoma
PC12 cells. These structures, referred to here
as tunneling nanotubes (TNTs), had a diameter of 50 to 200 nm and a length of up to
several cell diameters (Fig. 1, A to G). TNTs
rarely displayed a branched appearance (Fig.
1C, arrow). Furthermore, they were stretched
between interconnected cells attached at their
nearest distance and did not contact the substrate (Fig. 1D). TNTs were also observed in
human embryonic kidney (HEK) or normal
rat kidney (NRK) cells (fig. S1). TNTs displayed a pronounced sensitivity to prolonged
light excitation, leading to visible vibrations
and rupture (fig. S2). Mechanical stress and
chemical fixation also resulted in the rupture
of many TNTs. However, trypsin-EDTA
treatment did not disrupt TNTs (fig. S3).
TNTs contained F-actin but not microtubules
(Fig. 1E). Similar findings have been reported for cellular extensions termed cytonemes,
first observed in the Drosophila wing imaginal disc (2). When we performed scanning
electron microscopic (SEM) analysis, the
stretched shape and structure of TNTs could
be preserved, and their surface showed a
seamless transition to the surface of both
connected cells (Fig. 1F). Transmission electron microscopic (TEM) analysis changed the
stretched morphology of TNTs into a bent
Interdisciplinary Center of Neuroscience (IZN), Institute of Neurobiology, University of Heidelberg, INF
364, Heidelberg 69120, Germany. 2Otto-MeyerhoffZentrum, University of Heidelberg, INF 350, Heidelberg 69120, Germany. 3Institute of Biochemistry, Faculty of Medicine, University of Belgrade, Pasterova 2,
Belgrade 11000, Yugoslavia. 4Electron Microscopy Facility, University of Ulm, Albert-Einstein-Allee 11,
89069 Ulm, Germany. 5Institute for Biochemistry and
Molecular Biology, University of Bergen, Jonas Lies vei
91, Bergen 5009, Norway.
configuration presumably because of mechanical stress during sample preparation.
However, serial sectioning showed that, at
any given point along TNTs, their membrane
appeared to be continuous with the membranes of connected cells (Fig. 1G).
The peculiar morphology of TNTs raised
the question of how they are generated. Over
a 4-min period, a cell formed filopodia-like
protrusions seemingly directed toward a
neighboring cell (Fig. 2, A and B). One protrusion made contact (Fig. 2C), which resulted in TNT formation (Fig. 2D, arrow) and the
degeneration of remaining protrusions (Fig.
2, compare B and D). The number of TNTs
increased during the first 2 hours after plating
of single cells (fig. S4A). TNTs did not appear to be relics of incomplete cytokinesis
(fig. S5). Thus, our data strongly suggest that
TNTs are formed de novo. TNT formation
was not an event restricted to pairs of cells
but could lead to complex cellular networks
(Fig. 2, E and F). Because TNTs were frequently found between diverging cells (Fig.
2F, black arrows), they also may exist between associated cells. After treatment with
latrunculin-B, a substance that depolymerizes
F-actin, no TNTs were detectable (fig. S4B),
which suggested that actin-driven cellular
protrusions participate in TNT formation.
The existence of expanded intercellular networks prompted us to test whether TNTs can
participate in cell-to-cell communication.
Videomicroscopic analysis (1) revealed tubular
or vesicular objects moving in one direction at
a speed of 25.9 ⫾ 7.9 nm/s along given TNTs
(Fig. 2, G and H). This phenomenon showed
striking similarities to the demonstrated transfer
of lipid containers between liposomes via phos-
1
*To whom correspondence should be addressed. Email: [email protected]
Fig. 1. (A to D) Architecture of TNTs between cultured PC12 cells. Wheat germ agglutinin–stained
PC12 cells were analyzed by 3D live-cell microscopy. Cells are connected via one (A) or several
TNTs (B) with surrounding cells. Rarely, branched TNTs were observed [(C), arrow]. In (D) a selected
(x-z) section obtained from a confocal 3D reconstruction is shown. (E) TNTs contain actin but no
microtubules. Fixed PC12 cells were immunostained with an antibody against ␣-tubulin (green),
phalloidine–fluorescein isothiocyanate (FITC) (red), and DAPI (blue). A single (x-y) section of a
deconvolved 3D reconstruction is shown. The inset depicts the corresponding (x-z) section through
the marked TNT (arrow). (F and G) Ultrastructure of TNTs. PC12 cells analyzed by SEM (F) or TEM
(G) of consecutive 80-nm sections (G1, G2). For boxed areas, higher magnification images are
shown (F1 to F3, G1, G2). Open arrowhead, secretory granule. Scale bars: (A to E), 15 ␮m; (F), 10
␮m; (G), 2 ␮m; (F1 to F3, G1, G2), 200 nm.
www.sciencemag.org SCIENCE VOL 303 13 FEBRUARY 2004
1007
REPORTS
Fig. 2. De novo formation of TNTs.
Two hours after plating, PC12 cells
were imaged by time-lapse bright-field
microscopy. (A to D) Different stages
of TNT formation. Selected frames of a
video sequence (Movie S1, acquired at
2 frames per second over 4 min) are
shown (A). One protrusion [(B), arrowhead] makes contact with a neighboring cell (C), which resulted in TNT
formation [(D), arrow]. (E and F) TNTmediated network formation. Timelapse videomicroscopy (3D) shows that
TNTs become visible between diverging cells (black arrows) and are formed
de novo (white arrows). Shown are
overlays of three selected z-sections
each, displaying all detectable TNTs.
Numbers refer to corresponding cells. (G and H) Unidirectional translocation of an object (arrowhead) along TNTs. Selected frames of a video sequence
(Movie S2, acquired at 1 frame per min over 4 min) are shown. The arrow indicates the starting position in translocation. Time points of image
acquisition are indicated. Scale bars, 20 ␮m.
Fig. 3. Transfer of soluble and membrane marker molecules between TNT-connected cells. (A and
B) Transport of organelles. PC12 cells were stained with LysoTracker and analyzed by fluorescence
videomicroscopy. Two selected frames of a video sequence (Movie S3, acquired at 1 frame per 2 s,
over 4 min) are shown. Arrowheads mark LysoTracker-stained organelles; arrows indicate their
starting position in translocation. Time points of image acquisition are indicated. (C to E) Partial
colocalization of synaptophysin and myosin Va. Fixed PC12 cells analyzed by immunofluorescence
microscopy with antibodies against synaptophysin [(C and E), green] or myosin Va [(D), green; (E),
red] and tetramethyl rhodamine isothiocyanate (TRITC)–phalloidine [(C and D), red; (E), blue]. Open
arrowheads indicate selected punctuated signals of both proteins, and the open arrow indicates
their partial colocalization. Dashed lines indicate parts of the cell borders. (F to Q2) Transfer of EGFP
fusion proteins. PC12 cells stained with CellTracker (blue) were mixed with cells transfected with
synaptophysin-EGFP [syn-EGFP (F to I2)], EGFP-actin (J to M2), or farnesylated-EGFP [f-EGFP (N to
Q2)] and cocultured for 24 to 48 hours. Cells were processed for immunocytochemistry with
GFP-specific antibody (red) and phalloidine-FITC (green), and analyzed by 3D microscopy. Single
(x-y) sections of deconvolved 3D reconstructions are shown as indicated. The numbers in the
single-channel recordings (top rows) refer to the transfected cells (population 1) or CellTrackerstained cells (population 2). Boxed areas are magnified in (I1), (M1), (Q1), respectively. A second
z-section for the same area is shown (I2, M2, and Q2). Arrows, TNTs; arrowheads, transferred marker
proteins. Scale bars, 10 ␮m.
1008
pholipid bilayer nanotubes generated in vitro
(3). After labeling acidic organelles with LysoTracker (1), we monitored fluorescent structures as they traveled unidirectionally inside
TNTs (Fig. 3, A and B). Immunocytochemical
analysis (1) revealed that synaptophysin, a
marker for early endosomes and endosomederived, small synaptic-like microvesicles
(SLMVs) (4), was present as discrete signals
inside TNTs (Fig. 3C, open arrowheads). Myosin Va, a motor protein shown to facilitate
organelle transport (5), was also present inside
TNTs (Fig. 3D, open arrowheads) and partly
colocalized with SLMVs (Fig. 3E, open arrow),
which is consistent with an actin-dependent
transport mechanism.
To investigate whether membrane containers
could be exchanged between TNT-connected
cells, we analyzed the transport of synaptophysin
fused to enhanced green fluorescent protein
(EGFP) between two different cell populations.
One population transfected with synaptophysinEGFP (population 1) was cocultured for 24 to 48
hours with a second population labeled with
CellTracker (population 2). Organelles that
stained positive for synaptophysin-EGFP could
be detected selectively in those cells of population 2 that were connected via a TNT with
synaptophysin-EGFP-expressing cells (Fig. 3, F
to I2, arrowheads). To investigate whether cytoplasmic molecules could also be transferred between TNT-connected cells, we tested for the
distribution of cytoplasmically expressed actin
fused to EGFP. EGFP-actin could be detected in
or near the actin cortex of TNT-connected cells
of population 2 (Fig. 3, J to M2, arrowheads).
These signals were similar to the patchy signals
of EGFP-actin found in cells of population 1
(fig. S6). Patchy signals of EGFP-actin most
likely represent prominent actin-rich foci referred to as “actin patches” (6, 7). The TNTdependent transfer of a soluble marker protein is
consistent with the existence of membrane
continuity, as suggested by the ultrastructural
analyses. However, neither cytoplasmically ex-
13 FEBRUARY 2004 VOL 303 SCIENCE www.sciencemag.org
REPORTS
Fig. 4. TNT-based transfer of membrane components and dye-labeled organelles. (A to C)
Intercellular membrane flow. PC12 cells stained with CellTracker were mixed with cells transfected
with f-EGFP, cocultured for 48 hours, and analyzed by 3D videomicroscopy. Numbers in the
single-channel recordings (B and C) refer to the transfected cells (population 1) or CellTrackerstained cells (population 2). The inset in (A) depicts the corresponding (x-z) section through the
marked TNT (arrow). Arrowheads, staining derived from transferred f-EGFP. (D to E2) Induced
transfer of f-EGFP. PC12 cells were transfected with f-EGFP and incubated for 12 hours. A selected
f-EGFP–positive cell (labeled 1), connected via a TNT (arrow) with a nonexpressing cell (2), was
microinjected with hyperosmotic solution and analyzed 10 min thereafter. Asterisks mark neighboring nonexpressing cells. For quantification, line profiles displaying gray value intensities [(E1 and
E2), right] were performed as indicated (red arrows). Arrowheads, position of the cell border. (F and
G) Unidirectional transfer of DiI-labeled organelles. PC12 cells were stained with DiI and analyzed
by fluorescence videomicroscopy. A selected frame of three consecutive video sequences (Movies
S4 to S6, acquired at 4 frames per s) showing DiI-labeled organelles within a TNT [(F), arrow] is
shown. Broken lines indicate parts of the cell borders. Arrows 1 to 4 mark trajectories of four
selected organelles (for details, see fig. S11). (H to I1) Intercellular organelle transfer correlates with
the existence of a TNT connection. A mixed population of DiI- (red) and DiO-labeled PC12 cells
(green) was plated and incubated for 12 hours. Pairs of TNT-connected cells identified by
differential interference contrast microscopy [(H), arrow] were analyzed for organelle transfer by
3D fluorescence microscopy (I). The boxed areas in H and I are magnified in H1 and I1, respectively.
(J and K) Proposed model for TNT-mediated organelle transfer. A cell forms actin-driven protrusions directed toward a target cell [(J), arrowhead]. TNT formation results in membrane continuity
between connected cells. Organelles are unidirectionally transferred (arrow) via actin-mediated
mechanisms. Red, F-actin; green, organelles. Scale bars, 20 ␮m.
pressed green fluorescent protein nor the small
dye molecule calcein (fig. S7) were found to be
transferred in detectable amounts between TNTconnected cells. Thus, with the exception of
actin as a major structural component of TNTs,
the small inner diameter of the stretched-membrane tubes, filled with F-actin, appears largely
to impede the passive transfer of soluble cytoplasmic molecules. To test whether plasmamembrane components could be transferred between TNT-connected cells, we analyzed the
transfer of EGFP fused to the farnesylation signal of c-Ha-Ras (f-EGFP), a fusion protein that
associated tightly with the membrane and spe-
cifically localized to the plasma membrane (fig.
S8). f-EGFP was detected as discrete signals at
the plasma membrane of those cells of population 2 that were connected via a TNT to cells of
population 1 (Fig. 3, N to Q2). As in the fEGFP–expressing cells, transferred f-EGFP was
found exclusively at the plasma membrane and
displayed a partly continuous surface labeling, as
well as patchy or raft-like signals (Fig. 3, N to
Q2, arrowheads). Thus, plasma-membrane components can be indeed transferred between TNTconnected cells.
To get more insight into the transfer of EGFP
fusion proteins, we analyzed living cells of
mixed populations by fluorescence videomicroscopy (1). After 24 to 48 hours, a weak fluorescence of f-EGFP was detectable in TNT-connected cells of population 2 (Fig. 4, A to C). This
staining was in part continuous with the TNT
labeling and covered large surface areas of the
cells (fig. S9). To monitor the transfer of f-EGFP
directly, we enhanced its transition by increasing
the osmotic pressure selectively in one cell of the
TNT-connected cell pair by microinjecting hyperosmotic solution (1). During the observation
period of 10 min, this led to a continuous increase of f-EGFP fluorescence at the plasma
membrane of the connected cell (Fig. 4, D to E2;
see also fig. S10). Thus, plasma-membrane components could flow selectively between TNTconnected cells, which suggests that their membranes are continuously connected.
Because the EGFP fluorescence of transferred synaptophysin-EGFP was barely above
the detection limit, we analyzed the TNT-based
exchange of membrane containers by labeling
PC12 cells with green fluorescent 1,1⬘dioctadecyl-3,3,3⬘,3⬘-tetramethylindocarbocyanine perchlorate (DiI) or red fluorescent 3,3⬘dioctadecyloxacarbocyanine perchlorate (DiO)
(1). Both membrane-specific dyes, well retained in cells (8) and frequently used as longterm tracers (9), were observed to be efficiently
endocytosed in PC12 cells and thus served as
markers of the endosomal or lysosomal pathway. It was possible to detect TNTs displaying
unidirectional transfer of fluorescent organelles
(Fig. 4F). This enabled us to track the movement of distinct organelles; we saw partly
overlapping trajectories (Fig. 4G). We could
follow organelles entering a TNT on one side,
organelles being transported along the TNT,
and organelles exiting the TNT into the connected cell (Fig. 4G, tracks 1 to 4, respectively;
for details, see fig. S11). During this unidirectional transfer, accumulation and/or depletion
of organelles could not be observed. Furthermore, a continuous and rapid translocation of
organelles could be detected at any given point
along the TNT (fig. S11), which was consistent
with the existence of a direct intercellular transfer mechanism based on membrane continuity.
When we analyzed TNT-connected cell pairs of
mixed cultures consisting of one cell of each
population, it became apparent that in 74%
www.sciencemag.org SCIENCE VOL 303 13 FEBRUARY 2004
1009
REPORTS
(⫾6.1%) of all cases, one cell (Fig. 4, H and
H1) accumulated organelles fluorescing in the
second color (Fig. 4, I and I1). With respect to
the low percentage of 11.3 (⫾2.8%) of all cells
in culture displaying transfer, this value suggests a strong correlation between organelle
transfer and the existence of a TNT connection.
The transfer of organelles first became visible 2
hours after coculturing and, at later time points,
showed as distinct cells harboring mostly organelles fluorescing in both colors (fig. S12).
This indicates fusion of red- and green-labeled
structures, a result consistent with reports on
early endosomal fusion (10). The intercellular,
TNT-dependent transfer of labeled organelles
was also detected by analyzing mixed cultures
of DiI- and DiO-labeled NRK cells (fig. S13).
For a pair of TNT-connected cells, only one cell
displayed both colors, which suggested transfer
in one direction only. A quantitative analysis by
fluorescence-activated cell sorting (FACS) revealed that the increase in number of cells with
mixed fluorescence correlated with the increase
in number of TNTs between cells after plating
(fig. S4A). Performing transfer experiments
close to 0°C, conditions that block exo-, endo-,
or phagocytotic events (11, 12), we still detected organelle transfer between TNT-connected
cells (fig. S14). Thus, the observed transfer did
not depend on conventional exo-, endo-, or
phagocytotic events. Organelle exchange could
be blocked in the presence of latrunculin-B (fig.
S14E), which strongly supported the presence
of an actin-based transfer mechanism.
The observation that functional TNTs were
also found in cell cultures of lineages other than
neuroendocrine cells (figs. S1 and S13) raises the
possibility that TNTs represent a general cellular
phenomenon occurring in long-range cell-to-cell
communication. The transfer of endosomerelated structures through TNTs is consistent
with the finding that similar structures, termed
argosomes, facilitate the intercellular spread of
wingless morphogens (13). Argosomes are
thought to be exchanged between cells via sequential exo- and endocytotic events (14). The
transfer of melanosomes between melanocytes
and keratinocytes represents another riddle of
organelle exchange (15). It has been proposed
that this transfer occurs by means of local membrane fusion or phagocytotic mechanisms (15).
Our finding that cells can actively exchange
small membrane carriers through membrane
channels provides evidence for a new principle
of cell-to-cell communication based on membrane continuity between TNT-connected cells
(Fig. 4, J and K). Provided that TNTs are
present in tissue, reconsideration of previous
interpretations of intercellular communication may be necessary. In this respect, the
concept of membrane continuity between animal cells may also facilitate cell-to-cell transport of, e.g., transcription factors or ribonucleoparticles, as has been documented for
the plant kingdom (16, 17).
1010
References and Notes
1. Materials and methods are available as supporting
material on Science Online.
2. F. A. Ramirez-Weber, T. B. Kornberg, Cell 97, 599
(1999).
3. A. Karlsson et al., Nature 409, 150 (2001).
4. M. J. Hannah, A. A. Schmidt, W. B. Huttner, Annu. Rev.
Cell Dev. Biol. 15, 733 (1999).
5. V. Mermall, P. L. Post, M. S. Mooseker, Science 279,
527 (1998).
6. D. A. Schafer et al., J. Cell Biol. 143, 1919 (1998).
7. R. J. Pelham Jr., F. Chang, Nature Cell Biol. 3, 235
(2001).
8. M. G. Honig, R. I. Hume, Trends Neurosci. 12, 333
(1989).
9. D. P. Kuffler, J. Comp. Neurol. 302, 729 (1990).
10. J. P. Gorvel, P. Chavrier, M. Zerial, J. Gruenberg, Cell
64, 915 (1991).
11. I. H. Pastan, M. C. Willingham, Science 214, 504
(1981).
12. M. Desjardins, L. A. Huber, R. G. Parton, G. Griffiths,
J. Cell Biol. 124, 677 (1994).
13. V. Greco, M. Hannus, S. Eaton, Cell 106, 633
(2001).
14. K. Denzer, M. J. Kleijmeer, H. F. Heijnen, W. Stoorvogel, H. J. Geuze, J. Cell Sci. 113, 3365 (2000).
15. G. Scott, S. Leopardi, S. Printup, B. C. Madden, J. Cell
Sci. 115, 1441 (2002).
16. K. Nakajima, G. Sena, T. Nawy, P. N. Benfey, Nature
413, 307 (2001).
17. W. Lucas, B.-C. Yoo, F. Kragler, Nature Rev. Mol. Cell
Biol. 2, 849 (2001).
18. We thank A. Hellwig for generous help in electron
microscopy, A. Kehlenbach and B. Schwappach for
FACS analysis, J. Hammer for providing Dil2 antibody,
A. Matus for providing EGFP-actin, R. Leube for providing synaptophysin-EGFP, and W. Franke and J.
Leichtle for valuable comments on the manuscript.
I.M. was supported by the Coimbra Group Hospitality
Scheme and A.R. by the Landesgraduiertenstipendium Baden-Württemberg; H.-H.G. was a recipient of
grants from the Deutsche Forschungsgemeinschaft
(SFB 488/B2, GE 550/3-2).
Supporting Online Material
www.sciencemag.org/cgi/content/full/303/5660/1007/
DC1
Materials and Methods
Figs. S1 to S14
Movies S1 to S8
References
30 October 2003; accepted 10 December 2003
Direct Activation of Bax by p53
Mediates Mitochondrial Membrane
Permeabilization and Apoptosis
Jerry E. Chipuk,1 Tomomi Kuwana,1 Lisa Bouchier-Hayes,1
Nathalie M. Droin,1 Donald D. Newmeyer,1 Martin Schuler,2
Douglas R. Green1*
The tumor suppressor p53 exerts its anti-neoplastic activity primarily through
the induction of apoptosis. We found that cytosolic localization of endogenous
wild-type or trans-activation–deficient p53 was necessary and sufficient for
apoptosis. p53 directly activated the proapoptotic Bcl-2 protein Bax in the
absence of other proteins to permeabilize mitochondria and engage the apoptotic program. p53 also released both proapoptotic multidomain proteins and
BH3-only proteins [Proapoptotic Bcl-2 family proteins that share only the third
Bcl-2 homology domain (BH3)] that were sequestered by Bcl-xL. The transcription-independent activation of Bax by p53 occurred with similar kinetics and
concentrations to those produced by activated Bid. We propose that when p53
accumulates in the cytosol, it can function analogously to the BH3-only subset
of proapoptotic Bcl-2 proteins to activate Bax and trigger apoptosis.
The induction of apoptosis is central to the
tumor-suppressive activity of p53 (1). Upon
activation by DNA damage–induced or oncogene-induced signaling pathways, p53 promotes the expression of a number of genes
that are involved in apoptosis, including
those encoding death receptors (2, 3) and
proapoptotic members of the Bcl-2 family
(4, 5). In most cases, p53-induced apoptosis proceeds through mitochondrial release
1
Division of Cellular Immunology, La Jolla Institute for
Allergy and Immunology, 10355 Science Center Drive,
San Diego, CA 92121, USA. 2Department of Medicine
III, Johannes Gutenberg University, D-55101 Mainz,
Germany.
*To whom correspondence should be addressed. Email: [email protected]
of cytochrome c, which leads to caspase
activation (6).
Although most of the effects of p53 are
ascribed to its function as a transcription
factor, reports have suggested that the protein
can also induce apoptosis independently of
new protein synthesis (7–10). However, these
studies have relied on ectopic expression of
p53 or overexpression of mutants that lack
transcriptional activity. Transcription-independent induction of apoptosis by p53 requires Bax and involves cytochrome c release
and caspase activation, all of which occur in
the absence of a nucleus, suggesting that p53
has the capacity to engage the apoptotic program directly from the cytoplasm (11).
We therefore tested if endogenous p53
can engage the apoptotic program directly
13 FEBRUARY 2004 VOL 303 SCIENCE www.sciencemag.org